Academyof Acctg Financial Ratio Paper

Volume 19, Number 3 Print ISSN: 1096-3685
Online ISSN: 1528-2635
ACADEMY OF ACCOUNTING
AND FINANCIAL STUDIES
JOURNAL
Marianne James
California State University, Los Angeles
Editor
The Academy of Accounting and Financial Studies Journal is owned and published by Jordan
Whitney Enterprises, Inc. Editorial content is under the control of the Allied Academies, Inc., a
non-profit association of scholars, whose purpose is to support and encourage research and the
sharing and exchange of ideas and insights throughout the world.
Page ii
Authors execute a publication permission agreement and assume all liabilities.

Neither Jordan Whitney Enterprises nor Allied Academies is responsible for the
content of the individual manuscripts. Any omissions or errors are the sole
responsibility of the authors. The Editorial Board is responsible for the selection of
manuscripts for publication from among those submitted for consideration. The
Publishers accept final manuscripts in digital form and make adjustments solely for
the purposes of pagination and organization.
The Academy of Accounting and Financial Studies Journal is owned and

published by Jordan Whitney Enterprises, Inc., PO Box 1032, Weaverville, NC
28787 USA. Those interested in communicating with the Journal, should contact
the Executive Director of the Allied Academies at info@alliedacademies.org.
Copyright 2015 by Jordan Whitney Enterprises, Inc., Weaverville, NC, USA
Academy of Accounting and Financial Studies Journal, Volume 19, Number 3, 2015
Page iii
EDITORIAL REVIEW BOARD
Robert Marley Liang Song

University of Tampa Michigan Technological University
Robert Graber Steve Moss

University of Arkansas – Monticello Georgia Southern University
Michael Grayson Chris Harris

Brooklyn College Elon University
Sudip Ghosh Atul K. Saxena

Pennn State University, Berks campus Georgia Gwinnett College
Rufo R. Mendoza Anthony Yanxiang Gu

Certified Public Accountant State University of New York
Alkali Yusuf Hema Rao

Universiti Utara Malaysia SUNY Osweg
Junaid M. Shaikh Marek Gruszczynski

Curtin University of Technology Warsaw School of Economics Douglass
Linda Bressler Natalie Tatiana Churyk

University of Houston Downtown Northern Illinois University
Suzanne Pinac Ward Dan Ward

University of Louisiana at Lafayette University of Louisiana at Lafayette
Eugene Calvasina Ron Stunda

Southern University Valdosta State University
P.N. Saksena Martha Sale

Indiana University South Bend Sam Houston State University
Askar Choudhury Rafik Z. Elias

Illinois State University California State University, Los Angeles
Harold Little Dawn Mead Hukai

Western Kentucky University University of Wisconsin, River Falls
James A. DiGabriele Desti Kannaiah

Montclair State University Middlesex University London, Dubai Campus
Philip H. Siegel Treba Marsh

Augusta State University Stephen F. Austin State University
Frank Plewa Mukunthan Santhanakrishnan

Idaho State University Idaho State University
Jan L. Williams
University of Baltimore
Page iv
TABLE OF CONTENTS
PREDICTORS OF STOCK RETURNS: SOME EVIDENCE FROM AN EMERGING
MARKET………………………………………………………………………………………….1
Joseph Abrokwa, University of West Georgia
Paul Nkansah, Florida A&M University
THE PRIVATE SECURITIES LITIGATION REFORM ACT AND LITIGATION BACKED

INVESTOR SPECULATION…………………………………………………………………….9
Ansley Chua, Kansas State University
Bruce Haslem, Southern Utah University
POTENTIAL FINANCIAL EFFECTS OF THE 2015 CORPORATE TAX REFORM OF

JAPAN TO ELECTRONICS COMPANIES WITH COMPARISON TO SOUTH
KOREA…………………………………………………………………………………………..27
Abraham R. de Guzman, Ritsumeikan Asia Pacific University
Won Jae Kim, Ritsumeikan Asia Pacific University
DISENTANGLING THE IFRS 1 DISCLOSURE: OBSERVATIONS FROM THE EARLY

ADOPTION OF IFRS IN CANADA……………………………………………………………45
Theresa DiPonio, Robert Morris University
LIQUIDITY IN LEBANESE COMMERCIAL BANKS AND ITS DETERMINANTS……….57

Rim El Khoury, Notre Dame University
UNDERSTANDING THE LIMITATIONS OF FINANCIAL RATIOS………………….…….75

Joseph Faello, Mississippi State University
FINANCIAL REPORTING QUALITY OF FAMILY-CONTROLLED SMALL AND

MEDIUM-SIZED ENTITIES: A JAPANESE CASE STUDY……….…………………………87
Muneaki Fujibayashi, Osaka Prefecture University
Hirofumi Kojima, Osaka Prefecture University
Mineo Tsuji, Osaka Prefecture University
BIG 4 AUDITOR OFFICE SIZE, ANALYSTS’ ANNUAL EARNINGS FORECASTS AND

CLIENT EARNINGS MANAGEMENT BEHAVIOR……………………………..……….…103
Sarah Garven, Morehead State University
Gary Taylor, The University of Alabama
AUDIT FEES, EARNINGS MANAGEMENT, AND LITIGATION RISK: EVIDENCE FROM

JAPANESE FIRMS CROSS-LISTED ON U.S. MARKETS………………………………….125
Junjian Gu, Nagoya University Dan Hu, Nagoya University
Page v
ABNORMAL AUDIT FEES AND AUDITOR SIZE IN THE JAPANESE AUDIT

MARKET…………………………………………………………………………………….…141
Dan Hu, Nagoya University
EARNINGS MANAGEMENT TO REPORT AN ADDITIONAL CENT OF EPS: EVIDENCE
FROM PRE- AND POST-SOX PERIODS…………………………………………………….153
Charles E. Jordan, Florida State University-Panama City
Stanley J. Clark, Middle Tennessee State University
Gwen R. Pate, University of Southern Mississippi
TIER 1 CAPITAL RATIOS: THE REGULATION OF RISK BEHAVIOR

CONTINUES………………………………………………………………...…………………165
Ryan Laiola, California State University Fullerton
ASSOCIATION BETWEEN PRE IPO EARNINGS MANAGEMENTS AND POST IPO

INSTITUTIONAL OWNERSHIP DRIFTS……………………………………………………189
Kyung Joo Lee, University of Maryland-Eastern Shore
John Jongdae Jin, California State University-San Bernardino
Diane Li, University of Maryland-Eastern Shore
THE ASSOCIATION BETWEEN FINANCIAL LITERACY AND TRUST IN FINANCIAL

MARKETS AMONG NOVICE NONPROFESSIONAL INVESTORS………………………201
Lee M. Kersting, Northern Kentucky University
Robert N. Marley, University of Tampa
Mark J. Mellon, University of South Florida
THE EFFECTS OF COMPENSATION SCHEME, SOURCE CREDIBILITY, AND

RECEIVER INVOLVEMENT ON THE ORGANIZATIONAL BUDGETING
PROCESS………………………………………………………………………………………217
Gloria J. McVay, Winona State University
OWNERSHIP STRUCTURE AND INFORMATION DISCLOSURE: AN APPROACH AT

FIRM LEVEL IN VIETNAM………………………………………………………………….235
Pham T. B. Ngoc, Hoa Sen University
Quach M. Hung, University of Economics HCMC
DAILY PRICE CHANGES IN THE REAL ESTATE MARKET……………………………..253

David Rodriguez, Henderson State University
THE EFFECTS OF VOLATILITY AND LEVERAGE ON THE EARNINGS-GDP

RELATION……………………………………………………………………………………..267
Jagjit S. Saini, Western Michigan University
Todd P. White, University of New Mexico
CAPITAL BUDGETING PRINCIPLES: BRIDGING THEORY AND PRACTICE…………283

Atul K. Saxena, Georgia Gwinnett College
Page vi
AN EXPERIMENT TO ASSESS THE EFFECT ON INVESTING OF SWITCHING

AUDITORS FROM BIG 4 FIRMS TO REGIONAL FIRMS…………………………………295
Arnold Schneider, Scheller College of Business
BANK SIZE EFFECT IN THE DEPOSIT MARKET: EVIDENCE FROM BANK BRANCH
OWNERSHIP CHANGES…………………………………………………………………..…309
Huabing (Barbara) Wang, West Texas A&M University
EFFICIENCY AND CORPORATE BANKRUPTCY PREDICTION………………………...319

Domenico Piatti, University of Bergamo
Peter Cincinelli, University of Bergamo
Davide Castellani, University of Bergamo
Page 1
PREDICTORS OF STOCK RETURNS: SOME EVIDENCE

FROM AN EMERGING MARKET
Joseph Abrokwa, University of West Georgia
Paul Nkansah, Florida A&M University
ABSTRACT
This study documents the predictive ability of five variables (dividends per share, dividend
price ratio, price earnings ratio, book to market ratio, and size) on common stock returns using
evidence from the emerging market country of Ghana. It sought to determine if the predictive
ability of these variables documented in several international studies going back many decades
could be replicated using data from the Ghana Stock Exchange.
The study found that of the five variables in the study, only dividends per share had a
significant predictive effect on stock returns. From the study, it is concluded that higher stock
returns are associated with higher dividends in the emerging market environment of Ghana; this
result is consistent with prior research.
Also, the hypothesized relations between stock returns and dividend price ratio, price
earnings ratio, size and book to market ratio were confirmed, but these relations were not found to
be statistically significant.
Keywords: Stock returns, dividends, emerging markets, Ghana Stock Exchange.
INTRODUCTION
This study examines the relationship between stock returns and five variables, namely,
dividends per share, dividend price ratio, price earnings ratio, book to market ratio, and size
using data from emerging market country of Ghana. The study uses evidence from the Ghana
Stock Exchange from the period from 2007 to 2013.
By way of background, Ghana is located in West Africa and has an estimated population of
25 million (Central Intelligence Agency) [CIA], 2014). Ghana gained independence from England
in 1957 and has been striving to develop a market economy ever since. Its economy is dominated
by the production of gold, cocoa, and recently oil. Ghana’s brisk economic growth, averaging
more than 8 percent during the past five years, political stability, and oil discovery in 2007 has
raised its profile among investors (Reuters, 10/31/13).
The Ghana Stock Exchange is relatively small by international standards. It was
established in July 1989 and trading commenced in November 1990. As of October 2014, there
were 35 companies listed on the exchange. Listings on the exchange are regulated by the Listing
Rules of the Ghana Stock Exchange, The Companies Code, Act 179, and L. I. 1728 of the
Securities and Exchange Commission Regulations of Ghana, 2003.
The relationship between stock returns and dividends per share, dividend price ratio,
price earnings ratio, book to market ratio and size has been studied in several countries (Fama
and French, 1995; McManus et al, 2004; Pontiff and Schall, 1998; Jaffe et al, 1989). These studies
go back a number of years. However, no such studies have been done to date for Ghana, an
emerging economy that has seen a lot of local and foreign investor interest within the past five
Page 2
years owing to its rapid economic growth and discovery of oil. Therefore, this study seeks to
examine the effect of these variables on stock returns using evidence from the Ghana Stock
Exchange.
Specifically, the study seeks to determine if these variables can be used to predict stock
returns on the Ghana Stock Exchange. Indeed, if these variables can be used to predict stock
market returns, investors can use this information to make greater profits.
Also, whereas earlier studies focused primarily on the relationship between stock
performance and one or two of these variables, this study examines all these five variables
together. Furthermore, the study contributes to the literature on predictors of stock returns by
examining the issue from an emerging market perspective and examining the effect of several
predictors together.
LITERATURE REVIEW
There have been many studies going back several decades that have sought to identify the
variables that can predict stock returns. These variables have included dividends per share,
dividend price ratio, price earnings ratio, book to market ratio, and size. The studies have generally
examined the relation between one or two of these factors and stock returns, and have mainly
involved the industrial countries, like the United States, United Kingdom, Canada, Japan and
Australia.
Aggarwal et al. (1992) examined the price to book ratio effect in the Japanese market using
evidence from the Tokyo Stock Exchange from 1968 to 1983. They observed that stocks with high
price to book value ratios earn low returns whereas stocks with low price to book ratios have
high returns.
Aono, K. & T. Iwaisako (2010) examined the predictive ability of price earnings ratio
using data from the Tokyo Stock Exchange from 1997 to 2009. They found that the performance
of price earnings ratio in forecasting stock returns in Japan was weaker than that found in
comparable studies in the United States.
Using data from the New York Stock Exchange, American Stock Exchange and NASDAQ
from 1963 to 1992, Fama and French (1995) have studied the behavior of stock prices and
earnings in relation to size and book to market equity. They found that size and book to
market equity are related to profitability and returns; firms with low book to market equity (high
stock prices relative to book) are associated with sustained strong profitability and stock returns,
and the converse is also true.
Banz (1981) studied the relationship between return and market value of common stock
using data on all common stock quoted for at least five years on the New York Stock Exchange
from 1926 and 1975. He concluded that smaller firms have higher risk adjusted returns on average
than larger firms.
Jaffe et al. (1989), examined the relation between earnings yield, market values and stock
returns and concluded that, there was significant earning price and size effects in US market
when estimated across all months during the period from 1951-1986. The study also found that,
there was difference between January and the rest of the year as the coefficient on both earning
price ratio and size were significant in January but in the case of the other months, only earning
price coefficient was significant outside January.
Kheradyar S., I. Ibrahim & F. Mat Nor (2011) used data from the Malaysia Stock Exchange
from January 2000 to December 2009 to investigate whether financial ratios (including
dividend yield, earnings yield and book-to-market ratios) can predict stock returns.
Page 3
Their results revealed that financial ratios can predict stock returns and that the book-to-market
ratio had a higher predictive power than dividend yield and earnings yield respectively.
Lewellen, J, (2004), studied whether financial ratios like dividend yield can predict
aggregate stock returns using New York Stock Exchange data from 1946 to 2000. His results
showed that dividend yield predicted market returns during the period 1946-2000, and also that
book to market ratio and earnings yield could predict market returns.
Miller, J.E. & F.W. Bacon (2006) used information on 1059 stocks gathered from the
website http://multexinvestor.com, and regressed the percentage change in stock price over the
past 52 weeks on dividend yield, book-to-market ratio, Beta, market capitalization and change in
earnings per share. They found that dividend yield was useful in predicting the change in stock
price over the 52-week period at the 99% confidence level.
McManus et al. (2004) have also examined the role of the payout ratio in the relationship
between stock returns and dividend yield in the United Kingdom stock market. They examined
stock returns in the London Stock Exchange over the period from 1958 to 1997. They found that
payout ratio did indeed have an important impact on the significance of dividend yield in
explaining stock returns.
Finally, Pontiff and Schall (1998) have also examined book to market ratios as predictors of
market returns in the United States market. Using data for the Dow Jones industrial average from
1926 to 1994, they found that the predictive ability of book to market ratios appears to stem from
the relation between book value and future earnings.
VARIABLES AND THEIR DETERMINATION
The variables used in the study to explore stock returns are dividend price ratio, dividends
per share, price earnings ratio, book to market ratio, and size. These variables were identified
in the literature where previous studies used one or two of them to predict stock returns. In
this section, we define the variables and the theoretical reasons for their inclusion.
Dividend Price Ratio (DPRICER) – Dividend Yield
Dividend Price Ratio (dividend yield) is defined as the dividends per share divided by
price per share. Several studies have investigated the association between dividend yield and
stock returns (Lewellen, 2004; McManus et al, 2004; Pointiff and Schall, 1998; Miller and
Bacon, 2006). Based on these studies, it is hypothesized that the higher the dividend price ratio, the
higher the stock returns.
Dividends per Share (DPS)
The dividends per share is defined as: Total dividends divided by number of outstanding
shares. This variable, though not commonly used in the prior studies, is included in this study
because it is disclosed in the annual reports, and is therefore readily available to the average
investor to use in predicting stock returns. As with the dividend price ratio (dividend yield), it is
hypothesized that high dividends per share for the prior year are associated with high current
year stock returns.
Page 4
Price Earnings Ratio (PER)
The association between price earnings ratio and stock returns have been documented in
several international studies (Lewellen, 2004; Aono and Iwaisako, 2010; Aggarwal et al. 1992;
Miller and Bacon, 2006). In general, high price earnings ratios suggests that investors are
expecting high earning growth in future compared to the companies with low price earnings ratios.
It is computed as: Price Earnings Ratio = Price Per Share / Earnings Per Share.
In line with these studies, it is hypothesized that firms with low price earnings ratios will
outperform those with high price earnings ratios.
Book to Market Ratio (BMR)
Studies of the relationship between book to market ratio and stock returns include Fama
and French (1995), Aggarwal et al. (1992) and Kheradyar et al. (2011). From these studies, it is
hypothesized that firms with low book to market equity (high stock prices relative to book) are
associated with high stock returns, and firms with high book to market equity are associated with
low returns. The ratio is calculated as: Book to Market ratio = Book Value/ Market Value.
Size (SIZE)
Size is the market capitalization of the company and is defined as:

Size = Number of outstanding shares × Price per share.
Based on prior studies (Banz, 1981; Fama and French, 1995), it is postulated that small
firms have higher stock returns than large firms.
Stock Returns (STOCKR)
Based on prior studies, stock returns (on common stock) are defined as:
Total Stock Return = ( (P1 – P0 ) + D)/P0
Where
P0 = Initial Stock Price
P1 = Ending Stock Price (Period 1)
D = Dividends
HYPOTHESES AND METHODOLOGY
As alluded to earlier, this study seeks to examine the relationship between stock returns
and five variables, namely, dividends per share, dividend price ratio, price earnings ratio, book to
market ratio, and size using data from the emerging market country of Ghana. The study uses
evidence from the Ghana Stock Exchange from the period from 2007 to 2013. Specifically, the
following hypotheses are investigated:
H1 The higher the dividends per share, the higher the stock return.
H2 The higher the dividend price ratio, the higher the stock return.
H3 Firms with high price earnings ratios have low stock returns.
H4 The lower the book to market ratio, the higher the stock return.
Page 5
H5 The smaller the size of the firm, the higher the stock returns.
H6 At least one of the five variables examined in the study has significant effect on stock returns.
The first five hypotheses, H1 to H5, are tested using Pearson correlation. The last
hypothesis, H6, is tested using multiple linear regression analysis. Since the variable SIZE is
severely skewed (Banz, 1981), it is log-transformed for use in the multiple linear regression model
shown below.
Where
t = Time period
Y= Stock returns (STOCKR)
X1= Dividends per share (DPS)
X2 = Dividend price ratio (DPRICER)
X3 = Price earnings ratio (PER)
X4 = Book to market ratio (BMR)
X5 = lnSIZE
DATA
The source of the data for the study was the World Wide Web, the Ghana Stock Exchange
website, individual company websites and the website www.annualreportsghana.com. The study
uses stock return data for all companies listed continuously on the Ghana Stock Exchange from the
period 2007 through 2013. The sample size comprised of 7 years of data for each company, in all
the sample size was 154; though the sample size is smaller than that of previous studies, (due to
limited size of the Ghana Stock Exchange), it is enough to perform statistical analysis.
ANALYSIS OF RESULTS
Preliminary analysis of the data showed an extreme outlier for one of the values for stock
return (STOCKR). The COOK’s distance for this outlier was 1.178 which exceeded the
recommended cut-off point of 1 so the case corresponding to this outlier was excluded from further
analysis.
Table 1 shows the descriptive statistics of the variables used in the study. Table 2 shows the
results of the Pearson correlation analysis. From this table, it can be seen that dividends per share
(DPS) and dividend price ratio (DPRICER) have a significant positive relationship with stock
return (STOCKR). On the other hand, price earnings ratio (PER), book to market ratio (BMR), and
size (SIZE) have insignificant inverse relationship with stock return. However, in terms of the
nature of the relationship, all the hypothesized directions in hypotheses H1, H2, H3, H4, and H5
are supported. Similar results have been obtained in prior studies (Pontiff and Schall, 1998;
McManus et al, 2004).
Page 6
Table 1
DESCRIPTIVE STATISTICS
N Minimum Maximum Mean Std. Deviation
DPS 153 .000 4.503 .217 .526
DPRICER 153 .000 4.732 .226 .763
PER 153 -275.000 1200.000 23.273 111.611
BMR 153 .012 5.348 .653 .806
SIZE 153 4513.156 21930000.000 871681.608 2851149.117
STOCKR
153 -.949 27.511 .841 3.306
Table 2
CORRELATIONS
DPS DPRICER PER BMR SIZE STOCKR
DPS Pearson Correlation 1 .501** -.050 -.188* .076 .338**
Sig. (2-tailed) .000 .538 .020 .349 .000
N 153 153 153 153 153 153
DPRICER Pearson Correlation .501** 1 -.057 -.009 -.035 .239**
Sig. (2-tailed) .000 .482 .917 .670 .003
N 153 153 153 153 153 153
PER Pearson Correlation -.050 -.057 1 -.127 .033 -.033
Sig. (2-tailed) .538 .482 .117 .685 .682
N 153 153 153 153 153 153
BMR Pearson Correlation -.188* -.009 -.127 1 -.142 -.113
Sig. (2-tailed) .020 .917 .117 .079 .163
N 153 153 153 153 153 153
SIZE Pearson Correlation .076 -.035 .033 -.142 1 -.033
Sig. (2-tailed) .349 .670 .685 .079 .690
N 153 153 153 153 153 153
STOCKR Pearson Correlation .338** .239** -.033 -.113 -.033 1
Sig. (2-tailed) .000 .003 .682 .163 .690
N 153 153 153 153 153 153
** Correlation is significant at the 0.01 level (2-tailed).
* Correlation is significant at the 0.05 level (2-tailed).
Table 3
ANOVA
Model Sum of Squares df Mean Square F p-value
Regression 211.915 5 42.383 4.298 .001
Residual 1449.651 147 9.862
Total 1661.567 152
a. Dependent variable: STOCKR

b. Predictors: DPS, DPRICER, BMR, PER, lnSIZE
c. R2 = .128
d. Adjusted R2 = .098
Page 7
Table 4
COEFFICIENTS
Standardized
Unstandardized Coefficients Coefficients
Model B Std. Error Beta t p-value
(Constant) 2.268 2.420 .937 .350
DPS 1.859 .602 .296 3.090 .002
DPRICER .366 .398 .084 .919 .360
BMR -.412 .388 -.100 -1.061 .291
PER -.001 .002 -.023 -.297 .767
lnSIZE -.136 .190 -.071 -.716 .475
In regards to using the five variables to predict stock returns, Tables 3 and 4 show the
results of the multiple regression analysis. The overall regression with R2 = .128 is significant at
the 1% level but only dividends per share (DPS) has significant effect on stock returns. In addition,
a stepwise regression was run using all the five predictor variables to find the best variable to
predict stock returns. Again, the best predictive variable was dividends per share with an R2 = .114.
CONCLUSION
This study documents the predictive ability of five variables (dividends per share, dividend
price ratio, price earnings ratio, book to market ratio, and size) on common stock returns
using evidence from the emerging market country of Ghana. It sought to determine if the predictive
ability of these variables documented in several international studies going back many decades
could be replicated using data from the Ghana Stock Exchange.
The study found that of the five variables in the study, only dividends per share had a
significant predictive effect on stock returns. From the study, it is concluded that higher stock
returns are associated with higher dividends in the emerging market environment of Ghana; this
result is consistent with prior research and is a well-established fact in the literature (McManus et
al. 2004).
Also, the hypothesized directions in the first five hypotheses were all supported. That is, the
hypothesized relations between stock returns and dividend price ratio, price earnings ratio, size and
book to market ratio were confirmed, but these relations were not found to be statistically
significant.
The implications of these results are that the average investor on the Ghana Stock Exchange
can earn significantly higher returns by investing in stocks with high dividends per share or
high dividend yields.
The primary limitation of this study is the smallness of the sample. Owing to the small size
of the Ghana Stock Exchange and the economic environment prevailing in Ghana, the stock market
in Ghana is not very liquid. Because of this, the study uses annual data instead of the daily or
monthly data used for comparable studies. Despite these limitations, the central results of this
study remain valid.
Page 8
REFERENCES
Aggarwal, R., T. Hiraki & R.P. Rao (1992). Price/book value ratios and equity returns on the Tokyo Stock
Exchange:Empirical evidence of an anomalous regularity. The Financial Review, 27(4), 589-605.
Aono, K. & T. Iwaisako (2010). Forecasting Japanese Stock Returns with Financial Ratios and Other Variables.
Working Paper Series No. 63, Institute of Economic Research, Hitotsubashi University, 1 – 18.
Banz, R.W., (1981). The relationship between return and market value of common stock. Journal of Financial
Economics, 9, 3-18.
Central Intelligence Agency (2014), The World Fact Book. Retrieved on October 15, 2014 from the website:
https://www.cia.gov/library/publications/the-world-factbook/geos/gh.html
Fama, E.F. & K.R. French (1995). Size and book-to-market factors in earnings and return. Journal of Finance, 1,
131 – 155.
Jaffe, J., D.B. Keim, & R.Westerfield, (1989). Earnings yield, market values and stock returns. Journal of Finance,
XLIV(1), 135 – 148.
Kheradyar S., I. Ibrahim & F Mat Nor (2011). Stock Return Predictability with Financial Ratios. International
Journal of Trade, Economics and Finance, 2(5), 391 -396.
Lewellen, J, (2004). Predicting returns with financial ratios. Journal of Financial Economics, 74, 209-235.
McManus,I., O.A.P Gwilmy & S.Thomas (2004). The role of payout ratios in the relationship between stock returns
and dividend yield. Journal of Business Finance & Accounting, 31(9&10), 1355-1387.
Miller, J.E. & F.W. Bacon (2006). The usefulness of ratio analysis in predicting stock market returns. Proceedings of
the Academy of Accounting and Financial Studies, 2(1), 53 -57.
Pontiff, J. & L.D. Schall (1998). Book-to market ratios as predictors of market returns. Journal of Financial
Economics, 19, 141 -160.
Reuters, article entitled “AFRICA INVESTMENT – For Ghana, South Africa, investor perceptions may be at odds
with realty” by Tosin Sulaiman, October 31, 2013.
Page 9
THE PRIVATE SECURITIES LITIGATION REFORM

ACT AND LITIGATION BACKED INVESTOR
SPECULATION
Ansley Chua, Kansas State University
Bruce Haslem, Southern Utah University
ABSTRACT
For some investors, shareholder litigation is a real option that allows them to partially
recover losses, albeit at the expense of other shareholders. We examine the market response to
the passage of the PSLRA, which decreased the value of this option, to test whether shareholder
litigation affects the valuation of firms and the willingness of investors to speculate on a firm’s
market value. We find that in firms where speculation is most likely to exist, such as those
trading at abnormally high valuations; the reaction to these reforms was significantly more
negative and was related to factors that would affect the likelihood and payoffs of litigation.
From a policy standpoint, PSLRA had the benefit of decreasing the value of investor speculation.
INTRODUCTION
It is generally accepted that when you prevent investors from suffering negative
outcomes in their investments ex post, you lower the incentive they have to carefully investigate
those investments ex ante. This increase in moral hazard allows investors to pay less attention to
the risk of an investment; instead shifting the focus to only the potential upside payoff.
Unfortunately, this idea has come to the forefront of economic discussion as we debate the relative
costs and benefits of government bailouts for industry, or distortions in real estate prices due to
government backing of mortgages. In this paper, we investigate empirically whether changes in
one such ex post remedy available to investors, shareholder lawsuits, affects the price that
investors attach to a company’s equity and their willingness to speculate on market information.
Cornell (1990) shows that any litigation creates a valuable real option for a plaintiff,
regardless of the merits of the case. But unlike derivative lawsuits in state courts, where one
shareholder sues the firm on behalf of all shareholders of the firm, securities litigation only protects
those shareholders who purchased their shares after some event or announcement misleads the
market about the underlying value of the firm. For instance, if the CEO of a firm announces
quarterly results that are inflated by accounting manipulations, it is assumed that only those
investors who purchased the shares based on the false information were harmed. In effect, when
there is greater uncertainty surrounding the validity of information regarding the firm, securities
litigation makes the stock more valuable for new shareholders, who have the ability to recover
damages, than it is for existing shareholders who not only don’t have that option, but will also bear
the cost of the damages paid to the eligible shareholders. Coffee (2006) does an excellent job in
describing the limitations and wealth shifting aspects of shareholder litigation.
However, do investors consider the value of this option when setting the price they are
willing to pay for a firm’s equity? That remains an open question. Like all options, shareholder
Page 10
litigation is at best a zero sum game between the two types of investors in the absence of any
externalities, and at its worst, it is a net loss for both parties as the lawyers take their portion and
the firm suffers reputational losses. While the option might increase the willingness of new
investors to speculate on the stock given their higher valuation, it doesn’t guarantee that there
will be a change in overall firm value due to the option. The new equilibrium price would
depend on several factors, such as difference in the price elasticity of demand for the new buyers,
who have the option to sue, versus the price elasticity of supply for the existing shareholders who
do not, or the proportion of investors who will bear the cost of damages relative to the number of
eligible investors who will receive the payout. Moreover, it might be difficult to imagine
individual investors, who following the release of new information, consider the likelihood it is
fraudulent and what the payoff would be in a lawsuit if it were.
To examine the impact this option has on investor’s willingness to speculate, we examine
the market response to the Private Securities Litigation Reform Act (PSLRA), which limited the
ability of shareholder defendants to recover damages. If the option to sue encourages
speculation, we expect to find a more negative reaction to the passage of PSLRA for those firms
that are most likely to be over-valued at the time of its passage, and thus have the highest
potential payouts in litigation. While this seems obvious, the evidence from previous papers
examining the market reaction to the law’s passage has been mixed. We suggest that this is due
to the prior research being focused on the changes in valuation of firms in the four industries
with the highest incidence of litigation, as well as the quality of governance within those firms.
Certainly, there is a high correlation between a firm’s industry, its quality of governance, and the
likelihood of a lawsuit. However, like all options, the option to sue will only be exercised when
there are potential damages, which are based on overvaluation in the stock’s price. Therefore,
we predict that the market reaction to the passage of PSLRA will be primarily related to those
factors that define payoffs in litigation, even after controlling for other factors such as industry,
governance and information opacity that contribute to its likelihood. We also expect that the
willingness to speculate spans industries, and that if shareholder litigation does encourage
speculation, we will find the most negative reaction to the law’s passage in the most over-valued
firms, regardless of their industry.
In this paper we do not suggest that shareholder litigation is the cause of over-valuation.
Companies release information continuously, any of which might lead investors to change their
valuation of the firm and speculate on the information being accurate. Our contribution is that
we show shareholder litigation makes speculation less costly and exacerbates an existing price
bubble as the value of the option is also incorporated into the price of the shares. This is especially
true as the variance in signal quality increases, and this litigation option becomes more valuable to
new investors who are in a better position to speculate on it being accurate, knowing that they can
recover a portion of the cost of misrepresentation.
While the contribution of this paper is that we show shareholder litigation affects the ex-
ante pricing within markets and can exacerbate bubbles, the importance of our results lies in
highlighting the pricing distortions created by enabling speculation. Certainly, some existing
shareholders will correctly time the market and sell out at a higher price to a speculator, who is
willing to pay more based on the ability to recover damages, and they will receive an undeserved
benefit from shareholder litigation. Unfortunately, those who benefit from the artificially high
prices are not the ones who pay the damages. Instead, it is the firm’s long-term, buy and hold
investors who must pay the damages to those shareholders who purchased shares during the
fraud period. This reduces the profitability of long-term investing relative to active attempts at
Page 11
market timing. In addition, if market speculators push the price of securities above their true
intrinsic value, they implicitly lower the firm’s cost of equity financing; potentially distorting the
firm’s investment decisions and leading to over-investment. In light of recent papers, such as
Cheng, Huang, Li and Lobo (2010), that suggest shareholder lawsuits can be positive weapons in
the corporate governance arsenal, our paper highlights one negative consequence of this tool.
Our paper builds upon an existing literature that has examined the market’s response to
the passage of PSLRA. In the early 1990s, large decreases in a firms’ stock price were often
followed by a race to the courthouse, in the hopes of becoming the lead plaintiff in a class-action
lawsuit against the firm. It was a common view that lawsuits were often filed against firms, with
the plaintiffs hoping to find the evidence to support the claims through the discovery process.
Additionally, firms faced large legal costs when defending or settling these fraud lawsuits,
regardless of whether any fraud had been committed. To lessen the impact on firms, the Private
Securities Litigation Reform Act (PSLRA) was passed in 1995. This reform made four major
changes to the existing statutes. Each lessened the likelihood or value of a successful lawsuit;
effectively decreasing the value of the option to sue.
Certainly, for those investors with legitimate claims and injury from fraud, the reforms
made it more difficult to obtain redress. However, by making it easier for managers to issue
unbiased forecasts, rather than the worst case forecast which prevents liability, these reforms
may have improved the information available to investors, while also significantly decreasing the
risk and costs of frivolous litigation reaching court. Overall, the passage of the reform was thought
to be a positive event for the market in general, as documented by Spiess and Tkac (1997)
and Ali and Kallapur (2001), who show that there was a positive market reaction during the
legislative passage and signing of the reform. However, Spiess and Tkac (1997) also document an
industry effect in the initial reactions. Although disputed by Ali and Kallapur (2001), they find
negative abnormal returns for those firms in industries with the highest numbers of lawsuits before
the reforms. These findings were supported by Johnson, Kasznik and Nelson (2000), who find that
there was a positive reaction to the reforms for most of the high- tech firms in their sample, but
there was a negative reaction on average for those firms most likely to be sued.
Our paper diverges from this prior research by showing that when analyzing the impact
of these reforms, it is important to consider potential payoffs instead of focusing on industry and
governance characteristics. Even for companies that are likely to be sued, when the costs of
litigation exceed the potential payoffs, you may find a positive reaction to the passage of the law.
We find that the share price reaction to the legislative introduction of the law was significantly
related to factors that would influence the potential payoff in a lawsuit. Damages in shareholder
litigation are a function of two factors. First is the difference between the firm’s inflated value
and its “true” long-term value. We proxy for this by taking each firm’s market to book ratio at
the passage of PSLRA, and normalizing it by the average market to book ratio for that firm over
the five preceding years to get a measure of abnormal valuation. The second factor adjusts
damages based on “in and out” trades, which is the number of shares purchased during the class
period that were then subsequently sold during the class period. The estimate of these “in and
out” shares is based on daily volume as a percent of overall float during the class period. When
firms have a higher percentage of their outstanding shares trading daily, the estimate of the number
of shares damaged will be higher. This should lead to higher damage estimates in litigation.
However, it is also means that shares purchased early in the class period are more likely to be
treated as in and out trades, and to the extent that overvaluation is higher earlier in the class
period, then that could actually decrease damage estimates.
Page 12
We find that our proxies for these two values are significant predictors of the share price
reaction to the passage of PSLRA, even after controlling for other potential factors used to
predict litigation risk such as the governance characteristics of the firm, analyst coverage or
industry. In fact, after dividing our firms into high (and low) litigation risk portfolios based on
industry, we find that consistent with previous research, firms with greater litigation risk did
have more negative stock reactions, but it was mainly focused in those firms with abnormally
high valuations. Overall, firms trading at low valuation ratios relative to their historical averages
reacted positively on average, while there was a significantly negative price response for those
firms trading at high relative values. This result is consistent with investors incorporating the
option to sue into their valuations, and that this option becomes more valuable as the price rises
above its historical average valuation ratios. To the extent that these higher valuations are due to
speculation as commonly suggested in the media or literature, it appears that having shareholder
litigation as an option contributes to investors being willing to push these values higher. These
results are robust to other measures of over-valuation, such as scaled price to book or price to
cash flow measures.
An alternative hypothesis could be suggested that it is easier for firms with less analyst
coverage or worse governance to cheat. So by making it more difficult to sue, the reforms
increased the likelihood that managers of these firms would commit a fraud in the future. And
because the existing, long-run shareholders bear the cost of litigation, we would thus expect to
see more negative reactions on the part of investors in these firms. However, and contrary to this
alternative hypothesis, we do not find the quality of governance or analyst coverage to be
significant explanatory variables in any of the quartiles other than the most over-valued. This
result is consistent with our hypothesis, where uncertainty regarding the accuracy of the signal
increases the probability of a lawsuit and thus its expected value. But like an out of the money
option, it wouldn’t be exercised without a positive expected payoff based on overvaluation and is
therefore not significant in explaining the market response to the passage of PSLRA.
Finally, our results are also economically significant. Based on our overvaluation measure,
the average overvaluation for the firms in the highest scaled, market to book quartile was
$272.87 million. At the introduction of PSLRA, this overvaluation was reduced by an average of
$11.75 million, or -4.3 percent. As a comparison, Cox, Thomas and Bai (2008) calculate provable
losses of securities class actions prior to PSLRA to have a mean of $382 million (median equal to
$55 million) and the average settlement to be thirteen percent of provable losses (nine percent
median). And Bajaj, Mazumdar and Sarin (2000) find that the amount paid in a settlement
relative to the size of potential damages was as high as 15.73% for settlements greater than one
hundred million.
HYPOTHESIS ON THE IMPACT OF PSLRA ON FIRM VALUE
As a response to the common perception that nuisance litigation was imposing significant
costs on business, Congress introduced the Private Securities Litigation Reform Act on January
4, 1995 with broad, bipartisan support for the bill, including that of President Clinton. And after
Congress passed the PSLRA on December 6, 1995, it was expected that it would be quickly
enacted. However, on December 19, 1995, President Clinton unexpectedly vetoed the bill, even
though there was more than enough support within Congress to override the presidential veto on
Page 13
December 22, 1995, making it a law. Because of this widespread support, there was little doubt
that the bill would eventually pass, which is why we, as well as previous studies, have focused
on the introduction date of January 4th as the key date in measuring the market response to PSLRA.
The reforms contain four major changes that significantly affect the ability and incentives
of shareholders to sue the firm. The first change creates a “safe harbor” for forward looking
statements as long as these statements include adequate cautionary statements. This allows
managers to release more information to investors, while making it more difficult for
shareholders to use managerial announcements as the cause for a shareholder lawsuit. Second,
PSLRA delays the discovery process until after the defendant firm’s motion for dismissal. This
clause prevents so called “fishing expeditions”, where investor-plaintiffs used the discovery
process in order to find cause for their lawsuits. This decreases the probability that a suit will be
heard, effectively decreasing the likelihood of shareholders recovering damages from the firm.
Third, PSLRA limits the amount of liability of corporations. Prior to PSLRA, firms were
potentially liable up to three times the damages under the Racketeer Influenced and Corrupt
Organizations (RICO) Act. The passage of PSLRA limits the liability of a firm to only the actual
monetary damages of the investor. Fourth, under PSLRA, the plaintiff who has the largest stake
in the proceedings is given preference to become the lead plaintiff in a class action, and extra
compensation to lead plaintiffs beyond their costs is prohibited. By taking preference away from
the first plaintiff (and their law firm) who files, this clause was meant to eliminate the races to
the courthouse following a decrease in a firm’s stock price. Collectively, the passage of PSLRA
decreased both the probability of winning the litigation and the potential payoff, therefore
lowering the ex-ante value of the litigation option.
Following the release of noisy information into the market, each investor must decide
both the validity of the information and the impact it would have on firm value. If some
investors (who buy shares after the information release) are able to recover losses in the event
that the information is eventually revealed to be false, then the stock is more valuable to them
than it is to the existing shareholders of the firm who would have to pay those future damages. If
buyers set prices in the market, this could lead to the stock price being higher than its fundamental
value. However, it is not certain that this difference in valuation will affect the value of the
stock. Because the damages are just a shift in wealth from one set of a firm’s shareholders to
another set, in the absence of deadweight costs such as legal fees or reputational costs, the overall
value of the firm will be unchanged. Thus, the net impact of being able to recover damages
will be determined by the price elasticity of demand for new buyers versus the price elasticity of
supply for existing shareholders, as well as the proportion of each type. As these things are
unobservable, we instead test empirically whether the option to litigate affects market prices by
examining the impact the passage of PSLRA had on equity prices when the value of this option
was reduced.
If investors do consider this option to sue when determining the price of a firm’s equity, we
hypothesize that the market reaction to PSLRA will be related to factors that affect potential
damage payoffs in litigation. When determining damages in a shareholder lawsuit, participants
calculate the difference between the price of a firm’s stock during the class period, when the
fraudulent information was available to market participants but not yet known to be untrue, and the
average price of the stock measured in the period following the revelation of the fraud. The higher
price a stock trades at during the class period relative to its post-revelation average, the greater the
potential damages in a lawsuit.
Page 14
This damage estimate is then reduced based on a second factor, which is “in and out”
trading during the class period. Because many of the shares purchased at the beginning of the
class period will not be held until the end of the class period, these shares are not thought to be
damaged by the revelation of fraudulent information. Therefore, damage estimates are reduced
downward based on the percentage of trades thought to be in and out during the class period.
This reduction is a function of daily volume relative to the total number of shares the firm has
available for trading, or the float. When daily volume increases as a percentage of float, it is
considered more likely that shares purchased early in the class period would have been sold by the
end of the period and potential damages based on price differences early in the class period are
reduced. To the extent that overvaluation is greater early in the class period, this could have a
marginal negative impact on damage estimates.
Therefore, if investors do consider the option to litigate in setting prices, we hypothesize
that the market reaction to the passage of PSLRA will be significantly more negative when the firm
is over-valued relative to its historical average and when the firm has a higher percentage of shares
trading daily (higher potential damages). This will be true even after controlling for other variables
found to be significant predictors of abnormal returns in previous studies, such as industry,
governance quality or information opacity. The two main hypotheses we test in this paper are as
follows:
H1 Because shareholder litigation is a real option, the market reaction to the passage of the PSLRA
will be related to factors that affect payoffs in litigation.
H2 Because options become more valuable as they are deeper in the money, we expect that the market
response to the passage of the PSLRA (which reduced the value of this option) will be most
negative for those firms trading at the highest valuations relative to their long-term norms.
On the date PSLRA was introduced, we cannot identify specific firms that were over-
valued for use in our tests. Instead, we proxy for this using the market-to-book equity (M/B)
ratio, recognizing that previous studies have shown that the market to book ratio can reflect
many things, such as industry, potential firm growth or agency problems within the firm. To
control for this, we do not focus solely on whether the firm has a high market-to-book ratio;
which may or may not be higher than its usual multiple. Instead, we divide each firm’s M/B ratio
by its long-term valuation to find those firms where new information or speculation is more
likely to have inflated the value of the firm beyond its average levels. We would expect that the
option value to sue will be highest for those securities in which high value signals have lead
investors to price the shares at higher multiples than usual for that security. We also run the same
tests using price to earnings ratios, and our results are robust to the difference in valuation
measurement.
These higher multiples could be caused by the announcement of unexpectedly high
earnings, a change in perceived rate of growth, or a multitude of other information events to which
investors must assign a likelihood of being truthful. Whether caused by speculation, or by rational
analysis of the data, this overvaluation is a necessary condition for the option to be valuable. If
there is no overpricing, there are no damages to recover in a lawsuit and the option is worthless,
regardless of the industry, corporate governance or other factors that typically explain the
likelihood of a shareholder lawsuit. We also would suggest that if these high relative valuations are
not due to speculation within the market, there is no reason to predict differential market reactions
to the introduction of the PSLRA based on firm value.
Page 15
DATA AND RESEARCH METHODOLOGY
We begin by utilizing all firms listed in CRSP and Compustat at the time of the reform
passage in 1995, and then remove all financial and utility firms. We also remove those firms
whose price was below five dollars, and are left with a sample of 3,580 firms that we use in
measuring the market’s reaction to the reforms. We test the effect of the passage by first
determining which firms are overvalued relative to their historical levels. As in Barber and Lyon
(1997), we determine the M/B ratio of all of the firms by calculating the book-to-market ratio
based on market value of equity at the end of the previous calendar year, t-1, and book value of
equity reported on a firm's balance sheet in the same calendar year. We average these M/B ratios
for the five fiscal years prior to 1995 to estimate the historical M/B values for each of the firms.
We define the level of overvaluation as the scaled market to book for each firm at the beginning of
1995 as:
M / Bi ,1994
ScaledM / Bi  (1)
M / Bi , Historical
Where M/Bi,1994 is the M/Bi of firm i as of December of 1994 and M/Bi,historical is the average
M/B of firm i over the prior five years. If Scaled M/Bi is greater than one, then the firm’s
valuation level is higher than its long-term fundamental value. Since we group our firms based on
the Scaled M/B relative to other firms, the market level of over/under valuation will only increase
or decrease the Scaled M/B for all firms and not affect the relative over/under valuation of the
firms in relation to each other.
For robustness, we also sort the firms along two other dimensions. First, we sort the firms by
their actual M/B and group them into quartiles. And because the law was passed eleven
months after the initial introduction of the legislation, the level of overvaluation may have
changed during that time. In order to control for that, we also sort the firms by the November 30,
1995, M/B and scaled M/B characteristics and rerun the study. In each case, the firms with the
highest valuations are in the fourth quartile, while those with the lowest are in the first.
Panel A of Table 1 shows the descriptive statistics for each of the portfolios. We see that
after sorting our firms into quartiles based upon their scaled M/B, firms in the highest quartile have
market to book ratios 41 percent higher than their long-term average. There is significant
variation between quartiles, with the third quartile only slightly above its long-term average, while
the other two quartiles were trading below their long-term averages. We also find that when the
portfolios are sorted solely on M/B the scaled M/B increases monotonically and when the
portfolios are sorted by scaled M/B, the M/B of the portfolios likewise increases monotonically.
This suggests that even though the composition of the groups change, there is significant overlap
between the groups. However, it is worth noting that the firms with the largest market
capitalizations are in the third scaled M/B quartile; not the fourth. So although an alternative
explanation could be offered suggesting that larger firms are more complex and are thus more
likely to have fraud occur, we would then expect to see the most negative reaction to the reforms
occurring in the third quartile if that were the case.
In Panel B, we examine industry representation within the quartiles and find that three
specific industries dominate the highest valuation quartile. The first is the business equipment
industry, which contains firms involved with software development and personal computing,
Page 16
followed by the personal and business service industries. The third is the healthcare industry,
which includes the pharmaceutical and bio-tech sectors. These industries coincide with those
that Spiess and Tkac (1997) found to have had a negative reaction to passage of PSLRA.
Table 1
DESCRIPTIVE STATISTICS OF EACH OF THE QUARTILES
Panel A shows the equally weighted quartile portfolios when ranked as of year-end 1994. Market Value is the number of
shares outstanding multiplied by the share price on December 30, 1994. Book Value is the equity book value. M/B is the
market-to-book ratio of the firms based on the book value at the end of the prior fiscal year. Scaled M/B is the M/B scaled by the
firm specific average M/B over the previous 5 years. Each panel is broken down into M/B groupings as well as Scaled M/B
groupings. Panel B shows the number of firms from each industry that is in the specific quartile. Differences in number of
observations are due to some firms’ SIC code not corresponding to one of Fama-French’s 30 industry classification.
Panel A:
Quartile No. Obs. Market Value Book Value M/B Scaled M/B
Market-to-Book Low 894 461.57 399.47 1.058 0.857
2 895 1,300.93 721.46 1.775 0.923
3 895 1,356.88 493.14 2.722 0.955
High 896 1,717.13 332.97 8.642 1.098
Scaled Market-to-Book Low 895 621.05 285.16 2.611 0.563

2 894 1,140.97 461.82 2.846 0.847
3 896 1,764.21 680.50 2.842 1.018
High 895 1,311.00 519.24 5.907 1.406
Panel B:
Scaled M/B Quartile
Industry Low 2 3 High
Food Products 19 28 22 18
Recreation 38 30 19 21
Printing and Publishing 13 20 16 20
Consumer Goods 20 32 24 25
Apparel 19 11 12 10
Healthcare, Medical Equipment, Pharmaceutical Products 115 91 66 56
Chemicals 11 16 33 24
Construction and Construction Materials 33 49 47 47
Steel Works, Etc. 6 29 26 25
Fabricated Products and Machinery 32 36 42 55
Electrical Equipment 31 25 37 34
Automobiles and Trucks 8 21 20 30
Petroleum and Natural Gas 37 60 50 35
Communication 45 31 31 21
Personal and Business Services 118 94 102 97
Business Equipment 81 81 88 145
Business Supplies and Shipping Containers 5 17 32 26
Transportation 25 28 34 27
Wholesale 51 61 49 41
Retail 89 53 62 41
Restaurants, Hotels and Motels 42 27 14 27
Other 57 54 70 70
Consistent with Spiess and Tkac, we examine the market reaction around four event
dates. January 4, 1995 was the date the legislation was first introduced in Congress, and December
6, 1995 was the date of the vote passing the final version of the law. December 19, 1995 was
the date President Clinton unexpectedly vetoed the bill and on December 22, 1995 Congress
overrode the veto and passed the legislation into law. The a priori expectation of the reactions for
the high scaled market to book portfolio is: A negative reaction on January 4, December 6 and
December 22 and a positive reaction on December 19.
Page 17
The valuation impact on firms was measured using standard event study methodology.
For robustness, we report abnormal returns surrounding each event using the Fama French
(1993) and Carhart (1997) four factor models, the market model, as well as raw returns. Since
our quartiles are formed on the basis of the market to book ratio, we also estimate the abnormal
returns based on the four factor model without the inclusion of the HML loading factor. Other than
when calculating raw returns during our event windows, the parameters used for predicting stock
returns were estimated over 125 trading days beginning 140 days before the start of the event
window. For easy comparison to the results of prior studies, we estimate abnormal returns over
three event windows; (-1,1), (-1,0), (0,1), similar to prior studies examining the market response to
the PSLRA. Although we only report the results for the period beginning the day before the
event date to the day following the event date, our results were similar using any of the three
event windows.
We then examine whether the level of overvaluation retained an ability to explain market
returns after controlling for other factors previously linked to shareholder litigation, such as
industry, the quality of corporate governance and analyst coverage. As a measure of governance
quality, we obtain values from Gompers, Ishii and Metrick’s (2003) governance index. Peng and
Roell (2008) and Johnson, Ryan and Tian (2009) show that managers have more incentive to
disclose false information when a greater portion of their compensation is equity based. Therefore,
we include a variable measuring the percentage of each CEO’s salary that comes from option and
restricted stock grants from Execucomp. We then utilize IBES to obtain the number of analyst
estimates for each firm, as analyst coverage is often used as an indicator for the information
opacity of a firm. Finally, a growing literature is developing that uses the skew of stock returns
to measure the willingness of investors to take on additional risk through the purchase of stocks
with lottery type attributes. As we are also looking at speculative type behavior, we include the
skew of each firm’s return as well, measured using daily returns for the year prior to the
introduction of PSLRA.
The intersection between the Gomper’s, Ishii and Metrick data set, IBES, Execucomp
and our previous sample contains 821 firms. We utilize the four factor adjusted returns as the
dependent variable in each of our models, although the results are similar using our other returns
measures as well.
EMPIRICAL RESULTS
Table 2 reports the abnormal returns at the introduction of the PSLRA. Panel A shows the
overall market reaction. Contrary to previous studies, we find a negative reaction to the
introduction of the law using raw returns and the market model estimates of abnormal returns.
Panel B shows the abnormal returns of each of the quartiles when sorted by the standard market
to book ratio. We find, consistent with our hypothesis, that regardless of how abnormal returns
are calculated, there is a monotonic decrease in the returns from the smallest market to book
quartile to the largest. Additionally, we find that the difference between each of the three
quartiles and the fourth quartile is statistically significant. Panel C shows the abnormal returns of
each of the quartiles when sorted by the scaled market to book ratio. Again, we find consistent
with our hypothesis that there is a monotonic decrease in abnormal returns in the groups. We
also find that the differences between the first two quartiles and the fourth quartile are always
significant, but the significance is lost in the difference between the third and fourth quartiles.
These results are consistent with our hypothesis that there was a strong relationship between
Page 18
overvaluation and market reaction when there was a decrease in the value of the option to sue.
Firms with high M/B and high, scaled M/B reacted more negatively to the introduction of the
PSLRA.
Table 2
ABNORMAL RETURNS AT ANNOUNCEMENT OF LEGISLATION
Abnormal returns are calculated using the four factor model as specified by Fama and French (1993) and Carhart (1997), the fo ur factor
specification minus the HML factor, and the standard market model. Market-to-book quartiles for Panel B are calculated by ranking the
market-to-book ratios and grouping the firms within quartiles. Scaled market-to-book quartiles for Panel C are calculated by finding the
monthly average M/B for each firm from 1990-1994 and scaling the market to book on the announcement date by each firm’s average MB.
The scaled MB are then sorted and grouped into quartiles. In order to estimate the Fama-French Carhart factor loadings, we used daily returns
from June 1994 until November 1994 to estimate loading factors. The (-1,1) event window results are shown, although other event windows
had quantitatively similar results. Mean 4-x is the mean difference between the High quartile and the respective quartile. The final column
shows the p-value for a t-test of difference of means between the High quartile and the respective quartile. *,**, and *** represent 10 percent,
5 percent and 1 percent significance respectively.
Panel A: All Firms
Mean p-value
Four Factor 0.0014 0.1939
Four Factor - HML 0.0016 0.1410
Market Model -0.0042*** 0.0000
Unadjusted Returns -0.0032*** 0.0008
Panel B: M/B
Quartile Mean p-value Mean High - x p-value
Four Factor Low 0.0084*** 0.0001 -0.0145*** 0.0000
2 0.0036* 0.0676 -0.0098*** 0.0016
3 -0.0002 0.9131 -0.006* 0.0608
High -0.0062*** 0.0098
Four Factor - HML Low 0.0096*** 0.0000 -0.0177*** 0.0000

2 0.0049** 0.0113 -0.0129*** 0.0000
3 -0.0001 0.9620 -0.008** 0.0138
High -0.0081*** 0.0012
Market Model Low 0.0053*** 0.0059 -0.0199*** 0.0000

2 -0.0005 0.7491 -0.0141*** 0.0000
3 -0.0067*** 0.0003 -0.0079*** 0.0053
High -0.0146*** 0.0000
Unadjusted Low 0.0068*** 0.0004 -0.0209*** 0.0000

2 0.0004 0.8029 -0.0145*** 0.0000
3 -0.0059*** 0.0011 -0.0082*** 0.0034
High -0.0141*** 0.0000
Panel C: Scaled M/B

Quartile Mean p-value Mean High - x p-value
Four Factor Low 0.0177*** 0.0000 -0.0252*** 0.0000
2 0.0003 0.8601 -0.0078*** 0.0066
3 -0.0050*** 0.0080 -0.0025 0.3602
High -0.0075*** 0.0004
Four Factor - HML Low 0.0186*** 0.0000 -0.0259*** 0.0000

2 -0.0003 0.8690 -0.0070*** 0.0193
3 -0.0046** 0.0144 -0.0027 0.3244
High -0.0073*** 0.0005
Market Model Low 0.0118*** 0.0000 -0.0245*** 0.0000

2 -0.0047* 0.0881 -0.0080*** 0.0021
3 -0.0110*** 0.0000 -0.0017 0.4941
High -0.0127*** 0.0000
Unadjusted Low 0.0124*** 0.0000 -0.0243*** 0.0000

2 -0.0035* 0.0510 -0.0084*** 0.0012
3 -0.0097*** 0.0000 -0.0022 0.3883
High -0.0119*** 0.0000
Page 19
Table 3
ABNORMAL RETURNS AT ANNOUNCEMENT OF LEGISLATION SORTED BY LITIGATION RISK OF INDUSTRY
Abnormal returns are calculated using the four factor model as specified by Fama and French (1993) and Carhart (1997), the four factor
specification minus the HML factor, and the standard market model as well as raw returns. Scaled market-to-book quartiles are calculated by
finding the monthly average M/B for each firm from 1990-1994 and scaling the market to book on the announcement date by each firm’s average
MB. The scaled MB are then sorted and grouped into quartiles, and then separated based on whether they are in an industry found in previous
studies to have a higher risk of shareholder litigation. In order to estimate the Fama-French Carhart factor loadings, and the market model
estimation, we used daily returns from June 1994 until November 1994 to estimate loading factors. The (-1,1) event window results are shown,
although other event windows had quantitatively similar results. High – Low is the mean difference between the high litigation risk firms in that
quartile and the other firms in that quartile, with the p-value of a t-test of difference of means given. Mean Q4-x is the mean difference between
the High quartile and the respective quartile, with the p-value for a t-test of difference of means between the High quartile and the respective
quartile given. *,**, and *** represent 10 percent, 5 percent and 1 percent significance, respectively.
Panel A: Four Factor Model
Quartile N Mean High - Low p-value Mean Q4 - x p-value
High Lawsuit Risk Industries Low 476 0.0208 0.0049 0.327 -0.0303*** 0.000
2 398 -0.0029 -0.0047 0.265 -0.0065 0.166
3 403 -0.0050 -0.0007 0.846 -0.0045 0.306
High 501 -0.0095 -0.0049 0.237
Low Lawsuit Risk Industries Low 414 0.0159 -0.0205*** 0.000

2 487 0.0018 -0.0063* 0.088
3 483 -0.0042 -0.0003 0.927
High 384 -0.0046
Panel B: Four Factor - HML

2 398 0.0000 -0.0006 0.868 -0.0084* 0.063
3 403 -0.0041 0.0015 0.688 -0.0043 0.342
High 501 -0.0084 -0.0019 0.660

2 487 0.0006 -0.0071* 0.055
3 483 -0.0056 -0.0009 0.812
High 384 -0.0065
Panel C: Market Model

2 398 -0.0072 -0.0047 0.191 -0.0087** 0.031
3 403 -0.0144 -0.0063** 0.050 -0.0015 0.720
High 501 -0.0159 -0.0072* 0.061

2 487 -0.0025 -0.0062* 0.068
3 483 -0.0081 -0.0006 0.845
High 384 -0.0087
Panel D: Unadjusted Returns

2 398 -0.0065 -0.0055 0.121 -0.0088** 0.026
3 403 -0.0131 -0.0062* 0.055 -0.0022 0.582
High 501 -0.0153 -0.0077** 0.044

2 487 -0.0010 -0.0066** 0.048
3 483 -0.0069 -0.0007 0.827
High 384 -0.0076
Table 3 shows the returns for the same scaled, market to book quartiles, except each quartile
is split into two sub-portfolios based on whether the firm is in a high litigation risk industry. Previous
research has found that the market reaction was more negative for these firms, and in general,
our results would support that. However, we find that it again depends on the level of market
valuation. There is a positive reaction to the law in the lower scaled valuation quartiles
Page 20
whether you are in a high risk industry or not. However, among firms in the higher valuation
quartiles, those in the high risk portfolios suffered significantly more negative returns than the low
risk firms. These results are consistent with the option to sue increasing investor’s willingness
to speculate, and certainly it is worth more to those shareholders in industries with higher
occurrences of shareholder litigation. However, even among the high risk firms, the same
monotonic relationship exists that also is found among low risk firms. The higher the valuation
of the firm, the more negative the market response to the introduction of the PSLRA.
Table 4
ABNORMAL RETURNS AT KEY DATES FOR PASSAGE OF LEGISLATION
Abnormal returns are calculated using the four factor model as specified by Fama and French (1993) and Carhart (1997). Scaled market-to-
book quartiles for Panel A are calculated by finding the monthly average M/B for each firm from 1990-1994 and scaling the market to book
on the announcement date by each firm’s average MB. The scaled MB are then sorted and grouped into qu artiles. Panel B groups the firms
into returns skewness quartiles. The previous (I think you did this so I don’t know how long). In order to estimate the Fama -French Carhart
factor loadings, we used daily returns from May, 1995 until October, 1995 to estimate loading factors. The (-1,1) event window results are
shown, although other event windows had quantitatively similar results. Mean 4-x is the mean difference between the 4th quartile and the
respective quartile. The t-test column shows the p-value for a t-test of difference of means between the 4 th quartile and the respective quartile.
*,**, and *** represent 10 percent, 5 percent and 1 percent significance, respectively.
Panel A:
Event Date Scaled M/B Quartile Mean p-value Mean 4-x p-value
12/6/1995 Low -0.0044* 0.068 0.0046 0.144
2 -0.0031 0.103 0.0032 0.238
3 -0.0034* 0.054 0.0035 0.183
High 0.0002 0.917
12/19/1995 Low -0.0025 0.251 0.0038 0.198
2 0.0030* 0.076 -0.0017 0.542
3 0.0011 0.500 0.0002 0.920
High 0.0014 0.504
12/22/1995 Low -0.0011 0.653 -0.0012 0.700
2 -0.0011 0.587 -0.0012 0.665
3 -0.0012 0.502 -0.0011 0.677
High -0.0022 0.263
Panel B:
Event Date Scaled M/B Quartile Mean p-value Mean 4-x p-value
12/6/1995 Low -0.0004 0.862 -0.0020 0.477
2 -0.0040* 0.060 0.0016 0.561
3 -0.0039* 0.056 0.0014 0.585
High -0.0024 0.204
12/19/1995 Low -0.0015 0.427 0.0053** 0.033
2 -0.0010 0.652 0.0048* 0.073
3 0.0017 0.367 0.0021 0.397
High 0.0039** 0.021
12/22/1995 Low 0.0033 0.106 -0.0057** 0.029
2 -0.0011 0.622 -0.0013 0.633
3 -0.0053*** 0.009 0.0029 0.252
High -0.0024 0.140
Table 4 shows the abnormal returns around the three December event dates. We report only
abnormal returns using the four factor model, but other models were also estimated and were
quantitatively similar. As mentioned in the methodology section, these firms have been resorted
and grouped based on their M/B and scaled M/B at the end of November as the time between
introduction and passage was nearly a full year. We find the results are not as strong as at the
Page 21
Table 5
ANALYSIS OF SCALED MARKET TO BOOK ON ABNORMAL RETURNS AT INTRODUCTION OF PSLRA
Cross sectional regression of abnormal returns calculated using the Fama and French (1993) and Carhart (1997) four-factor specification,
against the value of scaled market to book ratio. Scaled market-to-book ratios are calculated by finding the monthly average M/B for each firm
from 1990-1994 and scaling the market to book on the announcement date by each firm’s average market to book ratio. Control variables
includes the Gompers, Ishii and Metrick (2003) governance index (GIM), the number of analysts (# Analysts), as well as indica tors for those
industries previously identified in other studies to have had significantly different reactions at the announcement of the PSLRA. p-values are in
parenthesis. *,**, and *** represent 10 percent, 5 percent and 1 percent significance, respectively.
Abnormal Returns: Fama-French four factor adjusted

1 2 3 4 5 6
Scaled M/B -0.0180*** -0.0179*** -0.0117*** -0.0239*** -0.0196*** -0.0123**
(0.000) (0.000) (0.004) (0.000) (0.000) (0.014)
Turnover Percent 0.2095 0.2355 2.017*** 0.6597** 1.3064*** 1.8113***
(0.207) (0.159) (0.000) (0.017) (0.000) (0.000)
Return Skewness -0.0002 -0.0002
(0.130) (0.282)
GIM 0.0003 -0.0007
(0.615) (0.263)
No. Analyst Estimates 0.0001 0.0003
(0.776) (0.150)
% Option Compensation 0.0015 0.0046
(0.783) (0.509)
Industry Fixed Effects:
Healthcare 0.0074* 0.0073* -0.0014 0.0082* 0.0055 0.0064
(0.052) (0.057) (0.795) (0.058) (0.303) (0.326)
Services -0.0013 -0.0015 0.0026 -0.0014 -0.0018 0.0043
(0.711) (0.659) (0.637) (0.741) (0.729) (0.499)
Electronics 0.0143*** 0.0142*** 0.0101** 0.0179*** 0.0195*** 0.0080
(0.000) (0.000) (0.046) (0.000) (0.000) (0.174)
Retail 0.0034 0.0033 -0.0030 0.0051 -0.0008 -0.0064
(0.425) (0.450) (0.587) (0.288) (0.894) (0.313)
Intercept 0.0153*** 0.0153*** 0.0008 0.0174*** 0.0096* 0.0043
(0.000) (0.000) (0.898) (0.000) (0.068) (0.588)
Observations 3,546 3,546 1,016 2,315 1,182 675
R-squared 0.018 0.019 0.055 0.032 0.052 0.071
introduction date. While the sign of the average reaction is what we predicted for the highest
valuation quartile, the monotonic relationship between quartiles disappears, as does the significance
of the difference between groups. This is likely due to the market already incorporating information
relevant to the legislation into prices over the course of the year. Despite this information leakage,
we still find in the cross-section that the level of overvaluation is still a significant determinant of
the market’s reaction.
However, the market reaction to President Clinton’s veto on December 19th is very
important. Because we are focusing on firms trading at very high valuations, it would be reasonable
to suggest that on any random day they would revert back towards their long-term means, and a
negative abnormal return over any window should not be a surprise. However, after initially
indicating his support for the PSLRA, President Clinton unexpectedly vetoed it instead. In this
case, our hypothesis would suggest that the reaction would be positive for speculators, and we
would see a negative reaction for low market to book stocks and a positive reaction for high
valuation firms. This is indeed what we find, and again, our results support our hypothesis that
shareholder litigation encourages speculation in the market.
We then analyze whether our univariate evidence is robust to the introduction of other
variables that have been found to be significant factors in explaining the market’s response to the
PSLRA. Table 5 reports the results of a regression of abnormal returns around the introduction date,
January 4, on the Scaled M/B and other explanatory variables. When Scaled M/B and share
turnover are the only variables included, the coefficient on Scaled M/B is negative and highly
significant, while contrary to our prediction, there is a positive, albeit insignificant, relationship
Page 22
Table 6
ANALYSIS OF HIGH SCALED MARKET TO BOOK QUARTILE DUMMY ON ABNORMAL RETURNS AT INTRODUCTION OF
PSLRA
Cross sectional regression of abnormal returns calculated using the Fama and French (1993) and Carhart (1997) four -factor specification,
against an indicator variable that equals one for those firms in the highest scaled, market to book quartile. Control variables includes the
Gompers, Ishii and Metrick (2003) governance index (GIM), the number of analysts (# Analysts), as well as indicators for thos e industries
previously identified in other studies to have had significantly different reactions at the announcement of the PSLRA. p-values are in
parenthesis. *,**, and *** represent 10 percent, 5 percent and 1 percent significance, respectively.
Abnormal Returns: Fama-French four factor adjusted

1 2 3 4 5 6
Scaled M/B -0.0126*** -0.0126*** -0.0060* -0.0127*** -0.0070** -0.0074*
(0.000) (0.000) (0.061) (0.000) (0.047) (0.051)
Turnover Percent 0.2179 0.2544 2.0407*** 0.7282** 1.4382*** 1.7908***
(0.191) (0.128) (0.000) (0.009) (0.000) (0.000)
Return Skewness -0.0058** -0.0029
(0.020) (0.396)
GIM 0.0002 -0.0008
(0.747) (0.192)
No. Analyst Estimates 0.0000 0.0003
(0.976) (0.144)
% Option Compensation 0.0002 0.0043
(0.969) (0.535)
Healthcare 0.0076** 0.0074* -0.0017 0.0096** 0.0062 0.0057
(0.048) (0.054) (0.767) (0.027) (0.244) (0.383)
Services -0.0010 -0.0013 0.0021 -0.0011 -0.0027 0.0039
(0.772) (0.701) (0.696) (0.793) (0.609) (0.542)
Electronics 0.0139*** 0.0137*** 0.0090* 0.0167*** 0.0172*** 0.0075
(0.000) (0.000) (0.073) (0.000) (0.001) (0.207)
Retail 0.0040 0.0039 -0.0026 0.0063 -0.0001 -0.0058
(0.356) (0.366) (0.634) (0.195) (0.980) (0.362)
Intercept 0.0011 0.0025 -0.0084 -0.0024 -0.0075** -0.0042
(0.506) (0.157) (0.110) (0.330) (0.014) (0.534)
Observations 3,546 3,546 1,016 2,315 1,182 675
R-squared 0.013 0.014 0.050 0.020 0.040 0.067
between turnover and the share price reaction at the introduction of PSLRA. This is consistent with
our hypothesis that when the overvaluation of the firm is higher, the more negative the impact on
firm price will be because of the decrease in the value of the option to litigate. We then add in
the various control variables separately, and finally, altogether in the model. And we find that
adding the additional explanatory variables changes neither the sign nor the significance of the
scaled market to book variable, while the coefficient on turnover retains its sign and gains in
significance. In addition, we find that firms in the healthcare and electronics sectors had a
significantly positive reaction, consistent with Ali and Kallapur (1997), and inconsistent with
Spiess and Tkac (1997). Also, we find that the reaction was more positive for firms with greater
analyst coverage. The quality of corporate governance was not a significant factor in any
specification.
The relationship between turnover and the reaction to the law is contrary to our predictions,
and we propose two potential explanations for why it exists. First, with higher turnover, the
likelihood that any one share will be treated as damaged from early in the fraud declines. This is
because it will be assumed that it was traded during the class period, and while it is still damaged,
the damage estimate will be based on the estimated overvaluation on that date. To the extent that
overvaluation is greater early in the fraud period relative to the price at the end, turnover
actually can decrease the potential damage estimates for that investor. Alternatively, if investors
expect to trade out of their speculative position before a fraud is revealed, then higher turnover
allows them to do so without having as much impact on prices for that security.
Page 23
Table 7
CROSS-SECTIONAL ANALYSIS OF SCALED MARKET TO BOOK ON ABNORMAL RETURNS AROUND FINAL PASSAGE OF
PSLRA
Cross sectional regression of abnormal returns calculated using the Fama and French (1993) and Carhart (1997) four -factor specification,
against the scaled market to book ratio for each firm. Scaled market-to-book ratios are calculated by finding the monthly average M/B for
each firm from 1990-1994 and scaling the market to book on the announcement date by each firm’s average market to book ratio. The market
to book ratio is calculated based on the price as of November 30, 1995. Control variables includes the Gompers, Ishii and Met rick (2003)
governance index (GIM), the number of analysts (# Analysts), as well as indicators for those industries previously identified in other studies to
have had significantly different reactions at the announcement of the PSLRA. p-values are in parenthesis. *,**, and *** represent 10 percent,
5 percent and 1 percent significance, respectively.
Event Date
12/6/1995 12/19/1995 12/22/1995
Scaled M/B -0.0005** 0.0112** 0.0064
(0.032) (0.438) (0.204)
Turnover Percent -1.6349*** 2.4322*** -0.0866
(0.000) (0.000) (0.798)
Return Skewness 0.0003* -0.0001 -0.0002
(0.091) (0.223) (0.754)
GIM 0.0008* -0.0007** -0.0001
(0.670) (0.035) (0.123)
No. Analyst Estimates 0.0001 0.0004 0.0003
(0.215) (0.387) (0.871)
% Option Compensation -0.0070* -0.0054*** -0.0009***
(0.074) (0.008) (0.004)
Healthcare 0.0096 -0.0160 0.0147
(0.149) (0.997) (0.826)
Services -0.0075** -0.0011 -0.0000
(0.021) (0.252) (0.972)
Electronics -0.0112 0.0062 -0.0002
(0.547) (0.120) (0.662)
Retail 0.0032 -0.0091* -0.0022
(0.894) (0.069) (0.122)
Intercept 0.0009* -0.0132 -0.0096**
(0.093) (0.111) (0.024)
Observations 673 673 673
R-squared 0.093 0.111 0.024
When examining the range of values for the Scaled M/B ratio, we find that there is not a lot
of dispersion in the measure of overvaluation. It could be possible that rather than the level of
valuation being important, it is whether or not you are overvalued (and thus in the money) that
matters. In order to test this, we create an indicator variable that is equal to one if the firm is in
the largest quartile, and zero otherwise. Table 6 reports the results using the same specifications
as before, with the exception that the high Scaled M/B dummy was used instead. Consistent with
our earlier results, we find that being in the highest valuation quartile lead to a significantly more
negative reaction on January 4. When we include the additional variables, the coefficient on the
dummy variable remains significant and negative, and the coefficient on turnover remains positive
and significant in most specifications. The coefficients on the control variables remained consistent
with the previous model specifications using the each firm’s actual Scaled M/B value.
In Table 7, we report the results of regressing our Scaled M/B ratio on the abnormal returns
surrounding the three December event dates. We find that the coefficient on the scaled market to
book ratio is, as predicted, significant and negative on December 16, the day of the final
congressional vote. Again, and consistent with our hypothesis, the reaction to the presidential veto
was significantly more positive for firms with higher levels of overvaluation. And on the date the
veto was over-ridden, they reacted more negatively, but not significantly so. These results clearly
support our hypothesis that when the probability of passage increased and the option value to
litigate decreased, the market reacted negatively for those firms which were overvalued. And when
Page 24
Table 8
ANALYSIS OF MARGINAL IMPACT OF GOVERNANCE ON ABNORMAL RETURNS FOR SCALED MARKET TO BOOK
QUARTILES
Cross sectional regression of abnormal returns calculated using the Fama and French (1993) and Carhart (1997) four-factor specification, on
portfolios sorted based on scaled market to book values of the firms. Control variables includes the Gompers, Ishii and Metrick (2003)
governance index (GIM), the number of analysts (# Analysts), as well as indicators for those industries previously identified in other studies to
have had significantly different reactions at the announcement of the PSLRA. p-values are in parenthesis. *,**, and *** represent 10 percent,
5 percent and 1 percent significance, respectively.
Scaled M/B Quartiles
Low 2 3 High
Turnover Percentage 2.4432* 0.5366 1.1048 2.2514***
(0.098) (0.601) (0.105) (0.006)
Return Skewness -0.0007 -0.0003 0.0000 0.0000
(0.365) (0.416) (0.874) (0.985)
GIM 0.0000 -0.0004 0.0004 -0.0027**
(0.0997) (0.715) (0.679) (0.011)
No. Analyst Estimates -0.0003 0.0003 0.0000 -0.0099***
(0.707) (0.391) (0.961) (0.006)
% Option Compensation 0.0099 0.0093 0.0117 -0.0083
(0.662) (0.454) (0.319) (0.500)
Healthcare 0.0126 0.0123 -0.0042 0.0070
(0.485) (0.297) (0.714) (0.620)
Services 0.0225 -0.0109 0.0020 0.0014
(0.340) (0.412) (0.836) (0.894)
Electronics 0.0447 0.0064 0.0145 0.0009
(0.151) (0.642) (0.129) (0.916)
Retail 0.0005 -0.0004 -0.0084 -0.0176
(0.978) (0.975) (0.417) (0.255)
Intercept -0.0047 -0.0067 -0.0172 0.0025
(0.838) (0.563) (0.120) (0.830)
Observations 105 166 218 186
R-squared 0.117 0.047 0.055 0.149
the probability of passage decreased and the option value to litigate increased, the firms which
were overvalued received a more positive market reaction.
Overall, these results confirm our two hypotheses that the market does incorporate the
option to sue into prices, and that the response to key events during the passage of the PSLRA
would be dependent on factors that determine the value of payouts in litigation. But prior
research has shown that corporate governance and the information environment of the firm will
play a role as well. Certainly, the option to sue should be most valuable when there is greater
uncertainty regarding the truthfulness of information released by the firm. But similar to high- risk
industries, an alternative hypothesis could be suggested that firms with worse corporate governance
or information environments would also be the most likely to cheat in general, and given that
PSLRA made it more difficult to punish them, investors of these firms would react more
negatively in general to the introduction of PSLRA. In this case, we should see a direct
relationship between a firm’s corporate governance and information environment and the initial
market reaction to the introduction of the law. However, our model would indicate that these
factors are marginal and most relevant when investors are speculating on the firm’s value to
begin with. If there is no mispricing, there are no damages to recover regardless of how bad
governance is.
Therefore, we predict that these factors are more important for firms in the highest
valuation quartile, where speculation is the most likely to be occurring. In Table 8, we report the
results of a regression of our return measure on the governance index variable, number of analysts,
and other control variables for the firms in each of the four quartiles. Consistent with our
hypothesis, we find that in our highest valuation quartile, firms with fewer analysts and worse
corporate governance had a significantly more negative reaction to the introduction of PSLRA.
Page 25
On the other hand, this was not the case in the other three quartiles where they were
insignificant factors in explaining the market’s response. This is again consistent with a real option
explanation, and when combined with the results in Table 3, does not support the market reaction
being based solely on litigation becoming more likely for firms in high risk industries, firms
having bad governance, or firms having poor information environments. While these factors are
relevant, there must be overvaluation for a shareholder lawsuit to have value.
CONCLUSION
Shareholder lawsuits allow speculators to invest in firms based on uncertain information

with the belief that they may be able to recover a portion of their losses through litigation. By
passing PSLRA, Congress effectively decreased the option value to sue, and reduced the
anticipated payoff from speculation. Our findings show that during the events leading up to the
passage of PSLRA, the market response depended a great deal on the level of overvaluation for a
firm. We find that when the reform bill was introduced, decreasing the value of the option to
litigate, firms who were more likely to be overvalued and had relatively low turnover suffered
worse losses in the market. As these factors are used to determine payoffs in shareholder litigation,
we conclude that the option to sue was incorporated into market prices. Not surprisingly, this
option was most valuable for investors in high litigation risk industries, and those with worse
information environments. Therefore the losses were exacerbated by these factors, but for the most
part only for those with the highest levels of overvaluation where there was the potential for a
positive payoff in litigation. For firms trading closer to their long-term valuation levels,
corporate governance or industry has little ability to predict the market response to the securities
litigation reforms.
These results give credence to the belief that the ability to sue ex post increases the
willingness to speculate on firm value ex ante. Potentially, this could also distort the cost of raising
equity for firms, leading firms to over-invest in marginal projects. Also, our results illustrate a
heretofore overlooked negative cost imposed upon markets by shareholder litigation, in
comparison to the potential benefits it creates as a deterrent against managerial malfeasance.
REFERENCES
Ali, A. & S. Kallapur (2001). Securities price consequences of the private securities litigation reform act of 1995 and
related events. The Accounting Review 76, 431-460.
Bajaj, M., S.C. Mazumdar & A. Sarin (2003). Securities class action settlements: An empirical analysis. Santa Clara
University Law Review 43, 101-132.
Barber, B.M. & J.D. Lyon (1997). Firm size, book-to-market ratio, and security returns: A holdout sample of financial
firms. The Journal of Finance 52, 875-883.
Campbell, J.Y., A.W. Lo & A.C. MacKinlay (1997). The Econometrics of Financial Markets. Princeton, NJ: Princeton
University Press.
Carhart, M.M (1997). On persistence in mutual fund performance. The Journal of Finance 52, 57-82.
Cheng, C.S.A., H. Huang, Y. Li & G. Lobo (2010). Institutional monitoring through shareholder litigation. Journal of
Financial Economics 95, 356-383.
Choi, S. (2007). Do the merits matter less after the private securities litigation reform act? Journal of Law, Economics
and Organization 23, 598-626.
Coffee, J.C. (2006). Reforming the securities class action: An essay on deterrence and its implementation. Columbia
Law Review 106, 1534-1586.
Cornell, B. (1990). The incentive to sue: an option-pricing approach. The Journal of Legal Studies 19, 173-187.
Page 26
Cox, J.D., R.S. Thomas & L. Bai (2008). There are plaintiffs and…there are plaintiffs: An empirical analysis of
securities class action lawsuits. Vanderbilt Law Review 61, 355-386.
Fama, E. & K. French (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial
Economics 33, 3-56.
Gompers, P., J. Ishii, and A. Metrick (2003). Corporate governance and equity prices. The Quarterly Journal of
Economics 118, 107-155.
Johnson, M.F., K. Nelson & A.C. Pritchard (2007). Do the merits matter more? The impact of the private securities
litigation reform act. Journal of Law, Economics and Organization 23, 627-652.
Johnson, M.F., R. Kasznik & K.K. Nelson (2000). Shareholder wealth effects of the Private Securities Litigation
Reform Act of 1995. Review of Accounting Studies 5, 217-233.
Johnson, S.A., H.E. Ryan, Jr., & Y.S. Tian (2009). Managerial incentives and corporate fraud: The sources of
incentives matter. Review of Finance 13, 115-145.
Kumar, A. (2009). Who gambles in the stock market? Journal of Finance 64(4), 1889-1933.
Peng, L. & A. Roell (2008). Executive Pay and Shareholder Litigation. Review of Finance 12, 141-184.
Perino, M. (2003). Did the private securities litigation reform act work? U. Illinois Law Review 913-977.
Rhodes–Kropf, M., D.T. Robinson & S. Viswanathan (2005). Valuation waves and merger activity: The empirical
evidence. Journal of Financial Economics 77, 561-603.
Spiess, D.K. & P. Tkac (1997). The private securities litigation reform act of 1995: The stock market casts its vote.
Managerial and Decision Economics 18, 545-561.
Page 27
POTENTIAL FINANCIAL EFFECTS OF THE 2015

CORPORATE TAX REFORM OF JAPAN TO
ELECTRONICS COMPANIES WITH COMPARISON TO
SOUTH KOREA
Abraham R. de Guzman, Ritsumeikan Asia Pacific University
Won Jae Kim, Ritsumeikan Asia Pacific University
ABSTRACT
In an effort by the Japanese government to restore its weak economy, majority of

corporate tax amendments and reforms since 2011 were driven towards reducing the financial
burden on companies. We focus on the potential financial effect(s) of the corporate tax reforms
mentioned in the following documents: (1) Basic Policies for the Economic and Fiscal
Management and Reform 2014 (the “Honebuto no Houshin”) and (2) the discussion paper
outlining views on future corporate tax reform (the “Discussion Paper”) released by the
Corporate Tax Discussion Group (CTDG) of Japan’s Government Tax Commission. We simulate
how the 2015 tax reforms for Japan impact its electronics-manufacturing industry financially by
assuming the specific tax changes on the reported historical financial information from 2002 to
2014 of the 12 actively traded electronics-manufacturing corporations (12EMC) in the Tokyo
Stock Exchange (TSE). Likewise, a similar simulation is performed for South Korea’s 2015 tax
reforms and selected electronics-manufacturing corporations.
We hypothesize that the tax reforms individually and/or collectively will result to: (H1)
potential tax benefits for Japan; and (H2) potential tax losses for South Korea, i.e. in line with
the amendments’ intentions to create economic virtuous cycles and encourage fair competition
for Japan and South Korea. Simulation results prove contrary to expectation for Japan as its
electronics-manufacturing industry suffered overall potential losses after applying the tax reform
items to 12EMCs’ historical financial data. Therefore, the expectations for the 2015 corporate
tax amendments do not show positive economic impact due to the two major amendments – (1)
the reduced effective corporate tax rate; and (2) the net operating losses (NOL) allowable
deduction – being dependent on the profitability of corporations. On the other hand, as expected
and inasmuch as the South Korean tax reforms are created to prevent monopoly by promoting
usage of excess earnings, overall potential losses were noted for its electronics-manufacturing
companies’ historical financial data.
Keywords: Japan, South Korea, tax reform, and electronics industry
BACKGROUND
Since the great Tohoku earthquake of 2011, the Japanese government’s corporate tax
reforms have been directed towards the recovery of Japan’s economy. Taxation aids are
particularly visible on how the government immediately approved a bill in 2011 to reduce the
effective tax rate for corporations despite the concurrent approval of the Special Restoration Tax
Law requiring the payment of surtaxes. Furthermore, upon the re-election of Shinzo Abe as
Page 28
Prime Minister on December 2012, contained in his Abenomics initiatives is the early
termination of the aforementioned corporate surtax. Moreover, the scheduled 35.64 percent
effective corporate tax rate was implemented a year ahead by April 1, 2015. Developments and
discussions about Japan’s corporate taxation reforms during 2014 also leaned towards simplified
and industry-neutral taxation rules.
The Japanese Cabinet approved the “Basic Policies for the Economic and Fiscal
Management and Reform 2014” or the Honebuto no Houshin on June 24, 2014. It targets to
reduce the effective corporate tax rate in the range of 20 to 30 percent over the next years with
the first reduction scheduled for the 2015 tax reform. A reduction in corporate tax rate must be
coupled with and increased taxable base for the Japanese government to maintain a consistent
stream of government revenues. As a result, the Corporate Tax Discussion Group (CTDG) of
Japan’s Government Tax Commission recommended the abolishment or reduction of the
following benefits being enjoyed by eligible corporations: (1) tax incentives (special depreciation,
tax credit and etc.); (2) net operating loss (“NOLs”) deductions; (3) dividend income non-
taxation; (4) depreciation methods; (5) deduction of local tax payment for corporate income tax
calculation; (6) preferential tax measures for SMEs; and (7) local tax regime.
Purpose of the study
We analyze how the 2015 tax reforms for Japan impact its electronics-manufacturing
companies financially by simulating the specific tax changes on the reported historical financial
information of corporations within the industry. Moreover, corporate tax reforms for South
Korea are examined and executed for its selected electronics-manufacturing companies. The
simulation results are then utilized to ascertain whether amendments intentions occurred for the
collected historical data. This aims to guide corporate decision makers as to the appropriate
preparation they may do in response to tax changes, and inspire further researches for items,
which may not be considered in this study.
Scope and limitation
We attempt to measure potential individual and net collective financial impacts of the
2015 corporate tax reform on Japanese companies by re-computing amended taxable items of
historical financial data from 2002 to 2014 for the top 12 actively trade electronics-
manufacturing corporations (12EMC). See Appendix 1 for the list of 12EMC’s financial
performance for 2014.
On the other hand, we also simulate the 2015 corporate tax reform of South Korea to its
two actively traded electronics-manufacturing companies (SKEMC): (1) LG Electronics, Inc.,
and (2) Samsung Electronics Co. Ltd.
REVIEW OF RELATED LITERATURE
Japan’s Electronics Industry
According to Japan Electronics and Information Technology Industries Association

(JEITA) statistics for 2014, the country’s total electronics production amounted to JPY11.80
trillion or six (6) percent of the JPY207.63 trillion global output.
Page 29
From 2002 to 2007, Japan’s electronics industry (JEI) maintained a relatively constant
level of production, exports, and imports. However, due to the adverse financial collapse
spawned by the 2008 Lehman shock to industries worldwide, JEI’s production plummeted from
JPY18.58 trillion to JPY13.61 trillion by 2009. In response to the previous, the Japanese
government approved an industrial competitiveness enhancement act to establish organizations
that will support companies who incurred sizable losses (Ono, 2010). However, this
revitalization for companies was short-lived due to the Tohoku Earthquake on March 2011. JEI’s
productivity went down from JPY15.33 trillion to JPY13.04 trillion. The 2011 earthquake halted
JEI’s operations for several months. Inasmuch as JEI’s production accounts for 20 percent of
global semiconductor sales as well as 60 percent of the total supply of wafers to manufacturers,
the material impact of the production stoppage was felt by the world (PwC Japan, 2011). The
following years, continuing production declines were recorded for JEI even amidst growth in
global demand for smartphones and tablets. Furthermore, market leaders for video screen and
television appliance – Sony, Panasonic, and Sharp – lose their shares to two growing
powerhouses from South Korea, Samsung and LG electronics, holding 23 percent and 13 percent
of the world’s total market coverage (Viera, G., 2012). This shift can also be seen in Japan’s
increasing imports for electronics from 2012, see Figure 1.
25
19.85 20.28 20.42

20 18.84 19.30 19.10 18.58
15.33
15 13.61 13.04
14.66 15.04 11.82 11.35 11.80
13.75 13.62 13.12
12.70
10 12.26
10.34 9.05
9.09 9.11 9.39
8.66
5
0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Production Exports Imports
Figure 1: Production, Exports, and Imports by Japan’s Electronics Industry

from 2002 to 2014 (in trillions JPY)
Japan’s 2015 Corporate Tax Reform
The ruling majority parties of the legislative body of Japan, i.e. the Liberal Democratic
Party and Komeito (The Liberal Democratic Party of Japan, 2015), approved the 2015 Tax
Reform Proposal (2015TRP) of Japan. It targets to provide measures to help reach economic
virtuous cycles for Japanese corporations by reducing the effective tax rate starting fiscal year
Page 30
2015 (PwC Japan, 2015). Cross matching the approved tax reforms with the proposed tax base
expansion measures by the CTDG after the approval of the 2014 Honebuto no Houshin on
24 June 2014 are presented on table 1 below.
Table 1
STATUS OF CTDG PROPOSED MEASURES AFTER APPROVAL OF JAPAN’S 2015 TAX REFORM
Tax measure Status
1. Tax incentives (special depreciation, tax credit and etc.) Approved
2. Net operating losses (NOL/s) deductions Approved
3. Dividend income non-taxation Approved
4. Depreciation methods Pending
5. Deduction of local tax payment for corporate income tax calculation Pending
6. Preferential tax measures for SMEs Approved
7. Local tax regime Approved
Provisions contained in the 2015TRP officially took effect by April 2015 as published in
Japan’s Ministry of Finance Tax Reform Policy web pages. Detailed discussion of the reduced
effective corporate tax rate and corresponding tax measures to increase the tax base follows and
are presented in appendix 2.
Reduced Effective Corporate Tax Rate (RECTR) and Local Tax Regime
Starting fiscal year 2015, the effective corporate tax rate of 35.67 percent drops to 33.06
percent. This new rate further decreases to 32.26 percent by fiscal year 2016. The local tax
regime base widening measure by the CTDG – increases in the size based taxes, i.e. value added
base, and capital base tax rates – are already considered in the aforementioned effective rates.
Net Operating Losses (NOL) Deduction
Under the old tax system, corporations incurring net losses from operations were allowed
80 percent of which to be deducted on the succeeding income earning years, with any creditable
excess permitted to be carried over within nine (9) years following the initial recognition of the
NOL credit. According to the tax reform, net operating losses by corporations are only 65
percent creditable for fiscal years 2015 and 2016 with the same carryover period provision for
any excess credits over nine years. Consequently, by fiscal year 2017, any operating losses
incurred will only be 50 percent deductible from succeeding income earning years matched by a
year increase in carryover period of up to 10 years.
Dividend Income Exclusion (DivEx)
For fiscal year 2014 and earlier, dividends income exclusion percentages almost parallel
to the equity method and market method for subsidiary accounting. Under the equity method, a
parent corporation holding 20 percent or more interest in a subsidiary has significant influence
over the financial and operating activities of the investee (Weygandt, Kimmel, Kieso, 2013).
Congruently, dividends from subsidiaries under significant influence are recorded as a deduction
Page 31
from the recognized investment amount in the balance sheet of the parent company.
Alternatively, market method has dividends received from investments without significant
influence posted as other income in the income statement of the parent (Macabacus LLC, 2015).
Specifically, under Japan’s old rule, 25 percent or more interest guarantees the complete
exclusion of dividends received from subsidiaries. On the other hand, dividends from
investments below 25 percent control are allowed 50 percent exclusion for tax purposes.
However, the new rule carrying the intention of increasing the taxable base for
corporations became stricter in defining subsidiary control requirement to allow 100 percent
dividends income exclusion. A parent corporation must now own more than one third interest
over its investee to claim an entire exclusion of dividends income. Furthermore, the new tax
reform creates two new levels for partial exclusion of dividends income, i.e. five (5) percent to
less than 1/3 ownership to claim for 50 percent exclusion, and less than five (5) percent
ownership for 20 percent exclusion.
Tax incentives changes – Permanent Research & Development Incentive (PR&D)
According to the old corporate tax rules, corporations were given incentives equivalent to
eight (8) percent to 10 percent of their Research & Development (R&D) expenditures subject to
a creditable limit amounting to 30 percent of the corporate tax obligation for the year under
consideration. Any excess incentives after reaching the creditable limit were then permitted to be
credited or carried-over towards the succeeding corporate income tax paying year. Conversely,
under the new tax reform, the aforementioned 30 percent creditable limit for R&D tax incentives
is further reduced to 25 percent, with any corresponding excess no longer valid for carry-over the
following year.
Tax incentives changes – Temporary Research & Development Incentive (TR&D)
On top of the PR&D, corporations are provided additional temporary R&D tax credits
valid within April 1, 2013 to March 31, 2017. These TR&Ds are equal to whichever is higher of
(a) five (5) percent to 30 percent of the corporation’s incremental R&D; or (b) R&D costs in
excess of 10 percent of average sales multiplied by a mechanically calculated tax credit ratio that
changes depending on the relationship of R&D expenditures and average sales. The higher
amount between the two aforementioned is then subject to a creditable limit matching to 10
percent of the corporate tax obligation for the year under consideration. Upon enforcement of the
tax amendments by fiscal year 2015, there are no changes concerning the rules of the TR&D
except for the non-renewability of its period of enforcement.
Japan and South Korea’s historical effective corporate tax rates
The comparison of historical effective corporate tax rates between Japan and South Korea
is presented in Figure 2. Japan’s trend reflects its government aim to reduce the effective
corporate tax rate. On the other hand, the lowest effective corporate tax rate recorded for South
Korea was for 22 percent in 2011, which immediately increased back to the current 24.20
percent by 2012.
Page 32
42.00% 42.00% 42.00% 42.00%

40.69% 40.69% 40.69% 40.69% 40.69% 40.69%
38.01% 38.01%
35.04%
29.70% 29.70% 29.70%

27.50% 27.50% 27.50% 27.50%
24.20% 24.20% 24.20% 24.20% 24.20%
22.00%
Japan South Korea
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Source: KPMG, 2015
Figure 2: Effective Corporate Tax Rates for Japan and South Korea from 2002 to 2014
South Korea’s 2015 Corporate Tax Reform
Features of the 2015 corporate tax reforms for South Korea include changes targeted to
increase government revenues via additional surtax, revision of penalties for noncompliance with
certain administrative procedures, and reduction of the tax incentive for R&D.
Additional Surtax to facilitate the use of Retained Earnings (ASRE)
To encourage utilization of excess retained earnings to fund facility investments, salary

increases, and more dividends distribution, the South Korean government levies a 10 percent
surtax to corporations with excess earnings amounting to 50 billion Korean Won (KRW) and
belonging to industries subject to the Act of Monopoly Regulation and Fair Trade (Samil PwC,
2014). Detailed information concerning the surtax computation is presented in Appendix 3.
Administrative Purposes Tax Reforms (APTR)
Tax administration reforms concerning changes in foreign tax credit rules, particularly,
credit limitation, indirect foreign tax credit, and extended period for submission of foreign tax
credit reports, were approved for tax year 2015 (Samil PwC, 2014). Moreover, penalty changes
relating to the extension of corporate tax return filing and non-compliance with reporting
requirement of overseas property holdings were added.
Page 33
Research and Development Tax Credit (R&DTC)
Starting tax year 2015, the operative tax credit rate for R&D has been lowered (Samil
PwC, 2014), so from three (3) percent tax credit, large corporations can only credit
approximately two (2) percent of their R&D expenditures.
HYPOTHESES
Considering the potential financial impact of the corporate tax reforms, the simulation
was performed using historical financial information from 2002 to 2014 of the corporations
within the electronics-manufacturing industries for the two countries. We, therefore hypothesize
that:
H1 the 2015 corporate tax reform of Japan, in line with the intention of creating economic virtuous
cycles, will bring an overall potential tax benefit for its electronics-manufacturing companies.
H2 the 2015 South Korean tax reform will result to potential tax losses for its electronics companies
in accordance with Mill’s 1848 concept of equality of sacrifice, wherein levied amount is
proportionate to earned benefits. Specifically, South Korean changes are promulgated to protect
its industry from monopoly by certain corporation/s and promote usage of excess corporate
earnings to invest in facilities, pay higher salaries, and distribute more dividends.
These are the a-priori expectations based on the primary intentions of the tax reforms.
Results contrary to expectations for Japan would mean that the approved tax base widening
measures by CTDG are then significant with regard to the historical financial data of the
12EMC. On the other hand, outcomes opposite the expectations for South Korea will not occur
because the tax reforms focuses on means to discourage monopoly via excess earnings, i.e.
obligations will either remain the same or become higher due to probable penalties or
noncompliance to certain rules in the amendments.
METHODOLOGY
To simulate the potential financial effect of the tax reforms presented in the previous
sections, we have performed the following simulation procedures:
1. Gathered the financial records of the 12EMC and SKEMC for the period covered, i.e.
from 2002 to 2014
2. Performed descriptive statistics to ensure integrity of data
3. Normalized data for consolidation purposes
4. Simulated the effects of applicable tax reforms:
a. Applied the reduced effective tax rate to historical taxable income
b. Computed the NOLs deductions for corporations incurring losses
c. Computed if dividends received from affiliates are deductible from tax
obligation
d. Computed applicable R&D incentives
5. Consolidated results and determined if the individual and combined effects of the tax
reform/s will support the creation of economic virtuous cycles for Japan’s 12EMC,
and promote usage of excess earnings for SKEMC.
Page 34
Financial records were derived from the S&P Capital IQ web-based online database using
Japanese Yen (JPY) as the functional currency for simulation purposes. Consequently, to ensure
completeness and integrity of information we used for the simulation, we performed descriptive
statistics of data inputs and had the information verified in the published reports of the respective
sampled corporations. See tables 6 to 9 for the variable label dictionaries and descriptive
statistics summaries.
Table 2
VARIABLE LABEL DICTIONARY FOR 12EMC SIMULATION DATA
Variable label Description
firm
1 Alps Electric Co. Ltd. (Alps)
2 Canon Inc. (Canon)
3 Fujifilm Holdings Corporation (Fujiflim)
4 Hitachi Ltd. (Hitachi)
5 Nikon Corporation (Nikon)
6 Panasonic Corporation (Panasonic)
7 Pioneer Corporation (Pioneer)
8 Ricoh Company, Ltd. (Ricoh)
9 Sharp Corporation (Sharp)
10 Sony Corporation (Sony)
11 Toshiba Corporation (Toshiba)
12 Yamaha Corporation (Yamaha)
year Tax year
nectr New effective corporate tax rate
oectr Historical effective corporate tax rate
tct Tax obligation
ni Net income
di Dividend income
r&d Research and development
rev Revenues
Table 3
DESCRIPTIVE STATISTICS FOR 12EMC SIMULATION DATA
Variable Obs Mean Std. Dev. Min Max
firm 156 6.5 3.46317 1 12
year 156 2008 3.753708 2002 2014
newectr 156 0.3232154 0.0021386 0.3226 0.3306
oldectr 156 0.4024615 0.0197906 0.3504 0.42
tct 156 57073.24 53446.2 1507 264258
ni 156 21444.16 187450.5 -787337 488332
di 156 6846.795 6737.063 281 34272
rd 156 221431.5 178277.8 21736 615524
revenue 156 3526646 2933116 356616 1.03E+07
Page 35
Table 4
VARIABLE LABEL DICTIONARY FOR SKEMC SIMULATION DATA
Variable label Description
firm
1 LG Electronics Inc. (LG)
2 Samsung Electronics Co. Ltd. (Samsung)
year Tax year
ectr Historical effective corporate tax rate
tct Tax obligation
payroll Salaries paid
tdp Dividends paid
r&d Research and development
Table 5
DESCRIPTIVE STATISTICS FOR SKEMC SIMULATION DATA
Variable Obs Mean Std. Dev. Min Max
firm 26 1.5 0.509902 1 2
year 26 2008 3.815757 2002 2014
ectr 26 0.2631538 0.0254522 0.22 0.297
tct 26 161495.1 191068.7 18154.56 742938.1
payroll 26 1531069 935316.3 546977 4110159
tdp 26 60484.29 60713.96 0 245197.3
rd 26 458792.9 393015.1 38619.45 1578978
Simulation procedures and assumptions - Japan
Reduced effective corporate tax rate simulation
To arrive at taxable income, we divided the historical effective corporate tax rate to the
applicable corporate tax obligation for the years covered by the simulation, i.e. 2002 to 2014.
Consequently, we multiplied the new effective corporate tax rate to the computed income for tax
purposes. The resulting difference between the actual income tax obligation and the computed
tax obligation under the new effective corporate tax rate is the recognized potential tax benefit
for the simulation.
Net operating losses (NOL) deduction
We applied the new rules for the previous year/s’ incurred losses. Moreover, we further
assumed that there is no government involvement under the process of tax credit timing. The
simulation was done under the assumption that corporations had received permission to use any
cumulative tax credit amount for the immediately following income reporting year.
Dividend Income
All dividends accounted for under the equity method, i.e. treated as a deduction in the
Investment account of the Balance Sheet, are 100 percent deductible. On the other hand,
Page 36
dividends income recognized in the Income Statement are considered as dividends from
corporations wherein the parent only holds below five (5) percent control of the investee.
Research & Development (R&D)
All reported R&Ds are applicable for permanent and temporary tax incentives. Although
the creditable amount for tax purposes is stipulated as within the eight (8) percent to 10 percent
range, we assumed that all corporations would receive the maximum creditable limit for R&D
incentives. This is due to the fact that electronics-manufacturing companies reported relatively
consistent R&D amounts annually based on their historical financial data. Moreover, their
industry is a significant contributor to Japan’s growth with production of JPY11.8 trillion or six
(6) percent of world’s electronics output (JEITA, 2014).
Simulation procedures and assumptions - South Korea
Additional surtax to facilitate use of corporate retained earnings
We performed the same procedures done for the Japanese corporations’ simulation in
computing for the taxable income of South Korean corporations. From the two methods
mentioned in the South Korean tax reform, we applied the second one in the simulation, wherein:
((Taxable income for the year x 30%) – (the total amount of payroll increases and dividend
payments)) x 10%
The second method is preferred due to the fact that the first method requires information
that are not available from public information, e.g. interest income received on refunds of
overpaid national taxes, statutory reserve transfers, disallowed donations, etc.
Administrative Purposes Tax Reforms (APTR)
Tax administration reforms were not considered in the quantitative simulation of

potential financial effects as any amount resulting from these penalties is contingent on the future
probabilities of delayed submission or noncompliance of the sampled corporations.
Research & Development (R&D)
All reported R&D expenditures are applicable for tax incentives simulation.
SIMULATION RESULTS
The consolidated corporate tax reform simulation results for Japanese electronics
corporations are presented in Table 6. As expected, reductions in the effective corporate tax rate
result to potential tax benefits for recomputed financial data from 2002 to 2014. Conversely,
reductions of allowable tax incentives or exclusions (limitation on NOL deduction, reduced
dividend income exclusion, and reduction of R&D tax incentive) result to potential tax losses.
Overall, 12EMC incurred a potential tax loss amounting to JPY971 billion, contrary to our first
hypothesis (H1) that the electronics-manufacturing companies of Japan will have an overall
Page 37
potential tax benefit after the simulation. Among the four major tax reform items of Japan, the
reduction of the effective corporate tax rate, and the limitation on the NOL deduction, are the
significant figures in the simulation. Both reforms resulted to more than JPY1.75 trillion, or four
(4) percent of the industries JPY42.77 trillion contributions to the Japanese economy for 2014.
From an individual company standpoint, Canon is the highest potential benefit gainer
with JPY263 billion. This is due to the fact that it is not affected by the reduction on the NOL
limit. In other words, Canon is the sole corporation within the industry that did not report any
loss from 2002 to 2014. Next to Canon are two corporations who only reported a loss for one
period within the simulation, Ricoh and Fujifilm. Since Canon, Ricoh, and Fujifilm are the top
performing corporations within the simulation period, it proves how greatly favorable the
reduced effective corporate tax rate reform is for profitable corporations.
On the other hand, Panasonic is the largest potential loss receiver with JPY545 billion.
This loss is attributable to Panasonic reporting net losses six times (2002 to 2003, 2009 to 2010,
and 2012 to 2013) for the past 13 years. Next to Panasonic are Sharp, Hitachi, and Sony*,
reporting large losses for over three (3) periods during the simulation period. These losses are
also reflected in the declining productivity of JEI, see figure 1. Results for Panasonic, Sharp,
Hitachi, and Sony exemplifies how significantly unfavorable the NOL tax reform is for
corporations that will report net losses in the coming years. Thus, the NOL reform heavily
contradicts the creation of economic virtuous cycles brought about by the reduced effective tax
rate for non-performing corporations.
*Note: Although historical data used in the simulation for Sony contains its other profitable segments like financial
services, music, and pictures. Majority of Sony’s results still relates to its core electronics-manufacturing segments.
Table 6
CONSOLIDATED CORPORATE TAX REFORM SIMULATION RESULTS FOR JAPAN’S 12EMC
from 2002 to 2014 (in millions JPY)
Reduced Potential
Reduced Limitation Change in
dividend Tax
Net assets corporate on NOL R&D tax
income Benefit
tax rate deduction incentive
exclusion (Losses)
Alps Electric Co. Ltd. 230,380 21,269 (23,142) (3,734) (5,464) (11,071)
Canon Inc. 3,140,758 414,002 - (48,060) (102,917) 263,025
FUJIFILM Holdings Corp. 2,198,223 124,025 (11,532) (30,809) (30,508) 51,176
Hitachi Ltd. 3,852,464 300,812 (368,142) (25,379) (78,094) (170,803)
Nikon Corporation 546,812 40,096 (5,907) (6,338) (10,412) 17,439
Panasonic Corporation 1,586,438 201,978 (669,739) (25,328) (52,047) (545,136)
Pioneer Corporation 77,816 20,988 (98,380) (9,618) (5,055) (92,065)
Ricoh Company, Ltd. 1,094,396 112,804 (13,368) (14,363) (28,165) 56,908
Sharp Corporation 207,173 88,168 (314,171) (19,654) (22,041) (267,698)
Sony Corporation 2,787,256 271,509 (294,820) (75,722) (68,900) (167,933)
Toshiba Corporation 1,652,327 146,765 (147,093) (56,049) (37,427) (93,804)
Yamaha Corporation 274,842 16,577 (18,016) (5,377) (4,141) (10,957)
Total 17,648,885 1,758,993 (1,964,310) (320,431) (445,171) (970,919)
Page 38
The results for the SKEMC tax reforms simulation are presented in table 7. The nature of
the tax reforms greatly affects the results as the major change, i.e. the additional surtax for
corporations with net equity in excess of more than KRW50 billion, is geared towards prevention
of monopoly by a single firm within the industry. In other words, Mill’s 1848 concept of equality
of sacrifice was observed under South Korea’s government revenue collection reform. As a
result, SKEMC posted potential losses amounting to JPY225 billion in our simulation, with
Samsung sharing over 80 percent or JPY186.94 billion of these potential losses.
Table 7
CONSOLIDATED CORPORATE TAX REFORM SIMULATION RESULTS FOR SKEMC
from 2002 to 2014 (in millions JPY)
Additional Change in R&D Potential Tax
Net asset
Surtax tax incentive Benefit (Losses)
LG Electronics Inc. 1,425,926 (15,599) (23,058) (38,657)
Samsung Electronics Co. Ltd. 18,449,653 (90,708) (96,228) (186,936)
Total 19,875,579 (106,307) (119,286) (225,593)
CONCLUSION
In conclusion, ceteris paribus, our simulation with potential tax losses of JPY971 billion
corroborates that the collective corporate tax reforms of Japan negatively impacts its electronics-
manufacturing companies. This is due to the fact that the two largest reforms are dependent on
the financial performance of the corporations. For the reduced effective corporate tax rate, more
benefits result for corporations reporting consistent positive bottom lines. On the other hand, the
NOL tax reform produce more disadvantages for corporations who suffer net losses. This is due
to the fact that a mere three percent reduction in effective corporate tax rate cannot offset the
impact of a 15 percent and 30 percent reduction in allowable tax deduction from reported net
losses.
As expected, South Korea’s tax reform aim on neutralizing certain corporation’s
monopoly of an industry is apparent in the negative impact of JPY225 billion potential losses for
its electronics-manufacturing industry players.
Furthermore, the simulation results are aimed at corporate decision makers to help them
plan their financial planning efforts in anticipation of the respective potential financial
consequences of the examined tax reforms. Accordingly, further investigation is recommended
as more changes are expected for Japan and the electronics industry. Additional research can
study on specific government policies and its impact on the performance of electronics
companies. These policies can include but are not limited to government subsidy, fiscal policies,
pump priming, income tax holiday(s) (ITH) and policy mix. On the other hand, an augmented
study by introducing more competition and looking into movements in companies’ financial
performance when the Hirschman Herfindahl index changes are applicable. For broader scope,
explore the role of Peoples Republic of China and Taiwan, Republic of China in the global
electronics arena. Last, investigate impulse response of the private sector on fiscal policies
amendments.
Page 39
REFERENCES
The Liberal Democratic Party of Japan (2015). About LDP. Retrieved from Strength of Political Groups in the
Houses: https://www.jimin.jp/english/about-ldp/strength/index.html
Alps Electric Co., Ltd. (2015). ALPS. Retrieved 2015, from ALPS - Annual Report:
http://www.alps.com/e/ir/ir_annual.html
Canon Inc. (2015). Canon Global. Retrieved 2015, from Canon Global - Annual Reports:
http://www.canon.com/ir/annual/index.html
Exchange Rates UK (2015). Exchange Rates UK. Retrieved 2015, from Exchange Rates UK - Currencies:
http://www.exchangerates.org.uk/USD-JPY-exchange-rate-history.html
FUJIFILM Holdings Corporation (2015). FUJIFILM Holdings. Retrieved 2015, from FUJIFILM Holdings –
Archives (Annual Reports):
http://www.fujifilmholdings.com/en/investors/ir_library/annual_reports/archives/index.html#h2-10
Hitachi, Ltd. (2015). Hitachi. Retrieved 2015, from Hitachi - Annual Report: http://www.hitachi.com/IR-
e/library/annual/index.html
International Monetary Fund (2014). International Monetary Fund. Retrieved 2015, from IMF - Data and Statistics:
http://www.imf.org/external/pubs/ft/weo/2014/02/weodata/weorept.aspx?sy=2014&ey=2014&scsm=1&ssd
=1&sort=country&ds=.&br=1&c=512%2C668%2C914%2C672%2C612%2C946%2C614%2C137%2C31
1%2C962%2C213%2C674%2C911%2C676%2C193%2C548%2C122%2C556%2C912%2C678%2C313
%2C181%2C
Japan Electronics and Information Technology Industries Association (2015). December 2014 Production by the
Japanese Electronics Industry. Retrieved 2015, from JEITA / Statistics:
http://www.jeita.or.jp/english/stat/electronic/2014/product/pro_12.html
Japan Electronics and Information Technology Industries Association (2015). 2015 Production Forecasts for the
Global Electronics and Information Technology Industries. Retrieved 2015, from
http://www.jeita.or.jp/english/topics/2014/1216/2015production_forecasts.pdf
KPMG International (2006). An international analysis of corporate tax rates. KPMG's Corporate Tax, 304-250.
Retrieved 2015, from https://www.lib.uwo.ca/files/business/KPMGCorporateTaxRateSurvey.pdf
KPMG (2015). KPMG Global. Retrieved 2014, from KPMG Global - Corporate tax rates table:
http://www.kpmg.com/global/en/services/tax/tax-tools-and-resources/pages/corporate-tax-rates-table.aspx
LG Electronics (2015). LG Global. Retrieved 2015, from LG Global - Financial Statements:
http://www.lg.com/global/investor-relations/reports/financial-statements
Macabacus LLC (2015). Subsidiary Accounting. Retrieved 2015, from Macabacus LLC – Subsidiary Accounting:
https://www.macabacus.com/accounting/subsidiary-accounting
Mill, J. S. (1848). Principles of Political Economy.
Ministry of Finance Japan (2015). Tax Reform. Retrieved 2015, from Ministry of Finance Japan:
http://www.mof.go.jp/english/tax_policy/tax_reform/index.html
Nikon Corporation (2015). Nikon Global. Retrieved 2015, from Nikon Global - NIKON REPORT/Annual Report:
http://www.nikon.com/about/ir/ir_library/ar/backnumber/index.htm
Ono, S. (2010). Wa ga Kuni M&A no Genjou to Kadai. Keizai no Prism, 20-21. Retrieved 2015, from
http://www.sangiin.go.jp/japanese/annai/chousa/keizai_prism/backnumber/h22pdf/20107602.pdf
Panasonic Corporation (2015). Panasonic. Retrieved 2015, from Panasonic - Annual Report:
http://www.panasonic.com/global/corporate/ir/annual.html
Pioneer Corporation (2015). Pioneer. Retrieved 2015, from Pioneer - Annual Reports:
http://pioneer.jp/en/ir/library/annual_reports/
PwC Japan (2010). Proposed 2011 Tax Reform. PwC Japan Tax Update 2010, Issue 56. Retrieved 2014, from
http://www.pwc.com/jp/en/taxnews/pdf/Proposed_2011_Tax_Reform_E.pdf
http://www.pwc.com/jp/en/taxnews/pdf/NEW3_Proposed_2011_Tax_Reform_E_57.pdf
PwC Japan (2011). 2011 Tax Reform. PwC Japan Tax Update 2011, Issue 66. Retrieved 2014, from
http://www.pwc.com/jp/en/taxnews/pdf/2011-tax-reform-june-30-e.pdf
PwC Japan (2011). PwC Industry Analysis - Consumer Electronics. Retrieved 2015, from
http://www.pwc.com/jp/en/japan-
knowledge/archive/assets/pdf/archive_research_analytics_japan_industry_consumer-electronics-en.pdf
Page 40
http://www.pwc.com/jp/en/taxnews/pdf/2011-tax-reform-corporate-tax-e.pdf
PwC Japan (2011). Recovering from the Tohoku Disaster – managing related tax issues. PwC Japan Tax Update
2011, Issue 63. Retrieved 2014, from
http://www.pwc.com/jp/en/taxnews/pdf/tax_issues_on_tohoku_disaster.pdf
PwC Japan (2011). Special Tax Measures for Tohoku Earthquake Restoration. PwC Japan Tax Update 2011, Issue
67. Retrieved 2014, from
http://www.pwc.com/jp/en/taxnews/pdf/special_tax_measures_for_tohoku_earthquake_restroation_e.pdf
PwC Japan (2011). Tax Legislation Update for 2011 & 2012 Tax Reform (Part 1). PwC Japan Tax Update 2011,
Issue 68. Retrieved 2014, from http://www.pwc.com/jp/en/taxnews/pdf/2011-2012-tax-reform-part1-e.pdf
PwC Japan (2011). Tax Legislation Update for 2011 & 2012 Tax Reform (Part 2). PwC Japan Tax Update 2011,
Issue 69. Retrieved 2014, from http://www.pwc.com/jp/en/taxnews/pdf/2012-tax-reform-part2-e.pdf
PwC Japan (2011). Tohoku Earthquake and special tax legislation. PwC Japan Tax Update 2011, Issue 64. Retrieved
2014, from http://www.pwc.com/jp/en/taxnews/pdf/Tohoku_Earthquake_special_tax_legislation_E.pdf
PwC Japan (2011). Tohoku Earthquake and tax implications. PwC Japan Tax Update 2011, Issue 62. Retrieved
2014, from http://www.pwc.com/jp/en/taxnews/pdf/Tohoku_Earthquake_tax_implications_E.pdf
PwC Japan (2012). Tax Legislation Update for 2012 Tax Reform (Part 3). PwC Japan Tax Update 2012, Issue 72.
Retrieved 2014, from http://www.pwc.com/jp/en/taxnews/pdf/2012-tax-reform-part3-e.pdf
PwC Japan (2012). Tax Legislation Update for 2012 Tax Reform (Part 4). PwC Japan Tax Update 2012, Issue 76.
Retrieved 2014, from http://www.pwc.com/jp/en/taxnews/pdf/2012-tax-reform-part4-e.pdf
PwC Japan (2012). Tohoku Earthquake Restoration Surtax implemented. PwC Japan Tax Update 2012, Issue 79.
Retrieved 2014, from http://www.pwc.com/jp/en/taxnews/pdf/tohoku-earthquake-restoration-surtax-e.pdf
PwC Japan (2014). Future corporate tax reform. PwC Japan Tax Update 2014, Issue 101. Retrieved 2014, from
http://www.pwc.com/jp/en/taxnews/pdf/jtu101-may-2014-e.pdf
PwC Japan (2014). Future corporate tax reform (part 2). PwC Japan Tax Update 2014, Issue 102. Retrieved 2014,
from http://www.pwc.com/jp/en/taxnews/pdf/jtu102-jul-2014-e.pdf
PwC Japan (2015). 2015 Tax Reform Proposal. PwC Japan Tax Update 2015, Issue 108. Retrieved 2015, from http:/
/www.pwc.com/jp/en/taxnews/pdf/jtu108-jan-2015_en.pdf
PwC Japan (2015). 2015 Tax Reform Act. PwC Japan Tax Update 2015, Issue 110. Retrieved 2015, from http://ww
w.pwc.com/jp/en/taxnews/pdf/jtu110-apr-2015-en.pdf
Ricoh Company, Ltd. (2015). Ricoh. Retrieved 2015, from Ricoh - IR Library:
https://www.ricoh.com/IR/financial_data/financial_result/
Samil PwC (2014). Tax News Flash Korean Tax Law Changes for 2015. Korean Tax Newsletter. Retrieved 2015,
from
http://www.samil.com/publication/filemng.nsf/0/7818D0CF1DE3415D49257DC4000A8EB8/$File/Tax_N
ewsflash_special_edition.pdf
Samsung Electronics (2015). Samsung Electronics. Retrieved 2015, from Samsung Electronics - Annual Reports:
http://www.samsung.com/us/aboutsamsung/investor_relations/financial_information/annual_reports.html
Sharp Corporation (2015). Sharp. Retrieved 2015, from Sharp - IR Library/Annual Report: http://www.sharp-
world.com/corporate/ir/library/annual/index.html
Sony Corporation (2015). Sony. Retrieved 2015, from Sony - Annual Report:
http://www.sony.net/SonyInfo/IR/financial/ar/Archive.html
SOZEISHIRYOKAN (2014). 2014 CORPORATION TAX ACT of Japan Including Chapter III of SPECIAL
TAXATION MEASURES ACT as of April 1, 2014. (2014). (Y. GOMI, & T. HONJO, Eds.) Tokyo, Japan:
Cetus & General Press Inc. Retrieved 2014, from
http://www.sozeishiryokan.or.jp/corporation_tax/z_pdf/corporation_tax2014.pdf
Toshiba Corporation (2015). Toshiba. Retrieved 2015, from Toshiba - Annual Reports:
http://www.toshiba.co.jp/about/ir/en/finance/ar/ar2014.htm
Vieira, G. (2012). Japan, Unplugged. Maclean's, pp. 45-45. Retrieved 2015, from
http://web.a.ebscohost.com/ehost/detail/detail?sid=e034750…107&bdata=JnNpdGU9ZWhvc3QtbGl2ZQ%
3d%3d#db=bth&AN=74602293
Weygandt, J.J., Kimmel, D.P., & Kieso, E.D. (2013). “Financial Accounting IFRS Edition.” John Wiley & Sons,
Inc.
Yamaha Corporation (2015). Yamaha. Retrieved 2015, from Yamaha -
IR Publications: http://www.yamaha.com/about_yamaha/ir/publications/past.html
Page 41
Appendix 1
12EMC's REPORTED REVENUES FOR 2014
in millions Japanese Yen (JPY)
Name of firm Revenues
1 Alps Electric Co. Ltd. (Alps) 684,362
2 Canon Inc. (Canon) 3,727,252
3 FUJIFILM Holdings Corporation (Fujiflim) 2,094,291
4 Hitachi Ltd. (Hitachi) 8,330,966
5 Nikon Corporation (Nikon) 980,556
6 Panasonic Corporation (Panasonic) 7,736,541
7 Pioneer Corporation (Pioneer) 498,051
8 Ricoh Company, Ltd. (Ricoh) 2,195,696
9 Sharp Corporation (Sharp) 2,927,186
10 Sony Corporation (Sony) 6,682,274
11 Toshiba Corporation (Toshiba) 6,502,543
12 Yamaha Corporation (Yamaha) 410,304
Total 42,770,022
Page 42
Appendix 2
DETAILS OF THE 2015 CORPORATE TAX REFORM OF JAPAN
PLOTTED IN THE CORPORATE TAX COMPUTATION FORMULA
Corporate tax item Ref Effect Old rule New Rule
Gross Income
Less: Allowable deductions NOL 1. Decrease 80% credit 2015 to 2016 - 65% credit
allowable 9 years carryover period 9 years carryover period
credit 2017 onwards - 50% credit
2. Increase 10 years carryover period
carryover
period
Exclusions/Exemptions DivEx Stricter Subsidiary ownership Subsidiary ownership

subsidiary 25% or more = Dividends 100% 1/3 or more = Dividends 100%
ownership excluded excluded
requirement
for higher less than 25% = Dividends 50% 5% to > 1/3 = 50% excluded
exclusion excluded
less than 5% = Dividends 20%
excluded
Equals Taxable Income
Multiply: Effective Corporate Tax Rate RECTR Decrease 35.67% 33.06% - 2015
32.26% - 2016 onwards
Equals Corporate Tax Obligation

Add: Surtax/es
Less: Tax incentives or credits PR&D Reduced Permanent R&D incentive Permanent R&D incentive
creditable Credit amount = 8% to 10% Credit amount = 8% to 10%
amount of R&D of R&D
Limitation of credit = 30% Limitation of credit = 25%
of Corporate Tax Obligation of Corporate Tax Obligation
Carryover of excess = one year Carryover of excess = no longer

applicable
TR&D Non- Temporary R&D incentive Temporary R&D incentive

extension Period: April 1, 2013 to March Period: Not extended
of 31, 2017
applicable
period Credit amount, whichever is Credit amount: No change
higher of:
a) 5% to 30% of incremental
R&D; or
b) R&D costs in excess of 10%
of average sales,
times the tax credit ratio*
Limitation of credit = 10% Limitation of credit = No change

of Corporate Tax Obligation
Equals Corporate Tax Obligation Payable

Source: PwC Japan and Ministry of Finance Japan, 2015
*Note: Ratio is a mechanical calculation which increases the credit depending upon the relationship between the amount of
R&D costs and average annual sales
Page 43
Appendix 3
DETAILS OF THE 2015 CORPORATE TAX REFORM OF SOUTH KOREA
PLOTTED IN THE CORPORATE TAX COMPUTATION FORMULA
Corporate tax item Ref Effect Old rule New Rule
Gross Income
Less: Allowable deductions
Exclusions/Exemptions
Equals Taxable Income
Multiply: Effective Corporate Tax Rate
Equals Corporate Tax Obligation

Add: Surtax/es ASRE Additional None Additional surtax to facilitate the
Tax use of Retained Earnings
Obligation
for Choose either of the two (2)
corporations methods with choice being
not meeting irrevocable for three (3)
the RE consecutive years
requirement
1. (taxable income (*Note 1) x
80% – Investments (*Note 2),
i.e. payroll increases and dividend
payments) x 10%; or
2. (taxable income x 30% –

Investments, i.e. payroll increases
and dividend payments) x 10%
Less: Tax incentives or credits R&DTC Reduced Credit amount = 3% of R&D Credit amount = 2% of R&D
creditable
amount
Equals Corporate Tax Obligation Payable

Source: Samil PwC, 2014
*Note 1: Taxable income shall be adjusted to: (1) include dividends received from subsidiaries, interest income received on
refunds of overpaid national taxes, and the amount of depreciation expenses incurred in facility investments made in the current
year; and (2) exclude corporate taxes (excluding this surtax), NOLs, statutory reserve transfers, disallowed donations, etc.
*Note 2: Investment will include tangible and intangible fixed assets for business use. However, overseas investment and
investment made to acquire equity will be excluded from the scope of investment for these purposes.
Page 44
Page 45
DISENTANGLING THE IFRS 1 DISCLOSURE:

OBSERVATIONS FROM THE EARLY ADOPTION OF
IFRS IN CANADA
Theresa DiPonio, Robert Morris University
ABSTRACT
The objective of this study is to investigate whether the method employed to analyze the
implementation of International Financial Reporting Standards (IFRS) is consequential as to our
ability to evaluate IFRS as a financial reporting vehicle. In this study, IFRS 1 reconciliations are
deconstructed to exhibit the financial magnitude of optional exemption choices permitted under
IFRS 1, standard-to-standard differences, and equity component switching. Findings from this
study demonstrate that optional exemption choices and equity component switching comprise the
larger part of the financial magnitude of IFRS adoption. Evidence from this study should prompt
standard setters, regulators, practitioners, investors, and researchers to carefully consider how
IFRS is being applied and the extent to which it is being adopted when assessing the standards
for any attainment of relevance, quality, and comparability.
INTRODUCTION
Studies of International Financial Reporting Standards (IFRS) adoption have

demonstrated an increase in accounting quality (Ball, Robin, Wu 2003; Capkun, Cazavan-Jeny,
Jeanjean, Weiss 2008; Gassen and Sellhorn 2006). High quality financial reporting has
demonstrated a reduction in information asymmetries for investors (Street and Bryant 2000;
Tarca 2004; Ashbaugh and Pincus 2001; Gordon, Jorgensen, and Linthicum 2010). The
International Accounting Standards Board (IASB) has designed standards intended to reduce
information asymmetries amongst external users of the financial statements, primarily investors
(Haller, Ernstberger, and Froschhammer 2009) and more broadly amongst countries (Barth,
Landsman, and Lang 2008). However, any assessment of IFRS as a reporting vehicle should
consider how the standards are applied (Kvaal and Nobes 2010; Schipper 2005) and the extent to
which IFRS is adopted.
The research objective of this study is to employ disaggregation methods to investigate
the implementation of IFRS. IFRS adoption is not a monolithic event. The implementation of
IFRS entails management choices permitted under in IFRS 1 First Time Adoption of
International Financial Reporting Standards, remeasurement of all accounts in accordance with
IFRS, and material reclassifications (equity component switching) of accounts within the
statement of financial position. The transition to IFRS is a significant disclosure-enhancing event
(Cormier 2013; Fifield, Finningham, Fox, Power, and Veneziani 2011; Karamanou and Nishiotis
2009) and as such any evaluation by standard setters, regulatory agencies, practitioners, and
researchers of the IFRS global framework should carefully consider management discretionary
choices as well as pronouncement differences upon transition to IFRS. Due to the complexity of
disentangling the IFRS 1 disclosure, the current body of literature is limited as to studies which
disaggregate the IFRS transition by IFRS 1 choices, standard-to-standard differences, and equity
component switching (material reclassifications).
Page 46
This study examines the transition to IFRS by voluntary adopters in Canada. Canada
provides an interesting platform for the study. Effective January 1, 2011, the Canadian
Accounting Standards Board (AcSB) required all Canadian Publicly Accountable Enterprises
(PAE) to adopt IFRS for financial reporting. With the 9th largest economy based on Gross
Domestic Product (GDP) (Economy Watch, 2012), Canada is a formidable economic force and
presents an opportunity to examine the implementation of IFRS in a large market-oriented
economy. Canada also provides an optimal setting to examine a country with long-term
convergence efforts as a precursor to the transition. Prior research has investigated country
contexts which are divergent from IFRS (Cormier, Demaria, Lapointe-Antunes, and Teller 2009;
Hung and Subramanyam 2007; Lantto and Sahlström 2009). Finally, the Canadian transition to
IFRS is of vital importance to the Financial Accounting Standards Board (FASB), the Securities
Exchange Commission (SEC), and U.S. constituents as Canadian GAAP was closely aligned
with U.S. GAAP.
Empirics are presented which disentangle the financial magnitude of implementation
choices, GAAP-to-GAAP differences, and equity component switching. Data hand-collected
from IFRS 1 reconciliations reveal business combinations, share-based payments, and
cumulative translation differences as the optional exemption choices most frequently exercised
by sample firms. The financial effect of the cumulative translation difference resulted in an
overall decrease to retained earnings of $13.4 million for firms in the sample. (All amounts are
reported in Canadian dollars.) In sample, firms experienced an average decrease to retained
earnings of $1.2 million per firm. The analysis of pronouncement differences revealed IFRS 2
Share-Based Payments and IAS 12 Income Taxes as the standards having the most frequent
effect on sample firms. The largest remeasurement effect on total stockholders’ equity was a
result of the application of IAS 16 Property, Plant, and Equipment, IAS 40 Investment Property,
and IAS 12 Income Taxes at an increase of $19.6 million, an increase of $4.9 million, and a
decrease of $6.7 million, respectively. For all sample firms, the change in total stockholders’
equity as reported totaled $16.4 million. However, disaggregation methods revealed that $13.6
million of the total adjustment to stockholders’ equity was attributed to equity component
switching which decreased retained earnings and bypassed the income statement.
In Henry’s 2009 study of SFAS 159 The Fair Value Option of Financial Assets and
Liabilities, firms avoided recognition of realized security losses on the income statement by
using the adoption of the provision to report the remeasurement to fair value as an adjustment to
the opening balance of retained earnings. Employing this finding analogously for the transition
to IFRS coupled with the evidence provided in the present study, these findings may alert
standard setters and regulators as to opportunistic equity component switching under the veil of
IFRS adoption.
As evidenced by the present study, optional exemption choices and equity component
switching comprise the larger part of the financial magnitude of IFRS adoption. Evidence
provided in this study demonstrates how optional exemption choices selected by first-time IFRS
adopters conceals the impact of IFRS which may compromise the comparability objective of the
IFRS Conceptual Framework (IASB 2010). Disaggregating the implementation of IFRS should
be of interest to standard setters and regulators as a critical technique to assess how IFRS is
being applied and which aspect of the implementation of IFRS – IFRS 1 optional exemption
choices, standard-to-standard differences, or equity component switching bears the greatest
financial impact of IFRS adoption. Understanding the true effects of IFRS are crucial as standard
setters, practitioners and firms weigh the benefits of IFRS against the costs of adoption.
Page 47
The remainder of this study is organized as follows. The next section discusses IFRS 1.
The following section describes the data collection and sample. The following section presents
analyses of equity components. The last section concludes the study.
IFRS 1
Authoritative Guidance
IFRS 1 First Time Adoption of International Financial Reporting Standards sets the
precedent for financial reporting under IFRS, overrides transitional provisions included in other
IFRS, and prescribes detailed disclosures. The IFRS 1 disclosure entails detailed reconciliations
and explanations of the transitory financial effects from Canadian GAAP (CA GAAP) to IFRS.
IFRS 1.39 requires the first IFRS financial statements to include a reconciliation of equity
reported under national GAAP to equity under IFRS at the date of transition to IFRS and at the end
of the latest period for comparative information presented in the first IFRS financial statements.
For this study, the reconciliation of equity is of particular interest. According to IFRS 1.40, the
reconciliations have to be sufficiently detailed in order to enable users to understand the material
adjustments to the balance sheet and income statement.
IFRS 1 requires entities to apply, retrospectively, all IFRS standards effective at the end
of their first IFRS reporting period. The standard requires the opening presentation of IFRS
statement of financial position and the comparative financial statements be prepared in
accordance with the recognition, measurement, presentation and disclosure requirements of these
standards. The Canadian Securities Administrators (CSA) requires the presentation of an opening
IFRS statement of financial position in the first IFRS interim financial report. In the opening
statement of financial position, a Canadian company must:
 Recognize all assets and liabilities required by IFRS

 Derecognize all assets and liabilities not permitted by IFRS
 Classify all assets, liabilities and components of equity in accordance with IFRS; and
 Measure all assets and liabilities in accordance with IFRS
All adjustments, when applicable, should be recognized through retained earnings, or

other equity items, at the transition date (CICA 2011).
IFRS 1 also establishes two categories of exceptions to the retrospective rule: mandatory
and optional exemptions. Mandatory exemptions prohibit retrospective application of IFRS due
to insufficient measurement reliability. Optional exemptions grant relief from IFRS requirements
in which the costs of compliance exceed the benefits to the users of the financial statements. The
optional exemptions represent choices of accounting policies under IFRS which may have
significant impact on an entity’s future financial results (Jermakowicz and Gornik-Tomaszewski
2006).
Optional Exemption Choices
As previously mentioned, firms adopting IFRS must comply with IFRS 1. IFRS 1 permits
the election of exemption choices in specific areas where the cost of complying with IFRS 1 may
exceed the benefit to financial reporting or where retrospective application is impractical. For
example, at the transition date to IFRS, IFRS 1 permits firms to elect to maintain assets at
historical cost, a previous GAAP valuation, or remeasure assets to fair value. If a firm exercises
Page 48
the option to remeasure a property, plant, or equipment asset to fair value, the fair value would
surrogate for the historical or depreciated cost of the asset as the deemed cost at the transition
date. These exemption choices represent compromises of the IFRS measurement system upon
adoption. Any compromises of the IFRS system upon adoption should be of concern to both
regulators and investors (Capkun, Cazavan-Jeny, Jeanjean, and Weiss 2011). In a 2007 report on
the European Union implementation of IFRS, the Institute of Chartered Accountants in England
and Wales (ICAEW) noted comparability was impeded among and between first-time adopters.
The report also stated that these implementation differences will have an effect on future periods
of financial reporting (ICAEW, 2007).
Table 1 presents the optional exemption choices by entity count extracted from the
disclosures. The optional exemption choices which were most frequently exercised by sample
firms were business combinations, share-based payments, and cumulative translation differences.
On average, firms exercised 2.72 optional exemptions. The choices selected as well as the
number of choices exercised bring into question the extent to which a firm adopts IFRS.
Table 1
OPTIONAL EXEMPTION CHOICES
Optional Exemptions Firm Count
Business combinations 25
Share-based payment transactions 25
Fair value or revaluation as deemed cost 7
Deemed cost of oil and gas assets 2
Leases 1
Employee Benefits 5
Cumulative translation differences 19
Investment in subsidiaries, jointly controlled entities, and associates 1
Compound financial instruments 1
Designation of previously recognized financial instruments 1
Decommissioning liabilities 5
Service concession arrangements 1
Borrowing Costs 10
Total Number of Optional Exemption Choices made by Sample 103
Firms
DATA COLLECTION AND SAMPLE
IFRS was mandated effective January 1, 2011. However, early adoption was permitted
subject to approval of the CSA. Quarterly financial statements, management discussion and
analysis reports, and annual financial statements were obtained from company websites,
SEDAR, EDGAR, and the TMX website. The audit opinion letter, accounting policy disclosure,
and required IFRS 1 disclosure were reviewed for explicit language regarding early adoption.
Reconciliation data for this study was hand-collected.
The sample consists of 39 Canadian PAEs deemed “pure” early adopters from a
population of 69 PAEs which sought early adoption of IFRS. “Pure” early adopters are defined
as those companies which met the following criteria:
 Audit opinion letter stated presentation, “in accordance with International Financial Reporting Standards.”
Page 49
 Financial statement note on “Basis of presentation” cited compliance and conversion to International
Financial Reporting Standards as issued by International Accounting Standards Board as well as the
entity’s transition date.
 Financial statement note disclosure on adoption of International Financial Reporting Standards contained
are conciliation from Canadian GAAP to IFRS of the statement of financial position at the transition date.
Although PAEs opting for early adoption were required to seek CSA permission, there
were no additional reporting requirements for early adopter firms. The process and reporting
requirements, for example adherence to IFRS 1, were the same for early adopter and compulsory
complaint firms.
The firms presented in this study were regulated by five provincial regulators: Alberta,
British Columbia, Ontario, Quebec, and Saskatchewan. Sample firms were overwhelmingly
represented by the mining industry which is consistent with prior literature on Canadian early
adopters (Blanchette, Racicot, and Girard 2011). The industry classifications represented in the
sample were: Mining (n=28), Utilities (n=2), Manufacturing (n=5), Information (n=1), Real
Estate, Rental, and Leasing (n=2), and Professional, Scientific, and Technical Services (n=1).
ANALYSES OF EQUITY COMPONENTS
The majority of research has examined IFRS by comparing national GAAPs to IFRS
through various analyses primarily examining earnings through comparability indices (Fifield et
al. 2011; Haller et al. 2009) and key financial ratios (Lantto and Sahlström 2009, Blanchette et
al. 2011). A growing body of literature has also tested the value relevance of accounting
information delivered by the IFRS reporting system (Horton and Serafeim 2010; Christensen,
Lee, and Walker 2009, Schadewitz and Vieru 2007; Gjerde, Knivsflå, and Sættem 2008).
However, the true market valuation of the earnings and book value reconciliations are observable
only in the year of transition when financial statements are prepared both under local GAAP and
IFRS. These differences are reported in aggregate in the change in total stockholders’ equity
(Hung and Subramanyam 2007). Without a complete deconstruction of the how IFRS is being
applied, evidence from the present study brings into question our ability to evaluate earnings as
reported under the IFRS system or interpret results from tests of market reaction.
Decomposition of Equity Adjustments
Table 2 presents the decomposition of reported adjustments to equity by IFRS standard.

The standard affecting the most firms in the sample (n=20) was IFRS 2 Share-Based Payments.
This standard had a magnitude effect on the adjustment to contributed capital of $41 million and
a corresponding effect of -$41 million to the adjustment to retained earnings. The primary
difference between IFRS 2 and CA GAAP 3870 was the way in which these options are
measured. CA GAAP required the entire award of options to be measured at the intrinsic value
whereas IFRS subjects these options to separate value tranches measured at fair value using an
option price model. However, the true effect of the standard-to-standard difference is altered by
an optional exemption permitted by IFRS 1 which allows the first-time adopter to not apply
IFRS 2 to equity instruments granted after November 7, 2002 that vested before transition to
IFRS. If a company applies the exemption, a reclassification adjustment is recorded which
transfers the amount out of contributed capital into retained earnings. As demonstrated by this
example, these equity components are not only affected by pronouncement differences, but
choices made at adoption and switching among equity components.
Page 50
IAS 12 Income Taxes, IAS 16 Property, Plant, and Equipment, IAS 40 Investment
Property, and IFRS 1 Cumulative Translation Differences represent the standards with the
largest magnitude effect on the retained earnings adjustment at -$6,898 billion, $19,580 billion,
$4,930 billion, and -$13,428 billion, respectively. The negative tax effect is consistent with
evidence from a study by Fifield et al. conducted in 2011 which examined IFRS reconciliations
in the context of the U.K., Italy, and Ireland. IFRS 1 Cumulative Translation Differences
represents the standard with the largest magnitude effect on the adjustment to accumulated other
comprehensive income at $13,426 billion.
Table 2
DECOMPOSITION OF REPORTED ADJUSTMENT TO EQUITY COMPONENTS BY
STANDARDS
Page 51
The Effect of Equity Component Switching
Table 3 eliminates the effect of equity reclassifications and presents the decomposition of
the adjustments to equity components without component switching. In comparing Tables 2 and
3, continuing with the example of IFRS 2 Share-based Payments, the number of firms affected
by the standard decreased from 20 to 2. After eliminating the switching effect, the magnitude
effect on the adjustment to contributed capital decreased from $41 million to -$2 million and the
cumulative adjustment to retained earnings increased from -41.464 million to 2.074 million.
Removing the equity component switching effect divulges the true standard-to-standard financial
effect.
IAS 12 Income Taxes, IAS 16 Property, Plant, and Equipment, and IAS 40 Investment
Property continued to represent the standards demonstrating the largest magnitude effect on the
retained earnings adjustment at -$6,741 billion, $19,580 billion, and $4,930 billion, respectively.
Analysis of the adjustment to accumulated other comprehensive income after eliminating the
switching effect revealed a decrease in the magnitude adjustment from $13,532 billion to -$7
million. This observation should put regulators, standard setters, practitioners, and researchers on
notice that the way in which we analyze and measure equity components could be consequential
to our ability to evaluate a GAAP change.
Table 3
DECOMPOSITION OF ADJUSTMENT TO EQUITY COMPONENTS BY STANDARDS
WITHOUT SWITCHING
Page 52
Summarizing the Switching Effect
Table 4 summarizes the switching effect. The largest switching effect was within the
adjustment to retained earnings at $13,582 billion. Of this amount, $13,428 billion (Table 2)
related to cumulative translation differences, an optional exemption choice exercised under IFRS
1. This exemption permits firms to zero out balances of cumulative translation differences for all
foreign operations at the transition date. Under Canadian GAAP, these differences were recorded
in accumulated other comprehensive income as unrealized gains and losses. Upon transitioning
to IFRS, the majority of the firms (n=16), in sample, elected to reclassify aggregated unrealized
gains and losses to retained earnings, an earned capital account. More specifically, for the firms
represented in the population of early adopters, $13,428 billion (Table 2) of unrealized
translation differences bypassed the income statement and were reclassified to retained earnings.
Table 4
VARIATIONS OF ADJUSTMENT TO EQUITY COMPONENTS AND SWITCHING EFFECT
Summary of Standard Effects on all Components of Equity as Reported
Table 5 presents a summary of the magnitude adjustments to the components of equity as

reported disaggregated by the IFRS standard effect. Table 5 combines the financial magnitude
adjustments by standard from Table 2 and reconciles these adjustments to the financial
magnitude on total stockholders’ equity. IAS 16 Property, Plant, and Equipment demonstrated
the largest percentage of change in total stockholders’ equity at 119 percent of the overall change
in stockholders’ equity. IAS 16 permits the revaluation of property, plant, and equipment to fair
value which explains the positive adjustment of remeasurement of $19,580 billion to total
stockholders’ equity.
As discussed in section titled, Summarizing the Switching Effect, the IFRS 1 election for
cumulative translation differences increases accumulated other comprehensive income by
$13,426 billion and decreases retained earnings by $13,428 billion. However, upon examination
of the effect of this standard on the adjustment to total stockholders’ equity, the remeasurement
effect is relatively minimal at a decrease of $1.7 million resulting in an overall decrease to total
stockholders’ equity of .01056 percent. More specifically, the real effect of transition is obscured
at the stockholder’s equity level of analysis.
CONCLUSION
The preceding deconstruction of the implementation of IFRS exhibits the complexity of

adopting IFRS. The decomposition of equity components provides an opportunity to observe
adoption effects concealed at aggregated stockholders’ equity. Magnitude adjustments
disaggregated by pronouncement differences permits a more comprehensive understanding of the
Page 53
particular standards which are associated with the adjustments to equity components. Further,
only when equity components are decomposed can reclassifications (component switching)
among the components be observed. Component switching reveals the implementation effects of
IFRS on equity. As demonstrated in the study, underlying the switching effect (implementation
choices) are the true GAAP-to-GAAP differences which can only be observed upon
disaggregation.
An examination of the exercised optional exemptions brings into question the extent to
which IFRS is adopted. The extent to which IFRS is adopted is crucial to the larger assessment
of whether the IFRS reporting system reduces information asymmetry and increases accounting
quality. Consideration of the modifications and limitations of the implementation of IFRS has
great bearing on our ability to measure any improvement IFRS may contribute to financial
reporting.
By examining the IFRS transition in a comparable country context to the United States,
this study provides preliminary evidence which compels attention from standard setters and
regulators as to how IFRS is being applied. If the magnitude of the IFRS transition is primarily
comprised of management choices and equity component switching as the evidence from this
study suggests, standard setters and regulators may want to investigate this trend further as they
weigh the costs of adopting IFRS versus the benefits the IFRS reporting system. At a minimum,
findings from this study should prompt standard setters, regulators, practitioners, investors, and
researchers to carefully consider how IFRS is being applied and the extent to which it is being
adopted when assessing the standards for any attainment of relevance, quality, and
comparability.
Table 5
SUMMARY OF STANDARD EFFECT ON ADJUSTMENT TO EQUITY COMPONENTS
Page 54
REFERENCES
Accounting Standards Board (AcSB). (2005). Accounting Standards in Canada: Building on New Directions.
Ashbaugh, H., and M. Pincus. (2001). Domestic accounting standards, international accounting standards, and the
predictability of earnings. Journal of Accounting Research 39(3), 417–34.
Ball, R., A. Robin, and J. Wu. (2003). Incentives versus standards: Properties of accounting income in four East
Asian countries and implications for the acceptance of IAS. Journal of Accounting and Economics 36(1–3),
235–70.
Barth, M., W. Landsman, and M. Lang. (2008). International accounting standards and accounting quality. Journal
of Accounting Research 46(3), 467–98.
Blanchette, M., F.-É. Racicot, and J.-Y Girard. (2011). The Effects of IFRS on Financial Ratios: Early Evidence in
Canada. Certified General Accountants Association of Canada.
Capkun, V., A. Cazavan-Jeny, T. Jeanjean, and L. A. Weiss. (2008). Earnings Management and Value Relevance
During the Mandatory Transition from Local GAAPs to IFRS in Europe. Working paper, available at
SSRN eLibrary. doi: 10.2139/ssrn.1125716.
Capkun, V., Cazavan-Jeny, T. Jeanjean, and L. A. Weiss. (2011). Setting the Bar: Earnings Management During a
Change in Accounting Standards. Working paper, available at SSRN eLibrary. doi: 10.2139/ssrn.1870007.
Canadian Institute of Chartered Accountants (CICA). (2011). Financial Reporting in Canada under IFRS.
Christensen, H. B., E. Lee, and M. Walker. (2009). Do IFRS Reconciliations Convey Information? The Effect of
Debt Contracting. Journal of Accounting Research, 47(5), 1167-1199.
Cormier, D. (2013). Relevance of Financial Reporting and Disclosure for Financial Markets under IFRS: Some
Canadian Evidence. Cahier de recherche, 04.
Cormier, D., S. Demaria, P. Lapointe-Antunes, and R. Teller. (2009). First-Time Adoption of IFRS, Managerial
Incentives, and Value-Relevance: Some French Evidence. Journal of International Accounting Research,
8(2), 1-22.
Economy Watch (2012) accessed at http://www.economywatch.com/worldeconomy/canada
Fifield, S., G. Finningham, A. Fox, D. Power, and M. Veneziani. (2011). A cross-country analysis of IFRS
reconciliation statements. Journal of Applied Accounting Research, 12(1), 26-42.
Gassen, J., and T. Sellhorn. (2006). Applying IFRS in Germany-determinants and consequences.
Betriebswirtschaftliche Forschung und Praxis, 4, 365.
Gjerde, Ø., K. Knivsflå, and F. Sættem. (2008). The value-relevance of adopting IFRS: Evidence from 145 NGAAP
restatements. Journal of International Accounting, Auditing and Taxation, 17(2), 92-112.
Gordon, E., B. Jorgensen, and C. Linthicum. (2010). Could IFRS replace US GAAP? A comparison of earnings
attributes and informativeness in the US market. UTSA College of Business, Working paper, No. 0028.
Haller, A., J. Ernstberger, and M. Froschhammer. (2009). Implications of the mandatory transition from national
GAAP to IFRS — Empirical evidence from Germany. Advances in Accounting 25(2), 226-236.
Henry, E. (2009). Early Adoption of SFAS No. 159: Lessons from Games (Almost) Played. Accounting Horizons,
23(2), 181-199.
Horton, J., and G. Serafeim. (2010). Market reaction to and valuation of IFRS reconciliation adjustments: First
evidence from the UK. Review of Accounting Studies 15(4), 725–51.
Hung, M., and K. R. Subramanyam. (2007). Financial statement effects of adopting international accounting
standards: the case of Germany. Review of Accounting Studies 12(4), 623-657.
Institute of Chartered Accountants in England and Wales (ICAEW). (2007). EU Implementation of the IFRS and
the fair value directive, report to the European Commission.
International Accounting Standards Board (IASB). (2010). Conceptual Framework for the Preparation and
Presentation of Financial Statements.
International Accounting Standards Board (IASB). (2003). First-Time Adoption of IFRS. IFRS 1. London, U.K.
Jermakowicz, E. K. and S. Gornik-Tomaszewski. (2006). Implementing IFRS from the perspective of EU publicly
traded companies. Journal of International Accounting, Auditing and Taxation, 15(2), 170-196.
Karamanou, I., and G. P. Nishiotis, (2009). ‘Disclosure and the cost of capital: evidence from the market's reaction
to firm voluntary adoption of IAS’, Journal of Business Finance & Accounting, 36(7‐8),793-821.
Kvaal, E., and C. W. Nobes. (2010). International differences in IFRS policy choice. Accounting and Business
Research,40(2), 173–87.
Lantto, A.-M., and P. Sahlström. (2009). Impact of International Financial Reporting Standard adoption on key
financial ratios. Accounting & Finance, 49(2), 341-361.
Page 55
Schadewitz, H. and M. Vieru. (2007). How markets valuate and response to IFRS reconciliations adjustments in
Finland. Available at SSRN 1005601.
Schipper, K. (2005). The introduction of International Accounting Standards in Europe: Implications for
international convergence. European Accounting Review, 14(1), 101–6.
Street, D.L. and S.M. Bryant. (2000). Disclosure level and compliance with IASs: a comparison of companies with
and without US listings and filings. The International Journal of Accounting, 35(3), 305–329.
Tarca, A. (2004). International Convergence of Accounting Practices: Choosing between IAS and US GAAP.
Journal of International Financial Management & Accounting, 15(1), 60-91.
Page 56
Page 57
LIQUIDITY IN LEBANESE COMMERCIAL BANKS AND

ITS DETERMINANTS
Rim El Khoury, Notre Dame University
ABSTRACT
The last global financial crisis reemphasized the importance of liquidity for the well-
functioning of the banking sector and the financial markets. Given the importance of banks in
the Lebanese financial markets, this paper aims to identify the determinants of liquidity of
Lebanese commercial banks for the period from 2005 to 2013.
Results show that bank liquidity is positively related to bank size and interbank rate, and
negatively related to loan growth rate, inflation and the financial crisis. The impact of capital,
economic growth, unemployment, short term interest rate, and lending interest rate on liquidity is
not conclusive. No difference was found between listed and unlisted banks.
INTRODUCTION
Banks, due to their role as financial intermediaries, are exposed to many types of risks such
as credit risk, liquidity risk, capital risk, and interest rate risk. Liquidity risk is defined as banks’
ability to fund their assets and meet their obligations, without incurring unacceptable level of
losses (BCBS, 2008). By transforming short term deposits into medium to long term loans,
banks are exposed to liquidity risk. Therefore, banks must be ready to meet the retirements of
deposits by holding some liquid assets. Although liquid assets reduce liquidity risk, they have an
opportunity costs since they generate low or no return.
Before the crisis, banks did not consider liquidity risk a priority as compared to other
types of risks such as default, capital, and interest rate risk. However, the financial crisis had
changed the whole picture by highlighting the importance of liquidity management, with the
failure, resolution, or forced merger of some banks (Teply, 2011). In response to the crisis, the
Basel committee on Banking Supervision (BCBS) - whose aim is to enhance the financial
stability and to improve the quality of banking supervision methods- issued, in December 2010,
new guidelines for managing liquidity risk and Basel III introduced new requirements which
force banks to hold higher level of capital and liquid assets.
In Lebanon, given the absence of a secondary market, banks dominate the financial
sector. Thus, the aim of this paper is to identify the determinants of liquidity of Lebanese
commercial banks, which is important to the well-being of banks’ operations, the economy, and the
country as a whole. This topic has returned to be a hot topic since the financial crisis.
The outline of this paper is as follows: Section 2 provides a literature review, while
section 3 presents the data, defines the variables, and specifies the econometrical model. Empirical
determinants of bank's liquidity are presented and discussed in Section 4. Finally, section 5
concludes.
Page 58
LITERATURE REVIEW
Overview of Banking Sector in Lebanon
Regulated by Banque du Liban (BDL), the banking sector continues to be the backbone
of the Lebanese economy, being ranked 12th worldwide in 2011 (World Economic Forum, 2011-
2012) and 29th in 2013 (World Economic Forum, 2013-2014) in terms of soundness. Total assets of
this sector grew by 8.5% in 2013 to represent 379% of the size of GDP at the end of the year,
while deposits increased by 9% to represent 312.2% of GDP. These two ratios are among the
highest in the world, highlighting the growth and the importance of this sector. Furthermore, this
sector is the major provider of capital to business, where loans accounted for 109% of GDP in
2013 (IDAL, 2014). It is also the supporter of the government’s debt through the purchase of
government Treasury bills. As of 2013, the total number of commercial banks operating in
Lebanon reached 56 (Financial Access Survey [FAS], 2014).
The Lebanese banks were able to remain shielded from the global financial crisis of
2008. One of the factors that enabled them to remain resilient and to operate normally is the
liquidity. Subject to the high reserve requirements set by BDL, the Lebanese banking sector was
able to enjoy high liquidity by all standards, well above regional and international benchmark.
Measured by net primary liquidity divided by total deposits, this ratio reported a high level of
30.9% in 2010, slightly decreasing to 29.1% in 2011, before reaching 31.6% and 30.7% in 2012
and 2013 respectively. This high liquidity ratio is mainly due to higher liquidity in foreign
currency as compared to domestic currency. For example, in 2013, this ratio reached 20% in
domestic currency versus 35.3% in foreign currency.
Liquidity can be also measured by a mirror image, which is loan to deposit ratio, whereby a
high ratio suggests low liquidity. This ratio reached 37.7% in 2013, lower than the regional
average of 70.2%, the emerging market average of 77.1%, and the global average of 83.1%
(Bank Audi, 2014). All these ratios are highlighting the strong liquidity position that the Lebanese
banking sector is enjoying.
Theories of Bank Liquidity
Liquidity is defined as the ability of the bank to meet its obligations and to finance any
increase in assets, without incurring unacceptable losses (BCBS, 2008). Banks’ primary function
is to collect deposits and lend them. Therefore, banks are transforming short term deposits into
long term and illiquid assets, exposing themselves to liquidity risk. If a large part of depositors
demanded their money, the bank might be forced to liquidate its illiquid assets at unfavorable price
or borrow at unfavorable costs. The result of a liquidity shortage is a loss of value, which might
lead to a solvency crisis (Aspachs, Nier, & Tiesset, 2005) or even to default (Ozdincer &
Ozyildirim, 2008). In fact, the absence of liquidity might lead to a liquidity risk. Liquidity risk
can be of two types: funding liquidity risk and market liquidity risk. First, funding liquidity risk is
defined as the bank’s risk of not being able to meet its future cash flows without affecting its
operations. Second, market liquidity risk is defined as the risk of not being able to offset or
eliminate a position because the market is not deep or disrupted. These two types of risks are
interacted in the crisis period (Drehman & Nikolaou, 2013); for example, the exposure to funding
liquidity risk might lead to asset sales or a decrease in the price of assets, leading to market
liquidity risk. Similarly, the exposure to market liquidity risk might lead to higher margin,
leading to funding liquidity risk. Thus, these two risks work together.
Page 59
To avoid liquidity risk, Aspachs et al. (2005) suggested three mechanisms that banks can
use. First, banks can hold a high amount of liquid assets, such as cash, balances with central
banks, short term securities, and reverse repo. Liquid assets serve as a buffer against liquidity
risk by reducing the probability that a deposit withdrawal will threaten the viability of the bank.
Second, banks can use their liability side of the balance sheet by borrowing from other banks in
case of liquidity demand, thus relying on the interbank market. Third, banks can use again their
liability side of the balance sheet by relying on the central bank as a lender of last resort.
Vento and La Ganga (2009) emphasized that liquidity risk is not an isolated risk. It is the
result and the cause of other risks within the banking sector. For example, credit risk might lead to
liquidity risk and liquidity risk might lead to legal risk.
Empirical Overview of Determinants of bank Liquidity
Although liquidity was an old topic, this theme regains its importance following the
financial crisis of 2007. Many researchers and international organizations tested the determinants
of liquidity which are listed in chronological order starting from 2005.
Aspachs et al. (2005) investigated the determinants of liquidity of 57 UK-resident banks,
using quarterly data from 1985 to 2003. They found that interest margin and loan growth rate
negatively affect bank liquidity, while profitability and bank size do not have a significant
impact. They also found that liquidity is negatively related to real GDP growth and the policy rate.
Fielding and Shortland (2005) examined the determinants of excess liquidity in the
Egyptian banking sector. They found that the level of economic output, discount rate, and the
violent political incidence have a positive effect on liquidity, and cash to deposit ratio and
economic reform have a negative effect.
Valla, Saes-Escorbiac, and Tiesset (2006) investigated the determinants of liquidity of
English banks using both bank specific and macroeconomic variables. Concerning the bank
specific variables, they found that liquidity is negatively related to the probability of obtaining
support from the central bank, the interest rate margin measuring the opportunity costs of holding
liquid assets, loan growth, and bank profitability. No clear relationship was found between bank
size and liquidity. Concerning the macroeconomic variables, they found that liquidity is negatively
related to the business cycle measured by GDP and the monetary policy effect measured by short
term interest rate.
Lucchetta (2007) investigated the importance of interest rate on bank's risk taking and the
decision to hold liquid assets in European countries and found a positive relationship between
interbank rate and liquidity and a negative relationship between monetary interest rate and
liquidity. Furthermore, he found a negative impact of loans divided by total assets, and loan loss
provision divided by net interest revenue, and a positive impact of bank size.
Bunda and Desquilbet (2008) analyzed the liquidity of 1107 commercial banks in 36
emerging countries between 1995 and 2000, using bank specific variables, market and
macroeconomic variables, and exchange regimes. They found that liquidity is negatively related to
(i) bank size as measured by total assets, (ii) lending interest rate, (iii) and the presence of
financial crisis. On the other side, liquidity is positively related to (i) capital adequacy ratio as
measured by equity divided by total assets, (ii) presence of regulation obliging banks to be
liquid, (iii) public expenditures divided by GDP, (iv) inflation rate, and (v) Exchange rate
regime. Banks in extreme regimes (floating or hard pegs) are more liquid than countries in
intermediate regimes.
Page 60
Rauch, Steffen, Hackethal, and Tyrrel (2010) investigated the determinants of liquidity of
German's 457 state owned saving banks over 1997 to 2006 and found that liquidity is negatively
related to monetary policy interest rate, level of unemployment, bank size measured by number
of customers, and bank profitability. At the same time, saving quotas and liquidity in the
previous period positively affect liquidity.
In the same year, Moore (2010) analyzed the liquidity of commercial banks in Latin
America and Caribbean countries, and found that liquidity positively depends on current
macroeconomic situation, and negatively depends on cash to deposit ratio and money market
interest rate.
Vodova (2011) investigated the factors affecting the liquidity of 22 banks operating in
Czech Republic using a panel data from 2001 to 2009. By considering four firm specific
variables and eight macroeconomic variables, he found that capital adequacy, lending interest
rate, interbank interest rate, and non-performing loans positively affect bank liquidity, while
inflation rate, GDP growth, and financial crisis negatively affect bank liquidity. However, bank
size has an ambiguous impact, and unemployment, interest margin, bank profitability, and repo rate
have no significant impact on banks’ liquidity.
The positive impact of capital adequacy on liquidity has been confirmed by Bonfim and
Kim (2012). They used a regression analysis on a panel data covering European and North
American banks in 2002-2009 and found that the impact of bank specific variables such as size,
performance, and loan deposit ratio depends on the type of liquidity measure used. However,
they found that the bank size has a positive impact on liquidity.
In the same year, Fadare (2011) aimed to identify the determinants of liquidity in Nigerian
banks from 1980 to 2009 by using a linear least square method. They found that monetary policy
rate and lagged loan to deposit rate significantly predict banking liquidity.
All these empirical evidences suggest that commercial bank liquidity is determined by
bank specific factors (capital adequacy ratio, profitability, size, etc…), macroeconomic factors
(interest rates, economic cycle, etc…) and other factors such as regulations, financial crisis, and
political accidents.
DATA
This study is based on the annual observations of 23 commercial banks covering a 9 year
period between 2005 and 2013. Bank specific data were obtained from BANKSCOPE, which
includes the financial statistics of all banks; the data was complemented with the annual reports
of Lebanese banks. As for macroeconomic data, they were obtained from many sources such as
International Financial Statistics of International Monetary Fund (IMF) and Banque du Liban
website (central bank of Lebanon). Although there exists 56 commercial banks in Lebanon in
the year 2013 (FAS, 2014), the banks included in this study are only those having observations
for at least 6 years on BANKSCOPE. Due to some missing information, the obtained data is
unbalanced panel data.
VARIABLES
Dependent Variable
Liquidity can be achieved by (1) holding a portfolio of assets that can be easily converted
into cash without a significant loss of value (cash, reserves, or government securities), (2)
Page 61
holding significant volume of stable core deposits (nonvolatile deposits), and (3) maintaining credit
line with financial institutions implying the ability to borrow anytime needed.
Because there is no consensus on the best way to measure liquidity, this study will use four
different liquidity ratios as the dependent variable, similar to many studies (Moore, 2010; Vodova,
2011).
1. Liquid assets/Total assets (L1): where liquid assets include cash, deposits with central banks and other
banks, short term government securities and reverse repos. The higher the ratio, the higher the
liquidity, and the higher the capacity of banks to absorb liquidity shocks. However, a higher ratio can
be interpreted as a measure of inefficiency, since liquid assets have lower yield. Therefore, it is
important to have a good balance between liquidity and profitability.
2. Liquid assets/ customer deposits (L2): this ratio captures the bank's ability to meet its obligations in
terms of funding and bank's sensitivity to deposit withdrawals. The denominator here is replaced by
only deposits ignoring the fact that banks can borrow from other banks in case of liquidity needed.
Although some studies used in the denominator deposits of households, deposits of banks and other
financial institutions, and short term debt securities, this ratio will not be used in this study due to
many missing information from Bankscope. Similar to L1, a higher ratio signals a better capacity to
absorb liquidity risk, and a lower sensitivity to deposits withdrawals.
3. Loans/ Assets (L3): This ratio measures the percentage of banks' assets tied up in illiquid loans.
Contrary to the above two measures, a higher ratio suggests a lower bank's liquidity.
4. Loans/ deposits and short term financing (L4): This ratio relates illiquid assets with liquid liability with
similar interpretation as L3, where a higher ratio suggests lower liquidity.
Although these ratios are not able to always capture all liquidity, they are widely used
because they are easy to calculate and interpret.
Independent Variables
The selection of variables was based on the cited empirical studies, limited by data
availability. The independent variables will be divided into two broad categories: (1) bank specific
determinants and (2) macroeconomic determinants. The bank specific factors include capital
adequacy ratio (+), bank size (?), loan growth (-), and nonperforming loans (-). The macroeconomic
factors include growth of real gross domestic product (-), inflation rate (+), liquidity premium (-),
short term interest rate (+), interbank rate (+), real interest rate on lending (-), and unemployment
(+). In addition, two dummy variables will be included, to represent the financial crisis period
and to differentiate between listed and unlisted banks. Other variables such as political events,
exchange rate regime and economic reform are excluded since these variables made no sense in
Lebanese conditions.
This study considers four bank specific factors as follows:
1. Capital Adequacy Ratio (CAP) (+): The impact of capital adequacy on liquidity and liquidity creation
is debatable, especially in emerging countries. Liquidity creation is defined as transforming less liquid
assets into more liquid liabilities. The more liquidity is created, the greater is the possibility and
magnitude of losses associated with meeting the liquidity demands of customers by disposing illiquid
assets. On one side, and under the hypothesis of ‘risk absorption’, higher capital ratio will improve
banks’ ability to absorb risks associated with liquidity creation, thus increasing the bank’s ability to
create more liquidity (Repullo, 2004). On the other side, under the ‘financial fragility-crowding out’
hypotheses, higher bank capital may hinder creation because it makes the bank’s capital structure less
fragile or because it crowds out deposits. Capital adequacy ratio is measured as Equity Capital/ Total
Assets (Bonfim & Kim, 2012) with a positive effect on liquidity ratios, since solvent banks are found
to be more liquid.
Page 62
2. Bank size (SIZE) (+/-): The impact of size on bank liquidity is not clear. On one side, according to the
‘too big to fail’ hypothesis, large banks tend to be less liquid. If large banks see themselves as too big to
fail, they will be less motivated to hold liquid assets. By benefiting from an implicit guarantee
(assistance of Lender of less resort), large banks tend to invest more in riskier assets and hold less
liquid assets (Lucchetta, 2007). On the other side, small banks are more likely to be involved in
traditional intermediation activities and hold small liquid assets (Rauch et al. 2010; Bunda &
Desquilbet, 2008). Bank size is measured as the natural logarithm of total assets following many
studies (Bonfim & Kim, 2012; Vodova, 2011; Horvath, Seidler, and Weill, 2014).
3. Loan growth (GROWTH) (-): Loans are considered as the principal activity of most commercial banks
as they generate the most important source of revenue. However, they are illiquid. Therefore, an
increase in the demand for loans will lead to less liquid assets, resulting in a negative relationship
between loan growth and banks’ liquidity (Pilbeam, 2005). Loan Growth is measured as the annual
growth rate of gross loans. Banks which specialize in lending activity tend to have higher exposure to
liquidity risk, thus a lower liquidity ratio (Bonfim & Kim, 2012; Valla et al. 2006).
4. Nonperforming loan Ratio (NPL) (-): Non performing loans are loans that are not up to date in terms of
payment of interest and principal. Thus, they measure the quality of bank assets. The presence of large
proportion of non-performing loans might lead to liquidity problem since depositors and foreign
investors might lose their confidence in the bank (Bloem & Gorter, 2001). Therefore, non-performing
loans as a proportion of gross loans have an expected negative impact on bank liquidity.
This study will not include any measure of profitability such as net interest margin, return
on equity or return on assets as independent variables since they are considered to be more or
less codetermined with asset liquidity. In fact, many studies found that liquidity has an impact on
bank profitability.
In addition to the bank specific variables, 7 macroeconomic variables are included in this
study.
1. Real GDP growth rate (GROWTH) (-): Economic cycle affects banks’ activities; demand for loans is
higher during expansion and lower during downturns. Therefore, in expansion, the number of
profitable investments is higher, which induces banks to lend more, resulting in less liquid assets
(Valla et al. 2006). This variable was used by many studies (Aspachs et al. 2005; Valla et al. 2006;
Vodova, 2011). It is measured as the percentage change in Gross Domestic Product (GDP) using
constant prices and is taken from the International Monetary Fund, World Economic Outlook
Database.
2. Inflation (INF) (+): An increase in inflation will reduce the real rate of return, creating market frictions
and credit rationing. The result is fewer loans, reduction in intermediary activity and a higher amount
of liquid assets held by banks. Therefore, a positive relationship is expected between inflation rate and
liquidity. Furthermore, since loans made by banks are long term loans, their nominal values are sticky
and highly affected by inflation. Therefore, a higher inflation will motivate banks to hold liquid asset
to reduce their vulnerabilities to inflation. This variable was used by Vodova (2011) and Bunda and
Desquilbet (2008). Inflation is defined as the percent change in the index using end of period consumer
prices and is taken from the International Monetary Fund, World Economic Outlook Database.
3. Liquidity premium (LP) (-) = Defined as the difference between interest rate on loans and interest rate
on deposit, LP is expected to have a negative impact on bank liquidity. Higher interest rate margin
will motivate banks to lend more and to hold less liquid assets (Aspach et al. 2005; Valla et al. 2006).
This variable is obtained from International Financial Statistics (IFS).
4. Short term interest rate (INT) (+): Short term interest rate is the rate paid on money market securities. A
higher short term interest rate will motivate banks to invest more in these short term instruments, which
will improve their liquidity positions. Therefore, a positive relationship between short term interest
rate and liquidity is expected (Pilbeam, 2005). Given that Treasury bills are considered as the most
liquid and safest assets, short term interest rate is measured as the interest rate on 3 month T-bill and is
obtained from BDL statistics. This variable is especially important in the case of Lebanon since
Lebanese banks are the main supporters of the government’s debt through the purchase of government
Treasury bills.
Page 63
5. Interbank Interest rate (IRB) (+): This variable represents the illiquidity cost since banks lacking
liquidity can borrow in the interbank market to meet their cash needs. It was used by Lucchetta (2007)
who argued that the higher this rate is, the more expensive is the cost of illiquidity and the more liquid
the banks are. This variable is obtained from BDL statistics.
6. Real Interest rate on loans (RL) (-): It is calculated as the lending interest rate adjusted for inflation as
measured by the GDP deflator and is obtained from World Development Indicators (WDI), which is
the primary World Bank database. A negative relationship is expected since the higher the lending
interest rate is, the more profitable the loans are, which will push banks to lend more and to maintain
less liquid assets. However, with the presence of asymmetric information, Stiglitz and Weiss (1981)
found that adverse selection will lead to credit rationing, so that banks’ liquidity might increase with
the presence of high interest rates.
7. Unemployment rate (+): This variable is obtained from World Development Indicators (WDI) and is
included because an increase in unemployment rate will reduce the demand for loans, enabling banks
to be more liquid.
Furthermore, a number of other macroeconomic variables were evaluated to be included in

the model, such as discount rate, deposit interest rate and lending interest rate. However, due to
significant correlation among the variables, they were dropped from the model in order to avoid
muticollinearity problem.
Table 1
EXPLANATORY VARIABLES AND EXPECTED SIGNS
EXPLANATORY EXPECTED
NOTATION DEFINITION AND SOURCE
VARIABLES SIGNS
Dependent variables
Liquid assets/Total Assets (L1) Liquid assets/Total assets (BANKSCOPE) NA
Liquid assets/ Deposits (L2) Liquid assets/ Total Deposits (BANKSCOPE) NA
Loans/ Assets (L3) Net Loans/ Total assets (BANKSCOPE) NA
Loans/Dep and Short term (L4) Net loans/ Deposits and Short term funding NA
Funding (BANKSCOPE)
Bank-specific variables
Capital Ratio (CAP) Equity/Assets (BANKSCOPE) Positive
Size (SIZE) Ln of Total Assets (BANKSCOPE) Positive/Negative
Loan Growth (GLOAN) Change in Gross loans (BANKSCOPE) Negative
Non-Performing Loans ratio (PL) Non-Performing Loans/ Gross Loans Negative
(BANKSCOPE)
Macroeconomic variables
Economic Growth (GROWTH) Real GDP Growth Rate (WEO) Negative
Inflation (INF) Percentage change in inflation using end of Positive
period prices (WEO)
Liquidity Premium (LP) Lending Interest Rate – Deposit Interest Rate Negative
(IFS)
Short term Interest Rate (INT) Interest rate on 3 month T-bills (BDL) Positive
Interbank Rate (IRB) Interbank (BDL) Negative
Real Interest on Loans (RL) Lending interest – % change in GDP deflator Negative
(WDI)
Unemployment (UMP) Unemployment Rate (WDI) Positive
Dummy variables
Dummy 1 (D1) 1 during the crisis (2008-2010), 0 otherwise Negative
Dummy 2 (D2) 1 if listed bank, 0 otherwise Positive
In addition to the above mentioned variables, two dummy variables are included:
Page 64
8. Dummy variable 1 (D1): This variable is used to distinguish the period before the financial crisis from
the period after. It is equal to 1 during the financial crisis (2008-2012), and 0 otherwise (2005-2007
period and 2013 period). The negative impact of the financial crisis on the liquidity ratio had been
highlighted in many studies (Bunda & Desquilbet, 2008; Vodova, 2011).
9. Dummy Variable 2 (D2): This variable is used to distinguish listed banks from unlisted banks. It is
equal to 1 if banks are listed, 0 otherwise. Nguyen, Skully, and Perera (2012) found that listed banks
usually hold more liquid assets than non-listed banks.
MODEL
In order to identify the determinants of the liquidity of Lebanese commercial banks, a

panel data regression analysis is used, which takes the following equation:
Y = α + βit Xit + ε it (1)
Where Y is the dependent variable measuring liquidity for bank i in time t, Xit is a vector of
explanatory variables for bank i in time t, α is a constant, β is slope of the variable and εi is the error
term.
Since we are dealing with a panel data, some tests using STATA software will be performed
in order to choose the suitable model for our data.
Descriptive Statistics
Table 2 presents the descriptive statistics for the dependent and independent variables
involved in the regression, including mean, standard deviation, minimum and maximum. The
results show that most variables comprise 170 observations except growth in loans, NPL (due to
missing reported figure from Bankscope), and unemployment (due to unavailable data). Variables
containing loans such as growth in loans and net loans divided by total assets present larger standard
deviation with 22.44112 and 10.19225 respectively as compared with other variables. It revealed that
the quantity of loans has more significant variance than other variables.
Table 2
VARIABLES OBS MEAN STD DEV MIN MAX
L1 170 .2899883 .1285224 .07225 .8395045
L2 170 .3541414 .1631921 .0905254 1.353085
L3 170 28.09981 10.19225 8.622 63.51
L4 170 32.09844 11.56946 11.318 68.69
CAP 170 8.592488 2.962659 3.494 35.773
SIZE 170 8.258256 1.328531 3.818459 10.90703
GLOANS 158 19.29766 22.44112 -49.68 122.05
NPL 149 11.88389 12.16183 .34 74.79
GROWTH 170 5.537059 3.847166 1 10.3
INF 170 5.354876 1.849306 .517 7.212
LP 170 2.00126 .342959 1.476667 2.490833
INT 170 4.67225 .5069485 3.926667 5.22
IRB 170 3.227941 .5532571 2.75 4
RL 170 4.660176 3.462081 -.8462127 11.8559
UMP 155 22.56516 .5413539 21 22.8
Page 65
Unit Root Test
Before proceeding with the regression, some tests are needed. First, the stationary of the
data will be tested using a Fisher test- a unit root test for unbalanced panel data as suggested by
Maddala and Wu (1999).
Results reported in Table 3 show that the null hypothesis of non-stationary (or presence
of unit root) is rejected for all variables except short term interest rate, interbank rate, and
liquidity premium. Since the dependent variable is stationary, we are less likely to get spurious
results even if some variables are not stationary.
Table 3
FISHER TEST FOR A PANEL UNIT ROOT TEST USING ADF
VARIABLES Chi2(46) P-Value
L1 138.0735 0.0000
L2 151.8448 0.0000
L3 237.2095 0.0000
L4 254.2093 0.0000
CAP 221.8591 0.0000
SIZE 109.1474 0.0000
GLOANS 82.2865 0.0008
NPL 146.7197 0.0000
GROWTH 87.8911 0.0002
INF 256.6205 0.0000
LP 0.6695 1.0000
INT 13.8646 1.0000
IRB 12.0708 1.0000
RL 108.2474 0.0000
UMP 971.2679 0.0000
To confirm this, stationary of errors will be tested using Fisher test. For example, using L1
as the dependent variable, results show a p-value of 0.0003, rejecting the null hypothesis of non-
stationary. Thus, we can conclude that there are no unit roots in the panel under the given test
conditions (included panel mean and time trend for bank variables and time trend for
macroeconomic variables).
Choice of Regression
The estimation used should take into consideration the special features of the panel
data. In static relationship, the literature applies pooled OLS, fixed effect, or random effect
model. To choose between the fixed effect (FE) and the random effect (RE), Hausman test (1978)
for the exogeneity of the unobserved error component is used. Rejecting the null hypothesis
suggests that RE is inconsistent and FE model is better. Results in Table 4 show that
accepting or rejecting the null hypothesis depends on the dependent variable. Thus, the
analysis is divided into 2 parts: L1 and L2 as the dependent variables on one side, and L3 and
L4 as the dependent variables on the other side.
First, using L1 and L2 as the dependent variables, the null hypothesis is rejected
(Prob>Chi2 is less than 0.05), concluding that the fixed effect is more efficient than random
Page 66
effects. The results suggest that each bank has its own individual characteristics that may have
an influence on the liquidity. Next, to choose between fixed effect and pooled OLS, the
Restricted F test reports a p-value of 0.0000, suggesting that fixed effect is better than pooled OLS.
Therefore, the choice of FE indicates the importance to control for all time-invariant difference
between banks. Given that fixed effect model will be used for L1 and L2, the next step is to use a
joint test to see if time fixed effects are needed. The null hypothesis is that all time fixed effects
coefficients are equal to zero. The Prob>F reported in Table 4 is lower than 0.05, rejecting the null
hypothesis and suggesting the need to include time fixed effect in our model.
Second, using L3 and L4 as the dependent variables, the null hypothesis cannot be rejected
(Prob>Chi2 is more than 0.05), indicating that the random effect is more efficient estimator. Then,
to choose between random effect and pooled OLS, Breush and Pagan LM test (1980) is used, with a
null hypothesis that variance across entities is zero. Prob>Chibar2 reported in the last row in
Table 4 is lower than 0.05, concluding that a simple pooled OLS regression cannot be used.
Therefore, a random effect model is run when L3 and L4 are used as dependent variables.
Since some studies suggest the presence of a dynamic model, where bank liquidity position
might persist over time, a dynamic model will be run by including the lagged dependent variable
among the independent variables.
Table 4
HAUSMAN TEST, F-TEST, TESTPARM AND LM TEST
TESTS L1 L2 L3 L4
Hausman Test: Ho: difference in coefficients not systematic
Chi2(8) 29.23 29.37 0.54 1.10
Prob 0.0003 0.0003 0.9998 0.9975
F-test that all u_i=0
F(21,95) 11.01 9.50 27.01 26.83
Prob>F 0.0000 0.0000 0.0000 0.0000
Testparm
F(3, 95) 8.40 10.59
Prob>F 0.0004 0.0001
Breusch and Pagan Lagrangian multiplier test for random effect
Chibar2(01) 117.96 116.35
Prob>chibar2 0.0000 0.0000
Multi-Collinearity
To detect multicollinearity, a correlation matrix for all independent variables is conducted.

Multicollinearity is considered a serious problem if the correlation coefficients are more than 0.8
(Cooper & Schindler, 2010). Results in Table 5 suggest that muticollinearity is not a problem
given that all coefficients are less than the cut-off point set by Cooper and Schindler (2010).
Page 67
Serial Correlation
Serial correlation, or correlations between errors, should be tested since it might cause
smaller standard errors and higher R-squared. Since the Durbin Watson test can be used only in
time series, Lagram-Multiplier test derived by Wooldridge (2002) is applied to test
autocorrelation in panel-data. Given the null of no serial correlation, the results in Table 6
show that the data has no first order autocorrelation when L1 and L2 are used as the dependent
variables since the probability is higher than 0.05. However, when L3 and L4 are used as the
dependent variables, results indicate the existence of first order autocorrelation (prob<0.05).
Since serial correlation is considered to be a problem in macro panels with long time
series, and since our data is made of only few years, we can conclude that autocorrelation is
not a problem. However, for more reliable results, regressions run will be adjusted for
autocorrelation, especially in the case of L3 and L4.
Table 6
WOOLDRIDGE TEST FOR AUTOCORRELATION IN PABEL DATA
WOOLRIDGE TEST L1 L2 L3 L4
F(1,21) 4.007 2.313 33.259 38.552
Prob>F 0.0584 0.1432 0.0000 0.0000
Heteroscedasticity
To detect the presence of heteroscedasticity, Wald Test as suggested by Baltagi (2008) is

employed. The null hypothesis is homoscedasticity, or constant variance, meaning that the variance of
the error is the same for all individuals. Table 7 reports a P-value lower than 0.05, rejecting the null
hypothesis and concluding the presence of heteroskedasdicity regardless of the dependent variables.
Therefore, the regression coefficients will be adjusted for heteroskedasdicity.
More specifically, when L1 and L2 are used as the dependent variables, the fixed effect
regression used will be adjusted for heterosckedasticy by using robust standard errors known as
Huber/White estimators and will be also adjusted for autocorrelation by including robust and clustered
standard errors (White, 1980; Rogers, 1993). As for L3 and L4, a fixed effect model is not the best
estimator. With the presence of heterosckedasticy and autocorrelation, the generalized least square
(GLS) can be used since standard errors estimates can be robust to disturbance as heteroscedastic,
contemporaneously cross-sectionally correlated, and autocorrelated of type AR(1). However, GLS is
feasible only if the number of entities (N) is less than number of periods (T). Given that our data is
made of 22 banks and only 9 years, the latter is not appropriate. Furthermore, Beck and Katz (1995)
argued that GLS tends to produce unacceptably small standard errors estimates. Instead, they advocate
the use of Prais-Wisnten regression, with panel corrected standard errors. Therefore, Prais-Winsten
regression will be used, with panel-specific AR (1) to adjust for autocorrelation and heteroskedastic
panels corrected standard errors.
Table 7
MODIFIED WALD TEST FOR GROUP-WISE HETEROSCEDASTICITY
TEST L1 L2 L3 L4
Chi2(7) 926.52 1622.55 11966.64 9563.15
Prob 0.0000 0.0000 0.0000 0.0000
Page 68
EMPIRICAL FINDINGS
Presentation of Findings
The empirical evidence on the determinants of banks’ liquidity is based on unbalanced

panel data. Although two models are presented for each dependent variable, the analysis will be
based on the best model.
Using L1 and L2 as the dependent variables, the first model reports the fixed effect
technique with robust and clustered standard errors to control for heterosckedasticity and
autocorrelation respectively and with the inclusion of time dummies (as shown by Testparm
test). The second model (Model 2) reports the dynamic model with robust standard errors where the
lagged dependent variable will be included within the independent variables to see whether banks’
liquidity exhibits certain degree of persistence over time. Table 8 suggests a high coefficient of
determination (within R2 of 0.4972 and 0.5376) and the estimated model fits the panel data as
shown by the high F test in case of model 1 and the high Wald test in case of model 2.
To test whether residuals are serially correlated at specified range (1-1; 1-2) and at
specific lag 2, Table 9 reports Cumby and Huizinga test (1992) where a p-value higher than 0.1
indicates the presence of serial correlation.
Results show that the residuals from Model 1 are not serially correlated when L1 and L2
are used as the dependent variable. Since the residuals of Model 2 appear to be serially
correlated, the analysis will be based on Model 1. Some variables were omitted due to collinearity
and they are represented by a coefficient of 0.
By comparing the results for L1 and L2, the regression using L2 as the dependent
variable has a higher explanatory power. The findings revealed that these two liquidity ratios are
affected by similar factors given a very close number and signs of significant variables. Thus,
the discussion of results based on L1 and L2 will be combined together.
Determinants of liquidity measured by L3 and L4 are shown in Table 10. These two
ratios have reverse interpretations than L1 and L2, since a higher value means lower liquidity.
Thus, a positive impact on liquidity is represented by a negative sign of the coefficient. Similarly,
two models are run with L3 and L4 as dependent variables. The first model reports the Prais-
Winsten regression with AR (1) and heteroskedastic panels corrected standard errors (Model 1) ,
while the Model 2 reports the dynamic model with robust standard errors.
Comparing the results for L3 and L4 suggests that these two liquidity ratios are affected
by similar factors. Therefore, the discussion of results for both L3 and L4 will be combined
together. Table 11 reports Cumby and Huizinga test (1992) when L3 and L4 are used as the
dependent variables. Although that the regressions were adjusted for autocorrelation, results
show that the residuals are serially correlated for all ranges and for all lags. Due to the presence
of serial correlation, results obtained using L1 and L2, are considered to be more reliable than those
obtained using L3 and L4.
Page 69
Table 8
DETERMINANTS OF BANK’S LIQUIDITY (L1/L2 AS DEPENDENT VARIABLES)
L1 as the dependent variable L2 as the dependent variable
Model 1 (Fixed Model 2 (Dynamic Model 1 (Fixed Effect, Model 2 (Dynamic
Effect, Robust and Model, Robust with Robust and clustered Model, Robust with time
Clustered with Time Time Dummies) with time dummies) dummies)
Dummies)
Coef. P>t Coef. P>t Coef. P>z Coef. P>z
CAP .0133667 0.016** .004366 0.421 .0161149 0.013** .0096844 0.194
NPL -.0012276 0.130 -.0008112 0.244 -.0014834 0.146 -.000691 0.367
SIZE .1786031 0.060* .1430619 0.045** .2602682 0.019** .2475066 0.018**
GLOANS 0.005**
-.0008658 -.0009595 0.006*** -.0010824 0.004*** -.001306 0.002***
*
GROWTH .0062644 0.145 -.0240094 0.000*** .0096887 0.070* -.0312075 0.000***
INF 0.003**
-.0237275 -.0792122 0.000*** -.030631 0.002*** -.1104813 0.000***
*
IRB 0.002**
.1148101 .1402029 0.004*** .1606745 0.000*** .2194733 0.003***
*
INT 0 0 0 0
LP 0 0 0 0
RL .0093243 0.012** .0234917 0.000*** .012162 0.006*** .0335087 0.000***
UMP 0 0 0 0
Dummy 1 -.0459299 0.072* .0462082 0.168 -.0615855 0.046** .0661668 0.119
Dummy 2 0 0 0 0
Lagged L .4117442 0.029** .3742853 0.067*
Year 1 0 0 0 0
Year 2 .1145303 0.019** 0 .1633077 0.005*** 0
Year 3 0 0 0 0
Year 4 0 0 0 0
Year 5 0 0 0 0
Year 6 0 0 0 0
Year 7 -.0301311 0.036** -.1029898 0.000*** -.0319471 0.102 -.1315789 0.000***
Year 8 0 0 0 0
Year 9 0 0 0 0
Cons -1.584576 0.073* -.9783167 0.132 -2.358101 0.022** -1.913112 0.047**
Obs 128 91 128 91
Within R2 0.4972 0.5376
0.000** 15.90 0.000***
F (12,21) 12.09
*
Wald Chi 93.82 0.000*** 81.90 0.0000
***, **, and * = significant at the 1% level, 5% level, and 10% level respectively
Variables with a coefficient of 0 were dropped because of collinearity.
Table 9
CUMBY-HUIZINGA TEST FOR AUTOCORRLEATION (L1/L2 AS DEPENDENT VARIABLES)
Model 1 (Fixed Effect, Model 2 (Dynamic Model 1 (Fixed Effect, Model 2 (Dynamic
Robust and Clustered Model, Robust with Robust and clustered Model, Robust with time
with Time Dummies) Time Dummies) with time dummies) dummies)
Range
Chi2 P-value Chi2 P-value Coef. P-value Coef. P-value
Specified
1-1 1.754 0.1854 16.091 0.0001 0.522 0.4699 16.868 0.0000
1-2 3.195 0.2024 18.097 0.0001 1.640 0.4405 19.117 0.0001
Lag
Specified
2 1.672 0.1959 11.379 0.0007 1.234 0.2667 12.566 0.0004
Page 70
Table 10
DETERMINANTS OF BANK’S LIQUIDITY (L3/L4 AS DEPENDENT VARIABLES)
Model (PW with Model 2 (Dynamic Model 1 (PW with Panel Model 2 (Dynamic
Panel AR1 and Model, Robust) AR1 and heteroskedastic) Model, Robust)
heteroskedastic)
Coef. P>z Coef. P>z Coef. P>z Coef. P>z
CAP -.2170271 0.456 -.11613 0.714 .0339292 0.917 .170694 0.638
NPL -.0665023 0.213 -.0064641 0.869 -.0848806 0.196 -.0084807 0.840
SIZE -2.008349 0.035** -7.863646 0.039** -1.452805 0.121 -9.265285 0.028**
GLOANS .040313 0.008** .1073009 0.000*** .0397814 0.030** .1119529 0.000***
GROWTH -.1462954 0.205 .0268978 0.906 -.0966553 0.460 .0763645 0.756
INF -.3486168 0.121 1.89368 0.002*** -.4296815 0.085 2.303425 0.001***
IRB -.6945789 0.511 -3.354049 0.023** -.192025 0.875 -3.519557 0.038**
INT 3.091329 0.045** -1.591749 0.259 3.178667 0.071* -2.336809 0.108
LP -12.47713 0.000*** 0 -14.26727 0.000*** 0
RL .436649 0.008*** -.5189146 0.035** .494083 0.008*** -.6344441 0.017**
UMP 2.6735 0.000*** 0 2.616408 0.000*** 0
Dummy 1 2.836048 0.007*** -1.379885 0. 338 3.181476 0.008*** -1.757885 0.244
Dummy 2 -2.314856 0.343 0 -.1552198 0.953 0
Lagged L .6971631 0.000*** .7154974 0.000***
Cons 0 82.99239 0.033** 0 94.77832 0.028
Obs 128 91 128 91
R2 0.8023 0.8585
Wald Chi 975.70 0.0000 168.01 0.0000 1923.43 0.0000 215.64 0.0000
***, **, and * = significant at the 1% level, 5% level, and 10% level respectively
Variables with a coefficient of 0 were dropped because of collinearity.
Table 11
CUMBY-HUIZINGA TEST FOR AUTOCORRLEATION (L3/L4 AS DEPENDENT VARIABLES)
Model (PW with Model 2 (Dynamic Model 1 (PW with Panel Model 2 (Dynamic
Panel AR1 and Model, Robust) AR1 and heteroskedastic) Model, Robust)
heteroskedastic)
Range
Chi2 P-value Chi2 P-value Coef. P-value Coef. P-value
Specified
1-1 14.786 0.0001 15.953 0.0001 12.858 0.0003 16.244 0.0001
1-2 14.820 0.0006 17.083 0.0002 12.863 0.0016 17.293 0.0002
Lag
Specified
2 12.883 0.0003 13.196 0.0001 11.566 0.0007 13.516 0.0002
Determinants of Bank liquidity as measured by L1 and L2
Starting with the bank specific variables, first, capital positively affects bank’s liquidity,
consistent with the assumption that banks with sufficient capital should be liquid, in line with
previous studies (Vodova, 2011). Higher capital ratio might act as a positive signal to the
external public, which will attract more deposits, enabling banks to be more liquid. Second, the
positive and statistically significant impact of bank size on liquidityis consistent with the
Page 71
assumption that small banks focus more on traditional activities such as transforming deposits
into loans. By focusing on loans, small banks tend to hold little investment securities, leading to
low cash and reserves balance, consistent with Bonfim and Kim (2012) and Lucchetta (2007).
Although non-performing loan ratio has the expected sign, the coefficient is not significant. As
for the last bank specific variable (loan growth), it has a negative and significant impact, in line
with our expectation and consistent with Aspachs et al. (2005) and Valla et al. (2006). The
higher the amount of loans provided, the more is the amount of illiquid assets, and the lower is
the liquidity. This significant impact reveals that the dependency of Lebanese commercial banks on
loans.
Moving to the macroeconomic variables, the economic growth is not significant. Second,
the negative significant coefficient of inflation rate might suggest that inflation lowers bank
liquidity because it deteriorates overall macroeconomic conditions, consistent with Vodova (2011).
Third, the interbank interest rate is positive and significant, suggesting that a higher rate encourage
banks to maintain their money in the interbank market as part of liquid assets. Furthermore, a
higher rate increases the illiquidity costs if banks need to borrow in the interbank market, pushing
banks to be more liquid, consistent with Vodova (2011) and Lucchetta (2007). Fourth, although we
expect the short term interest rate to have a positive and significant sign given the dominance of
Lebanese commercial banks participation in the Treasury bills market (banks are the major
financer of Lebanese government), this variable was dropped from the model because of
collinearity. Fifth, the positive effect of real interest rate on lending is surprising. While RL is
significant, LP’s significance could not be tested due to collinearity. Although it is expected
that higher rates on lending encourage banks to lend more and to hold less liquid assets, the
positive relationship is consistent with Bunda and Desquilbet (2008) and Vodova (2011), which can
be explained with the presence of credit crunch and credit rationing.
Moving to the dummy variables, dummy 1 is negative, indicating that the financial crisis
had a negative impact on bank liquidity.
More specifically, result shows that liquidity increases in year 2006 and decreases in year
2011. Dummy 2 is insignificant indicating that listed banks do not differ from unlisted banks.
Determinants of Bank liquidity as measured by L3 and L4
Though the residuals are serially correlated, it is still important to analyze the results
obtained in Table 10. The model included four bank-specific variables, with only bank size and
growth of loans are significant, while capital position and quality of loans are non-significant. Size
is significant with a negative impact on L3 and L4, consistent with the previous finding that
smaller banks tend to focus more on lending activities, which lead to lower liquidity. On the
other hand, larger banks tend to focus more on investment activities, which lead to higher liquidity.
The loan growth has a positive and significant sign, coherent with the fact that loans are illiquid
and the higher they grow, the less is the bank liquidity.
As for the macroeconomic variables, only four variables are significant. First, the rate on T-
bills has a positive and significant impact on bank loans, illustrating the role of T-bill rate as a
benchmark rate. A higher rate leads to an increase in cost of borrowings and lending rates which
motivates banks to lend more and to reduce the liquid assets.
Page 72
Liquidity premium has a negative and significant coefficient. Although unexpected, it

supports Bunda and Desquilbet (2008) findings that higher spread will not encourage banks to lend
more, due to the existence of credit rationing, similar to the findings obtained using L1 and L2 as
the dependent variables. Contrary, the real interest rate on lending has a positive and significant
sign, supporting the positive impact of T-bill rates. Finally, unemployment has a positive and
significant coefficient in Model 2, suggesting that bank liquidity decreases with unemployment.
As for the dummy variables, results in Table 10 are consistent with the results obtained in
Table 8. The financial crisis negatively affects the liquidity of banks, regardless of the measure
used.
CONCLUSION
In conclusion, the paper aims to identity the determinants of the liquidity of the Lebanese
banks between 2004 and 2013, by using a panel data regression, and four different measures of
liquidity. Using four bank specific variables, seven macroeconomic variables and two dummy
variables, only few variables proved to be significantly important in determining banks’ liquidity.
First, the results of this paper show that banks’ characteristics explain part of their liquidity.
Higher liquidity tends to be associated with bank size and lower growth of loan. These
variables are significant regardless of the dependent variables. Second, the paper finds that the
impact of macroeconomic indicators on bank liquidity depends on the dependent variables used.
Bank liquidity decreases with inflation and increases with real interest rate and interbank rate
when L1 and L2 are used as dependent variables. Moreover, bank liquidity decreases with short
term interest rate, real interest rate, and unemployment, and increases with liquidity premium when
L3 and L4 are used as dependent variables. The impact of real interest rate and liquidity premium
on liquidity supports the presence of credit rationing in the Lebanese banking sector, while the
impact of interbank rate proves the dependency of the Lebanese banks on the interbank market in
case of liquidity shortage. The impact of short term interest rate demonstrates the use of T-bill
as a benchmark rate.
Third, the paper supports the persistence of liquidity in the banking sector given the
significance of the lagged liquidity. Fourth, the paper finds that the financial crisis has a
negative impact on bank liquidity as shown by the significance of Dummy 1. Finally, no
significant difference exists between listed and unlisted banks as shown by the insignificance of
Dummy 2.
These findings are important in many aspects. The study concludes that bank specific
fundamentals must be monitored; since more liquid assets are required as the bank size increases.
The central bank regulations also greatly affect the liquidity of Lebanese commercial banks
(such as the interbank rate). Moreover, monetary policy needs be monitored due to the undesirable
effects of inflation on liquidity. Besides, the negative impact of financial crisis on banks liquidity
suggests the need for Lebanese banks to carefully forecast the liquidity requirements as
anticipation for future events. Lastly, the unstable political environment needs to be solved to
improve liquidity. The results suggest that some of the variables affecting Lebanese banks
liquidity may be controlled by the government. The paper is a just a stepping stone and future
researches are needed to focus on qualitative factors such as political instability, the currency
circulation and salary and wages levels as probable determinants.
Page 73
ENDNOTES
1 The GLS random model with robust and clustered standard errors adjusted for heterskedasticity and
autocorrelation was run. The explanatory power was 0.3220 for L3 and 0.3007 for L4 supporting Beck and
Katz (1995)’s recommendation to use Prais Winsten regression instead of GLS regression.
REFERENCES
Aspachs, O., E. Nier & M. Tiesset (2005). Liquidity, banking regulation and the macroeconomy: Evidence on bank
liquidity holdings from a panel of UK-resident banks. Mimeo, 1-26.
Baltagi, B. H. (2008). Econometric analysis of panel data. Chichester, UK: John Wiley & Sons.
Bank Audi (2014, September 11). Lebanon banking Sector report. Retrieved from
http://www.bankaudigroup.com/GroupWebsite/openAudiFile.aspx?id=2348
Bandue Du Liban (BDL). (2014). Interest Rates: Commercial banks -LBP: interbank rate: End of period rate.
Retrieved from Banque Du Liban website.
Bandue Du Liban (BDL). (2014). Interest Rates: Treasury Bills - 3 months. Retrieved from Banque Du Liban website.
Basel Committee on Banking Supervision (BCBS). (2008, September). Principles for sound liquidity risk
management and supervision. Basel: Bank for International Settlements. Retrieved from
www.bis.org/publ/bcbs144.pdf
Beck, N. L. & J.N. Katz (1995). What to do (and not to do) with time-series cross-section data. American Political
Science Review, 89 (3), 634–647.
Bloem, M.A. & N.C. Gorter (2001, December). Treatment of non-performing loans in macroeconomic statistics
(Report No. 209). Retrieved from https://www.imf.org/external/pubs/ft/wp/2001/wp01209.pdf
Bonfim, D. & M. Kim (2012). Liquidity risk in banking: Is there herding? (Working paper No 201218).
Lisboa: Banco de Portugal Economics and Research Department. Retrieved from https://www.bportugal.pt/pt-
PT/BdP%20Publicaes%20de%20Investigao/wp201218.pdf
Breusch, T. S. & A.R. Pagan (1979). A Simple test for heteroscedasticity and random coefficient variation.
Econometrica, 47 (5), 1287–1294.
Bunda I. & J. Desquilbet (2008). The bank liquidity smile across exchange rate regimes. International Economic
Journal, 22(3), 361-386.
Cumby, R.E. & J. Huizinga (1992). Testing the autocorrelation structure of disturbances in Ordinary Least Squares and
instrumental variables regressions. Econometrica, 60(1), 185-195.
Cooper, D. & P. Schindler (2010). Business research methods (Eleventh Edition). New York: McGraw Hill /Irwin.
Drehman, M. & K. Nikolaou (2013). Funding liquidity risk: definitions and measurement. Journal of
Banking and Finance, 37 (3), 2173-2182.
Fadare, S. O. (2011). Banking sector liquidity and financial crisis in Nigeria. International Journal of Economics
and Finance, 3(5), 3-11.
Fielding, D. & A. Shortland (2005) Political Violence and Excess Liquidity in Egypt. Journal of Development
Studies, 41(4), 542-557.
International Monetary Fund (IMF). (2014). Lebanon. Retrieved from the Financial Access Survey (FAS) database.
International Monetary Fund (IMF). (2014). Lebanon Economic Indicators. Retrieved from the International
Financial Statistics (IFS) database.
International Monetary Fund (IMF). (2014). Lebanon. Retrieved from the World Economic Outlook (WEO) database
Hausman, J. A. (1978). Specification tests in econometrics. Econometrica, 46 (6), 1251–1271.
Horváth, R., J. Seidler, & L. Weill (2014). Bank capital and liquidity creation Granger-Causality evidence. Journal of
Financial Services Research, 45(3), 341-361.
IDAL. (2014). Financial system overview. Retrieved from
http://www.investinlebanon.gov.lb/en/doing_business/financing_your_business/financial_system_overview.
Lucchetta, M. (2007). What do data say about monetary policy, bank liquidity and bank risk taking? Economic
Notes, 6 (2), 189-203.
Maddala, G.S. & S. Wu (1999). A Comparative study of unit root tests with panel data and a new simple test.
Oxford Bulletin of Economics and Statistics, 61, 631-652.
Page 74
Moore, W. (2010). How do financial crises affect commercial bank liquidity? Evidence from Latin America and the
Caribbean (MPRA Paper No. 21473). Munich, Germany: University Library of Munich. Retrieved from
https://mpra.ub.uni-muenchen.de/21473/1/MPRA_paper_21473.pdf
Nguyen M., M. T. Skully, & Perera S. (December, 2012). Bank market power and liquidity, evidence from 113
developed and developing countries. Presented to the 25th Australasian Finance and Banking Conference,
Sydney, Australia.
Ozdincer, B. & C. Ozyildirim (2008). Determining the Factors of Bank Performance with a Focus on Risk and
Technical Efficiency. Proceedings from 2nd WSEAS International Conference on Management, Marketing
and Finances (MMF'08), Harvard, 31-39.
Pilbeam, K. (2005). Finance and Financial Markets (Second Edition). New York: Palgrave Macmillan.
Rauch C., F. Steffen F., A. Hackethal, & M. Tyrrel. (2010). Determinants of bank liquidity creation. Retrieved from
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1343595˃
Repullo, R. (2004). Capital requirements, market power, and risk taking in banking. Journal of Financial
Intermediation, 13, 156-182.
Rogers, W. H. (1993). Regression standard errors in clustered samples. Stata Technical Bulletin, 13, 19–23.
Stiglitz, J. & A. Weiss (1981). Credit rationing in markets with imperfect information. American Economic Review,
71, 393-410.
Teply, P. (2011). The future regulatory challenges of liquidity risk management. World Academy of Science,
Engineering and Technology, 73, 945-949.
Valla, N., B. Saes-Escorbiac, & M. Tiesset (2006). Bank liquidity and financial stability. Banque de France
Financial Stability Review, 9, 89-104.
Vento, G. A. & P. La Ganga (2009). Bank liquidity risk management and supervision: which lessons from recent
market turmoil? Journal of Money, Investment and Banking, 10, 79-126.
Vodova, P. (2011). Liquidity of Czech commercial banks and its determinants. International Journal of
Mathematical Models and Methods in Applied Sciences, 6 (5), 1060-1067.
White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for
Heteroskedasticity. Econometrica, 48 (4), 817–838.
Wooldridge, J. M. (2002). Econometric analysis of cross-section and panel data. Cambridge, MA: MIT Press.
World Economic Forum. (2011). Global Competitiveness Report 2011-2012. Geneva, Switzerland: World Economic
Forum. Retrieved from www3.weforum.org/docs/WEF_GCR_Report_2011-12.pdf.
World Economic Forum. (2013). Global Competitiveness Report 2013-2014. Geneva, Switzerland: World Economic
Forum. Retrieved from http://www3.weforum.org/docs/WEF_GlobalCompetitivenessReport_2013-14.pdf.
World Bank. (2014). World Development Indicators. Retrieved from the World Development Indicators (WDI)
database
Page 75
UNDERSTANDING THE LIMITATIONS OF FINANCIAL

RATIOS
Joseph Faello, Mississippi State University
ABSTRACT
The purpose of this paper is to provide financial statement users and accounting
academics with some useful insights when working with financial ratios. Initially, the uses and
benefits of financial ratios and the limitations of using financial ratios are discussed from the
financial statement users’ and accounting academics’ perspectives. Then, practical advice is
provided to both financial statement users and accounting academics alike to mitigate the
limitations of using financial ratios. Financial statement users and accounting academics will
find the issues discussed in this paper useful in their work with financial ratios.
Keywords: Financial ratios, Comparability, Homoscedasticity, Outliers
INTRODUCTION
Financial ratios play an important role in the analysis of financial statements and
accounting research. However, the use of financial ratios comes with its hazards. Both
accounting academics and financial statements’ users need to understand the problems and
limitations in working with financial ratios. The purpose of this paper is to address these issues
and to provide guidance on how to mitigate the problems surrounding financial ratios. Both
accounting academics and financial statement users will find this study useful in their dealings
with financial ratios.
The study is organized as follows:
1. Uses and benefits of financial ratios;

2. Limitations of financial ratios;
3. Dealing with the limitations of financial ratios; and
4. Conclusion.
USES AND BENEFITS OF FINANCIAL RATIOS
Financial ratios play an important role in financial reporting. A ratio “expresses the
mathematical relationship between one quantity and another,” (Kieso et al. 2013, p. 245). A
financial ratio consists of a numerator and a denominator, relating two financial amounts. The
two financial amounts can be from the balance sheet (e.g. current ratio), or from the income
statement (e.g. times interest earned), or from both the balance sheet and the income statement
(e.g. return on total assets).
Financial ratios help explain financial statements. For example, financial ratios assist in
benchmarking a firm’s performance with other firms in the same industry. Further, financial
ratios help financial statement users in identifying problem areas with a company’s operations,
liquidity, debt position, or profitability. From this benchmarking and assessment of a firm’s
performance, financial ratios help in assessing the firm’s overall risk (CICA, 1993). Prior
Page 76
research supports the use of financial ratios as a means to predict firms’ performance,
specifically stock returns and return on assets (e.g., Soliman, 2008; Nissim & Penman, 2001;
Fairfield & Yohn, 2001).
Financial ratios are frequently used in loan contracts between a firm (borrower) and a
financial institution (lender) as a means to limit the firm’s activities. A borrower has an incentive
to engage in activities that benefit his or her self-interests at the expense of the firm’s overall
value, resulting in the lender inserting accounting numbers in the debt contract (i.e., debt
covenant) to restrict the borrower’s value-reducing activities (Watts & Zimmerman, 1986). For
example, the loan contract may stipulate that the firm must maintain a current ratio of at least
2:1. In this manner, the firm is encouraged to effectively manage its current assets and current
liabilities, for example, by collecting its accounts receivables on a timely basis.
For financial statement users, financial ratios not only provide information about where a
firm has been, but also provides guidance about where it is headed in the future. For example,
negative trends in financial ratios over time could indicate a firm is in decline and provide
insights into predicting corporate failure. The Canadian Institute of Chartered Accountants
(CICA, 1993) in their Research Report titled “Using Ratios and Graphics in Financial
Reporting,” summarizes these and additional benefits of financial ratio analysis (see Appendix
1).
From an academic perspective, financial ratios play an important role in modeling. A
variable of interest (dependent variable) is estimated in a linear regression model by key
independent variables that are frequently financial ratios. Many bankruptcy prediction models
utilize financial ratios (Altman & Hotchkiss, 2006).
In summary, financial ratios provide important information about a firm’s past
performance, predicting a firm’s future performance prospects, assessing management’s
decision-making, risk assessment, and are a critical tool employed in lending agreements to
control a firm’s activities. In addition, accounting academics use financial ratios in modeling the
key variable of interest in their research studies.
LIMITATIONS OF FINANCIAL RATIOS
Inherent in the use of financial ratios are limitations that if not understood could result in
drawing incorrect conclusions from the results. Significant limitations in the use of financial
ratios are discussed below.
A financial ratio consists of a numerator and a denominator. If either the numerator or
denominator is misstated, then the financial ratio will be in error. Misstatement of either the
numerator or denominator could be the result of human error. For example, an error could be
made in collecting the firms’ financial statement data. Alternatively, the firms could be
employing earnings management techniques (e.g., manipulating accruals) that results in the data
itself being misstated. Prior accounting research provides evidence that firms’ managers
manipulate accruals to portray firms’ financial results in a more positive manner (e.g., Healy &
Wahlen, 1999; McNichols, 2000). In either case, financial statement users and academics would
obtain results from their analyses that could lead to erroneous conclusions.
Financial ratios derived from a particular firm’s financial statements are based on
accounting principles, accounting methods, and accounting classifications chosen by the firm.
These choices may not be consistent over time or between firms, thus compromising
comparability. Financial statement users and researchers alike need to understand that the
availability of accounting choices under generally accepted accounting principles (GAAP) may
Page 77
provide firms with greater flexibility in financial reporting, but it also can lead to a decline in
comparability of a firm’s financial results over time and between other firms.
The relation between the numerator and denominator variables in financial ratios poses
special problems to accounting researchers. Often, the researcher assumes a relationship or
proportionality between the two variables. However, little attention is paid to the distributional
properties of ratios (Foster, 1986).
Prior research studies on the distributional properties of financial ratios find that financial
ratio data may not be normally distributed, but rather can be severely skewed (Deakin, 1976;
Frecka & Hopwood, 1983; Foster, 1986). The implications of non-normality are that the basic
assumptions of ordinary least squares (OLS) regression are not valid; that is, the OLS estimators
are not the best linear unbiased estimators (Gujarati, 2003). For example, one assumption is that
the variance is equal or constant for all observations (homoscedasticity). Skewness causes a
breakdown in homoscedasticity resulting in a condition known as heteroscedasticity (unequal
variance). The importance of maintaining the validity of OLS assumptions is that these
assumptions are necessary to draw inferences about the population under investigation that are
statistically valid with hypothesis testing. Otherwise, hypothesis testing becomes more complex
(e.g., use of non-parametric methods).
Another cause of heteroscedasticity is the effect of outliers. An outlier is defined as “an
observation (or subset of observations) which appears to be inconsistent with the remainder of
that set of data” (Barnett & Lewis, 1994, p. 7). Financial ratio data are prone to include outliers.
As the denominator of an observation for a particular financial ratio approaches zero, the
financial ratio becomes extremely large. Frecka’s and Hopwood’s Table 1 (Frecka & Hopwood,
1983, p. 117) provides 10 ratio values. With the numerator kept constant with a value of “2.0,”
the denominator is reduced from a value of “1.0” to “0.10.” The ratio value changes from a value
of 2.0 (i.e., 2.0 divided by 1.0) to a value of 20 (i.e., 2.0 divided by 0.10). Thus, a decline in
value of 0.90 in the denominator has a tenfold effect on the financial ratio. Frecka and Hopwood
(1983, pp. 126-127) find that including an outlier with a value of 100 in a sample that has a mean
equal to 4 results in weighing that observation 2,304 times more than the average observation.
The effect of an outlier on small samples is even more profound. Frecka and Hopwood (1983, p.
127) state that including an observation with a current ratio of 100:1 in an industry of 9 firms that
would normally have a mean of 2.0 and a variance of 1.0, changes the sample mean and variance
to 12.89 and 1,068, respectively.
Overall, the limitations of financial ratios must be addressed prior to drawing any
meaningful conclusions from the results. Methods on how to deal with these limitations are
discussed in the next section.
DEALING WITH THE LIMITATIONS OF FINANCIAL RATIOS
The methods on how to deal with the limitations of financial ratios are discussed from
two perspectives: (a) financial statement users’ perspective; and (b) accounting researchers’
perspective.
Financial Statement Users’ Perspective
From the perspective of financial statement users, the limitations of financial statement
ratios affect the comparability of a firm’s financial results. For example, a firm that changes its
inventory valuation method from last-in, first-out (LIFO) to first-in, first-out (FIFO) will lack
Page 78
comparability with other firms that use LIFO. Thus, comparing inventory turnover ratios
between firms is compromised. Financial statement users can address this issue by reviewing the
accounting principles of the firm with the accounting change. Generally, a summary of the firm’s
accounting principles is provided in the first or second notes to the financial statements. Further,
additional information about the accounting change would be described in a separate note (e.g.,
inventory valuation note). Comparability is enhanced because the effects of the accounting
change on earnings would be explained in the note, allowing the financial statement user to
adjust for these effects and compare inventory turnover ratios with other firms in the industry
using LIFO. Note that comparing the firm’s inventory turnover ratio with its own inventory
turnover ratio in the previous year is not compromised because changes in accounting principles
are applied retrospectively, meaning the comparative year’s amounts are prepared on the same
FIFO basis as the current year’s amounts. Thus, financial statement users should always review
the accounting principles included in the financial statements of firms whose financial amounts
are used to calculate and compare financial ratios.
However, not all accounting changes are accounted for retrospectively. A change in
accounting estimate is accounted for prospectively, meaning prior period results are not adjusted
for the change. The reasoning for this treatment is that a change in accounting estimate is based
on new information and should affect only reported results in the current period and going
forward. Examples of changes in accounting estimates are changes to the estimate for
uncollectible accounts receivable and changes to depreciation expense due to changes in the
useful lives or salvage values of property, plant, and equipment assets.
One effect of a change in accounting estimate is to either increase or decrease earnings.
For example, a decrease in the rates for uncollectible accounts receivable, an extension to an
asset’s useful life, and an increase to an asset’s salvage value, all have the effect of increasing
operating income. Financial statement users should review the firm’s notes to its financial
statements for additional information provided about any changes to accounting estimates.
Further, the changes to accounting estimates should be consistent with the economic substance of
transactions affecting the firm. For example, during a recessionary period, a firm’s customers
have a greater likelihood not to pay accounts receivable, thus one would expect the rate for
uncollectible accounts receivable to increase and not decrease.
Financial statement users should be concerned with the direction of accounting estimates
(i.e., either income increasing or income decreasing). Generally during the fourth quarter of a
firm’s annual fiscal period, the firm’s managers make decisions concerning accounting estimates
that affect the reported results for the annual fiscal period (e.g., determine allowance for doubtful
accounts receivable). If the adjusting entries at year-end are predominantly income-increasing,
then it is important that these adjustments faithfully represent the underlying economic substance
of events affecting the firm. Otherwise, the managerial intent for the adjustments comes into
question. For example, accounting managers can make income-increasing accruals sufficient
enough to earn income-based bonuses that are stipulated in their employment contracts. Oyer
(1998) finds a non-linear relationship between employees’ compensation and accounting
earnings. That is, employees earn their bonuses once a certain threshold of accounting earnings
has been attained, usually occurring during the fourth quarter. Thus, a strong incentive exists for
managers to record income-increasing accruals or utilize other income-increasing earnings
management techniques in order to obtain higher compensation.
Since financial statement users cannot observe a firm’s internal accounting-related
decisions, other techniques must be employed to reduce the risk of relying on unreliable
Page 79
accounting figures that are used in computing financial ratios. First, financial statement users
should have a general understanding of how economic and other factors affect a firm’s financial
results. For example, a home builder’s financial results is more likely affected by increasing
interest rates than firms in other industries or a firm operating in a highly regulated industry is
more likely affected by changes in governmental regulation. Information about these effects can
be found in the management discussion section of the firm’s annual report for a public company.
Second, the compensation agreements of key personnel need to be scrutinized for potential
incentives to manage earnings. This information is also provided in the firm’s annual report for
public companies. Third, key covenants in contracts entered into by the firm that use accounting
measures need to be reviewed with particular attention paid to “thresholds” that have been
marginally exceeded. Information concerning accounting thresholds used in contracts affecting
the financial statements should be disclosed in the firm’s financial statement notes. For example,
a bank loan covenant may stipulate the firm must maintain a current ratio of 2:1. If the year-end
current ratio is 2.05:1 and the firm has reduced its allowance for doubtful accounts receivable,
then the risk of earnings management increases.
Overall, the financial statement user needs to take a proactive approach when relying on
financial amounts employed in financial ratio analysis.
Accounting Researchers’ Perspective
From a researcher’s viewpoint, a major limitation in working with financial ratios is the
effect of outliers on the statistical results. An outlier represents an observation that is inconsistent
with the remainder of the data set (Barnett & Lewis, 1994, p.7). The inclusion of a few outliers
could result in the sample’s mean and variance and regression’s variable coefficients being
significantly different from the results that are truly representative of the data set.
In dealing with outliers, Foster (1986) provides five alternatives:
1. Deleting the outlier because it is unrepresentative of the data set;

2. Retaining the outlier because it represents an extreme example of an underlying characteristic of the data
set;
3. Making adjustments to the data for the underlying economic and accounting factors that are believed to
cause the extreme observation(s); e.g., impute interest payments for “off balance sheet financing” debt;
4. “Winsorizing” the data set’s extreme observations; and
5. “Trimming” the data set by deleting a certain percentage or a certain number of the top and bottom
observations.
In addition, data can be transformed to alleviate the harmful effects of outliers. When
applying these alternatives in a particular research study, a considerable amount of judgment is
required in order to not create results that are unrepresentative of the underlying economic
phenomena. Given the effect of a single outlier on the sample’s mean and variance, the
methodology in handling outliers becomes extremely important.
A first step is to understand the nature of outliers. The nature of an outlier can be
classified as either “random” or “deterministic” (Barnett & Lewis, 1994, p. 42). The source of a
deterministic outlier is either measurement error or in execution, and the proper response is to
correct it or reject it (Barnett & Lewis, 1994, p. 42).
For a random outlier, the response is more complex because its underlying nature is not
readily apparent. One response is to accommodate the outlier in the existing research model or to
revise the model to incorporate the effects of outliers (Barnett & Lewis, 1994). A logarithmic
Page 80
data transformation is an example of a model revision. Other responses are to winsorize or trim
the data set or to outright reject the outlier.
Winsorizing the data involves assigning the outlier a lesser weight. Ghosh and Vogt
(2012, p. 3456) state:
A common procedure has been to replace any data value above the ninety-fifth percentile of the sample
data by the ninety-fifth percentile and any value below the fifth percentile by the fifth percentile.
The assumption seems to be that the outlier does not look right and estimates will be improved if
the outlier is made to look like other data.
Further, Ghosh and Vogt (2012, p. 3456) state the implication of this line of reasoning as
“the outlier value must be incorrect” and “the value is replaced by a more plausible value.” The
fault with this logic occurs when the outlier does truly represent an important characteristic of
the sample data and thus, should not be altered.
Another winsorizing procedure is the “skipped Huber method.” Under this method, a
distance from the median is used as a basis to rein in outliers. Specifically, values that are greater
than 5.2 median absolute deviations from the median are adjusted to this threshold value (Connor
& Herbert, 1999; Martin & Simin, 2003).
Trimming the data set involves deleting a certain percentage or a certain number of the
top and bottom observations. For example, Ettredge et al. (2005) deleted the top 1% and bottom
1% of the observations in their study. Generally, this approach is based on rules of thumb that
are particular to the research discipline. Deleting a specifically identifiable outlier implies it does
not fit with the pattern of the data set. Thus, its presence in the sample increases the likelihood of
non-representative results.
When applying these procedures, it is important that the researcher exercises due care
because the procedures in dealing with outliers are not an exact science. Further, the researcher
should use a comprehensive approach in dealing with the effects of outliers. A comprehensive
approach entails examining the research study’s results in raw data form, with the deletion of
specifically identifiable outliers, and with the use of data trimming, winsorizing, or
transformation. The primary goal of this analysis is to generate a research model and results that
provide the researcher with the best linear unbiased estimators.
The researcher must identify the outliers as a first step in order to employ the
comprehensive approach. One method of identifying outliers is to generate a series of plots
involving the independent and dependent variables, predicted values, and residuals. The residuals
are the differences between the observed values and predicted values for the dependent variable.
Important plots to study are: dependent versus independent variables; residuals versus
independent variables; and residuals versus predicted values.
Plotting the dependent variable versus the independent variables is useful in identifying
gaps between a particular observation (i.e., potential outlier) and the primary cluster of
observations. Although useful, this analysis does not provide insights into how the observation in
isolation or jointly with other potential outliers influences the research model’s results (Belsley
et al. 1980). Further, the combined effects of more than one independent variable on the
dependent variable cannot be separated with this analysis meaning it is only relevant with a
simple regression model.
A starting point for checking violations of the model’s assumptions, including
homoscedasticity, linearity, and existence of outliers, is to plot residuals versus independent
variables or residuals versus predicted values (Chatterjee & Price, 1991, p. 24). Plotting residuals
Page 81
versus the independent variables is of value only with a simple regression model because of the
inability to separate the combined effects of the independent variables. The largest residuals,
both positive and negative, should be considered as outliers in this analysis. However, in
empirical accounting research, simple regression models are rare because most studies involve
multiple independent variables.
When examining residuals a form of standardization is recommended to provide greater
comparability of residuals. The studentized residuals are the most common form of standardized
residuals (Hair Jr. et al., 1998, p. 172). Examining studentized residuals versus predicted values
is the preferable plotting technique to identify outliers in a multiple regression study. The “ideal”
plot for the researcher is a “null plot.” As Hair Jr. et al. (1998, p. 173) state:
The null plot shows the residuals falling randomly, with relatively equal dispersion about zero and no
strong tendency to be either greater or less than zero.
A “pattern” in the plot indicates the violation of the constant variance of the error terms
assumption (homoscedasticity). The violation of this assumption is called heteroscedasticity and
results in a lack of statistically valid hypothesis testing. Two common patterns of
heteroscedasticity are diamond-shaped or triangle-shaped patterns (Hair Jr. et al., 1998, p.175). A
diamond-shaped pattern indicates more variation exists in the mid-range of the sample than at the
tails. A triangle-shaped pattern indicates the residuals increase or decrease with increases in the
predicted values. In both cases, the constant variance assumption is violated and the researcher
should question the validity of hypothesis tests.
As is the case with most accounting research studies, sample sizes can be quite large with
the number of observations in the thousands. Thus, visually identifying outliers can be difficult
through an analysis of plots. Three statistical tests provided by most statistical packages are the
Levene’s Test, Mahalanobis Distance, and Cook’s Distance.
The Levene’s Test is a test for homoscedasticity for groups of data. The null hypothesis
for this test is that the variances are equal across the groups. Field (2005, p. 98) explains:
Therefore, if Levene’s test is significant at ρ ≤ .05 then we can conclude that the null hypothesis is incorrect
and that the variances are significantly different ─ therefore, the assumption of homogeneity of
variances has been violated. If, however, Levene’s test is non-significant (i.e. ρ > .05) then we
must accept the null hypothesis that the difference between the variances is zero ─ the variances
are roughly equal and the assumption is tenable.
A breakdown of the homoscedasticity assumption (i.e., heteroscedasticity) could be

caused by the effects of outliers. Thus, a significant Levene’s Test indicates the existence of
heteroscedasticity and increases the likelihood of finding outliers in the sample.
In some cases, the effect of a single outlier can strongly influence the results of the
researcher’s multiple regression model. This observation is called an influential observation and
it is important that the researcher identifies it. The Mahalanobis Distance statistic can be used to
detect the outliers that are considered influential observations. The Mahalanobis Distances
“measure the distance of cases from the mean(s) of the predictor variable(s),” (Field, 2005, p.
165). Also, Mahalanobis Distances possess the statistical properties for statistical testing. Hair Jr.
et al. (1998, p. 66) recommend a conservative .001 significance level as a threshold to identify
these influential cases.
The Cook’s Distance is a statistic that measures the effect of a single observation on the
research model. Specifically, it measures the influence of a single observation “based on the total
Page 82
changes in all other residuals when the case is deleted from the estimation process,” (Hair et al.
1998, p. 218). Hair Jr. et al. (1998, p. 225) prescribe a rule of thumb of greater than 4 / (n – k –
1), (where “k” is the number of independent variables and “n” is the sample size), as a threshold
to identify influential observations for small or large sample sizes. Other researchers use as a rule
of thumb a Cook’s Distance of ≥ 1.0 as an indicator of an influential observation (Field, 2005, p.
165).
After identifying the outliers by the procedures indicated above, the researcher can apply
the following techniques in dealing with the outliers’ effects on financial ratio data:
1. If the financial ratios used as independent variables in the research model exhibit skewness, then the
researcher should consider a data transformation. Foster (1986, p. 103) and Frecka and Hopwood
(1983, p. 122) find logarithmic transformations reduce skewness for certain financial ratios. However,
Frecka and Hopwood (1983, p. 117) find the maintenance of OLS normality assumptions are best
achieved by deleting the outliers.
2. A comparison of the statistical results using the raw data, transformed data, raw data with specific
outliers deleted, and trimmed data or winsorized data needs to be conducted. The researcher needs to
carefully analyze the results to find the best fitting research model.
Lien and Balakrishnan (2005) conduct a simulation exercise of a simple regression model
and provide results for the raw data, trimmed data, and winsorized data. On the one hand, they
find that the trimmed data results show little change in the value of the independent variable’s
co-efficient from that of the raw data’s results, but there is a substantial reduction in the
explanatory power of the model (i.e., lower R2). On the other hand, the winsorized data results
show a higher value for the independent variable’s co-efficient, when compared to the raw data’s
and trimmed data’s results. However, the explanatory power of the winsorized data model is
maintained (i.e., higher R2 than trimmed data model). Overall, these results indicate that the
researcher needs to exercise judgment and make trade-offs when deciding on how to deal with
the effect of outliers in financial ratio data.
CONCLUSION
Financial statement users and accounting researchers need to understand the problems
and limitations in working with financial ratios. The purpose of this study is to recognize these
issues and to provide both financial statement users and accounting researchers with useful hints
when working with financial ratios.
From the financial statement users’ perspective, financial ratios provide important
information about a firm’s past performance, predicting a firm’s future performance prospects,
assessing management’s decision-making, risk assessment, and are a critical tool employed in
lending agreements to control a firm’s activities. All of these uses are dependent on the
comparability of a firm’s financial ratios with itself and between firms over time. Comparability
is enhanced when financial statement users take a proactive approach in their work with financial
ratios. A proactive approach involves scrutinizing firms’ financial statement notes for the
consistent application of accounting principles and estimates. Changes to either accounting
principles or accounting estimates should be properly disclosed and be supported by the
underlying economic substance of events affecting the firm. Special attention should be directed
at the relation between income increasing accruals and managers’ performance contracts based
on earnings.
Page 83
From the accounting researchers’ perspective, a major limitation in working with

financial ratios is the effect of outliers on the statistical results. The inclusion of a few outliers
could result in the sample’s mean and variance and regression’s variable co-efficients being
significantly different from the results that are truly representative of the data set. Specifically,
the effect of outliers is to invalidate the constant variance assumption (homoscedasticity) of OLS
regression. This condition is known as heteroscedasticity. The importance of maintaining the
validity of OLS assumptions is that these assumptions are necessary to draw inferences about the
population under investigation that are statistically valid with hypothesis testing.
The accounting researcher needs to develop a methodology to first identify the outliers.
Plotting studentized residuals versus predicted values is a first step to identify outliers in a
multiple regression study. Hair Jr. et al (1998) provide guidance in this area. For large sample
sizes, statistical techniques such as Levene’s Test, Mahalanobis Distance, and Cook’s Distance
are available with most statistical packages. Field (2005) and Hair Jr. et al. (1998) provide
guidance on the use of these techniques.
Next, after identifying the outliers, a comprehensive approach is required in dealing with
the effect of outliers. A comprehensive approach entails examining the research study’s results in
raw data form, with the deletion of specifically identifiable outliers, and with the use of data
trimming, winsorizing, or transformation. The primary goal of this analysis is to generate a
research model and results that provide the researcher with the best unbiased estimators. Overall,
the accounting researcher needs to exercise considerable professional judgment and due care
throughout this process.
Page 84
REFERENCES
Altman, E., and Hotchkiss, E. (2006). Corporate Financial Distress and Bankruptcy, Third Edition. Hoboken, NJ:
John Wiley & Sons, Inc.
Barnett, V., and Lewis, T. (1994). Outliers in Statistical Data, Third Edition. Hoboken, NJ: John Wiley & Sons, Inc.
Belsley, D., Kuh, E., and Welsch, R. (1980). Regression Diagnostics ─ Identifying Influential Data and Sources of
Collinearity, Hoboken, NJ: John Wiley & Sons, Inc.
Canadian Institute of Chartered Accountants (CICA), (1993). Using Ratios and Graphics in Financial Reporting.
Toronto, Canada: Canadian Institute of Chartered Accountants.
Chatterjee, S., and Price, B. (1991). Regression Analysis by Example, Second Edition, Hoboken, NJ: John Wiley &
Sons, Inc.
Connor, G., and Herbert, N. (1999, Winter). “Estimation of the European Equity Model,” Horizon: The Barra
Newsletter, 169, Retrieved January 14, 2015, from
http://www.barra.com/Newsletter/nl169/estimationeuro169.asp.
Deakin, E. (1976, January). “Distributions of Financial Accounting Ratios: Some Empirical Evidence,” The
Accounting Review, pp. 90-96.
Ettredge, M., Kwon, S., Smith, D., and Zarowin, P. (2005, July). “The Impact of SFAS No. 131 Business Segment
Data on the Market’s Ability to Anticipate Future Earnings,” The Accounting Review, 80(3), pp. 773-804.
Fairfield, P., and Yohn, T. (2001). “Using Asset Turnover and Profit Margin to Forecast Changes in Profitability,”
Review of Accounting Studies, 6, pp. 371-385.
Field, A. (2005). Discovering Statistics Using SPSS, Second Edition. Thousand Oaks, CA: Sage Publications Inc.
Foster, G. (1986). Financial Statement Analysis, Second Edition. Englewood Cliffs, NJ: Prentice-Hall, Inc.
Frecka, T., and Hopwood, W. (1983, January). “The Effects of Outliers on the Cross-Sectional Distributional
Properties of Financial Ratios,” The Accounting Review, Vol. LVIII(1), pp. 115-128.
Ghosh, D., and Vogt, A. (2012). “Outliers: An Evaluation of Methodologies,” Section on Survey Research Methods
─ Joint Statistical Meetings, American Statistical Association, pp. 3455-3460.
Gujarati, D. (2003). Basic Econometrics, 4th Edition. New York, NY: The McGraw-Hill Companies, Inc.
Hair Jr., J., Anderson, R., Tatham, R., and Black, W. (1998). Multivariate Data Analysis, Fifth Edition. Upper
Saddle River, NJ: Prentice-Hall, Inc.
Healy, P., and Wahlen, J. (1999, December). “Commentary: A Review of the Earnings Management Literature and
Its Implications for Standard Setting,” Accounting Horizons, 13(4), pp. 365-383.
Kieso, D., Weygandt, J., and Warfield, T. (2013). Intermediate Accounting, Fifteenth Edition. Hoboken, NJ: John
Wiley & Sons, Inc.
Lien, D., and Balakrishnan, N. (2005, November). “On Regression Analysis with Data Cleaning via Trimming,
Winsorization, and Dichotomization,” Communications in Statistics – Simulation and Computation, Vol.
34(4), pp. 839-849.
Martin, R., and Simin, T. (2003, September/October). “Outlier-Resistant Estimates of Beta,” Financial Analysts
Journal, 59(5), pp. 56-69.
McNichols, M. (2000, Winter). “Research Design Issues in Earnings Management Studies,” Journal of Accounting
and Public Policy, 19(4/5), pp.313-345.
Nissim, D., and Penman, S. (2001). “Ratio Analysis and Equity Valuation: From Research to Practice,” Review of
Accounting Studies, 6, pp. 109-154.
Oyer, P. (1998, February). “Fiscal Year Ends and Nonlinear Incentive Contracts: The Effect of Business
Seasonality,” The Quarterly Journal of Economics, 113(1), pp. 149-185.
Soliman, M. (2008). “The Use of DuPont Analysis by Market Participants,” The Accounting Review, Vol. 83(3), pp.
823-853.
Watts, R., and Zimmerman, J. (1986). Positive Accounting Theory, Englewood Cliffs, NJ: Prentice-Hall, Inc.
Page 85
Appendix 1
BENEFITS OF USING FINANCIAL RATIOS

1. Ratios can provide a greater awareness of the interrelationship of individual financial statements and a
greater understanding of the financial statements themselves by summarizing data in a form that is more
easily understood, interpreted and compared.
2. Ratios can facilitate comparisons with external measurements, such as industry-wide norms, and help
identify an entity’s strengths and weaknesses.
3. Ratios can facilitate comparisons over time, such as changes in long-term trends of financial position,
operations and cash flows.
4. Ratios can provide an informed basis for making investment-related decisions by comparing an entity’s
financial performance in relation to another entity in the same industry and the industry as a whole.
5. Because they are determined by dividing one financial variable by another, ratios downplay the impact of
size and allow evaluation over time or across companies without undue concern for the effect of size
differences.
6. Ratios can identify the stability of relationships within an entity over time and common relationships
among entities within a given industry.
7. Ratios can be used to support the viability of extending credit. For example, a small closely-held entity may
ask a supplier for trade credit or approach a bank for a loan. In such cases, credit managers and bank loan
officers may be able to use ratio analysis on the financial statements to gain insights into the past, present,
and future prospects of the individual applicants. These insights may help potential creditors predict the
future cash flows of prospective borrowers and make credit decisions on a more rational basis.
8. Ratios can serve as benchmarks against which the financial strength of an entity is measured; for example,
debt covenants often specify limits in terms of financial ratios.
9. Ratios can be used as an initial indicator to determine if there are specific problems that need to be
investigated further, such as aging of accounts receivable and inventory turnover.
10. Ratios can assist auditors and other users in evaluating the reasonableness of the amounts shown in the
financial statements. Analytical review procedures can help the auditor in gaining an understanding of the
client’s business, plan the engagement, identify unexpected relationships among accounting data and
provide substantive audit evidence.
Reproduced from: The Canadian Institute of Chartered Accountants, “Research Report: Using Ratios and
Graphics in Financial Reporting,” The Canadian Institute of Chartered Accountants, 1993, p. 13.
Page 86
Page 87
FINANCIAL REPORTING QUALITY OF FAMILY-

CONTROLLED SMALL AND MEDIUM-SIZED
ENTITIES: A JAPANESE CASE STUDY
Muneaki Fujibayashi, Osaka Prefecture University
Hirofumi Kojima, Osaka Prefecture University
Mineo Tsuji, Osaka Prefecture University
ABSTRACT
Japan has not yet adopted International Financial Reporting Standards (IFRS) for
small and medium-sized entities (SMEs). Consequently, differences in accounting standards
and practices prevail between large corporations and SMEs in Japan. We argue that the
apparent auditor role of the National Tax Agency of Japan could drive some of the observed
differences in earnings quality between large corporations and SMEs. To address this, we
examine the financial reporting quality (FRQ) of Japanese SMEs and consider whether they
engage in accrual- or tax/cash-based earnings management activity. We first examine the
characteristics of 153 Japanese firms conforming to the European Union definition of an
SME. We then consider a smaller sample of 20 firm-year observations and formulate a
hypothesis regarding a particular insurance contract. We find that the FRQ of family-
controlled SMEs tends to be higher if they purchase insurance with cash after forecasting
their insurance payments. This is because SMEs can more easily make the decision to
purchase insurance than can large corporations, plus Japan’s accounting standards for
SMEs accord with the Corporation Tax Act, and therefore their current net income is their
taxable income. As a result, good performing SMEs tend to choose the managerial
accounting activity that allows their accounting profits to decrease as a means of reducing
taxes payable.
INTRODUCTION
According to the IASB (2014), Japan has not yet adopted International Financial
Reporting Standards (IFRS) for small and medium-sized entities (also enterprises) (SMEs).
Nor have there been any deliberations on whether unlisted companies should use IFRS for
SMEs in their financial reporting, such that SMEs not required to use IFRS currently employ
Japanese accounting standards. As a result, differences in accounting standards and practices
between large corporations and SMEs prevail in Japan (Tsuji & Fujibayashi, 2013, 18) such
that financial institutions often have no faith in the financial statements prepared by SMEs.
According to the Small and Medium Enterprise Agency (2010, 92–99), the most
common potential user requiring SMEs to disclose their financial statements is a financial
institution (78.7%), followed by the Credit Guarantee Corporation (38.3%) and the company
president (12.1%). Financial institutions in Japan usually request that SMEs in need of
finance submit a copy of their financial statements annexed to their final tax returns to which
a tax office superintendent has affixed a receipt seal. Considering there is no requirement
Page 88
under the Companies Act for auditors to audit the financial statements, we consider that one
of the practical roles of the National Tax Agency is that of an auditing firm.
In this paper, we argue that the apparent auditor role of the National Tax Agency of
Japan could drive some of the differences in earnings quality between large corporations and
SMEs. To do this, we examine the financial reporting quality (FRQ) of Japanese SMEs and
consider whether they engage in accrual- or tax/cash-based earnings management activity.
LITERATURE REVIEW AND HYPOTHESES
Even though they are not necessarily SMEs, Prencipe et al. (2014, 363) categorize
prior studies of family firms in the areas of management and business across four theoretical
frameworks; agency theory, stewardship theory, the resource-based view of the firm, and
socioemotional wealth theory.
An agency relationship is a contractual arrangement under which one or more
persons, or principal(s), engage another person, the agent, to perform some service on their
behalf, a process that invariably involves delegating some decision-making authority to the
agent (Jensen & Meckling, 1976, 308). Villalonga & Amit (2006, 387) refer to the potential
owner–manager conflict as Agency Problem I, and point out that in the case of a firm with a
single large shareholder and a fringe of small shareholders, this can be mitigated because of
the greater incentive of the large shareholder to monitor the manager. There is also the
possibility of a second type of conflict, that is, the conflict between controlling family owners
and other owners, known as Agency Problem II. If the single large shareholder is an
individual or a family, there is a greater incentive for both expropriation and monitoring,
which is likely to lead to Agency Problem II overshadowing Agency Problem I.
Using an agency theory framework, some studies have considered the FRQ of family
firms and/or SMEs (Prencipe et al., 2014). For example, Wang (2006), Ali et al. (2007), and
Cascino et al. (2010) find that family firms report better-quality earnings. Concerning the
earnings management of family firms, Jaggi & Leung (2007) and Jaggi et al. (2009) show
that audit committees or independent corporate boards constrain earnings management, while
Jiraporn & DaDalt (2009) suggest that family firms are less likely to manage earnings.
Elsewhere, Tong (2010) reveals that family firms have higher-quality reporting, as reflected
by lower absolute discretionary accruals, whereas Yang (2010) argues that earnings
management increases with the level of insider ownership in family-controlled firms. Lastly,
Prencipe & Bar-Yosef (2011) show that in family-controlled firms, independent board
directors have a weaker effect on earnings management.
In the mid-1990s, some Japanese listed company studies considered the relationship
between the firm owner–manager and the choice of accounting method. For instance,
Holthausen & Leftwich (1983), Watts & Zimmerman (1986), and Okabe (1994, 93–106)
hypothesize that in terms of a closer relationship between financial accounting and taxation, a
firm controlled by managers will adopt the accounting method that enables it to delay
(hasten) the recognition of revenues (expenses). Empirically, this should correspond with a
firm with a relatively large proportion of shares held by its directors. The findings in these
studies indicate that while the first hypothesis was not rejected, the second hypothesis was, as
was a third hypothesis that a manager-controlled firm will adopt an accounting method that
enables the firm to increase its income figures because of its adoption of the diminishing
balance method.
Page 89
Earlier studies on earnings quality in Japan largely concern the cash flow statement.
Kamata (1998) explains that the calculation of cash flow rarely depends on subjective
judgments, and that because cash flow is not as intricate as income, it is a comparable
measure for the performance of an entity. To assess the ability of an entity to circulate funds,
Sato (1998) places a special emphasis on the cost of inventories and receivables, and their
provisions and the depreciation method. It is necessary to understand these inflows to analyze
earnings quality, after which it then becomes possible to assess the profits available for
dividends.
In terms of the institutional context, under the Stock Exchange Law, Japanese
corporations have disclosed a form of cash flow statement since April 1987. However, this
was prepared merely on an individual basis, treated as an element outside the financial
statements, and accordingly was not subject to auditing. The major amendment of the
Ministry of Finance Consolidation Regulation, in accordance with the Business Accounting
Deliberation Council Opinion, is the legal requirement for consolidated statements from
fiscal year 1999-2000 onward (Articles 8-2 & 76 Ministry of Finance Consolidation
Regulation) (Kuroda, 2001, 1857–1858). Nonetheless, neither the Commercial Code nor the
Corporation Tax Act requires all businesses/companies to prepare cash flow statements
(Sakurai, 2001, 1693). In fact, the legal requirements concerning the preparation of the cash
flow statement remain unchanged, even though the titles of the acts that govern them have
changed (to the Financial Instruments and Exchange Act and the Companies Act,
respectively).
It is common to conceive of the legal framework for accounting in Japan as a
triangular legal system (Arai & Shiratori, 1991, 5). In this regard, the principle of the definite
settlement of accounts has resulted in negative effects on financial reporting. It is natural that
managers attach importance to tax savings and that they attempt to minimize costs on tax
return adjustments regarding any difference in taxable income and net income as reported in
the financial statements. The accounting behavior of managers thus likely results in a
phenomenon whereby financial statements for reporting purposes are prepared in conformity
with the accounting rules in the Corporation Tax Act. As a result, the amount of reported
income in financial statements under the Commercial Code may be lower. This argument for
the reduction in reported income finds support from the viewpoint of the protection of
creditors. However, despite much criticism, the principle continues at least partly because of
its economic rationality (Sakurai, 2001, 1724–1725).
The Commissioner’s Directive on Interpretation of the Corporation Tax Act regulates
the taxable treatments concerning payable insurance fees for term insurance (Article 9-3-5).
The taxable treatment of fees for long-term smoothing insurance paid by corporations is by
the Direct Corporation Taxation Division 2-2, as issued on June 16 1987, and most recently
amended by the Corporation Taxation Division 2-3 issued on February 28 2008. Such a term
insurance contract stipulates that the corporation itself is the contractor and that a director or
an employee is the insured. Accordingly, corporations capitalize half of their fees and
recognize the remainder as expenses in financial accounting and as deductible from taxable
income.
Based on the above discussion, our first hypothesis is:
H1: The FRQ of family-controlled SMEs tends to be higher if they purchase insurance with cash.
Page 90
In practice, a corporation must pay cash for term insurance. This is because in
accordance with the Ordinance for Enforcement of the Insurance Business Act in Japan,
banks must refrain from providing an agency or intermediary service for the conclusion of an
insurance contract, wherein the policyholder or insured person is the borrower of a business
loan, in consideration of any fees or other remuneration (Article 212; Japanese Bankers
Association, 2012). Therefore, SMEs must either purchase an insurance contract for cash
monthly or prepay it in cash yearly. Besides insurance contracts, Japanese SMEs can adopt a
small-scale enterprise mutual aid system provided, for example, by the Organization for
Small & Medium Enterprises and Regional Innovation (2015), founded the act of the same
name by the Independent Administrative Agency in 2004. Through this, an SME is able to
adjust its annual insurance fees each fiscal year. The interest on the term insurance, some part
of which is off-balanced, then appears identical to a term deposit.
The Commissioner’s Directive on the Interpretation of the Corporate Income Tax
Law allows all or part of any (pre)paid insurance cost to be deductible from taxable income
(9-3-4, etc.). In relation to the deduction, in 2012 the Supreme Court decided a case to seek
the revocation of reassessments, etc. regarding income tax (Case No. 2009 (Gyo-Hi) 404).
This is summarized such that “…where companies paid premiums under endowment
insurance contracts, which had been concluded by designating the beneficiaries of death
benefits as the companies and the beneficiaries of maturity insurance benefits as the
representatives of the companies indicated in the judgements such as where half of paid
premiums was treated as loans to the representatives, whereas the remaining half was treated
as premiums and included in deductible expenses in the companies’ accounting, the latter half
cannot be regarded as the amount expensed for the obtaining such revenue …” (Supreme
Court, 2012). Following this judgement, entities increasingly appeared to purchase insurance
contracts for tax saving purposes.
Using a sample of both public companies and family firms, given the Corporation Tax
Act (Articles 22 and 74(1)), Suzuki (2013, 287) hypothesizes that the tax accountants of
family firms tend to decrease taxable income more than in public companies. One result
shows that regardless of the shareholder structure, tax accountants tend to suggest that clients
with good performance should choose an accounting procedure for the repair of fixed assets
that enables both accounting profit and taxable income to decrease. In contrast, they tend to
suggest that clients with poor performance should choose a method that enables both
accounting profit and taxable income to increase.
Based on the above discussion, our second hypothesis is:
H2: Good performing SMEs tend to choose managerial accounting activity that allows their
accounting profits to decrease as a means to reduce taxes payable.
Suda (2004, 487–488) considers taxes as the chief consideration in the role of political
costs affecting the choice of accounting procedures in Japan, and argues that in contrast to the
scale hypothesis confirmed in the US, the significant relationship between firm size and the
choice of accounting procedure would disappear with a reduction in the incentive for tax
reductions. In particular, Suda (2004, 60–65) argues that a firm with a higher debt-to-equity
ratio would prefer an accounting procedure that enables it to establish and maintain trust with
a financial institution and, at the same time, increase the amount of income.
Page 91
Based on the above discussion, our third hypothesis is:
H3: SMEs never decrease their accounting profits (if any) beyond the range of trust with their
lenders, particularly financial institutions.
Nakakuki (2002) reviews empirical research on the relationship between accounting

information and managerial incentives, chiefly in the US. Overall, the evidence suggests that
firms that frequently manage their earnings are larger, have smaller debt-to-equity ratios, and
a higher proportion of manager-held shares. Further, smaller firms rarely manage earnings to
reduce taxes because they do not have access to suitable financial management techniques
and because they emphasize financial reporting. According to Tomioka (2014), some
Japanese corporations with more than a billion yen in capital have a much lower ratio of taxes
to net income. The Act on Special Measures Concerning Taxation enables these very large
firms to reduce their taxes.
However, the Act on Special Measures Concerning Taxation is not only applicable to
large companies but also to SMEs. In addition, the Japanese relationship between financial
accounting and taxation is very different from that found in the US (OECD, 1987). Even
today, managers include many elements, e.g. depreciation, necessary for the calculation of
taxable income in the financial statements prepared in advance under the Companies Act.
Moreover, Japanese SMEs do not have to prepare cash flow statements. Concerning the
disclosure of family firms (Prencipe et al., 2014), Chen & Jaggi (2000) reveal that family-
controlled firms exhibit a weaker positive association between the comprehensiveness of
financial disclosures and the percentage of independent directors on boards. In other work,
Chen et al. (2008) indicate that family firms provide fewer earnings forecasts, but more
earnings warnings, while Chau & Gray (2010) demonstrate that there is more (less) voluntary
disclosure by firms with larger (smaller) family shareholdings. Based on three Japanese SME
cases, Michie (2014) suggests that the cash flow statement is useful for determining whether
monies owed to suppliers will be paid when due and for deciding whether the business
continues as an SME. A study on the FRQ of US private firms specifies cash flow from
operations as an independent variable (Hope et al., 2013, 1724).
Based on the above discussion, our fourth hypothesis is:
H4: Good performing SMEs tend to choose an accounting procedure that allows their accounting
profits to decrease based on their cash flows.
Additionally, there is no obligation governing the auditing of SMEs. Although private

firms in the UK undergo voluntary audits to reduce agency costs (Dedman et al., 2014), only
6.9% of Japanese SME respondents to a survey indicated that they would hire an accounting
advisor ‘now’ or ‘in the future’ (Small and Medium Enterprise Agency, 2010, 47). In the next
section, we empirically investigate the financial reporting quality of Japanese SMEs.
SAMPLE AND METHODOLOGY
In this case study, we conduct two analyses. The first analysis reveals the general
characteristics of Japanese SMEs, while the second advances a novel hypothesis that these
SMEs purchase insurance contracts and other services as a means to reduce taxes. For the
Page 92
first analysis, we sample 153 SMEs using the following criteria: (a) listed in Japan on March
31 2014, (b) balance day of March 31 in 2014, (c) meets the European Union’s definition of
an SME as a company or group, in line with the Commission Recommendation of May 6
2003 concerning the definition of micro, small, and medium-sized enterprises (2003/361/EC),
with fewer than 250 employees and an annual turnover of €50 million or lower or assets of
€43 million or lower (€1 = ¥142.13 as at March 31 2014) (EU, 2005, 14), and (d) neither
operating profit nor operating cash flow is less than zero.
Table 1 provides selected descriptive statistics of our sample. Appendix A lists the
variable definitions. We categorize the variables into four groups: SME size, mandatory
financial statement figures and financial ratios, statement of cash flow figures, and
shareholders’ components.
Table 1
DESCRIPTIVE STATISTICS (153 Japanese SMEs)
Unit: JPY millions
Variable Min. Max. Mean SD
#EMPLOY 7.00 246.00 124.24 59.18
SALES 366.00 15909.00 4350.51 2655.62
TOTALAS 444.00 58776.00 5327.57 5347.38
COST 52.00 12677.00 2804.74 2172.59
SELLING 149.00 9456.00 1274.71 1236.14
DEPRECI 1.00 1911.00 137.74 183.84
NONOPER 0.00 19105.00 155.67 1542.81
EXTRAOR 0.00 221.00 19.56 34.99
INCOMEBE –79.00 1477.00 336.78 287.73
TAXES 1.00 689.00 113.75 119.11
INCOME –95.00 1833.00 226.87 235.85
CFOPER 2.30 2224.00 364.01 335.53
CFINVES –2076.00 3621.00 –129.91 459.45
CFFINANC –1795.00 1449.00 –52.99 382.42
CASH 48.00 5919.00 1213.61 936.69
#SHARE 168.00 19671.00 2568.86 3222.02
FOREIGN (%) 0.00 46.50 3.31 7.94
INDIVID (%) 19.30 95.30 60.18 19.32
OPEINCO 0.00 1419.00 324.49 289.03
ORDIINCO 3.00 1477.00 333.56 288.37
NETAS 183.00 13933.00 2917.33 2479.63
EQUITY (%) 4.30 93.90 57.16 20.68
CAPITAL 90.00 4097.00 938.44 710.96
RETEARN –3157.00 16490.00 1363.92 2315.37
BORROW 0.00 19500.00 939.26 1903.84
ROE (%) –21.90 122.00 9.84 12.93
ROA (%) –3.40 29.60 5.00 4.68
For the second analysis, we employ a smaller sample of 20 firm-year observations

where monthly financial data are available. We obtain the financial statements for four SMEs
operating in a single industry (warehousing) over the period 2009–2013. Each SME is family
owned, controlled, and operated, with all shares held by the same family, and all directors
from the same family.
Page 93
Table 2 provides selected descriptive statistics of this smaller sample. Appendix A

lists the variable definitions. We categorize the variables into three groups: SME size,
mandatory financial statement figures, and statement of cash flow figures. The variables in
the first group relate to SME size, namely, #EMPLOY, SALES, and NETAS. All of our
sample firms are unlisted and therefore not required to comply with the Financial Instruments
and Exchange Act, but do satisfy the definitions in the Companies Act (Article 2), the
Corporate Income Tax Law (Article 66(2)), and the Small and Medium-sized Enterprise
Basic Act (Article 2).
Table 2
DESCRIPTIVE STATISTICS (20 company-years)
Unit: JPY
Variable Min Max Mean SD
#EMPLOY 12.00 26.00 20.50 5.30640
SALES 107828432.00 217833196.00 166807400.15 34022574.26
NETAS 372001249.00 655481160.00 530393231.20 107648616.67
NETAS –25700739.00 20203036.00 –291535.85 12582558.44
LIQAS 22400190.00 241105724.00 127585439.10 66159462.33
LIQAS –63682345.00 45617454.00 2383984.85 24408394.56
FIXEDAS 273363176.00 441232540.00 375096588.45 61806952.53
FIXEDAS –26600376.00 16210184.00 –4027845.00 10591206.31
COST 51060306.00 173043094.00 116256664.30 41166018.84
DEPRECI 2862292.00 23256039.00 11176904.30 5505660.35
INSUR 197300.00 2747325.00 785004.15 838277.16
SELLING 22925224.00 65703699.00 43522881.25 14739256.08
NONOPER 0.00 2361677.00 476953.75 596425.58
EXTRAOR 0.00 38041292.00 5738855.15 11215971.65
TAXES 70000.00 13156100.00 2254813.25 3509833.31
INCOMEBE –21365223.00 34471929.00 4363223.85 15351957.18
INCOME –21611623.00 21315829.00 2108410.60 12976799.26
CFOPER –51423.00 121089.00 20741.05 35108.60
CFINVES –127287.00 12871.00 –18122.60 32182.33
CFFINANC –12000.00 0.00 –3467.50 4283.56
CASH –62655.00 48591.00 3129.70 24555.64
CASH 11013.00 188051.00 101053.10 54359.27
CFFREE –62654.78 27948.30 4356.37 25158.94
The variables in the second group relate to the mandatory financial statements
(balance sheet and income statement), including NETAS, LIQAS, LIQAS, FIXEDAS,
FIXEDAS, COST, DEPRECI, INSUR, SELLING, NONOPER, EXTRAOR, TAXES,
INCOMEBE, and INCOME. For the balance sheet items, we use not only LIQAS but also
FIXEDAS when considering the provisions relating to depreciation in the Corporation Tax
Act. For the statement of income, we specify expense items such as INSUR and TAXES as
variables, along with INCOMEBE and INCOME. The variables in the third group relate to
the statement of cash flows, including CFOPER, CFINVES, CFFINANC, CASH, CASH,
Page 94
and CFFREE. Because SMEs are not legally required to prepare a statement of cash flows,
we prepared these ourselves. Along with the three kinds of cash flow activities, we specify
CFFREE, CASH, and CASH.
For our first analysis of the 153 Japanese SMEs, we analyze the data underlying the
descriptive statistics in Table 1 and test our hypotheses using a one-sided test of the Pearson
correlation coefficients. In our second analysis of the four Japanese SMEs, we analyze the
data underlying the descriptive statistics in Table 2 and test our hypotheses using a one-sided
test of the Pearson correlation coefficients. We then see if monthly INSUR forecasts net and
taxable income and therefore allows the SMEs to manage their earnings. Finally, to illustrate
the purchasing behavior of managers immediately before the end of the accounting period,
we conducted several interviews in the last half of 2014 with two national insurance
companies and a multinational equipment corporation. Table 3 details the questions used in
these interviews.
Table 3
QUESTIONNAIRE ABOUT COMMERCIAL INSURANCE OR PRODUCTS
1 How many commercial insurance contracts or equipment products did you sell in the preceding three years?
When your clients bought your goods or services applicable to Q1 in their balance sheet month, how many
2
were commercial insurance contracts or equipment products?
How many commercial insurance contracts or equipment products applicable to Q1 have already been
3
surrendered?
RESULTS AND DISCUSSION
Correlation in the first analysis
The results for the 153 firms display some general characteristics of Japanese SMEs.
Regarding the three size variables, #EMPLOY and SALES are positively associated with
each other (0.518**), while TOTALAS is associated with #EMPLOY (0.163*) and SALES
(0.218**), which are both relatively smaller.
Of the expenditure variables in the financial statement figures, #EMPLOY is
positively related to COST (0.441**) and SELLING (0.303**). Similarly, SALES is
positively related to COST (0.905**) and SELLING (0.483**). In contrast, TOTALAS is
positively related to OPEINCO (0.426**), ORDIINCO (0.371**), INCOMEBE (0.353**),
and INCOME (0.621**) among the income statement variables, and to NETAS (0.520**),
CAPITAL (0.526**), and BORROW (0.776**) among the income statement variables, and
to #SHARE (0.330**) among the shareholders’ components variables.
Of the cash flow variables, three kinds of cash flows, CFOPER, CFINVES, and
CFFINANC, are negatively related with one another; CFOPER–CFINVES, –0.221**,
CFINVES–CFFINANC, –0.493**, and CFOPER–CFFINANC, –0.227**. CFOPER is
positively related to CASH (0.407**). Concerning shareholders’ components, #SHARE is
positively related to FOREIGN (0.392**) and negatively related to INDIVID (–0.152**).
TAXES is positively related to all four income measures, OPEINCO (0.820**),
ORDIINCO (0.861**), INCOMEBE (0.820**), and INCOME (0.489**), but less so to the
three size variables, #EMPLOY (0.213**), SALES (0.287**), and TOTALAS (0.228**).
TAXES is also positively related to SELLING (0.216**), which is the only related variable
among the expenditure items, including COST. TAXES is positively associated with CASH
Page 95
(0.529**), RETEARN (0.569**), and CFOPER (0.541**), while it is negatively but weakly
associated with CFINVES (–0.145*). TAXES is also weakly but positively associated with
SHARE (0.143*) and FOREIGN (0.205**).
In Japan, the traditional definition of an SME places an emphasis on capital stock.
The EU’s standards for SMEs indicate that the criteria of employee headcount and annual
turnover are innovative, although the balance sheet criterion is more conventional. Company
size relates to the amount of corporate income, inhabitants, and business taxes. It is natural
that the amount of taxes payable relates to income, however measured, but it is interesting
that taxes payable is associated with cash and cash equivalents at the end of the accounting
period. Taxation demands high financial reporting quality. Additionally, taxes payable is
associated with selling and general administration expenses. In order to illustrate their role in
reducing taxes, we carry out our second analysis to examine a novel hypothesis concerning
insurance.
Correlation in the second analysis
Table 4 details the one-sided test results for the Pearson correlation coefficients in the
second analysis. For H4, it is important that CFOPER is negatively related to CFINVES (–
0.684**). This indicates that a decision by a manager to invest in fixed assets, that is, in the
cash flows from investing activities, follows the forecast of cash flows from operating
activities. As a result, CFOPER is negatively related to FIXEDAS (–0.443*). Japanese tax
regulations for fixed assets are mostly the same as the accounting standards. Therefore,
managers recognize that this sort of investment will enable their taxable income to decrease.
FIXEDAS is negatively associated with both INCOMEBE (–0.570**) and INCOME
(–0.505*). This indicates that tangible fixed assets are an essential element of taxation and
that taxes payable will increase unless SMEs invest in plant and equipment under the national
policy. For example, SMEs can participate in the SME Investment Promotion Tax System in
accordance with the Small and Medium-sized Enterprise Basic Act (Article 42-6). However,
not all tangible assets are expensed and therefore deductible from taxable income in the
current accounting period. The Corporation Tax Act (Article 33(2)) includes a provision that
the depreciation of an asset should be on a systematic basis over its useful life. Here, the
Corporation Tax Act prescribes the useful life.
These investments in current and fixed assets lead to H2 and H3. While positively
related to COST (0.457*) and CFFINANC (0.624**), FIXEDAS is negatively related to
TAXES (–0.627**). In contrast, TAXES is positively related to SELLING (0.616**), as well
as INCOMEBE (0.739**) and INCOME (0.603**), and negatively related to CFFINANC
(0.624**), as well as FIXEDAS. Consequently, we suggest that before the end of each
accounting period, managers forecast earnings and then purchase items included in selling
and general administration expenses, which are also deductible from taxable income. It is
likely then that the use of selling and general administration expenses, apart from their
obvious role, is to reduce taxes. More particularly, we need to know the precise day when
each item was paid for or accrued.
In Japan, in accordance with the Ordinance for Enforcement of the Insurance
Business Act, banks must ensure that they will refrain from providing an agency or
intermediary service for the conclusion of an insurance contract, wherein the policyholder or
insured person is a borrower of a business loan, in consideration of any fees or any other
Page 96
remuneration (Article 212). Therefore, SMEs must either purchase an insurance contract for
cash monthly or prepay it in cash yearly. Besides these insurance contracts, Japanese SMEs
can adopt a small-scale enterprises mutual aid system provided by the Organization for Small
and Medium Enterprises and Regional Innovation, Japan. The Commissioner’s Directive on
Interpretation of the Corporate Income Tax Law (9-3-4, etc.) allows all or part of any paid or
pre-paid insurance cost to be deductible from taxable income. The Supreme Court has
decided this matter (Supreme Court, 2012, 1). The national financial system then enables
SMEs to defer income taxes and/or to smooth income in financial accounting, as well as
reducing taxes in tax accounting.
However, the amount of managed earnings should not be so excessive that SMEs lose
the trust of their stakeholders, especially financial institutions and the tax authorities. As
mentioned, financial institutions in Japan usually request that SMEs in need of finance submit
a copy of their financial statements annexed to their final tax returns to which a tax office
superintendent has already affixed a receipt seal. As long as a systematic relationship exists
between these interested parties, it is natural that SMEs will continue to decrease their taxable
income in a limited fashion to retain good long-term relations with their lenders in the
Japanese main bank system.
For H1, it is important that INSUR negatively relates to CFFREE (–0.803**). Free
cash flow is the cash remaining after investment activities and is likely taxable. While an
investment in fixed assets is the preferred and most significant way to decrease cash,
insurance and small equipment purchases are other measures also available to manage cash.
This is shown by the negative relationship between INSUR and CFINVES (–0.569**) and
between INSUR and CASH (–0.727**).
Page 97
Monthly cash payments for expenses
To see whether SME expenses ‘arise in the course of the ordinary activities’ (IASB,
1989, par. 4.33) or are artificially inflated to manage earnings, we investigate the payments in
cash for expenses by month. All of the SMEs in our sample close their books in December.
According to the directors, tax accountants forecast current net income by August. If the forecast
income suggests the imposition of higher tax, it is natural that they then attempt to decrease
income by December in order to lower taxes. Accordingly, we investigate the payments in cash
for expenses by month. Based on H2, our expectation is that payments in the fourth quarter
should be relatively larger.
However, because of the accrual principle, payments made in January will also
potentially include those from the fourth quarter. Accordingly, we include not only cash
payments but also those payments accrued: January plus the fourth quarter in the previous
financial year and the fourth quarter plus January in the following financial year. Table 5
provides descriptive statistics of the payments for insurance and other items by month as a
percentage of cash flows. As shown, both are relatively larger, and therefore appear to
demonstrate the buying behavior of managers immediately before the end of the accounting
period as a means of decreasing net and taxable income.
Table 5
DESCRIPTIVE STATISTICS OF THE PAYMENT OF INSURANCE BY MONTH AS A
PERCENTAGE OF CASH FLOWS
Min. Max. Mean SD
Cash outflows on the basis of cash and 20.90 73.70 47.2650 13.14455
cash equivalents
Cash outflows on the basis of the 15.00 100.00 47.8500 22.37309
accrual principle
Interviews with two insurance companies and a manufacturing corporation
To be confident that the activities of our sample SMEs are not exceptional in Japan, we
conducted interviews concerning the relationships between account days and the date of the
insurance contract or delivery of equipment in the last half of 2014. We conducted the interviews
with the managers of the branches of two national insurance companies and a subsidiary of a
multinational manufacturing corporation.
One insurance company responded to the three questions (see Table 3) as follows. (a) Q1:
there were 90 commercial insurance contracts during the preceding three fiscal years; (b) Q2: 21
of the 90 contracts (23.3%) were sold to customers in the same month as their balance sheet
month; and (c) Q3: only two of the 90 contracts were dissolved. The responses of the other
insurance company were as follows. (a) Q1: there were 18 commercial insurance contracts sold
during the preceding three years; (b) Q2: 12 of the 18 contracts (66.6%) were sold to customers
Page 98
in the same month as their balance sheet month; and (c) Q3: none of the 18 contracts was
dissolved.
For the three questions, the manager of the manufacturing corporation provided us with a
table of the number of orders received by month in 2013 and the number and percentage of
customers purchasing in the same month as their balance sheet month in both money and order
terms. Table 6 provides the essence of these figures together with the monthly percentages of the
balance sheet months of Japanese corporations (NTA, 1997).
Table 6
MANUFACTURER PERCENTAGE OF SALES BY MONTH IN THE SAME MONTH AS
CLIENT BALANCE SHEET-CLOSING MONTH
Month (i) Monthly percentage (ii) Monthly percentage (iii) Monthly percentage (iv) Monthly percentage
of the number of orders of the amount of orders of the number of of the number of all
received in the same received in the same Japanese companies Japanese companies2)
1) 1)
balance sheet month balance sheet month with capital of less than
one million yen2)
1 7.0 12.6 4.5 3.6
2 8.1 11.7 8.0 6.7
3 25.8 37.3 21.5 20.6
4 8.2 9.4 5.8 7.4
5 17.6 11.2 6.3 8.2
6 7.7 6.8 12.9 9.6
7 7.0 7.0 5.7 7.7
8 25.0 3.2 7.1 9.0
9 9.6 6.3 7.6 11.0
10 5.3 4.4 3.7 4.3
11 7.5 26.2 3.4 2.7
12 29.3 45.4 13.3 9.3
Total 100.0 100.0 100.0 100.0
1) During fiscal year 2013.
2) Source is NTA (1997), the most recent year available.
The responses of the manufacturing company to the questionnaire, as shown in Table 6,

indicate that the leading month for (i) was December (29.3%) followed by March (25.8%), while
that for (ii) was December (45.4%) followed by March (37.3%). In particular, based on the
amount of (ii), both November (26.2%) (i.e., just before December) and January (12.6%) and
February (11.7%) (i.e., just before March) were larger. Table 7 provides the correlation
coefficients between these percentages. This suggests two possibilities: that the interviewee dealt
mainly with SMEs, or that only SMEs tend to manage earnings in this way.
Page 99
Table 7
PEARSON CORRELATION COEFFICIENTS FOR THE ITEMS IN TABLE 6
(i) (ii) (iii) (iv)
(i) Monthly percentage of the number of orders received
1
in the same balance sheet month
(ii) Monthly percentage of the amount of orders received
0.621* 1
in the same balance sheet month
(iii) Monthly percentage of the number of Japanese
0.636* 0.604* 1
companies with capital of less than one million yen
(iv) Monthly percentage of the number of all Japanese
0.609* 0.385 0.901** 1
companies
**. Statistically significant at the 1 percent level (one-sided test).
*. Statistically significant at the 5 percent level (one-sided test).
It is interesting that one insurance company interviewee told us that their customers who
purchased commercial insurance in their balance sheet month usually paid their insurance
premiums in annual installments in cash and that they never fell behind in payments. As shown
in Table 3, cash flow from operating activities is negatively related to cash flow from investing
activities. This obviously suggests that even though SMEs do not typically prepare cash flow
statements, managers will still attempt to ensure that cash flow from investing activities moves in
the opposite direction to cash flow from operating activities so as to reduce taxes payable (H4).
Such tax-based earnings management activities require the same amount of money as the
decrease in taxes payable based on cash. As a result, these activities must enable SMEs to
maintain a level of cash and cash equivalents consistent with the previous fiscal year. Overall,
our findings suggest that the FRQ of SMEs is higher (H1).
SUMMARY AND CONCLUSION
This case study first examined the characteristics of Japanese SMEs and then revealed
that the FRQ of family-controlled SMEs tends to be higher if they purchase more insurance with
cash, with insurance payments forecast around the previous August. Importantly, it is easier for
SMEs to make the decision to purchase insurance than for large corporations. Further, Japan’s
accounting standards for SMEs align with the Corporation Tax Act, and therefore their current
net income is taxable income. As a result, good performing SMEs tend to choose the managerial
accounting activity that allows their accounting profits to decrease in order to reduce taxes
payable.
However, SMEs do not decrease their accounting profits, if any, beyond the range of trust
with their lenders, in particular, financial institutions. As is often the case with SMEs, there is a
long-term relationship between the commercial entity and its bank. In the Japanese financial
system, each type of financial institution plays a specific role, so a local financial institution will
emerge as a vital stakeholder, often requesting SMEs to maintain the same level of cash and cash
equivalents as in the previous accounting period.
Page 100
Consequently, only good performing SMEs can choose an accounting procedure that
allows their accounting profits to decrease based on cash flows. It is certain that revenue will
decrease because of their tax-saving management. However, after considering the role of
insurance companies in the financial system, Japanese governance mechanisms can partly justify
SME earnings management. This is why the yield from insurance from the SME perspective is
the same as a bank deposit.
The results of our study contrast with those of Hope et al. (2013). One reason is that we
do not have accounting information for as many SMEs. In addition, we do not fully reveal the
relationship between SMEs and their stakeholders outside banks and the government. For
instance, former government officials often take up positions in private enterprise, while
government corporations are a hotbed for severance agreements. Moreover, we do not
investigate whether individual managers play a significant role in determining the level of tax
avoidance that SMEs undertake. These represent promising research subjects for the future.
REFERENCES
Ali, A., T. Chen & S. Radhakrishman (2007). Corporate disclosures by family firms. Journal of Accounting and
Economics, 44(1/2), 238–286.
Arai, K. & S. Shratori (1991). Legal and conceptual framework of accounting in Japan. Kaikei, 140(3), 4–17.
Cascino, S., A. Pugliese, D. Mussolino & C. Sansone (2010). The influence of family ownership on the quality of
accounting information. Family Business Review, 23(3), 246–265.
Chau, G. & S.J. Gray (2010). Family ownership, board independence and voluntary disclosure: Evidence from Hong
Kong. Journal of International Accounting, Auditing and Taxation, 19(2), 93–109.
Chen, C.J.P. & B. Jaggi (2000). Association between independent non-executive directors, family control and
financial disclosures in Hong Kong. Journal of Accounting and Public Policy, 19(4/5), 283–310.
Chen, S., X. Chen & Q. Chen (2008). Do family firms provide more or less voluntary disclosure? Journal of
Accounting Research, 46(3), 499–536.
Dedman, E., A. Kausar & C. Lennox (2014). The demand for audit in private firms: recent large-sample evidence
from the UK. European Accounting Review, 23(1), 1–23.
EU (2005). The new SME definition, User guide and model declaration.
Financial Services Agency (2011). Access FAS, No. 98 (Japanese). Retrieved July 28, 2015, from
http://www.fsa.go.jp/access/23/201109b.html
Holthausen, R.W. & R.W. Leftwich (1983). The economic consequences of accounting choice: Implications of
costly contracting and monitoring. Journal of Accounting and Economics, 5(2), 77–117.
Hope, O.-K., W.B. Thomas & D. Vyas (2013). Financial reporting quality of US private and public firms. The
Accounting Review, 88(5), 1715–1742.
IASB (1989). The conceptual framework for financial reporting, Chapter 4: 1989 Framework: The remaining text.
IASB (2014). IFRS application around the world, Jurisdictional profile: Japan, last updated 4 March 2014.
Retrieved January 28, 2015, from http://www.ifrs.org/Use-around-the-world/Documents/Jurisdiction-
profiles/Japan-IFRS-Profile.pdf
Jaggi, B. & S. Leung (2007). Impact of family dominance on monitoring of earnings management by audit
committees: Evidence from Hong Kong. Journal of International Accounting, Auditing and Taxation,
16(1), 27–50.
Jaggi, B., S. Leung & F. Gul (2009). Family control, board independence and earnings management: Evidence based
on Hong Kong firms. Journal of Accounting and Public Policy, 28(4), 281–300.
Page 101
Japanese Bankers Association (2012). Life insurance & accident insurance: Guidance notes on compliance
(Japanese). Retrieved July 29, 2015, from
http://www.zenginkyo.or.jp/fileadmin/res/abstract/efforts/contribution/compliance/guidance.pdf
Jensen, M.C. & W.H. Meckling (1976). Theory of the firm: Managerial behavior, agency theory and ownership
structure. Journal of Financial Economics, 3, 305–360.
Jiraporn, P. & P. J. DaDalt (2009). Does founding family control affect earnings management? Applied Economics
Letters, 16(2), 113–119.
Kamata, N. (1998). Cash flow and income (Japanese). Kigyo Kaikei, 50(8), 4–11.
Kuroda, M. (2001). Japan – Group accounts. In Ordelheide, D. and KPMG (Eds.), Transnational Accounting
(Second Edition) (pp. 1807–1907). Basingstoke: Palgrave.
Michie, N. (2014). Possibility of financial statements for deciding whether a supplier will request its business
continuance of an SME (Japanese). Unpublished master’s thesis. Osaka Prefecture University.
Nakakuki, M. (2002). Review of empirical studies on the relationship between accounting information and
managerial incentives (Japanese). Discussion paper No. 2002-J-36. IMES, Bank of Japan.
National Tax Agency, Japan (1997). The number of corporations by the closing month (Japanese). Retrieved March
9, 2015, from http://www.nta.go.jp/kohyo/tokei/kokuzeicho/hojin1997/11.htm
OECD (1987). Accounting standards harmonization, No. 3: The relationship between taxation and financial
reporting, income tax accounting. Paris: OECD.
Okabe, Y. (1994). Theory as to financial reporting (Japanese). Tokyo: Moriyama Shoten.
Organization for Small & Medium Enterprises and Regional Innovation, JAPAN (2015). About SME Support,
JAPAN. Retrieved March 9, 2015, from http://www.smrj.go.jp/english/index.html
Prencipe, A. & S. Bar-Yosef (2011). Corporate governance and earnings management in family-controlled
companies. Journal of Accounting, Auditing and Finance, 26(2), 199–227.
Prencipe, A., S. Bar-Yosef & H.C. Dekker (2014). Accounting research in family firms: Theoretical and empirical
challenges. European Accounting Review, 23(3), 361–385.
Sakurai, H. (2001). Japan – Individual accounts. In Ordelheide, D. and KPMG (Eds.), Transnational Accounting
(Second Edition) (pp. 1685–1805). Basingstoke: Palgrave.
Sato, M. (1998). What is the cash flow statement? (Japanese). Kigyo Kaikei, 50(10), 38–44.
Small and Medium Enterprise Agency (2010). Basic research on SMEs in Japan: Fiscal year 2012 (Japanese).
Retrieved December 10, 2014, from http://www.e-stat.go.jp/SG1/estat/List.do?lid=000001119762
Suda, K. (2004). Positive theory of financial accounting (Japanese). Tokyo: Hakuto Shobo.
Supreme Court (2012). January 13, 2012 Judgement. Minshu (Japanese), 66(1). Retrieved July 1, 2015, from
http://www.courts.go.jp/app/hanrei_en/detail?id=1134
Suzuki, K. (2013). Analysis of tax accounting (Japanese). Tokyo: Moriyama Shoten.
Tomioka, Y. (2014). Largest corporations do not pay taxes (Japanese). Tokyo: Bungei Shunjyu.
Tong, Y.H. (2008). Financial reporting practices of family firms. Advances in Accounting, 23, 231–261.
Toyo Keizai (2014). Japan company handbook (Japanese), 2014(3). Tokyo: Toyo Keizai.
Tsuji, M. & M. Fujibayashi (2013). Environmental accounting of small and medium-sized entities following the
2011 Tohoku–Pacific Ocean Earthquake. Journal of Legal, Ethical & Regulatory Issues, 16(2), 17–28.
Villalonga, B. & R. Amit (2006). How do family ownership, control and management affect firm value? Journal of
Financial Economics, 80(2), 385–417.
Wang, D. (2006). Founding family ownership and earnings quality. Journal of Accounting Research, 44(3), 619–
656.
Watts, R.L. & J.L. Zimmerman (1986). Positive accounting theory. Englewood Cliffs, NJ: Prentice-Hall.
Yang, M.L. (2010). The impact of controlling families and family CEOs on earnings management. Family Business
Review, 23(3), 266–279.
Page 102
Appendix A
VARIABLE DEFINITIONS
Variables Definitions
#EMPLOY Number of employees
SALES Total sales
NETAS Net assets
TOTALAS Total assets
NETAS Change in NETAS in the current year from the previous year
LIQAS Liquid assets
LIQAS Change in LIQAS in the current year from the previous year
FIXEDAS Fixed assets
FIXEDAS Change in FIXEDAS in the current year from the previous year
COST Cost of SALES
DEPRECI Depreciation
INSUR Insurance payable
SELLING Selling and general administration expenses
NONOPER Nonoperating expenses
EXTRAOR Extraordinary losses
TAXES Corporate income, inhabitants, and business taxes
INCOMEBE Current net income before TAXES and tax adjustments
INCOME Current net income
CFOPER Cash flow from operating activities
CFINVES Cash flow from investing activities
CFFINANC Cash flow from financing activities
CASH Net increase (decrease) in cash and cash equivalents
CASH Cash and cash equivalents at end of period
CFFREE CFOPER plus CFINVES
#SHARE Number of shareholders
FOREIGN Percentage of foreign shareholders
INDIVID Percentage of individual shareholders and others
OPEINCO Operating income
ORDIINCO Ordinary income
EQUITY Percentage of shareholders’ equity
CAPITAL Capital stocks
RETEARN Retained earnings
BORROW Borrowings and bonds
Page 103
BIG 4 AUDITOR OFFICE SIZE, ANALYSTS’ ANNUAL

EARNINGS FORECASTS AND CLIENT EARNINGS
MANAGEMENT BEHAVIOR
Sarah Garven, Morehead State University
Gary Taylor, The University of Alabama
ABSTRACT
Prior research finds support for clients of larger Big 4 audit offices having less
aggressively managed earnings as evidenced by a lower likelihood of meeting the earnings
benchmarks of small profits and small earnings increases. We extend this research by examining
the analysts’ forecasts earnings benchmark. The extant literature provides evidence that failing
to meet analysts’ forecasts results in severe negative market reactions and auditors are less
likely to book audit differences that affect the client’s ability to meet analysts’ expectations.
Specifically, we examine the relationship between Big 4 auditor office size and two measures of
earnings management relative to analysts’ earnings forecasts: (1) minimization of analysts’
forecast error and (2) just meeting or beating analysts’ forecasts. Using a sample of 9,789 U.S.
companies audited by the Big 4 auditors, we find that the reported earnings of clients audited by
larger Big 4 offices are associated with increased absolute levels of analysts’ forecast error and
a decreased likelihood of just meeting or beating analysts’ earnings forecasts. These results
suggest that auditors in larger Big 4 offices are more likely to constrain management’s ability to
manage earnings to meet analysts’ forecasts, providing further evidence that auditors of the
same Big 4 accounting firm are not achieving consistent audit quality across the firm’s offices.
Our results can be useful for investors, researchers, and regulators interested in auditor
influence on analysts’ forecast error as well as Big 4 accounting firms seeking to ensure
consistency of audit outcomes across offices.
INTRODUCTION
The extant literature provides evidence to suggest that larger audit firm offices provide
higher quality audits than smaller audit offices (Francis & Yu, 2009; Choi, Kim, Kim & Zang,
2010; Francis, Michas & Yu, 2013). This study extends the office-level audit quality literature by
investigating the association between Big 4 office size and client earnings management related to
analysts’ annual earnings forecasts. The question of whether larger Big 4 audit offices are more
likely than smaller Big 4 audit offices to constrain earnings management around analysts’
forecasts is important given that (1) management’s zeal to meet consensus analysts’ forecasts in
recent years has called into question the quality of SEC registrants’ reported earnings and the
quality of their auditors (Levitt, 1998) and (2) the Big 4 firms audit “more than 98 percent of the
global market capitalization of U.S. issuers” (Franzel, 2013).1
The office-level audit quality literature has examined audit quality in terms of whether
auditors seem to influence earnings management behavior relative to two earnings
benchmarks—small positive profits and small earnings increases (Francis & Yu, 2009). Based on
these benchmarks, the larger Big 4 offices appear to provide higher quality audits than the
smaller Big 4 offices. Specifically, companies audited by auditors associated with the larger Big
Page 104
4 offices are less likely to report small profits or small earnings increases, indicating less
aggressive earnings management behavior. While prior literature indicates that auditors place
great importance on preventing management from managing earnings around these two earnings
benchmarks, auditors appear to be less concerned about preventing earnings management around
the analysts’ annual earnings forecasts benchmark (Nelson, Elliott & Tarpley, 2002; Frankel,
Johnson & Nelson, 2002; Ng, 2007). Conversely, management appears to place greater
importance on meeting analysts’ forecasts than avoiding losses and/or reporting small earnings
increases (Dechow, Richardson & Tuna, 2003; Brown & Caylor, 2005; Graham, Harvey &
Rajgopal, 2005 & 2006). Thus, it is an empirical question whether the association between Big 4
office size and earnings management around earnings benchmarks found in prior literature will
hold for the analysts’ forecasts benchmark.
We use two measures to assess whether the positive association between Big 4 office size
and earnings quality found by Francis and Yu (2009) translates to analysts’ annual earnings
forecasts: (1) minimization of the absolute value of analysts’ forecast error; and (2) just meeting
or beating analysts’ forecasts. Payne (2008) argues that the absolute value of forecast error will
capture management’s overall tendency to manage earnings around forecasts. It is very possible
that firms seek to minimize positive, as well as negative, forecast error (Payne & Robb, 2000;
Graham et al., 2005; Payne, 2008). For example, firms may want to minimize positive forecast
errors in order to create accrual reserves for the future (Graham et al., 2005; Payne, 2008). In this
study, higher absolute levels of analysts’ forecast errors represent lower levels of earnings
management.
Because missing analysts’ forecasts yields adverse consequences, firms that see that
unmanaged earnings will fall short of analysts’ annual earnings forecasts may attempt to “push”
earnings upward to meet the forecast. Consequently, some firms that just meet or beat analysts’
earnings expectations may have managed earnings upwards when the firm may have fallen short
of the analysts’ annual earnings benchmark otherwise (Reichelt & Wang, 2010). In our study, we
use firms that just meet or beat analysts’ annual earnings forecast benchmarks to represent higher
levels of earnings management.
To test our hypotheses, we conduct OLS and logistic regressions of SEC registrants
audited by domestic Big 4 auditors from 2003-2008. We find that larger Big 4 audit offices are
associated with increased absolute levels of analysts’ forecast error and a decreased likelihood of
just meeting or beating analysts’ annual earnings forecasts. We interpret these results to suggest
that firms audited by larger Big 4 audit offices are less likely to manage earnings to analysts’
annual earnings forecasts. In other words, the larger the Big 4 office, the more likely it is that the
auditor constrains earnings management related to analysts’ forecasts. The results from this
paper contribute to our understanding of auditor influence on analysts’ forecast error, reinforce
the relevance of office-level audit analysis, and provide additional evidence that audit quality
does not appear to be uniform across the varying local offices of the Big 4. These results are
robust to extensive sensitivity analyses.
BACKGROUND AND HYPOTHESIS DEVELOPMENT
In this section, we review related research and develop our hypotheses. We first discuss
the motivations behind earnings management, particularly as they relate to analysts’ forecasts.
We then consider the association between auditor industry specialization and analysts’ forecast
error. The audit literature is mixed as to whether or not auditors who are industry specialists are
more likely to constrain earnings management relative to analysts’ earnings forecasts.2 Next, we
Page 105
discuss audit firm office-level literature, which in general suggests that audit quality may not be
consistent across offices within the same firm. Finally, we state our hypotheses concerning the
relationship between Big 4 audit firm office size and client earnings management around
analysts’ annual earnings forecasts.
Earnings Management Incentives
Earnings management is “a purposeful intervention in the external financial reporting

process, with the intent of obtaining some private gain (as opposed to, say, merely facilitating the
neutral operation of the process)” (Schipper, 1989, 92). Healy and Wahlen (1999, 370) note that
incentives to manage earnings relate to anti-trust or other government regulation, lending and
compensation contracts written in terms of accounting numbers, and capital market expectations
and valuation. Earnings management can be used to allow the firm/management to: 1) appear
less profitable if under regulatory scrutiny by the government, 2) circumvent industry regulation,
3) avoid potential violation of lending contracts, and 4) increase earnings-based bonus awards
(Healy & Wahlen, 1999). With respect to capital market motivations, Healy and Wahlen point
out that “the widespread use of accounting information by investors and financial analysts to
help value stocks can create an incentive for managers to manipulate earnings in an attempt to
influence short-term stock price performance” (370-371). Consistent with this capital market
incentive, survey studies find that management often engages in earnings management to meet or
beat analysts’ forecasts (Graham et al., 2005 & 2006).
Empirical and anecdotal evidence indicate firms missing analysts’ earnings forecasts can
incur substantial penalties. These penalties can include, but are not limited to, lower stock
returns, reduced credibility, and reduced CEO compensation (see Levitt, 1998; Matsunaga &
Park, 2001; Bartov, Givoly & Hayn, 2002; Skinner & Sloan, 2002; Lopez & Rees, 2002; Graham
et al., 2005; Brown & Caylor, 2005; Koh, Matsumoto & Rajgopal, 2008). Failure to meet
earnings expectations can also result in costly earnings-based securities litigation (Francis,
Philbrick & Schipper, 1998; Payne, 2008). Auditors are aware of management’s incentives to
manage earnings to forecasted levels (SEC, 1999; Libby & Kinney, 2000; Habib & Hansen,
2008), and can potentially constrain management’s ability to do so.3
Analysts’ Forecast Error and Industry Specialization
Archival research examining auditors who are industry specialists and their influence on
constraining earnings management around the analysts’ forecast benchmark yields mixed results.
Payne (2008) finds that national industry specialist auditors constrain management’s ability to
manage earnings around analysts’ forecasts; specifically, clients of Big N national industry
specialists are associated with earnings that increase absolute levels of analysts’ forecast error
and are less likely to meet or to beat analysts’ forecasts. Reichelt and Wang (2010) examine not
only national-level industry expertise, but also city-level and joint national- and city-level
industry expertise. They find that, when the auditor is a city-level industry expert, either alone or
in conjunction with a national industry expert, its clients are less likely to meet or just beat
analysts’ earnings forecasts by one penny per share. However, their results do not extend to
auditors who are only national experts. Behn, Choi, and Kang (2008) do not find an association
between national industry specialists and analysts’ forecast error in their Big N sample but do
find that non-Big N national industry specialists are associated with earnings that decrease
absolute levels of analysts’ forecast error.
Page 106
Office-Level Analysis
Individual offices appear to play an important role in determining audit quality (see
Ferguson, Francis & Stokes, 2003; Francis, Reichelt & Wang, 2005; Krishnan, 2005a; Jones,
Krishnan & Melendez, 2008; Francis & Yu, 2009; Reichelt & Wang, 2010; Choi et al. 2010;
Choi, Kim, Qiu & Zang, 2012; Francis et al. 2013). Many argue that, because most audit
decisions with respect to a particular client are made by individual practice offices, analyses in
audit research should focus on the office-level, rather than the national-level (Wallman, 1996;
Francis, Stokes & Anderson, 1999; Reynolds & Francis, 2000; Craswell, Stokes & Laughton,
2002; DeFond & Francis, 2005). Office-level analysis is especially important if audit quality is
not uniform across different practice offices within the same audit firm.
Among prior office-level studies is evidence that, within the Big 4, audit quality differs
based on office size, with larger offices providing higher quality audits than smaller offices
(Francis & Yu, 2009; Notbohm, 2010; Choi et al. 2010; Francis et al. 2013). Francis and Yu
(2009) suggest that this association exists because the larger Big 4 audit offices (1) have more
collective experience in administering public company (SEC registrant) audits, (2) are expected
to have greater expertise than smaller Big 4 audit offices in addressing complex accounting
issues and in detecting and deterring aggressive earnings management behavior, (3) have more
office-level peers with whom to consult, resulting in a better local support network, and (4) are
better equipped to handle staff turnover.
Francis and Yu (2009) investigate the association of office size with client earnings
properties (abnormal accruals and likelihood of meeting the small profits and small earnings
increases benchmarks) with the objective of determining if there is variation in audit quality
across Big 4 practice offices of different sizes. They find that larger offices are less likely to have
large abnormal accruals and less likely to report small profits and small earnings increases.
These results indicate clients of larger Big 4 audit offices are associated with less aggressively
managed earnings, consistent with larger offices providing higher quality auditing. Choi et al.
(2010) also examine the relationship between office size and abnormal accruals and, similar to
Francis and Yu (2009), find that abnormal accruals are negatively related to office size. Francis
et al. (2013) investigate the association between office size and client restatements. They argue
that client restatements are a more direct measure of low quality audits than the measures used in
the earlier studies. Francis et al. (2013) find that Big 4 office size is negatively associated with
client restatements, suggesting that larger offices provide higher quality audits through better
enforcement of the correct application of generally accepted accounting principles (GAAP).
Lastly, Notbohm (2010) examines if auditor office size is a factor used by the market to assess
the assurance value component of auditor reputation. He reports higher earnings response
coefficients and lower cost of equity for clients with auditors from larger offices of the Big 4,
suggesting that investors perceive that these larger offices provide a higher level of assurance.
The above studies provide evidence that differences in auditor attributes, namely office
size, can result in variation in earnings quality and perceptions of earning quality. However, one
question regarding audit and earnings quality that studies have not yet addressed is whether the
size of a Big 4 audit office influences client earnings management behavior around analysts’
forecasts. Specifically, is office size associated with the absolute value of analysts’ forecast error
and the likelihood of just meeting or beating analysts’ forecasts? This is an especially important
question given that several studies provide evidence to suggest that auditors focus on
constraining earnings management around the small profits and small earnings increases
benchmarks (Nelson et al. 2002; Frankel et al. 2002; Ng, 2007) and management focuses on
Page 107
attempting to meet or beat the analysts’ earnings forecasts benchmark (Dechow et al. 2003;
Brown & Caylor, 2005; Graham et al. 2005 & 2006).
Hypothesis Development
There are three primary observable earnings benchmarks: (1) analysts’ forecasts, (2) loss
avoidance (small profits), and (3) earnings increases (Graham et al. 2005 & 2006). Recent
research suggests that management has much stronger incentives to manage earnings to achieve
analysts’ forecasts than to achieve small profits and small earnings increases (Dechow et al.
2003; Brown & Caylor, 2005; Graham et al. 2005 & 2006). Management’s focus on analysts’
annual earnings forecasts likely is related to the potential for a severe market reaction to missing
the consensus forecast (Graham et al. 2005 & 2006). In fact, executives believe that missing the
consensus forecast leads to a larger negative stock price reaction than when they miss the other
two earnings benchmarks (Graham et al. 2006). Brown and Caylor (2005) find evidence
consistent with this belief as there is a greater positive (negative) reaction for firms that meet
(miss) analysts’ forecasts than for firms that meet (miss) the earnings benchmarks of loss
avoidance or earnings increases.
Unlike management, auditors, however, appear to be more focused on constraining
earnings management related to the reporting of small profits or small earnings increases than for
meeting analysts’ forecasts (Nelson et al. 2002; Frankel et al. 2002; Ng, 2007). For example,
Nelson et al. (2002) interview Big N auditors and report that auditors are more likely to require
an adjustment related to an earnings management attempt to increase earnings or avoid a loss
than an earnings management attempt to meet analysts’ forecasts, despite identifying meeting
analyst expectations as the most common incentive for earnings management attempts.
Similarly, Ng (2007) finds that Big 4 auditors are more likely to book an audit adjustment
when the adjustment affects the client’s ability to achieve positive earnings or earnings that
exceed prior year’s earnings, and less likely when it affects the client’s ability to meet the
analysts’ consensus forecast. Frankel et al. (2002), in a study of the relation between non-audit
fees and earnings management to achieve small earnings increases and to just meet or beat
analysts’ consensus forecasts, find a significant positive relationship between non-audit fees and
just meeting or beating analysts’ forecasts but no relationship between non-audit fees and
reporting a small earnings increase. The authors suggest that the difference in findings may be
due to “managers have[ing] greater incentives to manage earnings to meet market expectations
and/or auditors have[ing] less incentives to prevent this behavior” (p. 91).
Audit literature posits various factors that may contribute to influencing the auditors’
decision as to whether to book or waive audit differences as the client’s earnings approach
different earnings benchmarks (Libby & Kinney, 2000; Frankel et al., 2002; Nelson et al., 2002;
Ng, 2007; Ng & Tan, 2007). These studies suggest the following as to why auditors may be more
likely to waive audit differences that allow the client to meet the analysts’ earnings forecasts
benchmark: 1) economic incentives/sympathy to client’s preferences, i.e., auditors believe
managers have a very strong preference for meeting the analysts’ forecasts; 2) differential
salience of the different thresholds, i.e., the auditor is less likely to focus on the analysts’
earnings forecasts threshold when making materiality assessments; 3) differential risk associated
with the different thresholds, i.e., auditors perceive a lower risk of litigation associated with
missing the analysts’ earnings forecasts benchmark relative to the profit/loss benchmark; and 4)
differential materiality implications associated with the different thresholds, i.e., the audit
Page 108
difference is less likely to be considered quantitatively material relative to the earnings base in
the analysts’ forecasts condition than in the small profits condition.
There are two perspectives on the effect of office size on client earnings management
behavior around analysts’ forecasts. As noted earlier, Francis and Yu (2009) find that clients of
larger offices are less likely to achieve the small profits and small earnings benchmarks. Based
on Francis and Yu’s (2009) arguments and findings, one would expect that larger Big 4 offices
are more likely than the smaller offices to identify and report problems in the financial
statements and insist on corrections, even if these corrections result in management failing to
meet analysts’ earnings forecasts. At the same time, as discussed above, the evidence suggests
that management has much stronger incentives to manage earnings to achieve analysts’ forecasts
than to achieve small profits and small earnings increases (Dechow et al. 2003; Brown & Caylor,
2005; Graham et al. 2005 & 2006) and auditors are less likely to insist on corrections that result
in missed analysts’ forecasts than on those that result in failing to report positive earnings or
small earnings increases (Nelson et al. 2002; Frankel et al. 2002; Ng, 2007). Consequently, the
positive association between office size and earnings quality found by Francis and Yu (2009)
may not translate to a setting in which management appears to have stronger incentives to
manage earnings and auditors appear to have a lower propensity to prevent them from managing
earnings. Thus, one could argue that auditor influence on the absolute value of analysts’ forecast
error and just meeting or beating analysts’ forecasts will be uniform across the Big 4 offices and
that office size will not be associated with these two earnings quality measures.
Given these two different perspectives on the effect of office size on client earnings
management behavior around analysts’ forecasts, we test the following hypotheses, presented in
null form:
H10: There is no association between Big 4 office size and absolute levels of analysts’ forecast error.
H20: There is no association between Big 4 office size and reported earnings that just meet or beat
analysts’ earnings forecasts.
RESEARCH DESIGN AND VARIABLE MEASUREMENT
The following models are used empirically to test H1 and H2, respectively
ERROR = β₀ + β₁OFFICE + β₂SIZE + β₃LOSS + β₄NUMEST + β₅ERNCHG + β₆PERSIST +

β₇DISP + β8HORIZON + β9SHARES + β10ABSACC + β11MTB + β12REVGROW + β13ZFC +
β14TENURE + β15CITYEXP + β16NATEXP + β17INFLUENCE + β18GEOSEGS + β19OPSEGS +
β20YEAR + β21IND + ε (1)
MEETBEAT= β₀ + β₁OFFICE + β₂SIZE + β₃LOSS + β₄NUMEST + β₅ERNCHG + β₆PERSIST

+ β₇DISP + β8HORIZON + β9SHARES + β10ABSACC + β11MTB + β12REVGROW + β13ZFC +
β14TENURE + β15CITYEXP + β16NATEXP + β17INFLUENCE + β18GEOSEGS + β19OPSEGS +
β20YEAR + β21IND + ε (2)
Page 109
where:
ERROR = absolute value of analysts’ forecast error defined as the difference between
I/B/E/S actual earnings per share in the current year and the most recent mean
I/B/E/S consensus analysts’ EPS forecast prior to the earnings announcement
date;4
MEETBEAT = 1 when actual earnings per share in the current year exactly meet or beat the
analysts’ mean consensus forecast by one cent per share, and 0 otherwise;
OFFICE = natural log of total office-specific audit fees of all SEC-registrant clients per
fiscal year;
SIZE = natural log of the market value of equity at the beginning of the year;5
LOSS = 1 if net income is less than zero, and 0 otherwise;
NUMEST = number of analysts following the company;
ERNCHG = absolute value of the change in annual earnings scaled by beginning of year
stock price;
PERSIST = 1 if ERNCHG is within the 20-80th percentile of earnings changes by year,
and 0 otherwise;
DISP = a measure of analysts’ forecast dispersion calculated as the standard deviation
across analysts’ forecasts divided by the absolute value of the mean EPS
forecast;
HORIZON = number of calendar days between the forecast announcement date and the
subsequent actual earnings announcement date;
SHARES = natural log of common shares outstanding at the end of the year;
ABSACC = absolute value of accruals calculated as net income minus cash flow from
operations scaled by lagged total assets;
MTB = market-to-book ratio at the end of the year;
REVGROW = change in sales deflated by lagged total assets;
ZFC = an index of financial condition based on Zmijewski’s (1984) weighted probit
bankruptcy prediction model. The financial condition measure is computed
using the weighted probit coefficients from Zmijewski’s Table 3, Panel B:
( ) ( ) ( )
TENURE = 1 if auditor tenure is three years or less, and 0 otherwise;

CITYEXP = 1 if an auditor is the number one auditor in terms of aggregated client audit
fees in an industry within that city in a specific fiscal year, and 0 otherwise;
NATEXP = 1 if an auditor is the number one auditor in terms of aggregated audit fees in
an industry within a specific fiscal year, and 0 otherwise;
INFLUENCE = ratio of a specific client’s total fees (audit fees plus nonaudit fees) relative to
aggregated annual fees generated by the practice office that audits the client;
GEOSEGS = number of geographic segments reported in the Compustat segments database
for the firm in year t;6
OPSEGS = number of operating segments reported in the Compustat segments database
for the firm in year t;
YEAR = dummy variables for each year from 2003-2007; and
IND = dummy variables representing two-digit SIC codes.
Page 110
In Equation (1), the dependent variable, ERROR, is measured by the absolute value of
forecast error, following Payne (2008). This variable captures management’s overall propensity
to manage earnings around analysts’ forecasts, as managers have incentives to manage earnings
both up to and down to the forecasted level (Payne, 2008). The dependent variable,
MEETBEAT, in Equation (2) is an indicator variable for reported earnings that exactly meet or
beat analysts’ earnings forecasts by one cent per share, following Reichelt and Wang (2010) and
Payne (2008). MEETBEAT is intended to capture management’s propensity to “push” earnings
upward when the firm otherwise may have fallen short of its benchmark earnings target (Reichelt
and Wang, 2010). Auditor office size (OFFICE) is the variable of interest in the study and,
consistent with Francis and Yu (2009) and Francis et al. (2013), is measured by the natural log of
7
total office-specific audit fees of all clients per fiscal year. Similar to these studies, the log of
audit fees, rather than aggregate audit fees, is used in the analysis due to skewness in the
distribution and some extreme values of aggregate audit fees. As noted by Francis and Yu (2009,
1524), “audit fees are directly related to engagement hours, and offices with higher fees will
therefore have more hours of experience in the audits of SEC registrants.”
The following variables are included in the models to control for factors shown in prior
research to influence analysts’ forecast error. SIZE is included to control for information
environment asymmetry between large and small firms (for example, large firms are scrutinized
more by investors and analysts) (Payne, 2008). The analyst forecast literature has also used SIZE
to control for numerous omitted variables (Becker, DeFond, Jiambalvo & Subramanyam, 1998).
There is disagreement in prior research on the size effect on analysts’ forecast error and meeting
or beating forecasted earnings, so we do not make a prediction as to its effect on the dependent
variables.
Firms that report losses (LOSS), have significant changes in earnings (ERNCHG), have
higher accrual levels (ABSACC), and are financially distressed (ZFC) are expected to provide
less predictable earnings (Payne, 2008; Behn et al. 2008).8 Less predictable earnings would
likely lead to increased forecast error and a reduced frequency of just meeting or beating
analysts’ forecasts. The opposite is expected for high-growth companies (MTB), companies with
strong sales growth (REVGROW), and companies with consistent earnings (PERSIST) (Barton
& Simko, 2002; Payne, 2008; Reichelt & Wang, 2010; Hill, Johnson, Liu & Lopez, 2015).
Consistent with Barton and Simko (2002) and Hill et al. (2015), the natural log of
common shares outstanding (SHARES) is included since firms with a larger number of shares
may find it more difficult to meet analysts’ forecasts and thus have greater forecast error. As
noted by Barton and Simko (2002), a $.01 deficit in EPS translates into more dollars of actual
earnings for firms with many shares as opposed to firms with fewer shares. The number of
calendar days between the forecast announcement date and the corresponding actual earnings
announcement date (HORIZON) is included because it is expected that a shorter forecast horizon
will be associated with less forecast error and thus an increased likelihood of just meeting or
beating forecasted earnings (Brown, 2001; Behn et al. 2008).
Finally, we include the dispersion of the analysts’ forecasts (DISP) and the number of
analysts following the company (NUMEST). Payne (2008) argues that when analysts are in
agreement (low dispersion) management is more likely to utilize their discretion over earnings to
minimize analysts’ forecast error and their firm’s earnings will be more likely either to meet or
beat analyst expectations as a result. Higher analyst following indicates that companies are
followed more closely by Wall Street and thus expected to be more likely just to meet or beat
Page 111
analysts’ expectations and to have lower analysts’ forecast error (Barton & Simko, 2002; Payne,
2008; Behn et al. 2008; Reichelt & Wang, 2010).
We also include the following six variables as recommended by Francis and Yu (2009) to
ensure that the results for office size are not the consequence of correlated omitted auditor –
related variables. Client influence (INFLUENCE) is included to control for the auditor’s office-
level incentives regarding influential clients. DeAngelo’s (1981) economic bonding theory
indicates that auditors may compromise audit quality in order to retain clients that are
economically important to the audit firm. However, Reynolds and Francis (2000) find that Big 4
auditors actually report more conservatively for their larger clients, as the incentive to avoid
litigation from misreporting outweighs the incentive to report in a client’s favor in order to retain
them as a client. Thus, the predicted effect of client influence on the two earnings quality
measures is unclear. Auditor tenure (TENURE) is included to ensure that systematic differences
in auditor tenure across practice offices do not influence the results associated with office size.
Short auditor tenure has been found in previous studies to be associated with lower quality audits
(Geiger & Raghunandan, 2002; Johnson, Khurana & Reynolds, 2002).
Auditor national industry expertise (NATEXP) and auditor city-specific industry
expertise (CITYEXP) are included to ensure that office size is not capturing an omitted variable
regarding auditor industry expertise. Numerous studies have demonstrated that national and city-
specific industry experts are associated with higher quality audits (Gramling, Johnson &
Khurana, 2001; Owhoso, Messier & Lynch, 2002; Krishnan, 2003; Carcello & Nagy, 2004;
Dunn & Mayhew, 2004; Krishnan, 2005b; Payne, 2008; Jones et al. 2008; Reichelt & Wang,
2010).9 The number of client geographic (GEOSEGS) segments and the number of operating
segments (OPSEGS) are included, as engagement offices with clients with multiple geographical
segments or operating units are more likely to require assistance from multiple audit practice
offices to assist them in completing the audit, which may, in turn, result in a higher quality audit.
Therefore, these two variables are included to control for any possible confounding effects of
other offices participating with the lead office in the audit engagement.
Consistent with previous studies (e.g., Payne, 2008; Reichelt & Wang, 2010; Choi et al.
2012), we include dummy variables for both year and industry. As noted by Payne (2008), these
year and industry dummy variables control for time- and industry-specific commonalities in the
coefficients in order to reduce correlation among the regression residuals.
SAMPLE AND RESULTS
The sample covers the six-year period from 2003 through 2008 based on Compustat year
definitions.10 The audit engagement office is determined from the audit report letterhead from the
10-K SEC filing as reported in Audit Analytics. Following Francis and Yu (2009), annual office
size is based on total yearly audit fees for each office in Audit Analytics, and the full population
of observations is used for the calculation. Additionally, auditor industry leadership is based on
the full Audit Analytics population with fee data. However, as in Francis et al. (2005), Jones et
al. (2008), and Reichelt and Wang (2010), to ensure that city industry leadership is not
determined by too few observations in a city-industry-fiscal year combination, we require a
minimum of two observations per city-industry-year combination and further delete observations
for which we could not identify metropolitan statistical area information needed for
determination of city industry leadership.11 Additional data reductions include non-Big 4
auditors, financial institutions (SIC codes between 6000 and 6999), observations without IBES
analysts’ forecasts and actual values or with forecasts announced more than two months before
Page 112
the earnings announcement date12,13, observations without a matching CIK code, and
observations without control variable values. Consistent with Payne (2008), we eliminate non-
Big 4 auditors to control for auditor type, and we also eliminate financial institutions as their
accrual structure is qualitatively different.
The final sample has 9,789 firm-year observations, representing 2,598 different
companies, 255 individual practice audit offices, and 1,309 office-year observations.14,15
Following Teoh and Wong (1993) and Frankel et al. (2002), we focus on annual earnings
because auditors’ involvement with quarterly financial statements is limited. All non-
dichotomous variables in the models were winsorized at the 1st and 99th percentiles to mitigate
the effect of outliers.
Descriptive statistics for this sample are presented in Table 1. The mean absolute forecast
error (ERROR) is $0.084. Approximately 13.3 percent of observations just meet or beat analysts’
earnings forecasts (MEETBEAT). On average, almost nine analysts follow a firm in our sample
(NUMEST). The mean number of calendar days between the forecast announcement date and
actual earnings announcement (HORIZON) is 16.765 days.
Average audit fees (AUDFEES) are $83.462 million and 23.7 percent of the sample firms
have an auditor-client relationship of three years or less (TENURE). A total of 51.9 percent of
the sample firms are audited by city industry leaders (CITYEXP), 31.4 percent of the sample
firms are audited by national industry leaders (NATEXP), and, on average, a client’s total fees
represent 6.7 percent of office-level fees (INFLUENCE). Additionally, sample firms have an
average of 3.051 geographic segments (GEOSEGS) and 1.486 operating segments (OPSEGS).
On average, our firms are not in financial distress as the mean financial distress score
(ZFC) is -2.432. The average price-to-book ratio (MTB) is 3.233, the average dispersion measure
(DISP) is 0.068, and the absolute value of total accruals (ABSACC) represents approximately
8.6 percent of total assets. The mean natural log of market value of equity at the beginning of the
year (SIZE) is $7.014 billion and the mean natural log of common shares outstanding at the end
of the year (SHARES) is 4.144 million. Approximately 25.4 percent of the firm-year
observations reported a loss (LOSS). The mean increase in total revenue deflated by total assets
(REVGROW) is 13.2 percent while the average change in earnings (ERNCHG) is 6.5 percent.
Additionally, 60 percent of firms are classified as having persistent earnings (PERSIST).
Page 113
Table 1
POOLED FIRM-YEAR DESCRIPTIVE STATISTICS (2003-2008); N=9,789
Variables Mean St. Dev. 25th Percentile Median 75th Percentile
ERROR 0.084 0.163 0.010 0.030 0.080
MEETBEAT 0.133 0.339 0.000 0.000 0.000
AUDFEES 83.462 98.565 24.808 55.763 106.700
OFFICE 17.696 1.116 17.027 17.837 18.486
SIZE 7.014 1.557 5.891 6.870 7.975
LOSS 0.254 0.436 0.000 0.000 1.000
NUMEST 8.786 6.318 4.000 7.000 12.000
ERNCHG 0.065 0.135 0.009 0.021 0.056
PERSIST 0.600 0.490 0.000 1.000 1.000
DISP 0.068 0.149 0.010 0.022 0.054
HORIZON 16.765 8.670 11.000 14.000 25.000
SHARES 4.144 1.145 3.343 3.936 4.762
ABSACC 0.086 0.085 0.033 0.062 0.109
MTB 3.233 4.218 1.581 2.414 3.909
REVGROW 0.132 0.210 0.019 0.090 0.201
ZFC -2.432 1.618 -3.551 -2.466 -1.462
TENURE 0.237 0.425 0.000 0.000 0.000
CITYEXP 0.519 0.500 0.000 1.000 1.000
NATEXP 0.314 0.464 0.000 0.000 1.000
INFLUENCE 0.067 0.115 0.010 0.025 0.069
GEOSEGS 3.051 2.486 1.000 2.000 4.000
OPSEGS 1.486 1.612 1.000 1.000 1.000
Variable definitions are contained in Table 2.
Table 2
VARIABLE DESCRIPTIONS
Variable Name Description
ERROR Absolute value of analyst forecast error defined as the difference between I/B/E/S actual
earnings per share in the current year and the most recent mean I/B/E/S consensus analysts’
EPS forecast prior to the earnings announcement date.
MEETBEAT 1 when actual earnings per share in the current year exactly meet or beat the analysts’ mean
consensus forecast by one cent per share, and 0 otherwise.
AUDFEES Total office-specific audit fees (in millions) of all SEC registrant clients per fiscal year.
OFFICE Natural log of total office-specific audit fees of all SEC registrant clients per fiscal year.
SIZE Natural log of the market value of equity at the beginning of the year.
LOSS 1 if net income is less than zero, and 0 otherwise.
NUMEST Number of analysts following the company.
ERNCHG Absolute value of the change in earnings per share divided by beginning of year stock price.
PERSIST 1 if ERNCHG is within the 20-80th percentile of earnings changes by year, and 0 otherwise.
DISP Measure of analysts’ forecast dispersion calculated as the standard deviation across analysts’
forecasts divided by the absolute value of the mean EPS forecast.
HORIZON Number of calendar days between the forecast announcement date and the subsequent actual
earnings announcement date.
SHARES Natural log of common shares outstanding at the end of the year.
ABSACC Absolute value of accruals calculated as net income-cash flow from operations scaled by
lagged total assets.
MTB Market-to-book ratio at the end of the year.
REVGROW Change in sales deflated by lagged total assets.
Page 114
Table 2
VARIABLE DESCRIPTIONS (Continued)
Variable Name Description
ZFC An index of financial condition based on Zmijewski’s (1984) weighted probit bankruptcy
prediction model. The financial condition is computed using the weighted probit coefficients
from Zmijewski’s Table 3, Panel B:
( ) ( ) ( )
TENURE 1 if auditor tenure is three years or less, and 0 otherwise.

CITYEXP 1 if an auditor is the number one auditor in terms of aggregated client audit fees in an industry
within that city in a specific fiscal year, and 0 otherwise.
NATEXP 1 if an auditor is the number one auditor in terms of aggregated audit fees in an industry within
a specific fiscal year, and 0 otherwise.
INFLUENCE Ratio of a specific client’s total fees (audit fees plus nonaudit fees) relative to aggregated
annual fees generated by the practice office that audits the client.
GEOSEGS Number of geographic segments reported in the Compustat segments database for the firm in
year t.
OPSEGS Number of operating segments reported in the Compustat segments database for the firm in
year t.
Pearson correlations (not reported) were calculated. The results indicate that office size is
significantly correlated with ERROR and MEETBEAT in the expected directions. Although the
correlations among a few variables are above 0.35, an analysis of the variance inflationary
factors (VIFs) for each model reveals that all are well below the commonly recommended 10.00
threshold, indicating that multicollinearity is not a concern.
Table 3
OLS REGRESSION RESULTS
Panel A: Dependent Variable Analysts’ Forecast Error (Model 1)
Variable Pred. Coefficient p-value
Intercept N/A - 0.164 <0.001
OFFICE (+) 0.008 <0.001
SIZE (?) 0.046 <0.001
LOSS (+) 0.037 <0.001
NUMEST (-) -0.003 <0.001
ERNCHG (+) 0.219 <0.001
PERSIST (-) 0.005 0.129
DISP (+) 0.171 <0.001
HORIZON (+) 0.000 0.069
SHARES (+) -0.055 <0.001
ABSACC (+) 0.132 <0.001
MTB (-) -0.001 <0.001
REVGROW (-) 0.020 0.010
ZFC (+) 0.008 <0.001
TENURE (-) 0.001 0.738
CITYEXP (+) 0.004 0.160
NATEXP (+) -0.004 0.224
INFLUENCE (?) 0.075 <0.001
GEOSEGS (?) 0.001 0.180
OPSEGS (?) 0.000 0.679
Page 115
Table 3
OLS REGRESSION RESULTS (Continued)
Model F-value = 27.92; p <0.001. Adjusted R2 = 17.85%. n=9,789.
Note: Coefficient p-values represent two-tailed significance. Estimates on fiscal year dummies and industry
dummies are not reported for brevity.
Panel B: Dependent Variable Analysts’ Forecast Error (Model 1)
Alternate Office Size Measures
Alternate Office Size Pred. Coefficient p-value
Measure
OFFICE2 (+) 0.008 <0.001
OFFICE3 (+) 0.007 <0.001
OFFICE4 (+) 0.005 <0.001
OFFICE2=natural log of total (audit plus non-audit) office-specific fees (in millions) of all SEC registrant clients
per fiscal year;
OFFICE3=natural log of total number of SEC registrant clients audited by each office per fiscal year.
OFFICE4=natural log of total sales of SEC registrant clients by each office per fiscal year.
Table 3 presents the OLS regression results for the analysts’ forecast error test.16 All p-
values are reported as two-tailed probabilities. The model is statistically significant (F-
value=27.92; p < 0.001) with an adjusted R2 of 17.85%. Panel A of Table 3 reports results for the
main office size variable of interest, OFFICE, the natural log of office-specific audit fees. The
coefficient for OFFICE is positive and significant at the p < 0.001 level. This result indicates that
clients audited by larger Big 4 audit offices are more likely to have increased absolute levels of
analysts’ forecast error. Among the control variables, of those for which directional predictions
are made, eight are significant in the expected direction (LOSS, NUMEST, ERNCHG, DISP,
HORIZON, ABSACC, MTB, and ZFC), two are significant in the opposite direction (SHARES
and REVGROW), and the remaining are not significant at the p < 0.10 level. Of the four
variables for which we did not make a directional prediction, SIZE and INFLUENCE are
positively significant while GEOSEGS and OPSEGS are insignificant at the p < 0.10 level.
Panel B of Table 3 reports results using three alternative measures of office size.
OFFICE2 is the natural log of office-specific total (audit plus non-audit) fees. OFFICE3 is the
natural log of the total number of SEC registrant clients audited by an office in a fiscal year and
OFFICE4 is the natural log of total sales of SEC registrant clients by each office per fiscal year.
The coefficients for all these measures are positive and significant at the p < 0.001 level,
indicating that the office size results demonstrated in Panel A are robust to using alternative
office size measures.
Page 116
Table 4
LOGISTIC REGRESSION RESULTS
Panel A: Dependent Variable Meet/Beat Forecast (Model 2)
Variable Pred. Coefficient p-value
Intercept N/A 0.369 0.592
OFFICE (-) -0.121 0.001
SIZE (?) -0.297 <0.001
LOSS (-) -0.316 0.002
NUMEST (+) 0.005 0.476
ERNCHG (-) -1.475 <0.001
PERSIST (+) -0.077 0.243
DISP (-) -0.354 0.156
HORIZON (-) -0.003 0.436
SHARES (-) 0.570 <0.001
ABSACC (-) -0.477 0.304
MTB (+) 0.026 <0.001
REVGROW (+) -0.022 0.891
ZFC (-) -0.115 <0.001
TENURE (+) -0.066 0.384
CITYEXP (-) -0.040 0.539
NATEXP (-) 0.030 0.654
INFLUENCE (?) -0.913 0.021
GEOSEGS (?) 0.008 0.558
OPSEGS (?) 0.002 0.937
Model χ2 = 350.41; p <0.001. Pseudo R2 = 7.58%. n=9,789.
Note: Coefficient p-values represent two-tailed significance. Estimates on fiscal year dummies and industry
dummies are not reported for brevity.
Panel B: Dependent Variable Meet/Beat Forecast (Model 2)
Alternate Office Size Measures
Alternate Office Size Pred. Coefficient p-value
Measure
OFFICE2 (-) -0.136 <0.001
OFFICE3 (-) -0.093 0.008
OFFICE4 (-) -0.096 0.001
OFFICE2=natural log of total (audit plus non-audit) office-specific fees (in millions) of all SEC registrant clients
per fiscal year;
OFFICE3=natural log of total number of SEC registrant clients audited by each office per fiscal year.
OFFICE4=natural log of total sales of SEC registrant clients by each office per fiscal year.
Table 4 presents the logistic regression results for the just meeting or beating analysts’
earnings forecasts test. P-values are reported as two-tailed probabilities. The model is statistically
significant (χ2=350.41; p < 0.001) with a pseudo R2 of 7.58%. Panel A of Table 4 reports results
for the main office size variable of interest, OFFICE, the natural log of office-specific audit fees.
The coefficient for OFFICE is negative and significant at the 0.001 level. This indicates that
clients audited by larger Big 4 audit offices are less likely just to meet or beat analysts’ earnings
forecasts.
Among the control variables, of those for which directional predictions are made, four are
significant in the expected direction (LOSS, ERNCHG, MTB, and ZFC), one is significant in the
opposite direction (SHARES), and the remaining are not significant at the p < 0.10 level. Of the
four variables for which we did not make a directional prediction, SIZE and INFLUENCE are
negatively significant while GEOSEGS and OPSEGS are insignificant at the p < 0.10 level.
Page 117
Panel B of Table 4 reports results using the three alternative measures of office size. The
coefficients for these measures are all negative and significant at the p < 0.01 level. The results
using these alternative office size measures indicate that the office size results demonstrated in
Panel A are robust to alternative office size measures.
In summary, the results of the two models indicate that the earnings of SEC clients
audited by larger Big 4 audit offices are of higher quality, as evidenced by audited reported
earnings with increased absolute levels of analysts’ forecast error and with a reduced likelihood
of just meeting or beating analysts’ earnings forecasts. Thus, larger audit offices appear to be
more likely to constrain earnings management behavior around analysts’ forecasts.
SENSITIVITY ANALYSES
We perform numerous sensitivity analyses to verify the robustness of the findings. In

untabulated results, we determine that the main results are robust to (1) using median analysts’
EPS forecasts in lieu of mean analysts’ forecasts, (2) using actual national and city level market
shares of audit fees instead of indicator variables to measure industry leadership (3) performing
Fama-MacBeth regressions to control for cross-sectional correlation of residuals, and (4) using
various specifications of industry classification schemes, namely Fama-French 5, 12, 17, 30, and
48 industry groupings along with one-digit SIC codes.17
Additionally, similar to Payne (2008) and Hill et al. (2015), to ensure that the results are
not altered by expectations management, we perform an untabulated analysis including an
indicator variable for settings in which it appears that analysts’ forecasts have been guided
downward. After controlling for this effect, the reported results do not change with respect to
office size. Also, since it is possible that firms just meeting or beating analysts’ forecasts are
generating small profits or small increases, the office size results from Equation (2) may simply
be mimicking the results from Francis and Yu (2009). To ensure this is not the case, we
eliminate from the sample firms that generated small profits or small earnings increases (as
defined by Francis and Yu 2009) and re-estimate the logistic equation. Additionally, in a separate
analysis, we rerun the logistic equation with the inclusion of indicator variables for small
earnings increases and small profits. In both cases, the untabulated results are qualitatively
similar to those presented.
In order to ensure that the main results are not associated with the economic downturn of
2008, we run both regression equations for the years 2003-2007. In addition, we rerun both main
regressions controlling for client sales volatility. Again, we do not find any qualitative change to
the reported results.
Habib and Hansen (2008) propose that additional research is needed to examine earnings
benchmarks based on firm characteristics. They suggest examining these benchmarks based on
prominent firm characteristics such as small versus large firms, high leverage versus low
leverage firms, and high versus low analyst following firms. Based on their suggestion, we
perform tests separately for (1) large clients (i.e., firms greater than or equal to median size)
versus small clients, (2) high leverage firms (i.e., firms with leverage greater than or equal to
median leverage) versus low leverage firms, (3) profit-reporting firms versus loss-reporting
firms, and (4) firms with a high analyst following (i.e., firms with analyst followings greater than
or equal to median analyst followings) versus firms with a low analyst following.18
Not all companies and audit offices appear in all six years of the sample. In order to
ensure that the results are not affected by survivorship bias, we run tests separately for (1)
companies that appear in all six years and (2) audit offices that appear in all six years. Finally, to
Page 118
examine whether the effect of office size on analysts’ forecast error differs systematically
between clients with positive forecast errors and negative forecast errors, we split the full sample
into two sub-samples based on the sign of the forecast error, and then re-estimate Equation (1)
separately for each sub-sample.
Table 5
SENSITIVITY ANALYSES
COEFFICIENT AND P-VALUE FOR OFFICE VARIABLE
Panel A: Analysts’ Forecast Error Panel B: Just Meeting or Beating
Analysts’ Earnings Forecasts
Sub-Sample n Coefficient p-value Coefficient p-value
Large Clients 4,895 0.014 <0.001 -0.143 0.014
Small Clients 4,894 0.008 0.001 -0.140 0.008
High Leverage 4,895 0.012 0.000 -0.194 0.001
Low Leverage 4,894 0.007 <0.001 -0.112 0.027
Profit Firms 7,298 0.007 <0.001 -0.132 0.002
Loss Firms 2,491 0.014 0.006 -0.155 0.070
High Follow 5,234 0.010 <0.001 -0.138 0.007
Low Follow 4,555 0.010 0.002 -0.128 0.027
Same Firm 4,818 0.011 <0.001 -0.151 0.007
Same Office 9,507 0.008 <0.001 -0.141 <0.001
Positive Error 6,656 0.008 <0.001 N/A N/A
Negative Error 3,133 0.012 0.004 N/A N/A
Table 5 presents the estimation results for the large versus small, high versus low
leverage, high versus low analyst following, and profit versus loss-reporting subsamples, as well
as the subsamples for the companies and audit offices included in all years of the sample, along
with companies with positive or negative forecast errors. For the sake of brevity, only results for
the office size variable, OFFICE, are presented. Panel A of Table 5 reports the OLS regression
results for the analysts’ forecast error tests. In all twelve tests, OFFICE is positive and significant
at the p < 0.01 level. Panel B of Table 5 presents the logistic regression results for the just
meeting or beating analysts’ earnings forecasts tests. In all ten tests, OFFICE is negative and
significant at the 0.07 level or less.
The results from the sensitivity analyses are all consistent with the main results in that the
earnings of clients audited by larger offices are associated with increased absolute levels of
analysts’ forecast error and reduced likelihood of just meeting or beating analysts’ earnings
forecasts.
SUMMARY AND CONCLUSIONS
Our study extends prior audit office-level research by examining the relationship between
Big 4 auditor office size and two management incentives that influence earnings management
behavior around analysts’ earnings forecasts: (1) minimization of analysts’ forecast error and (2)
just meeting or beating analysts’ forecasts. The findings are consistent with larger offices
producing higher quality audits. Specifically, we find that the earnings of clients audited by
larger Big 4 audit offices are associated with increased absolute levels of analysts’ forecast error
and a decreased likelihood of just meeting or beating analysts’ forecasts, suggesting that the
larger Big 4 offices are associated with higher client earnings quality. These results support
Francis and Yu’s (2009) argument that larger Big 4 offices have greater in–house expertise and
Page 119
experience in auditing SEC registrants and thus are more likely to identify, report, and insist on
correcting problems in their financial statements.
This study contributes to our understanding of auditor influence on client earnings quality
by providing evidence that larger Big 4 offices are more likely than smaller offices to constrain
management’s ability to manage earnings to analysts’ forecasts. As mentioned earlier, several
studies provide evidence suggesting that auditors are not as concerned about the analysts’
earnings forecasts benchmark. We provide evidence that, in reality, the auditors of larger Big 4
audit offices are concerned about the analysts’ earnings forecasts benchmark. This finding is
important given that management’s zeal to meet consensus analysts’ forecasts in recent years has
called into question the quality of SEC registrants’ reported earnings and the quality of their
auditors (Levitt, 1998) and the Big 4 firms audit most of the global market capitalization of U.S.
issuers (Franzel, 2013). Furthermore, the study reinforces the relevance of office-level audit
analysis and provides additional evidence that audit quality does not appear to be uniform across
the different local offices of the Big 4, supporting suggestions that the Big 4 accounting firms
need to mandate stricter policies to ensure audit quality consistency across offices and the
PCAOB should consider increasing the number of inspections they perform on audits conducted
by smaller offices.
Our paper is subject to the following limitations. First, since neither audit quality nor
earnings management is observable, we place reliance on proxy measures that, although used in
prior literature, are not without criticism. For example, since earnings management is
unobservable, we cannot determine if all firms in the sample with low levels of absolute
analysts’ forecast error or that just met or beat analysts’ earnings forecasts achieved these results
by managing their earnings. Second, as Francis and Yu (2009) note, although multiple offices
may contribute to an audit, the audit is attributed only to the engagement office of record;
however, if small offices participate on audits of large offices (and vice versa), such instances
would counteract office-size differences and would bias against finding an association between
office size and earnings quality. Third, one can argue that the total of office-specific audit fees is
a noisy measure of office-size due to the confounding of fees with client characteristics.
Although client count and client sales are used in the study as alternative measures of office size
and produce similar results, there potentially may be more precise office size measures or
measures superior to office size to capture differences in audit quality provided by local auditing
firms that to-date have not yet been identified or for which the data is not publicly available at
the present time.
Opportunities for future research include examining the association between Big 4
auditor office size and other facets of financial reporting quality, including conservatism, SEC
enforcement, auditor litigation, or the cost of debt financing. Future studies could also investigate
the relationship between second-tier auditor office size and various measures associated with
financial reporting quality. Finally, future research could attempt to determine if the influence of
auditor office size on earnings benchmarks extends to interim quarters.
Page 120
ENDNOTES
1. Based on 2012 year-end data.
2. It is important to note that industry specialization and office size are two different dimensions on which
auditors can be sorted.
3. For example, in an experiment, Libby and Kinney (2000) examine whether Big N audit managers expect
management to correct earnings after auditors discover quantitatively immaterial earnings misstatements.
They find that auditors believe that management has strong preferences for achieving the analysts’
consensus forecast as their results indicate that auditors believe that misstatements are less likely to be
corrected if they cause earnings to fall below analysts’ forecasts. This result implies “opportunistic
correction of quantitatively immaterial misstatements to manage earnings to forecasts, and auditor
acceptance of this practice” (Libby & Kinney, 2000, 385).
4. Results are qualitatively similar in the OLS and logistic regressions if we use the median instead of the
mean consensus analysts’ forecast. Additionally, by using the alternative measure of error found in Behn et
al. (2008) in which the absolute value of the error is deflated by stock price, we obtain similar results.
5. Results are qualitatively similar in Models 1 and 2 if we use the natural log of beginning year’s total assets
in place of the market value of equity.
6. Following Francis and Yu (2009), we assign GEOSEGS and OPSEGS a value of 1 if no segment data is
reported in Compustat.
7. We also examine three alternative measures of office size: natural log of total fees, natural log of number
of clients, and natural log of total sales of clients. See Panel B of Tables 3 and 4 for results.
8. Higher values of the Zmijewski financial condition (ZFC) index indicate more financial distress.
9. Results are qualitatively similar in Models 1 and 2 if we include an indicator variable for joint national and
city-specific industry expertise.
10. Fee data is only available in Audit Analytics for fiscal year 2000 financial statements and later. The SEC,
in 2000, required all registrants to disclose the most recent year’s audit and nonaudit fees in annual proxy
statements filed on or after February 5, 2001. We begin the sample period with 2003, the first year of the
Big 4, to remove the potential effect of Arthur Andersen.
11. Following Reichelt and Wang (2010) and Francis et al. (2005), in determining city-specific industry
expertise, a city is defined as a Metropolitan Statistical Area (MSA). We obtained the geographical city
from Audit Analytics and determined its MSA by hand-collecting from the following U.S. Census Bureau
website: http://www.census.gov/population/www/metroareas/lists/2007/List4.txt. For the majority of
cities not listed on this website, we were able to look up their county on the following website:
http://www.citycountyxref.info/index.lasso and then determine their MSA from the U.S. Census Bureau
website. Forty-seven different geographical cities, representing 464 observations, are not located within any
MSA and were thus removed from the sample before merging with Compustat.
12. As in Reichelt and Wang (2010) and Lim and Tan (2008), this is to ensure that forecasts are not stale-dated.
13. Consistent with Reichelt and Wang (2010), Lim and Tan (2008), and Payne (2008), to calculate analysts’
forecast error, we obtain analysts’ forecasts from the I/B/E/S unadjusted summary file to avoid problems
of lost precision in the EPS decimal places from stock splits (Payne & Thomas, 2003). Actual earnings
also come from I/B/E/S to ensure consistency of the earnings construct.
Page 121
14. The 9,789 firm-years in the sample are audited by 255 different Big 4 offices, which are distributed as
follows: PricewaterhouseCoopers (61), Ernst & Young (67), Deloitte (61), and KPMG (66). Each audit
office can appear up to six times in the sample (2003, 2004, 2005, 2006, 2007, and 2008).
15. There were 66,355 domestic companies on Audit Analytics from 2003-2008. Observations were excluded
because they were (1) audited by non-Big 4 auditors (26,703), (2) financial institutions (18,997), (3)
without analysts’ annual earnings forecasts (4,188), (4), without matching CIK code (2,275), (5) missing
MSA information or had fewer than two observations (1,958), or (6) missing control variables (2,445).
16. Estimates on fiscal year dummies and industry dummies are not reported for the sake of brevity.
17. Fama-French industry definitions were obtained from Kenneth R. French’s website:
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/Data_Library/changes_ind.html.
18. For the meet or beat loss firms sample, rather than use two-digit SIC codes to represent industry
classifications, we assign industries based on one-digit SIC codes. We do this to prevent quasi-separation
of data points in the logistic regressions for these two models. As noted by Paul D. Allison in his 2008
paper, “Convergence Failures in Logistic Regression”, “the most common cause of quasi-complete
separation is a dummy predictor variable such that, for one level of the variable, either every observation
has the event or no observation has the event. For those cases in which the problem variable is one of a set
of variables representing a single categorical variable, the problem can often be easily solved by combining
categories” (p.7).
Page 122
REFERENCES
Allison, P. (March, 2008). Convergence failures in logistic regression. Presented at the SAS Global Forum, San
Antonio, TX. Retrieved February 18, 2015, from http://www2.sas.com/proceedings/ forum2008/360-
2008.pdf.
Barton, J. & P. Simko. (2002). The balance sheet as an earnings management constraint. The Accounting Review,
77(Supplement), 1-27.
Bartov, E., D. Givoly, & C. Hayn. (2002). The reward to meeting or beating earnings expectations. Journal of
Accounting and Economics, 33, 173-204.
Becker, C., M. DeFond, J. Jiambalvo, & K. Subramanyam. (1998). The effect of audit quality on earnings
management. Contemporary Accounting Research, 15(1), 1-24.
Behn, B., J. Choi, & T. Kang. (2008). Audit quality and properties of analyst earnings forecasts. The Accounting
Review, 83(2), 327-349.
Brown, L. (2001). A temporal analysis of earnings surprises: Profits versus losses. Journal of Accounting Research,
39(2), 221-241.
Brown, L. & M. Caylor. (2005). A temporal analysis of quarterly earnings thresholds: Propensities and valuation
consequences. The Accounting Review, 80(2), 423-440.
Carcello, J. & A. Nagy. (2004). Client size, auditor industry specialization and fraudulent financial reporting.
Managerial Auditing Journal, 19(5), 651-668.
Choi, J., C. Kim, J. Kim, & Y. Zang. (2010). Audit office size, audit quality and audit pricing. Auditing: A Journal
of Practice & Theory, 29(1), 73-97.
Choi, J., J. Kim, A. Qiu, & Y. Zang. (2012). Geographic proximity between auditor and client: How does it impact
audit quality? Auditing: A Journal of Practice & Theory, 31(2), 43-72.
Craswell, A., D. Stokes, & J. Laughton. (2002). Auditor independence and fee dependence. Journal of Accounting
and Economics, 33(2), 253-275.
DeAngelo, L. (1981). Auditor size and audit quality. Journal of Accounting and Economics, 3(3), 183-199.
Dechow, P., S. Richardson, & I. Tuna. (2003). Why are earnings kinky? An examination of the earnings
management explanation. Review of Accounting Studies, 8(2-3), 355-384.
DeFond, M. & J.R. Francis. (2005). Audit research after Sarbanes-Oxley. Auditing: A Journal of Practice & Theory,
Dunn, K. & B. Mayhew. (2004). Audit firm industry specialization and client disclosure quality. Review of
Accounting Studies, 9(1), 35-58.
Ferguson, A., J. R. Francis, & D. Stokes. (2003). The effects of firm-wide and office-level industry expertise on
audit pricing. The Accounting Review, 78(2), 429-448.
Francis, J., D. Philbrick, & K, Schipper. (1998). Earnings surprises and litigation risk. Journal of Financial
Statement Analysis, 98(3), 15-27.
Francis, J. R., D. Stokes, & D. Anderson. (1999). City markets as a unit of analysis in audit research and the re-
examination of Big 6 market shares. Abacus, 35(2), 185-206.
Francis, J. R., K. Reichelt, & D. Wang. (2005). The pricing of national and city-specific reputations for industry
expertise in the U.S. audit market. The Accounting Review, 80(1), 113-136.
Francis, J. R. & M. Yu. (2009). Big 4 office size and audit quality. The Accounting Review, 84(5), 1521-1552.
Francis, J. R., P. Michas, & M. Yu. (2013). Office size of Big 4 auditors and client restatements. Contemporary
Accounting Research, 30(4), 1626-1661.
Frankel, R., M. Johnson, & K. Nelson. (2002). The relation between auditors’ fees for nonaudit services and
earnings management. The Accounting Review, 77(Supplement), 71-105.
Franzel, J. (July, 2013). Accountability: Protecting Investors, the Public Interest and Prosperity. Presented at the
Association of Government Accountants 62nd Annual PDC: Big Challenges, Bigger Thinking, Dallas, TX.
Retrieved September 20, 2013, from http://pcaobus.org/News/Speech/Pages/07172013_AGA.aspx.
Geiger, M, & K. Raghunandan. (2002). Auditor tenure and audit reporting failures. Auditing: A Journal of Practice
& Theory, 21(1), 67-78.
Graham, J., C. R. Harvey, & S. Rajgopal. (2005). The economic implications of corporate financial reporting.
Journal of Accounting and Economics, 40(1-3), 3-73.
Graham, J., C. R. Harvey, & S. Rajgopal. (2006). Value destruction and financial reporting decisions. Financial
Analysts Journal, 62(6), 27-39.
Page 123
Gramling, A., V. Johnson, & I. Khurana. (2001). The association between audit firm industry experience and
financial reporting quality. Working paper, Georgia State University.
Habib, A. & J. Hansen. (2008). Target shooting: Review of earnings management around earnings benchmarks.
Journal of Accounting Literature, 27, 25-70.
Healy, P. & J. Wahlen. (1999). A review of the earnings management literature and its implications for standard
setting. Accounting Horizons, 13(4), 365-383.
Hill, M., P. Johnson, K. Liu, & T. Lopez. (2015). Operational restructurings: Where’s the beef? Forthcoming,
Review of Quantitative Finance and Accounting.
Johnson, V., I. Khurana, & J. Reynolds. (2002). Audit firm tenure and the quality of financial reports.
Contemporary Accounting Research, 19(4), 637-660.
Jones, K., G. Krishnan, & K. Melendez. (2008). The impact of office-level vs. firm-level auditors’ industry expertise
on conservatism. Working paper, George Mason University, Lehigh University, and New Mexico State
University.
Koh, K., D. Matsumoto, & S. Rajgopal. (2008). Meeting or beating analyst expectations in the post-scandals world:
Changes in stock market rewards and managerial actions. Contemporary Accounting Research, 25(4),
1067-1098.
Krishnan, G. (2003). Does Big 6 auditor industry expertise constrain earnings management? Accounting Horizons,
Krishnan, G. (2005a). Did Houston clients of Arthur Andersen recognize publicly available bad news in a timely
manner? Contemporary Accounting Research, 22(1), 165-193.
Krishnan, G. (2005b). The association between Big 6 auditor industry expertise and the asymmetric timeliness of
earnings. Journal of Accounting, Auditing, and Finance, 20(3), 209-228.
Levitt, A. (September, 1998). The numbers game. Presented at New York University, Center for Law and Business,
New York, NY. Retrieved September 20, 2013, from http://www.sec.gov/news/speech/speecharchive/1998/
spch220.txt.
Libby, R. & W. Kinney, Jr. (2000). Does mandated audit communication reduce opportunistic corrections to manage
earnings to forecasts? The Accounting Review, 75(4), 383-404.
Lim, C. & H. Tan. (2008). Non-audit service fees and audit quality: The impact of auditor specialization. Journal
of Accounting Research, 46(1), 199-246.
Lopez, T. & L. Rees. (2002). The effect of beating and missing analysts’ forecasts on the information content of
unexpected earnings. Journal of Accounting, Auditing, and Finance, 17(2), 155-184.
Matsunaga, S. & C. Park. (2001). The effect of missing a quarterly earnings benchmark on the CEO’s annual bonus.
The Accounting Review, 76(3), 313-332.
Nelson, M., J. Elliott, & R. Tarpley. (2002). Evidence from auditors about managers’ and auditors’ earnings
management decisions. The Accounting Review, 77(Supplement), 175-202.
Ng, T. (2007). Auditors’ decisions on audit differences that affect significant earnings thresholds. Auditing: A
Journal of Practice & Theory, 26(1), 71-89.
Ng, T. & H. Tan. (2007). Effects of qualitative factor salience, expressed client concern, and qualitative materiality
thresholds on auditors’ audit adjustment decisions. Contemporary Accounting Research, 24(4), 1171-1192.
Notbohm, M. (2010). Auditor office level size and auditor reputation. Unpublished doctoral dissertation, Florida
State University.
Owhoso, V., W. Messier, Jr., & J. Lynch. (2002). Error detection by industry-specialized teams during sequential
audit review. Journal of Accounting Research, 40(5), 883-900.
Payne, J. & S. Robb. (2000). Earnings management: the effect of ex ante earnings expectations. Journal of
Accounting, Auditing, and Finance, 15(4), 371-392.
Payne, J. & W. Thomas. (2003). The implications of using stock-split adjusted I/B/E/S data in empirical research.
The Accounting Review, 78(4), 1049-1067.
Payne, J. (2008). The influence of audit firm specialization on analysts’ forecast errors. Auditing: A Journal of
Practice & Theory, 27(2), 109-136.
Reichelt, K. & D. Wang. (2010). National and office-specific measures of auditor industry expertise and effects on
audit quality. Journal of Accounting Research, 48(3), 647-686.
Reynolds, J. & J. R. Francis. (2000). Does size matter? The influence of large clients on office-level auditor
reporting decisions. Journal of Accounting and Economics, 30(3), 375-400.
Schipper, K. (1989). Commentary on earnings management. Accounting Horizons, 3(4), 91-102.
Page 124
Securities and Exchange Commission (SEC). (1999). Staff Accounting Bulletin No. 99: Materiality. Washington,
DC: SEC. Retrieved December 1, 2013 from http://www.sec.gov/interps/account/sab99.htm.
Skinner, D. & R. Sloan. (2002). Earnings surprises, growth expectations, and stock returns or don’t let an earnings
torpedo sink your portfolio. Review of Accounting Studies, 7, 289-312.
Teoh, S. & T. Wong. (1993). Perceived auditor quality and the earnings response coefficient. The Accounting
Review, 68(2), 346-366.
Wallman, S. (1996). The future of accounting, Part III: Reliability and auditor independence. Accounting Horizons,
10(4), 76-97.
Zmijewski, M. (1984). Methodological issues related to the estimation of financial distress prediction models.
Journal of Accounting Research, 22(Supplement), 59-82.
Page 125
AUDIT FEES, EARNINGS MANAGEMENT, AND

LITIGATION RISK: EVIDENCE FROM JAPANESE
FIRMS CROSS-LISTED ON U.S. MARKETS
Junjian Gu, Nagoya University
ABSTRACT
Many studies on audit fees have verified that audit fees reflect risk differences across
litigation regimes. However, research on the association between audit fees and earnings
management from the perspective of different litigation environments has been lacking. Using a
sample of Japanese firms cross-listed in U.S. markets, this paper aims to determine the
correlation between audit fees and different litigation environments and whether the extent of the
correlation between audit fees and earnings management in Japanese litigation environment is
different from the U.S. litigation environment. In this paper, we use the propensity score
matching model to control for differences in firm characteristics between two litigation
environments (Japan and U.S.) and the regression model to test three hypotheses on correlations
among audit fees, earnings management risk, and litigation risk. We found that there is a
difference in audit fees under different litigation environment and audit fees increase with higher
litigation risk. Further analyses show that high earnings management risk is correlated with
high audit fees and audit fees originated from earnings management risk is reduced under
greater litigation risk environment.
INTRODUCTION
Accounting research on litigation risk and audit fees beganin 1980. Simunic (1980) found
that audit fees reflect risk differences across litigation regimes. From 1984 to 1999, researchers
had done several empirical research to find evidence linking litigation risk to audit fees.
However, their findings are inconclusive due to lack of a large sample of audit fees data (Francis,
1984; Chung & Lindsay, 1988; Chan, Ezzamel & Gwilliam, 1993; Johnson, Walker &
Westergaard, 1995; Craswell & Francis, 1999). Seetharaman, Gul & Lynn (2002) used
cross-listed firms data and found that U.K. auditors charge higher fees for their services when
their U.K. based clients are cross-listed in U.S. markets, but not for U.K. firms cross-listed in
non-U.S. markets. The finding suggests that audit fees reflect risk differences across different
litigation regimes. Choi, Kim, Liu & Simunic (2009) examined data from 14 countries and
asserted that auditors charge higher fees for firms that are cross-listed in stronger legal regimes.
Based on previous literature, we can observe that audit fees are influenced by the difference in
litigation environment. In other words, litigation risk is an important determinant of audit fees.
Another fundamental determinant of audit fees is earnings management risk. Bedard &
Johnstone (2004) found that auditors respond to earnings management risk with ex ante increase
in planned audit hours and billing rates. Abbott, Parker & Peters (2006) found that audit fees are
Page 126
inversely related to income-decreasing discretionary accruals. They also found that the positive
relationship between audit fees and positive discretionary accruals is magnified for high P/E
firms. Our paper uses both of these aspects ‒ litigation and earnings management risk ‒ to
analyze audit fees.
We find methodology limitation in existing audit fees research when analyzing samples
of cross-listed and non-cross-listed firms. For examples, Seetharaman, Gul & Lynn (2002)
matched the sample by size and industry, while Choi, Kim, Liu & Simunic (2009) matched the
sample by country, year, and industry. They did not analyze the sample using a more rigorous
matching model. Recent audit research suggests that the potential threat of selection bias in
auditing research is likely to occur when researchers compare large and small accounting firms
that have different characteristics. Lawrence, Minutti-Meza, & Zhang (2011) found that after
using the propensity score matching (PSM) models, the treatment effect of Big 4 auditors turned
to be insignificantly different from those of non-Big 4 auditors with respect to their clients’
discretionary accruals.
To the best of our knowledge, empirical research on the association between audit fees
and earnings management using Japanese data is still limited. Yazawa (2011) performed research
on the association between audit fees and earnings management, but he did not consider the
effect of litigation environment on audit fees. In this paper, we consider the effect of audit fees,
earnings management risk, and litigation risk using a sample of Japanese firms cross-listed in the
U.S. markets to answer several questions: what is the correlation between audit fees and
difference litigation environments and whether the extent of the correlation between audit fees
and earnings management in one litigation environment is different from that in another
litigation environment.
To answer these questions, we use the PSM model to control for differences in firm
characteristics between two litigation environments and carried multivariate regression tests to
verify three hypotheses about the correlations among audit fees, earnings management risk and
litigation risk.
Based on our results, we found that there are differences in audit fees under different
litigation environments after using the PSM model, and that audit fees increase with litigation
risk. Further analyses showed that the high risk of earnings management is associated with
higher audit fees. We also found that the effect of audit fees resulting from earnings management
risk (i.e., absolute discretionary accruals) is reduced under higher litigation risk environment.
Compared with prior research, this paper has two distinct characteristics: first, we provide
empirical evidence on the relationship between audit fees and earnings management considered
from the perspective of different litigation environments; second, we use a sample of Japanese
firms cross-listed in U.S. markets and Japanese firms listed only in Japan, which is then matched
by the PSM model.
Page 127
HYPOTHESES DEVELOPMENT
Audit Fees and Litigation Risk
Houston, Peters & Pratt (1999) found that the presence of accounting choices reflects
higher risks of accounting irregularities which contribute to higher litigation risk assessments and
fee premiums. Seetharaman, Gul & Lynn (2002) documented that U.K. auditors charge higher
audit fees when their U.K. clients are cross-listed in U.S. markets, and Seetharaman, Gul & Lynn
(2002) attribute this premium mostly to the U.S. high litigation environment. Choi, Kim, Liu &
Simunic (2009) asserted that auditors charge higher fees for firms that are cross-listed in stronger
legal regimes than for non-cross-listed firms. Using the above rationale, we formulate the first
hypothesis:
H1 Audit fees are higher in a higher litigation risk environment.
Previous research in sample matching method did not analyze the sample using a more
rigorous matching model (Seetharaman, Gul & Lynn, 2002; Choi, Kim, Liu & Simunic, 2009).
This paper resolves the selection bias problem by using the propensity score matching (PSM)
method as our research method.
Audit Fees and Earnings Management Risk
Several empirical papers suggested that audit fees are influenced by earnings
management risk. Heninger (2001) argued for a positive association between income-increasing
abnormal accruals and ex-post auditor litigation. Bedard & Johnstone (2004) found that auditors
respond to earnings management risk with ex ante increase in planned audit hours and billing
rates. Alali (2011) mentioned that income-increasing discretionary accruals are positively and
significantly related with audit fees and that increase in CFO’s bonuses represents this positive
relationship. These findings indicate that earnings management risk increases would require
more audit work, more extensive reviews and closer supervision of staff which would result in
higher audit fees. Using the above rationale, we formulate the second hypothesis:
H2 Audit fees are higher in an environment with higher earnings management risk.
We use absolute discretionary accrual to represent the risk of earnings management. This
paper uses performance-adjusted discretionary accruals of the modified Jones model as
recommended by Kothari, Leone & Wasley (2005).
Audit Fees, Earnings Management Risk and Litigation Risk
Pagano et al. (2000) finds that European firms are more likely to cross-list in the U.S. due
to better investor protection, more efficient courts and bureaucracy. Several studies suggested
that there is a negative correlation between investor protection and earnings management (Shen
& Chih, 2007; Cahan, Liu & Sun, 2008), that is, better investor protection can mitigate earnings
Page 128
management. Japanese firms cross-listed on the U.S. major exchanges are subject to more
stringent U.S. laws and Securities and Exchange Commission (SEC) regulations, and are
therefore exposed to a stronger investor protection environment. It follows that the degree of
earnings management of these firms will be diminished. Thus this leads to our third hypothesis:
H3 Audit fees resulting from earnings management risk is lower under greater litigation risk
environments.
While the hypothesis above is similar in spirit to the earnings management risk
hypothesis presented by Abbott, Parker & Peters (2006), there is one important difference that
merit discussion. Abbott, Parker & Peters (2006) used high (low) earnings-to-price multiples to
represent the high (low) litigation risk. Based on the finding that high-growth firms are more
likely to use large accruals to manage earnings (Dechow & Skinner, 2000), they hypothesized
that the effect of audit fees resulting from earnings management risk is magnified under greater
litigation risk environment. In comparison, we use a firm’s cross-listing status to represent both
the high litigation risk environment (U.S.) and low litigation risk environment (Japan). The
different proxy for litigation risk changes the direction of the hypothesis.
METHOD
Matching Process
The propensity score matching (PSM) model is becoming an increasingly popular

research method adopted in the auditing literature. In this paper, we use the PSM model to match
the sample on a broad range of firm characteristics to examine whether the difference in audit
fees exists under different litigation environment. The PSM model matches observations based
on the probability of undergoing a treatment, which in our case is the probability of Japanese
firms’ public offering to sell their securities in the United States. This matching process has two
advantages. First, this model generates a sample in which both the U.S.-listed Japanese firms and
non U.S.-listed Japanese firms are matched to have similar characteristics. Second, Li &
Prabhala (2007) argued that matching models do not rely on a specific functional form and
provide a more direct estimate of the treatment effects.
We use a logistic model to estimate the probability of Japanese firms offering to sell their
securities in the United States, as it is the most prevailing approach for estimating propensity
scores (Guo & Fraser, 2010; Lawrence, Minutti-Meza & Zhang, 2011). Following prior research
(Karolyi 1998; De Medeiros, Erkens & Zhang, 2005; Kung & Cheng, 2012) we estimate the
propensity score using the following logistic regression model:
0  1DMARTi ,t   2 FMARTi ,t  3 STOCKi ,t  4 SALESi ,t

USi ,t 
(1)
 5 PROFITi ,t  6 LNTAi ,t  7CURRi ,t  8 LEVi ,t  Industry _ d   i ,t
Page 129
For firm i and fiscal year t, where:

US: Dummy variable equaling 1 if a firm was listed on the U.S. markets and 0 otherwise;
DMART: Number of shares issued at the end of period t, multiplied by the closing share price on the
same date, divided by the total capitalization, which is the sum of markets values of all sample
firms;
FMART: Overseas sales t / net sales t;
STOCK: Stock return, measured by log (stock price t / stock price t-1);
SALES: Net operating revenue growth rate t;
PROFIT: Operating Income t / net operating revenue t;
LNTA: Natural logarithm of total assets at the end of year t;
CURR: Current assets t / current liabilities t;
LEV: (Long term debt t plus debt in current liabilities t) / total assets t-1;
The estimation results are robust to the inclusion of all redundant variables
simultaneously or one redundant variable at a time (Lawrence, Minutti-Meza, & Zhang, 2011).
As a result, we estimate the propensity score model by including all control variables. In addition,
we use propensity score analysis with nonparametric regression (i.e. kernel-based matching).
This method uses propensity scores derived from multiple matches to calculate a weighted mean
that is used as a counterfactual. As such, kernel-based matching is a robust estimator (Guo &
Fraser, 2009).
Regression Model
Our tests on three hypotheses are based on cross-sectional regressions of the natural
logarithm of disclosed audit fees on a number of variables, including dummy variables to
identify Japanese firms trading on U.S. markets. We use the following cross-sectional regression
model:
 0  1USi ,t   2 DAi ,t  3US * DAi ,t   4 ROAi ,t  5CRATIOi ,t

LAFi ,t 
  6 ATURNi ,t   7 DEi ,t  8 LOSSi ,t  9 LNTAi ,t  10CURRi ,t (2)
 11 LEVi ,t  Industry _ d  Year _ d   i ,t

LAF: The natural log of audit fees t;
US: Dummy variable equaling 1 if a firm was listed on the U.S. markets and 0 otherwise;
DA: Absolute discretionary accruals t;
US*DA: The value of DA times 1 if a firm was listed on the U.S. markets and 0 otherwise;
ROA: Net income t / total assets t-1;
DRATIO: Current asset t / total assets t;
ATURN: Sales t / total assets t-1;
DE: Long term debt t / total assets t;
LOSS: Dummy variable equaling 1 if a firm was net loss at the end of year t;
LNTA: Natural logarithm of total assets at the end of year t;
CURR: Current assets t / current liabilities t;
LEV: (Long term debt t plus debt in current liabilities t) / total assets t-1;
Page 130
Performance-matched Modified Jones Model
Discretionary accruals are estimated with the performance-matched modified Jones

model (Kothari, Leone & Wasley, 2005). The model is estimated by year and by industry, scaled
by lagged assets:
TAi ,
t 0  1 (1/ ASSETSi ,t )  2 (SALESi ,t  RECi ,t )  3 PPEi ,t   i ,t (3)

TA: Total accruals equals (net income before extraordinary items minus operating cash flows from
continuing operations t) / total assets t-1;
ASSETS: Total assets t-1;
△SALSE: Change in sales from year t-1 to year t / total assets t-1;
△REC: Change in accounts receivable from year t-1 to year t / total assets t-1;
PPE: (Net property, plant and equipment in year t) / total assets t-1;
ε: Equals the estimated modified discretionary accrual;
From model (3) we can get each firm’s estimated modified discretionary accrual and we
subtract the estimated modified discretionary accrual of the closest Return on Assets (ROA) firm
in the same industry and year. The resulting error term is the performance-matched modified
discretionary accrual measure.
Variable definitions are summarized in Table 1.
Page 131
Table 1
DEFINITION OF VARIABLES
Sample Selection
La Porta et al. (1998) found that common-law countries (e.g. United States, United
Kingdom, Australia, etc.) generally have the strongest legal protection for investors, while
French-civil-law countries (e.g. France, Brazil, etc.) and German and Scandinavian-civil-law
countries (Japan, Germany, Finland, etc.) have weaker legal regime. In this paper, the litigation
environment of United States is stronger than that of Japan. This paper is different from the
Seetharaman, Gul & Lynn’s (2002) research on audit fees and litigation risk since our study
employs sample of Japanese firms which belong to the German and Scandinavian-civil-law
countries, which is different from the common-law countries like United States and United
Page 132
Kingdom. This paper provides a new research perspective on the topic of audit fees and earnings
management.
Non-US firms selling their securities in the United States are exposed to liability under
U.S. securities laws. As a result, Japanese firms offering to sell their securities publicly in the
U.S. provide an ideal opportunity to research firms operating under different litigation
environment. We use a sample of Japanese firms cross-listed on U.S. markets and Japanese firms
not cross-listed on U.S. markets to account for different litigation characteristics between Japan
and United States. Using this data allow our paper to provide an insight for discussing the
association between audit fees and earnings management under different litigation risk.
Table 2
SAMPLE SELECTION AND DESCRIPTION
Table 2 presents information on the sample’s selection process and descriptive statistics.
For our analyses, we use firm-year data from 2005 to 2013. Information about audit fee and U.S.
listed Japanese firms is obtained from the EOL database. Other variables are obtained from the
NEEDS Financial Quest database. After excluding financial companies, missing data, and
restricting our sample to firms with fiscal year ended as of March 31; our final sample consists of
12,255 firm-years. There are only four types (i.e. Machinery, Electric Appliances, Transportation
Page 133
Equipment, and Information & Communication) of Japanese industry cross-listed on U.S.

markets. The percentage of Japanese firm cross-listed on U.S. markets is around 1 percent of
total Japanese listed companies in each year.
Table 3
CORRELATIONS AMONG THE VARIABLES
Table 3 presents the Pearson correlations among the variables. From Panel A, there is a
low level of correlation among the all variables, which means there is no multicollinearity
problem. From Panel B, all variables except LNTA have low level of correlation.
Page 134
RESULTS
Logistic model
Table 4
LOGISTIC REGRESSION FOR ESTIMATING
PROPENSITY SCORE
Table 4 shows the results of the logistic regression from estimating the propensity score.
The logistic model results indicate that 6 variables are significant: DMART, FMART, SALES,
LNTA, CURR, and LEV. It shows that Japanese firms with higher participation in the domestic
stock markets (DMART), greater exposure to the foreign markets (FMART), higher sales growth
(SALES), larger size (LNTA), lower liquidity (CURR), and lower debt ratio (LEV), have a higher
tendency to cross-list in U.S. markets.
Table 5
Page 135
Table 5 presents the descriptive statistics for both the full sample and the PSM sample.
The full sample consists of 12,255 firm-years, out of which 117(1.0%) and 12,138 (99.0%)
firm-years represent U.S.-listed Japanese firms and non U.S.-listed Japanese firms, respectively.
The PSM sample consists of 3,736 firm-years, out of which 45(1.2%) and 3,691 (98.8%)
firm-years represent U.S.-listed Japanese firms and non U.S.-listed Japanese firms, respectively.
The descriptive statistics for the full sample and PSM sample indicate that U.S.-listed Japanese
firms and non U.S.-listed Japanese firms have significantly different audit fees.
Regression Results
Table 6
AUDIT FEES, EARNINGS MANAGEMENT AND LITIGATION ENVIRONMENT
ANALYSIS
In Table 6, we find a positive and significant US coefficient of 0.747 in the full sample
column and a positive significant multivariate US coefficient of 0.599 in the PSM sample,
suggesting that the treatment effects of U.S.-listed Japanese firms are significantly different from
those of non U.S.-listed Japanese firms with respect to audit fees, even after controlling for client
characteristics for both sample groups. We can also confirm the result of Seetharaman, Gul &
Lynn (2002), as the H1 is supported that audit fees increase in higher litigation risk environment.
Page 136
We also find a positive and significant DA coefficient of 0.069 in the full sample column,
thus the results from the full sample support the H2 that audit fees increase with higher earnings
management risk. In addition, we find a negative and significant US*DA coefficient of -1.324 in
the full sample column, thus the results from the full sample support the H3 that audit fees
resulting from earnings management risk are lower under greater litigation risk environments.
The right column of Table 6 presents the results of the PSM sample. We find a positive
significant multivariate DA coefficient of 0.092, confirming hypothesis that audit fees increasing
with high earnings management. Also we find a negative and significant US*DA coefficient of
-1.195 in the PSM sample column. Although the significance of these two coefficients are a
matter of judgment, after client characteristics are controlled for the two sample groups, the
results at least provide limited empirical support for H2 and H3.
Robust tests
We also use an alternative matching procedure, coarsened exact matching (CEM). King
et al. (2011) mention that CEM is a more robust matching technique that is not subject to random
matching, because CEM directly matches on all covariates and does not rely on a first-stage
propensity score model. DeFond, Erkens & Zhang (2014) document that in their setting, CEM
indeed dominates PSM in providing more balanced matches across all matched sample sizes.
We examine the model (1) with CEM method. The dichotomous treatment variable is US.
We match 9 pretreatment variables (i.e. DMART, FMART, STOCK, SALES, PROFIT, LNTA,
CURR, LEV, and Industry dummy). The result of the CEM sample is presented in Table 7.
Table 7
COARSENED EXACT MATCHING
Page 137
The left column of table 7 presents the descriptive statistics for CEM sample. The CEM
sample consists of 96 firm-years, out of which 19(19.8%) and 77 (80.2%) firm-years represent
US-listed Japanese firms and non US-listed Japanese firms, respectively. The descriptive
statistics for the CEM sample also indicate that US-listed Japanese firms and non US-listed
Japanese firms have significantly different audit fees.
We find a positive and significant US coefficient of 0.741, a positive and significant DA
coefficient of 0.840, and a negative and significant US*DA coefficient of -3.368 in the right
column of Table 7, suggesting that the results from CEM sample still support the H1, H2, and H3.
As a side note, by comparing the adjusted R2 of full sample (0.601), PSM sample (0.579), and
CEM sample (0.626), we conclude that the CEM method can improve the results of our
multivariate tests.
CONCLUSION
This paper examines the correlations among audit fees, earnings management, and
litigation risk, using discretionary accruals as a proxy for earnings management risk and
cross-listing status of Japanese firms in U.S. markets as a proxy for two different litigation
environments. We hypothesize that audit fees increase with high litigation risk. We also
hypothesize that audit fees increase with higher earnings management risk. We finally
hypothesize that audit fees resulting from earnings management risk is lower under greater
litigation risk environment. We test our hypotheses using the propensity score matching sample
of 3,736 firm-year data for the year 2005 to 2013.
We found that after using the PSM method, audit fees are still different in each litigation
environment and positive relationship exists between audit fees and litigation risk. We also found
that higher earnings management risk is associated with higher audit fees. We also found that the
audit fees resulting from earnings management risk is lower under greater litigation risk
environment. Lastly, we use the coarsened exact matching (CEM) method for the robustness
check and the conclusions from three hypotheses are supported. By comparing the adjusted R2 of
the full sample, the PSM sample, and the CEM sample; we concluded that CEM method improve
the results of the multivariate tests.
As for the limitation of this paper, we use a single proxy to estimate discretionary accrual.
Additionally, earnings management is just a single aspect of audit quality. This paper does not
investigate other factors of audit quality such as auditor’s professional experience, financial
statements restatements, and economic dependency. For future study, it would be useful to
analyze the association between audit fees and other factors of audit quality under different
litigation environments.
AUTHORS’ NOTE
The authors wish to thank Otsuka Toshimi Scholarship Foundation for financial support
towards this research.
Page 138
REFERENCES
Abbott, L.J., Parker. S., & Peters. G.F. (2006). Earnings management, litigation risk, and asymmetric audit fee
responses. Auditing: A Journal of Practice and Theory, (25), 85-98.
Alali, Fatima. (2011). Audit Fees and Discretionary Accruals: Compensation Structure Effect. Managerial Auditing
Journal, 26 (2), 90–113.
Bedard, J.C. & Jonstone, K.M. (2004). Earnings management risk, corporate governance risk, and auditors’ planning
and pricing decisions. The Accounting Review, 79 (April), 277-304.
Cahan, Steven F., Liu, Guoping, & Sun, Jerry. (2008). Investor Protection, Income Smoothing, and Earnings
Informativeness. Journal of International Accounting Research, 7 (1), 1–24.
Chan, P., Ezzamel, M., & Gwilliam, D. (1993). Determinants of audit fees for quoted UK companies. Journal of
Business Finance and Accounting, (20), 765-786.
Choi, J.H., Kim, J.B., Liu, X.H., & Simunic, D.A. (2009). Cross-Listing Audit Fee Premiums: Theory and Evidence.
The Accounting Review, 84 (May), 1429-1463.
Chung, D. & Lindsy, W.D. (1988). The pricing of audit services: the Canadian perspective. Contemporary
Accounting Research, (5), 19-46.
Craswell, A.T. & Francis, J.R. (1999). Pricing initial audit engagements: A test of competing theories. The
Accounting Review, 74 (April), 201-216.
De Medeiros, Otavio Ribeiro, & Carmem S. B. Tiberio. (2005). Factors Influencing Brazilian Firms in Their
Decision to List on Foreign Stock Exchanges. Working Paper.
DeFond, Mark, David Hendrik Erkens, & Jieying Zhang. (2014). Do Client Characteristics Really Drive the Big N
Effect? Evidence from Matching Methods. Working Paper.
Dechow, P.M. & Skinner, D.J. (2000). Earnings management: Reconciling the views of accounting academics
practitioners, and regulators. Accounting Horizons, 14 (June), 235-250.
Francis, J.R. (1984). The effect of audit firm size on audit price: a study of the Australian markets. Journal of
Accounting and Economics, (6), 135-151.
Guo, S. & Fraser. M. (2010). Propensity Score Analysis: Statistical Methods and Applications. Thousand Oaks, CA:
SAGE Publications, Inc.
Guo, Shenyang Y. & Fraser, Mark W. (2009). Propensity Score Analysis: Statistical Methods and Applications 12th
ed. SAGE Publications, Inc.
Heninger, W. G. (2001). The association between auditor litigation and abnormal accruals. The Accounting Review,
76 (January), 111-126.
Houston, R.W., Peters, M.F, & Pratt, J.H. (1999). The audit risk model, business risk, and audit planning decisions.
The Accounting Review, 74 (July), 281-298.
Johnson, E.N., Walker, K.B, & Westergaard, E. (1995). Supplier concentration and pricing of audit services in New
Zealand. Auditing: A Journal of Practice and Theory (14), 74-89.
Karolyi, G. Andrew. (1998). Why Do Companies List Shares Abroad?: A Survey of the Evidence and Its
Managerial Implications. Financial Markets, Institutions & Instruments, 7 (1), 1-60.
King, Gary, Richard Nielsen, Carter Coberley, James Pope, & Aaron Wells. (2011). Comparative Effectiveness of
Matching Methods for Causal Inference. Working Paper.
Kothari, S. P., Leone, Andrew J, & Wasley, Charles E. (2005). Performance Matched Discretionary Accrual
Measures. Journal of Accounting & Economics, 39 (1), 163–197.
Kung, Fan-hua, & Cheng, Chia-ling. (2012). The Determinants of Overseas Listing Decisions: Evidence from
Chinese H-Share Companies. Asian Business & Management, 11 (5), 591–613.
La Porta. R., Lopez-de-Silanes. F., Shleifer. A., & Vishny. R.W. (1998). Law and finance. Journal of Political
Economy, 106 (December), 1113-1155.
Lawrence. A., Minutti-Meza. M., & Zhang. P. (2011). Can big 4 versus non-big 4 differences in audit-quality
proxies be attributed to client characteristics? The Accounting Review, 86 (January), 259-286.
Li, K. & Prabhala. N. (2007). Handbook of Corporate Finance: Empirical Corporate Finance, edited by B. E. Eckso.
Amsterdam. The Netherlands: Elsevier Science B.V.
Page 139
Pagano, Marco, Otto Randl, Ailsa A. Röell, & Josef Zechner. (2000). What Makes Stock Exchanges Succeed?
Evidence from Cross-Listing Decisions. Working Paper.
Seetharaman, A., Gul, F. A., & Lynn, S. G. (2002). Litigation risk and audit fees: Evidence from UK firms
cross-listed on U.S. markets. Journal of Accounting and Economics, 33(1), 91–115.
Shen, Chung-Hua, & Chih, Hsiang-Lin. (2007). Earnings Management and Corporate Governance in Asia’s
Emerging Markets. Corporate Governance: An International Review, 15 (5), 999–1021.
Simunic, D. A. (1980). The pricing of audit services: Theory and evidence. Journal of Accounting Research, 18(1),
161–190.
Yazawa. K. (2011). Empirical research on relationship among corporate governance, audit fees, and earnings
management. Accounting Progress, (5), 28-44.
Page 140
Page 141
ABNORMAL AUDIT FEES AND AUDITOR SIZE IN THE

JAPANESE AUDIT MARKET
ABSTRACT
Using 9,716 firm-years samples of Japanese listed firms between 2005 and 2011, this
study tests the association between auditor size and abnormal audit fees as this is currently
missing from the literature. The study utilizes abnormal audit fees as an audit quality proxy,
where higher abnormal audit fees are associated with greater auditing efforts that contribute to
a higher quality audit. The empirical results show that Big N auditors are positively and
significantly associated with a higher level of abnormal audit fees, indicating that the Japanese
Big 3 auditors provide higher quality audits than non-Big 3 firms.
INTRODUCTION
Many empirical studies use auditor size as proxy for audit quality (e.g., Becker et al.
1998; Francis, Maydew & Sparks, 1999; Behn, Choi & Kang, 2008). Recent empirical findings
show that auditor size (Big N and non-Big N audit firms) does not explain differences in audit
quality after controlling for clients’ characteristics (Lawrence, Minutti-Meza, & Zhang, 2011).
However, Lawrence, Minutti-Meza, & Zhang’s (2011) conclusions are disputed by Eshleman &
Guo’s (2014b) finding that Big N auditors do perform higher quality audits than both non-Big N
and mid-tier auditors. In another study exploring the empirical relationship between audit pricing
and audit quality, Blankley, Hurtt, & MacGregor (2012) employed financial statement
restatements, the same proxy for audit quality as Eshleman & Guo (2014b), and show that audit
quality is positively associated with abnormal audit fees.
There is a lack of empirical studies investigating the effect of auditor size on abnormal
audit fees as a measure of audit quality. Lawrence, Minutti-Meza, & Zhang (2011) do not
employ abnormal audit fees as one of the proxies for audit quality. Other studies, e.g., Hoitash,
Markelevich, & Barragato (2007), Choi, Kim, & Zang (2010), Eshleman & Guo (2014a),
Asthana & Boone (2012), Blankley, Hurtt, & MacGregor (2012) examine the association
between abnormal audit fees and discretional accruals or restatements as a proxy for audit
quality. To the best of the author’s knowledge, there has not been a study that investigates
whether the pricing-based proxy for audit quality ‒ abnormal audit fee ‒ is empirically consistent
with the characteristics-based measure for audit quality ‒ auditor size in the setting of Japanese
environment.
This study reveals that abnormal audit fees are positively related to auditor size in the
Japanese context, suggesting that Japanese Big N auditors provide higher quality audits than
non-Big N firms from the perspective of audit pricing. This study contributes to the auditing
literature in two ways. First, it attempts to contribute some insight into audit quality in Japan.
Despite Japan’s economic significance to global investors and its possession of one of the largest
Page 142
stock markets, audit studies do not use its data enough (Hu & Kato, 2015). Furthermore, the
cross-country auditing literature excludes Japan due to the miscoding of the auditor identification
variable in the global database (Francis & Wang, 2008; Francis, Michas, & Seavey, 2013; Gul,
Zhou, & Zhu, 2013).
Second, this study takes place in the unique Japanese setting: low litigation risk but
emphasis on reputation, “face,” and “trust”; about 15 percent the companies operate on a
Keiretsu system; companies tend to rely on debt financing, hold high cash levels, exhibit a lower
ROA, and are not easily bankrupted compared to companies in western countries (Hu & Kato,
2014). Francis & Wang (2008) find that only for Big N auditors is the earnings quality (abnormal
accruals among others) better when investor protections are stronger, using a sample without
Japanese data. This study provides evidence from a country with a lighter regulatory
environment compared to Anglo-Saxon countries (Shima & Gordon, 2011), with a lower
litigation risk (Skinner & Srinivasan, 2012), and a heavier emphasis on reputation.
LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT
Abnormal Audit Fee and Audit Quality
Audit fee represents the value of the resources and costs required to conduct a proper
audit that is agreed upon by an audit firm and its client. Abnormal audit fees are defined as the
difference between the audit fee paid and the normal or expected audit fee (Knechel et al. 2013).
In empirical literature, abnormal audit fees are estimated as residuals from an audit fee
regression model using variables representing the clients’ characteristics. Abnormally high or
low audit fees might indicate the extent of the economic bond between the auditor and its client
that potentially decreases the auditor’s independence (Blankley, Hurtt, & MacGregor, 2012).
However, this study argues that abnormal audit fee is a good indicator of audit quality and I
explain the rationale of this argument in the following paragraphs.
The use of abnormal (unexplained) audit fees variable as a proxy for audit quality has
several positive characteristics (Hribar, Kravet, & Wilson, 2014). First, audit fees not only
represent the estimated cost necessary to conduct an audit, but also partly contain the auditor’s
estimate on the clients’ general business risks. Auditors must factor their clients’ internal control
quality, business, and litigation risk into their audit fee. Abnormal audit fees can capture audit
quality information that is otherwise not reflected in audit opinions or the total audit fee. Second,
other commonly used measures of audit quality rely on accounting earnings, which capture other
aspects of a firm’s characteristics besides audit quality. Using abnormal audit fees attempts to
isolate other firms’ characteristics in order to provide a better measure of audit quality. From this
perspective, abnormal audit fees may be a superior proxy of audit quality.
Additionally, this study adopts abnormal audit fees as an audit quality proxy because of
the positive association between abnormal audit fees and accounting quality. Using financial
statement restatements as a proxy for audit quality, Blankley, Hurtt, & MacGregor (2012)
concluded that there is a positive relationship between abnormal audit fees and audit/accounting
quality after controlling for the internal control variable post-SOX. Their study found that higher
abnormal audit fees are negatively related to the extent of subsequent audit restatements.
Page 143
Contrary to previous studies, Blankley et al.’s (2012) findings suggest that higher abnormal audit
fees are not always associated with impaired auditor independence.
Auditor Size and Audit Quality
Early empirical research on the audit quality of Big N firms shows that clients of Big N
firms tend to have lower levels of discretionary accruals, lower cost of capital, and more credible
earnings reports (Teoh & Wong, 1993; Knechel et al. 2013). These findings suggest that Big N
firms have higher quality audits than non-Big N firms. The higher quality provided by Big N
firms is generally associated with the bigger size of the firms, which allows them to invest more
resources in better training materials, standardized audit methodologies, and access to industry
and multinational specialists (Eshleman & Guo, 2014b; Lawrence, Minutti-Meza, & Zhang,
2011). The association between Big N firms and higher audit quality is also supported by
empirical evidence demonstrating the audit fee premium enjoyed by Big N auditors (Knechel et
al., 2013). A more recent study shows that the audit quality of U.S. Big 4 firms is higher than
that of mid-tier auditors (Eshleman & Guo, 2014b).
Big N distinction is often used as an audit quality proxy, with Big N firms having higher
audit quality than non-Big N firms; however, the significant difference between client
characteristics might also have an impact. Using measures of real and perceived audit quality,
Boone, Khurana, & Raman (2010) show that Big N audit quality superiority is more likely
rooted in investors’ perceptions rather than from tangible audit quality differences. Lawrence,
Minutti-Meza, & Zhang (2011) attempt to mitigate the endogeneity problem of audit quality by
using a propensity matching model to match client characteristics. They find that differences in
audit quality proxies between Big 4 and non-Big 4 auditors are largely attributed to clients’
characteristics, particularly client size, and argue that many factors contribute to the similar audit
quality between Big N and non-Big N firms.
Hypothesis
Empirical models of the relationship between auditor size and audit quality assume that
auditors’ loss functions determine the degree of audit quality (Sirois & Simunic, 2011). The
auditors’ loss functions are commonly associated with litigation or reputational loss, and they
have more incentives to provide higher quality audits in an environment where the audit failure
consequences are severe, for example, in an environment with strong investors’ protection, legal
enforcement, or in a litigious society. However, the Japanese business environment emphasizes
the importance of reputation, as “face” or “trust” (e.g., Fukuyama, 1995); larger companies or
auditors have more to lose in case they lose “face” and might have more incentives to hire better
auditors or perform higher quality audits.
A fundamental difference between Big N and non-Big N auditors is the different
investment strategies with respect to audit technology. Sirois & Simunic (2011) argue that given
the same level of audit hours for a comparable audit engagement, the Big N audit firms’ superior
technology allows them to offer superior audit value. Abnormal audit fees represent the level of
auditing effort (and technology) expended by audit firms, where a higher abnormal audit fee is
Page 144
associated with greater auditing effort (and technology) contributing to higher audit quality.
Thus, this study hypothesizes that the Japanese Big 3 auditors are positively associated with
higher audit quality as indicated by higher abnormal audit fees than non-Big 3 auditors. Notice
that From the client’s perspective, there are only 3 Big N auditors in Japan (Hu & Kato, 2014;
2015), with a market shares of 26.7%, 24.8%, and 19.3%. The fourth largest has a market share
of only 2.4%.
H1: Japanese Big 3 Auditors earn higher abnormal audit fees than non-Big 3 auditors.
RESEARCH METHOD AND DATA
Table 1
SAMPLE SELECTION
Listed companies for fiscal years 2005 to 2012 (ending in March) 18,792
(less) financial companies -1,255
(less) financial or stock price data unavailable -5,666
11,871
(less)Auditor data unavailable 1,174
(less) joint auditor clients -981
Total samples 9,716
This study uses the sample of Japanese listed companies using Japanese accounting
standards from fiscal year 2005 to 2011 obtained from the Nikkei NEEDS Financial Quest
database. The sample excludes Japanese firms with a fiscal year ending on a date other than
March 31, financial companies, companies with missing data, and companies with joint auditors.
The final sample size is 9,716 firm years. Table 1 describes the sample selection procedure.
0  1 LTAi ,t   2CRi ,t  3CA _ TAi ,t   4 ARINVi ,t  5 ROAi ,t  6 LOSSi ,t

LAFi ,t 
(1)
  7 LEVi ,t  8 INANGi ,t  9 POLICY   i ,t
Where for firm i and fiscal year t: LAF = logarithm of audit fees; LTA = logarithm of end of year total assets;
CR = current assets divided by current liabilities; CA_TA = current assets divided by total assets;
ARINV = sum of accounts receivable and inventory divided by total assets;
ROA = earnings before interest and taxes divided by total assets; LOSS = 1 if the firm incurred a loss, 0 otherwise;
LEV = long-term debt divided by total assets; INTANG = ratio of intangible assets to total assets;
POLICY = 1 if the firm changes accounting principles, 0 otherwise.
This study employs a modified version of the audit fee model in Blankley, Hurtt, &
MacGregor (2012) as in equation (1). Abnormal audit fees are first calculated by estimating an
audit fee regression model using variables that capture the scope of an audit. Next, the residuals
of the estimated audit fee regression are subtracted from the total audit fee to obtain the abnormal
audit fee.
After calculating the abnormal audit fee, this study investigates the relationship between
audit quality (proxied by abnormal audit fees) and auditor size using the model described in
Page 145
equation (2) using the weighed data sample. This is Rubin’s (1985) propensity score, or inverse
probability weighting (hereafter, “IPW”; see Hu & Kato, 2015 for details).
ABAFEE i ,t     0 BigN i ,t   i ,t (2)

(See Table 2 for the details of variables.)
Table 2
DEFINITION OF REGRESSION VARIABLES
Japanese companies' P redicted sign to
Variable Description Reference specific characteristics choose Big N
Dependent
value of abnormal audit fees
ABAFEE Blankley et al.(2012)
estimated by themodel developed by Blankley et al.(2012)
BigN 1 if the client has a BIGn auditor, and 0 otherwise. Blankley et al.(2012)
Independent (size perspective)
Chaney et al. (2004)
lnASSET natural logarithm of total asset Francis and Yu (2009) +
Behn et al. (2008)
client’s total liabilities deflated by total
LIAB Francis and Yu (2009) 〇 ?
equity
Khurana and Raman (2004)
asset turnover ratio calculated by sales
ATURN Chaney et al. (2004) ?
devided by average total asset
current ratio calculated by current asset
CURR Chaney et al. (2004) +
devided by current liability
Francis and Yu (2009)
operating cash flow devided by average
CFO Choi et al. (2010) ?
total asset
Reichelt and Wang (2010)
total accruals for the last fiscal year calculated by ordinary
lgACCR profit after tax minus operating cash flow devided by average Reichelt and Wang (2010) ?
total asset for the last fiscal year
sum of a client’s total cash deflated by
CASH Louis (2005) 〇 -
total assets
Independent (risk perspective)
dummy variable that takes the value of 1 Chaney et al. (2004)
LOSS if a firm reports a net loss for the current fiscal year, and 0 Francis and Yu (2009) ?
otherwise Reichelt and Wang (2010)
stock beta clculated over 36 month
BETA Khurana and Raman (2004) ?
ending in the month of the fiscal year end
client’s stock volatility calculated by the
VOLATILITY standard deviation of 12 monthly stock returns for the current Francis and Yu (2009) ?
fiscal year
dummy variable that takes the value of 1
lgLOSS if a firm reports a net loss for the last fiscal year, and 0 Francis and Yu (2009) ?
otherwise
standard deviation of sales revenue calculated by using a rolling
SALESVOLATILITY Francis and Yu (2009) ?
window for three years of data
SALESGROWTH one-year growth rate of a firm’s sales revenue Francis and Yu (2009) ?
standard deviation of CFO calculated by using a rolling window
CFOVOLATILITY Francis and Yu (2009) ?
for three years of data to estimate
Shirata(2003)'s SAF2002 value that
SAF2002 Original 〇 -
indicates the probability of a firm's bankruptcy in Japan
CONSOL the number of the consolidated firms of a firm Yazawa (2010) +
return on asset calculated by net income
ROA Butler et al. (2004) 〇 ?
/loss devided by average total asset
Choi et al. (2010)
foreign sales ratio calculated by foreign
SALESABROAD Original 〇 +
sales devided by total sales
KEIRETSU Original 〇 -
if a firm is classified as Keiretsu company
IPO9 Original +
if a firm made initial pubulic offering last 9 years
dummy variable that takes the value of 1if a firm
TSE1 Original 〇 +
are listed in the 1st section of Tokyo Stock Exchange(TSE)
Variables are based on data in Nikkei NEEDS Data-Base.
Page 146
The advantage of the IPW is that it can be included a lot of covariates that effect audit
quality proxy (see equation (3).), while conventional matching method, like nearest neighbor, are
limited to use small number of covariates because of the curse of the dimensionality (Li &
Prabhala, 2007; Tucker, 2010; Armstrong, Jagolinzer, & Larcker, 2010) .
There are three steps to determine the weighed number used in equation (2). First, use
logit regression (3) to estimate  , then use  to estimate BigN i ,t ( ei , “propensity score” as the
estimated probability of receiving the select treatment, that is, the probability of selecting a Big
N as its auditor.). Second, use equation (4) to calculate iei , which is the weighted number used to
weigh the sample. Third, run equation (2) again after employing the propensity score weighting
( iei ). Note that this study attempts to control several perspectives, including Japanese
company/business’ unique characteristics, using equation (3). Table 2 provides a detailed
description of the regression variables.
BigN i ,t   0   1 ln ASSET i ,t   2 LIABi ,t   3 ATURN i ,t   4 CURRi ,t   5 DOCFi ,t   6 lg ACCR i ,t

  7 CASH i ,t   8 LOSSi ,t   9 BETAi ,t   10VOLATILITYi ,t   11 lg LOSSi ,t (3)
  12 SALESVOLATILITYi ,t   13 SALESGROWTH i ,t   14CFOVOLATILITYi ,t
  15 SAF 2002 i ,t   16CONSOLi ,t   17 ROAi ,t   18 SALESABROA Di ,t   19 KEIRETSU i ,t   20 IPOi ,t
  21TSEi ,t  iD  yD   i ,t
(See Table 2 for the details of the variables.)
zi N1 1  zi N2 (4)
iei   
ei zi 1  ei N 1  zi
 i 1 1  e
N
i
ei
1
i
ie denotes the sampling weights, z denotes the treatment and control group (1 if it is Big N, and 0 otherwise),
e denotes the predicted propensity score, N1 denotes the number of Big N samples and N2 denotes the number of
non-Big N samples. N=N1+N2
RESULTS
Table 3
SAMPLE DESCRIPTION AND BIG N PROPORTION FOR EACH INDUSTRY
Industry firm-years Big N Proportion (%)
Construction 804 80.08%
Foods 357 74.91%
Textiles & Apparels 255 77.67%
Chemicals 1,019 75.60%
Pharmaceutical 199 82.84%
Glass & Ceramics Products 245 88.21%
Iron & Steel 292 67.17%
Nonferrous Metals 202 84.52%
Metal products 333 65.24%
Machinery 948 71.16%
Electric Appliances 1,040 73.70%
Transportation Equipments 577 74.00%
Precision Instruments 211 81.68%
Other Products 321 70.12%
Land Transportation 370 88.10%
Warehousing & Harbor Transportation Services 151 80.00%
Information & Communication 458 82.34%
Wholesale Trade 972 75.58%
Retail Trade 358 76.76%
Real Estate 213 65.62%
Services 391 77.80%
Total 9,716
Page 147
Table 3 provides the sample distribution by industry group and their respective Big N
market shares (by client number). The five largest industry groups in the sample belong to the
electric appliances, chemicals, wholesale trade, machinery, and construction industries. The
average Big N industry market share for the sample is 77%, which indicates the market
dominance of Big N auditors across Japanese industries.
Table 4
DESCRIPTIVE STATISTICS OF BIG NS AND NON-BIG NS' CLIENT COMPANIES
All samples 　（ N=9716 ） BigN (N=7358) Non-BigN (N=2358) Difference in Mean Difference in Median
Mean Std. Dev. 10% Median 90% Mean Std. Dev. 10% Median 90% Mean Std. Dev. 10% Median 90% t value z value
ABAFEE -0.020 0.204 -0.251 -0.020 0.210 -0.004 0.210 -0.235 -0.002 0.225 -0.069 0.177 -0.293 -0.074 0.152 -13.642 *** -14.537 ***
lnASSET 11.194 1.394 9.580 11.029 13.117 11.311 1.390 9.736 11.111 13.267 10.828 1.340 9.192 10.765 12.527 -14.791 *** -13.569 ***
LIAB 1.927 2.331 0.366 1.233 4.007 1.937 2.326 0.363 1.246 4.013 1.896 2.345 0.374 1.188 3.990 -0.742 -1.496
ATURN 1.119 0.565 0.578 0.985 1.811 1.122 0.564 0.586 0.991 1.802 1.107 0.571 0.557 0.969 1.838 -1.140 -2.174 **
CURR 1.900 1.491 0.866 1.499 3.197 1.909 1.532 0.867 1.489 3.247 1.872 1.355 0.863 1.535 3.116 -1.048 1.359
CFO 0.001 0.061 -0.066 0.000 0.070 0.002 0.061 -0.064 0.001 0.070 0.000 0.064 -0.072 -0.002 0.072 -0.874 -1.390
lgACCR -0.007 0.012 -0.021 -0.007 0.006 -0.007 0.012 -0.021 -0.007 0.006 -0.007 0.013 -0.022 -0.007 0.007 -1.129 -0.810
CASH 0.131 0.091 0.039 0.110 0.250 0.127 0.091 0.037 0.105 0.247 0.141 0.093 0.044 0.123 0.259 6.609 *** 8.226 ***
LOSS 0.155 0.362 0.000 0.000 1.000 0.144 0.351 0.000 0.000 1.000 0.192 0.394 0.000 0.000 1.000 5.678 *** 5.669 ***
BETA -0.002 0.459 -0.527 0.006 0.507 0.001 0.452 -0.528 0.012 0.502 -0.011 0.480 -0.526 -0.016 0.523 -1.047 -1.958 *
VOLATILITY 0.098 0.056 0.046 0.085 0.164 0.097 0.053 0.046 0.084 0.161 0.103 0.063 0.046 0.088 0.176 5.147 *** 3.550 ***
lgLOSS 0.154 0.361 0.000 0.000 1.000 0.141 0.348 0.000 0.000 1.000 0.197 0.398 0.000 0.000 1.000 6.609 *** 6.594 ***
SALESVOLATILITY 0.094 0.094 0.018 0.065 0.202 0.091 0.090 0.018 0.064 0.197 0.101 0.104 0.020 0.070 0.215 4.598 *** 4.209 ***
SALESGROWTH 0.026 0.148 -0.142 0.022 0.178 0.028 0.143 -0.135 0.023 0.176 0.020 0.163 -0.169 0.016 0.186 -2.262 ** -3.462 ***
CFOVOLATILITY 0.035 0.030 0.009 0.027 0.071 0.035 0.030 0.009 0.026 0.070 0.037 0.032 0.009 0.028 0.075 3.402 *** 3.285 ***
SAF2002 0.953 0.347 0.558 0.964 1.377 0.968 0.332 0.583 0.972 1.387 0.905 0.389 0.489 0.940 1.345 -7.725 *** -5.533 ***
CONSOL 4.876 17.375 0.000 0.000 12.000 5.737 19.286 0.000 0.000 14.000 2.191 8.597 0.000 0.000 6.000 -8.655 *** -9.415 ***
ROA 0.020 0.043 -0.018 0.022 0.064 0.022 0.041 -0.014 0.023 0.065 0.015 0.048 -0.032 0.019 0.060 -7.306 *** -6.922 ***
SALESABROAD 7.917 15.351 0.000 0.000 35.170 8.016 15.476 0.000 0.000 35.700 7.609 14.955 0.000 0.000 31.640 -1.122 -0.697
KEIRETSU 0.144 0.351 0.000 0.000 1.000 0.136 0.343 0.000 0.000 1.000 0.169 0.375 0.000 0.000 1.000 3.944 *** 3.941 ***
IPO9 0.128 0.334 0.000 0.000 1.000 0.132 0.339 0.000 0.000 1.000 0.115 0.318 0.000 0.000 1.000 -2.244 ** -2.243 **
TSE1 0.501 0.500 0.000 1.000 1.000 0.504 0.500 0.000 1.000 1.000 0.492 0.500 0.000 0.000 1.000 -1.049 -1.049
***, ** and * indicate significance at the 0.01, 0.05, and 0.10 respectively.
Table 4 provides descriptive statistics for the regression variables and t-test value of
difference in mean between Big N and non-Big N clients. From Table 4, it can be observed that
the mean and median value for the ABAFEE variable is negative for all firms, which indicate that
much of auditors regardless of their size earn discounted abnormal audit fee. Furthermore, the t-
test result for difference in means between Big N and non-Big N shows that the amount of
discounted abnormal audit fee (negative sign) of Big N clients is significantly smaller than that
of non-Big N at 1% significance level (0.004<0.069). This initial result indicates that clients of
Big N firms incur smaller discount of abnormal audit fee (negative) compared to the non-Big N
firms, without controlling for other variables.
The characteristics of the sampled firms are also shown in Table 4. 17% of the sampled
Japanese firms are classified as keiretsu companies, 12 % went public during the last 9 years, and
about 50% of the samples are listed on the first section of the Tokyo Stock Exchange. The
comparison of the mean differences between Big N and non-Big N clients shows that 15 out of
21 regression covariates are significantly different, suggesting that it is necessary to employ
regression analysis to control for the differences in clients’ characteristics.
Page 148
Table 5
CORRELATION MATRICES: PEARSON CORRELATIONS LEFT BELOW AND SPEARMAN
CORRELATIONS RIGHT ABOVE
SALESV
VOLATIL SALESG CFOVOL SALESAB KEIRETS
ABAFEE lnASSET LIAB ATURN CURR DOCF lgACCR CASH LOSS BETA lgLOSS OLATILI SAF2002 CONSOL ROA IPO9 TSE1
ITY ROWTH ATILITY ROAD U
TY
ABAFEE 1 -0.081 0.050 0.056 -0.026 -0.024 -0.118 0.060 0.039 0.067 0.167 0.119 0.089 -0.237 0.127 -0.059 -0.408 -0.090 -0.048 -0.070 0.031 0.042
lnASSET -0.013 1 0.105 -0.100 -0.077 0.003 -0.035 -0.213 -0.131 0.018 -0.074 -0.137 -0.130 0.062 -0.218 0.047 0.067 0.081 0.127 0.163 -0.158 -0.024
LIAB 0.044 0.075 1 0.216 -0.762 0.030 0.010 -0.338 0.072 -0.030 0.176 0.127 0.125 -0.029 0.053 -0.649 0.064 -0.260 -0.097 -0.101 0.018 -0.052
ATURN 0.008 -0.115 0.120 1 -0.153 0.038 0.033 0.019 -0.078 0.000 0.017 -0.059 0.430 0.153 0.196 0.135 0.003 0.125 -0.209 -0.098 0.152 -0.124
CURR -0.027 -0.063 -0.346 -0.176 1 -0.004 0.058 0.487 -0.137 0.016 -0.078 -0.098 -0.030 0.052 0.072 0.520 -0.051 0.323 0.119 0.068 -0.002 0.045
DOCF -0.016 -0.005 0.026 0.040 0.021 1 0.479 0.061 -0.088 -0.057 0.020 0.092 0.046 0.131 0.023 0.043 0.019 0.113 -0.007 -0.007 0.022 0.002
lgACCR -0.111 -0.035 -0.017 0.071 0.024 0.519 1 -0.018 -0.052 0.034 -0.036 -0.130 0.005 0.069 0.014 -0.003 0.041 0.039 -0.007 0.001 0.037 0.071
CASH 0.027 -0.211 -0.193 -0.007 0.461 0.084 -0.016 1 -0.031 0.018 0.075 0.025 0.129 0.027 0.182 0.275 -0.043 0.193 -0.034 -0.055 0.130 -0.048
LOSS 0.039 -0.131 0.079 -0.065 -0.109 -0.076 -0.056 -0.043 1 0.057 0.208 0.280 0.114 -0.304 0.101 -0.429 -0.093 -0.627 -0.028 -0.021 0.008 -0.013
BETA 0.038 0.029 -0.034 0.000 0.007 -0.045 0.023 0.002 0.050 1 0.024 -0.064 -0.004 -0.024 -0.002 0.006 0.041 -0.048 0.001 0.005 -0.003 0.004
VOLATILITY 0.137 -0.141 0.151 -0.002 -0.070 0.043 -0.056 0.084 0.216 -0.018 1 0.207 0.287 -0.096 0.233 -0.245 -0.101 -0.100 0.031 -0.039 0.094 0.023
lgLOSS 0.113 -0.136 0.176 -0.043 -0.054 0.085 -0.147 0.011 0.280 -0.052 0.229 1 0.141 -0.136 0.122 -0.315 -0.092 -0.302 -0.025 -0.026 -0.003 -0.009
SALESVOLATILITY 0.046 -0.142 0.116 0.475 -0.049 0.043 0.032 0.127 0.098 -0.008 0.233 0.121 1 0.126 0.350 -0.046 -0.066 0.026 -0.006 -0.073 0.153 -0.026
SALESGROWTH -0.170 0.038 -0.039 0.127 0.068 0.139 0.078 0.075 -0.280 -0.026 -0.064 -0.084 0.187 1 -0.063 0.187 0.215 0.419 0.072 0.011 0.061 0.039
CFOVOLATILITY 0.083 -0.228 0.089 0.186 0.019 0.053 0.038 0.210 0.092 -0.004 0.249 0.120 0.359 -0.019 1 -0.070 -0.110 -0.004 -0.037 -0.057 0.119 0.037
SAF2002 -0.064 0.073 -0.466 0.097 0.390 0.041 0.007 0.267 -0.464 0.001 -0.275 -0.336 -0.085 0.173 -0.132 1 -0.013 0.656 0.042 0.051 0.038 0.020
CONSOL -0.146 0.300 0.100 -0.021 -0.057 0.007 -0.020 -0.074 -0.062 0.052 -0.079 -0.050 -0.040 0.107 -0.106 -0.023 1 0.117 0.022 0.021 0.022 -0.017
ROA -0.071 0.122 -0.167 0.072 0.244 0.116 0.063 0.187 -0.699 -0.052 -0.162 -0.315 -0.041 0.355 -0.036 0.703 0.055 1 0.104 0.038 0.059 0.040
SALESABROAD -0.014 0.148 -0.105 -0.179 0.117 0.000 -0.031 0.002 -0.018 0.015 0.011 -0.017 -0.003 0.047 -0.038 0.072 0.045 0.097 1 0.088 -0.078 0.592
KEIRETSU -0.058 0.140 -0.078 -0.094 0.037 -0.006 -0.005 -0.061 -0.021 0.006 -0.052 -0.026 -0.077 0.000 -0.059 0.058 0.012 0.039 0.074 1 -0.149 0.068
IPO9 0.036 -0.158 0.037 0.154 0.004 0.032 0.044 0.155 0.008 -0.017 0.134 -0.003 0.190 0.074 0.164 0.006 -0.020 0.013 -0.064 -0.149 1 0.060
TSE1 0.039 -0.024 -0.048 -0.147 0.037 0.007 0.063 -0.052 -0.013 -0.001 0.031 -0.009 -0.038 0.033 0.004 0.033 -0.024 0.040 0.497 0.068 0.060 1
Table 5 presents the Pearson and Spearman correlation coefficients among the variables
used in this study. I conclude that the direction of correlations coefficients signs is consistent
with the expectation. It also does not find strong multicollinearity effect among the variables.
Results
Table 6 shows the estimation results of the logistic regression model (3), in which 11 out
of 21 variables are significant at least the 5% level. These results show that companies with the
following characteristics have a higher propensity to select Big N auditors: larger size
(lnASSET), higher assets turnover (ATURN), higher liquidity (CURR), hold more cash (CASH),
have higher sales (SALESVOLATILITY) and cash flow volatility (CFOVOLATILITY), less
potential for bankruptcy (SAF2002), have more complex businesses (CONSOL), belong to a
keiretsu (KEIRETSU), recently listed (IPO9), and are listed in the Tokyo Stock Exchange
(TSE1).
Table 6
RESULTS OF LOGISTIC REGRESSION (3) TO CALCULATE PROPENSITY SCORE
Big N Predict ed Sign Coefficient z-Value
_cons ? -2.061 -6.410 ***
lnASSET + 0.257 11.460 ***
LIAB ? 0.013 0.930
ATURN ? 0.142 2.200 **
CURR + 0.064 2.890 ***
DOCF ? 0.246 0.500
lgACCR ? 0.862 0.330
CASH - -2.343 -7.020 ***
LOSS ? 0.151 1.550
BETA ? -0.011 -0.190
VOLATILITY ? -0.684 -1.380
lgLOSS ? -0.059 -0.770
SALESVOLATILITY ? -1.000 -3.050 ***
SALESGROWTH ? 0.102 0.520
CFOVOLATILITY ? 2.583 2.760 ***
SAF2002 - 0.444 3.420 ***
CONSOL + 0.007 2.320 **
ROA ? 1.300 1.220
SALESABROAD + 0.000 -0.030
KEIRETSU - -0.418 -5.910 ***
IPO9 + 0.325 3.750 ***
TSE1 + 0.186 2.140 **
iD included
yD included
lik elihood -5033.9588
Pseudo R2 0.065
No. Obs 9,716
***, ** and * indicat e significance at t he 0.01, 0.05, and 0.10 resp ect ively .
Page 149
Table 7
RESULTS OF THE LOGISTIC REGRESSION (2)
Model 1 Model 2
without weight with weight
ABAFEE Coef. t-Stat Coef. z-Stat
Intercept 0.078 3.060 *** 0.108 3.350 ***
BigN 0.075 18.080 *** 0.078 18.250 ***
No. Obs 9,716 9,716
Models 1 and 2 of Table 7 describe the estimation results of Equation (2) before and after
employing IPW, respectively. The BigN variable coefficients of both models are significantly
positive, suggesting that auditor size is positively associated with abnormal audit fees and
demonstrating that the Japanese Big 3 auditors earn higher abnormal audit fees compared to non-
Big 3 firms. This result supports my hypothesis which in turn indicates that Big N auditors have
higher audit quality compared to non-Big N firms.
ROBUSTNESS TEST
Table 8
RESULTS OF THE ROBUSTNESS ANALYSIS
Panel A: Analysis using positive and negative abnormal audit fees dataset
Model 1 Model 2
without weight with weight
ABAFEE (+) Coef. t-Stat Coef. z-Stat
Intercept -0.178 -7.410 *** -0.129 -4.080 ***
BigN 0.018 4.150 *** 0.023 5.460 ***
No. Obs 4,475 4,475
ABAFEE (-) Coef. t-Stat Coef. z-Stat
Intercept 0.109 3.470 *** 0.108 3.020 ***
BigN 0.041 8.740 *** 0.045 9.950 ***
No. Obs 5,241 5,241
Panel B: Propensity Score Marching (PSM) *
Model 1 Model 2
Total Sample PSM Sample
Difference in mean t-Stat Difference in mean t-Stat
ABAFEE 0.065 13.642 *** 0.067 12.768 ***
No. Obs 9,716 9,220
* winsorized at 1% & 99%; Kernel matching-epanechnikov kernel
Panel C: Coarsened Exact Matching (CEM)
Model 1 Model 2
Total Sample CEM Sample
Difference in mean t-Stat Difference in mean t-Stat
ABAFEE 0.065 13.642 *** 0.026 1.721 *
No. Obs 9,716 170
Note: Model 1 and 2 in Panel B and C is different from Panel A.
Page 150
Choi, Kim, & Zang (2010) find that when they classify the sample into negative and
positive abnormal audit fee data-sets, the positive data-set shows a negative relationship between
abnormal audit fees and audit quality (absolute discretionary accruals), while the negative data-
set does not show any significance. Hence, this study classifies the sample into the positive and
negative abnormal audit fee groups and estimate Equation 2 using those two groups. The results
are presented listed in Panel A of Table 8.
To further robustness test of the empirical results, this study applies two matching
procedures. First, I employ the Rosenbaum and Rubin’s (1983) propensity score matching (PSM)
method that is commonly used in accounting literature (e.g., Armstrong, Jagolinzer, & Larcker,
2010). Panel B of Table 8 shows the results of the mean difference of discounted abnormal audit
fee between Big N and non-Big N for the PSM matched sample (Model 2) is consistent with the
results from the descriptive statistics (Model 1 – see Table 4).
For the second matching procedure, I use the coarsened exact matching (CEM) method
(DeFond, Erkens, & Zhang, 2014). In the CEM model, Big N is used as the treatment variable
and then matched with the 11 significant pretreatment variables from the logistic model (see
Table 6). Panel C of Table 8 shows the results of using 170 CEM matched samples as a
robustness test. The t-test difference in means for the CEM matched sample shows that the
abnormal audit fees of the Big N firms are statistically higher than those of non-Big N firms,
which also supports my prior hypothesis.
DISCUSSION OF RESULTS
Consistent with prior empirical literature (Blankley, Hurtt, & MacGregor, 2012; Hribar,
Kravet, & Wilson, 2014), I assume that higher abnormal audit fee is associated with higher audit
effort which translates to higher audit quality. Using a logistic regression model, the results of
the logistic regression estimates after considering the inverse probability weighting (IPW)
method (Table 7) support this study’s hypothesis that Big N firms earn higher abnormal audit fee
compared to non-Big N firms. The evidence of this study implies that both the pricing-based and
characteristics-based proxy for audit quality ‒ abnormal audit fee and auditor size, respectively ‒
is empirically consistent with each other.
In addition, I perform three additional robustness tests (Table 8) and results of the tests
support my hypothesis. In Table 8 Panel A, the sample is split into two groups according to the
sign of the ABAFEE variable: positive (+) ABFEE group and negative (-) ABFEE group. For
both of these groups, the positive and significant value for BIGN coefficient indicates that the
higher abnormal audit fee paid to Big N firms is consistent in both positive (premium) abnormal
audit fee and negative (discount) abnormal audit fee conditions. Panel B and C show the results
of sample matching analysis using the propensity score matching (PSM) and coarsened exact
matching (CEM) method respectively. The results from both Panel B and C indicate that Big N
firms earn more abnormal audit fee compared to non-Big N firms from their clients. Overall, the
results of the robustness tests presented in Table 8 are consistent with my prior hypothesis and
provide additional empirical evidence that Big N firms earned higher abnormal audit fee than
non-Big N firms which in turn indicates a higher audit quality.
Page 151
CONCLUSIONS
This study empirically investigates the audit quality difference measured by abnormal
audit fees, a pricing-based proxy for audit quality, between the Big 3 and non-Big 3 auditors
using a sample of Japanese listed firms between 2005 and 2011. There has been a lack of
literature analyzing of the association between the pricing-based audit quality proxy‒ abnormal
audit fee ‒ with the characteristics-based audit quality proxy ‒ auditor size ‒ and this study
attempts to contribute empirical evidence to the literature. The IPW method and robustness test
results show that auditor size is positively associated with abnormal audit fees, indicating that
Big N auditors do earn higher fees than non-Big N auditors. Assuming that the audit fee
premium motivates and expands audit efforts, this study empirically demonstrates that large
Japanese auditors do provide higher audit quality.
This study’s sample is limited to Japanese firms which limit the external validity of my
empirical findings to other countries that share similar characteristics with Japanese audit and
accounting environment.
AUTHOR’S NOTE
The author wishes to thank No. 24730385 of Grants-in-Aid for Young Scientists (B) and
No. 15K03768 of Grant-in-Aid for Scientific Research (C) in Japan for financial support towards
this research.
REFERENCES
Armstrong, S. C., Jagolinzer, A. D., & D. F Larcker (2010). Chief executive officer equity incentives and accounting
irregularities. Journal of Accounting Research, 48(2), 225-271.
Asthana, S. C. & J. P. Boone (2012). Abnormal audit fee and audit quality. Auditing: A Journal of Practice &
Theory, 31(3), 1-22.
Becker, C. L., DeFond, M. L., Jiambalvo, J. & K. R. Subramanyam (1998). The effect of audit quality on earnings
management. Contemporary Accounting Research, 15(1), 1-24.
Behn, B., Choi, J. H., & T.Kang (2008). Audit quality and properties of analyst earnings forecasts. The Accounting
Review, 83(2), 327–359.
Blankley, A. I., Hurtt, D. N., & J. E.MacGregor (2012). Abnormal audit fees and restatements. Auditing: A Journal
of Practice & Theory, 31(1), 79–96.
Boone, J. P., Khurana, I. K., & K. K. Raman(2010). Do the Big 4 and the second-tier firms provide audits of similar
quality? Journal of Accounting and Public Policy, 29(4), 330–352.
Chaney, P.K., Jeter, D.C., & L.Shivakumar (2004). Self-selection of auditors and audit pricing in private firms. The
Accounting Review, 79(1), 51-72.
Choi, J., Kim, C., Kim, J., & Y. Zang(2010). Audit office size, audit quality, and audit pricing. Auditing: A Journal
of Practice & Theory, 29(1), 73–97.
Choi, J., Kim, J., & Zang, Y(2010). Do abnormally high audit fees impair audit quality? Auditing: A Journal of
DeFond, Mark, David Hendrik Erkens, & J. Zhang (2014). Do Client Characteristics Really Drive the Big N Effect?
Evidence from Matching Methods. Working Paper.
Eshleman, J. D. & P.Guo (2014a). Abnormal audit fees and audit quality: The importance of considering managerial
incentives in tests of earnings management. Auditing: A Journal of Practice & Theory, 33(1), 117-138.
Page 152
Eshleman, J. D., & P.Guo (2014b). Do Big 4 auditors provide higher audit quality after controlling for the
endogenous choice of auditor? Auditing: A Journal of Practice & Theory, 33(4), 197–219.
Francis, J., & D. Wang (2008). The joint effect of investor protection and big 4 audits on earnings quality around the
world. Contemporary Accounting Research, 25(1), 1-39.
Francis, J. R., Maydew, E. L. & H. C. Sparks (1999). The role of big 6 auditors in the credible reporting of accruals.
Auditing: A Journal of Practice & Theory, 18(2), 17-34.
Francis, J. R., Michas, P. N., & S. E. Seavey (2013). Does audit market concentration harm the quality of audited
earnings? Evidence from audit markets in 42 countries. Contemporary Accounting Research, 30(1), 325–
355.
Francis, J. R., & M.Yu(2009). Big four office size and audit quality. The Accounting Review, 84(5), 1521–1552.
Fukuyama, F. (1995). Trust: The social virtues and the creation of prosperity. London: Hamish Hamilton.
Gul, F. A., Zhou, G., & X. Zhu (2013). Investor protection, firm informational problems, Big N auditors, and cost of
debt around the world. Auditing: A Journal of Practice & Theory, 32(3), 1–30.
Hoitash, R., Markelevich, A. & C. A. Barragato (2007). Auditor fees and audit quality. Managerial Auditing
Journal, 22(8), 761-786.
Hribar, P., Kravet, T., & R.Wilson (2014). A new measure of accounting quality. Review of Accounting Studies,
19(1), 506–538.
Hu, D. & R. Kato (2014). Auditor size as a measure for audit quality: A Japanese Study. Working paper. Nagoya
University.
Hu, D. & R. Kato (2015). Accruals-based audit quality in the Japanese audit market.” Academy of Accounting and
Financial Studies Journal, 19(1), 186-197.
Khurana, I., & K.Raman (2004). Litigation risk and the financial reporting credibility of Big 4 versus non-Big 4
audits: Evidence from anglo-american countries. The Accounting Review, 79(2), 473-495.
Knechel, W. R., Krishnan, G. V., Pevzner, M., Shefchik, L. B., & U.K.Velury (2013). Audit quality: Insights from
the academic literature. Auditing: A Journal of Practice & Theory, 32(Supplement 1), 385–421.
Lawrence, A., Minutti-Meza, M., & P.Zhang(2011). Can Big 4 versus non-Big 4 differences in audit-quality proxies
be attributed to client characteristics? The Accounting Review, 86(1), 259–286.
Li, K. & N. Prabhala (2007). Self-selection models in corporate finance. in Eckbo, B. Espen (Ed.), Handbook of
Corporate Finance: Empirical Corporate Finance Vol. 1, Elsevier Science B.V., Amsterdam, 37-86.
Louis, H. (2005). Acquirers’ abnormal returns and the non-Big 4 auditor clientele effect. Journal of Accounting and
Economics, 40(1–3), 75–99.
Reichelt, J. K., & D. Wang (2010). National and office-specific measures of auditor industry expertise and effects on
audit quality. Journal of Accounting Research, 48(3), 647–686.
Rosenbaum, P.R. & D. B. Rubin (1983). The central role of the propensity score in observational studies for causal
effects. Biometrika, 70(1), 41-55.
Rubin, D. B. (1985). The use of propensity scores in applied bayesian inference. Bayesian Statistics, 2, 463-472.
Skinner, D. J., & S. Srinivasan (2012). Audit quality and auditor reputation: evidence from Japan. The Accounting
Review, 87(5), 1737-1765.
Shima, K., & E. Gordon (2011). IFRS and the regulatory environment: the case of U.S. investor allocation choice.
Journal of Accounting and Public Policy, 30 (5), 481-500.
Shirata, Y. (2003). Corporate bankruptcy prediction model. Tokyo: Chuokeizai Sha (in Japanese).
Sirois, L.-P., & D.A.Simunic (2011, July). Auditor Size and Audit Quality Revisited: The Importance of Audit
Technology. Working Paper, University of British Columbia.
Teoh, S. H., & T.J.Wong (1993). Perceived auditor quality and the earnings response coefficient. The Accounting
Review, 68(2), 346–366.
Tucker, J. W. (2010). Selection bias and econometric remedies in accounting and finance research. Journal of
Accounting Literature, Vol. 29, 31-57.
Yazawa, K. (2010). Big 4 and audit quality: Audit cost hypothesis and conservative accounting hypothesis. Aoyama
Journal of Business, 44(4), 167-181 (in Japanese).
Page 153
EARNINGS MANAGEMENT TO REPORT AN

ADDITIONAL CENT OF EPS: EVIDENCE FROM PRE-
AND POST-SOX PERIODS
Charles E. Jordan, Florida State University-Panama City
Stanley J. Clark, Middle Tennessee State University
Gwen R. Pate, University of Southern Mississippi
ABSTRACT
Prior research suggests certain types of earnings management (e.g., discretionary

earnings management) declined noticeably after implementation of the Sarbanes-Oxley Act
(SOX) while other forms (e.g., real earnings management) increased in the post-SOX era.
The current research examines a form of biased reporting documented in the literature whereby
management manipulates income so that EPS can be rounded up to present an additional cent of
EPS. In particular, prior researchers found that when they re-calculated firms’ EPS amounts,
the third digit to the right of the decimal point fell in the numerical range five through nine
inordinately more often than would be expected. Thus, the rounding of these calculated EPS
figures by management to report them in the financial statements in only two places to the right
of the decimal point resulted in EPS being increased by an additional cent much more frequently
than should have occurred by chance. We replicate the previous research in this area but do so
for distinct pre- and post-SOX periods. The findings reveal clear evidence of this manipulative
behavior prior to SOX’s implementation but no signs of it afterward, thus suggesting SOX played
a key role in the demise of this form of biased financial reporting.
INTRODUCTION
Schipper (1989, p. 92) takes a relatively broad view of earnings management by

indicating that the notion really embodies disclosure management “in the sense of a purposeful
intervention in the external financial reporting process, with the intent of obtaining personal
gain (as opposed to say, merely facilitating the neutral operation of the process).” This view
suggests various financial statement items could be the object of manipulation. For example,
Cullen et al. (2008) provide evidence that companies with a history of reporting net losses and
negative operating cash flows often manipulate sales revenue (rather than earnings) to boost
market value.
Even if one confines disclosure management strictly to earnings management,
research shows that various forms of earnings are manipulated and for different reasons. For
example, income may be managed to ensure that reported earnings per share (EPS) meets
analysts’ forecasts (e.g., Payne & Robb, 2000). Alternatively, studies (e.g., Guan et al., 2006;
Lin et al. 2014) suggest that net income is frequently manipulated upward to reach cognitive
reference points for users (e.g., unmanipulated earnings of $3.94 million would be enhanced to
just above $4.00 million). Regardless of the particular financial statement item being
manipulated, a key point of Schipper’s (1989) definition of disclosure (earnings)
management is management’s intervention in the outcome of the financial reporting process
rather than their neutral facilitation of it. That is, disclosure (earnings) management results in
Page 154
biased financial reporting.

Recent research suggests that at least some forms of earnings management declined
significantly in the U.S. following the implementation of the Sarbanes-Oxley Act of 2002
(SOX). For example, Bartev and Cohen (2009) and Cohen et al. (2008) present evidence that
accrual-based earnings management declined markedly after SOX. Aono and Guan (2008), Lin
and Wu (2014), and Wilson (2012) indicate that earnings manipulation to reach user reference
points in income largely disappeared after SOX. Similarly, Jordan and Clark (2013) provide
results suggesting the upward manipulation of sales revenue to reach user reference points
vanished after SOX as well.
Although some studies show that particular forms of earnings management declined
precipitously after SOX, research also indicates that other types (e.g., real earnings management)
actually increased in the post-SOX era (e.g., see Bartev & Cohen, 2009; Cohen et al. 2008).
Still, other research suggests “that the regulatory changes initiated under SOX to empower
boards and audit committees” produced little, if any, effect on the incidence of earnings
management (Ghosh et al. 2010, p. 1145). Thus, SOX’s impact on earnings management appears
inconclusive at this point, and further research in the area is warranted.
The current article examines a specific form of earnings management documented in the
literature whereby income is manipulated so that EPS can be rounded up to report one more cent
of EPS (Das & Zhang, 2003; Jorgensen et al. 2014; Miller et al. 2012). The present research
replicates the key portion of these studies but does so in both pre- and post-SOX eras to
determine if this corporate governance legislation produced any apparent effects on
management’s propensity to manipulate income to report an additional cent of EPS. The results
provide clear evidence of this manipulative behavior prior to SOX but no indications of it in the
post-SOX period, thus implying SOX produced a positive impact by contributing to the
elimination of this form of biased financial reporting.
The next section discusses the literature relative to the intentional rounding up of EPS
and also develops the research question for the study. This is followed by sections on
methodology and data and results. The final section contains our discussion and conclusion.
LITERATURE REVIEW
Degeorge et al. (1999) suggest that earnings are managed in order to meet three
thresholds considered important to users. For example, the researchers note that earnings might
be manipulated to ensure achieving positive income, rather than reporting negative or zero
earnings. Another threshold to achieve is generating earnings at least equal to that o f the prior
period; finally, Degeorge et al. (1999) note that analyst earnings forecasts represent a threshold
to which income is managed.
In addition to the three rather obvious earnings thresholds indicated above, other
benchmarks have been discovered as well. For example, research (e.g., Carslaw, 1988;
Kinnunen & Koskela, 2003; Lin et al. 2014; Skousen et al. 2004; Van Caneghem, 2002) has
shown that income has frequently been managed upward to increase the first (left-most) digit in
the earnings number by one when the second-from-the-left digit of unmanipulated income is
relatively high (e.g., unmanipulated earnings of $5.94 million would be managed upward to
slightly above $6.00 million). Thomas (1989) notes that even this relatively small upward
manipulation around a user reference point (threshold) can produce a significant impact on firm
value.
Page 155
Thomas (1989) observed that EPS numbers were manipulated (rounded) upward to
thresholds more frequently than net income figures. As an example, in a large sample of listed
U.S. companies, he discovered unusually high rates of EPS numbers reported in multiples of 5
cents and 10 cents; nines occurred in the right-most EPS digit (i.e., the cent position) much less
often than anticipated. These findings suggested earnings were often manipulated upward so that
EPS could be reported in round numbers (i.e., user thresholds).
Das and Zhang (2003) noted that very small upward manipulations of income can
frequently lead to an additional cent of EPS through the rounding of the third digit to the right of
the decimal point in calculated EPS numbers. For example, a company with net income of
$3,868,800 and 5,200,000 shares of common stock outstanding would have a calculated EPS of
$.744. Since EPS numbers are reported only to the cent position (i.e., two digits to the right
of the decimal point), traditional rounding would dictate that this calculated EPS be rounded
down and reported as $.74 in the financial statements. Yet, if net earnings were manipulated
upward by as little as $5,200, the calculated EPS would be $.745, which would be rounded up
and reported as $.75 in the financial statements. Das and Zhang (2003) pointed out that the
rounding of EPS is not manipulation per se; however, managing earnings upward to allow the
rounding up of EPS clearly represents biased financial reporting, especially when the rounding
up to report an additional cent of EPS results in meeting analysts’ forecasts or other thresholds.
Das and Zhang (2003) stated that in the absence of earnings manipulation, the third digit
to the right of the decimal point in calculated EPS (hereafter referred to as the third decimal
digit) should occur in the numerical range zero through four 50% of the time and in the range
five through nine 50% of the time as well. They posited that if management manipulates
income upward so that the rounding of EPS results in an additional cent of EPS, the third
decimal dig it in calculated EPS would appear in the five through nine range significantly more
often than expected and in the zero through four range at an abnormally low rate.
Using quarterly data from listed U.S. companies for the period 1989-1998, Das and
Zhang (2003) calculated primary (basic) EPS before extraordinary items using the firms’
appropriate income and weighted average number of shares provided in COMPUSTAT. In
essence, they recalculated the companies’ reported EPS, but carried the computation to the third
decimal digit. For the firms reporting profits, 54.6% of them had calculated EPS with the third
decimal digit falling in the five through nine range. This was significantly higher than the
expected proportion (50%), and of course means these firms were able to round up EPS and
report one more cent of EPS. Conversely, an abnormally low portion (45.4% versus 50%) of the
companies had a calculated EPS with the third decimal digit in the zero through four range.
These firms did not get to report an additional cent of EPS.
To ensure their expected proportions (i.e., 50% each for the numerical ranges zero
through four and five through nine appearing in the third decimal digit) were appropriate,
Das and Zhang (2003) calculated per share amounts for sales, operating income before
depreciation, and operating cash flows as control variables. They selected these per share
computations as, unlike EPS, there would be no incentive to round them up. The researchers
found that the calculated per share amounts for the three control variables all produced third
decimal digits that fell in the zero through four and five through nine ranges at frequencies
approximating 50%, which suggested that 50% is the appropriate expected percentage for
unmanipulated data. Their overall results indicated and the researchers concluded that
management frequently manipulates earnings so that EPS can be rounded up to report an
additional cent of EPS.
Page 156
Similar to Das and Zhang (2003), Miller et al. (2012) calculated the quarterly EPS
numbers for listed U.S. companies but did so for the period 1995-2010. Not surprisingly, their
results closely resembled those of Das and Zhang (2003) in that for firms with positive income,
the third decimal digit in calculated EPS ranged from five through nine 53% of the time, with
47% of them falling in the zero through four range. Both of these proportions differed
significantly from the expected rate of 50%.
Miller et al.’s (2012) primary contribution to the literature in this area was their discovery
that the propensity to manipulate earnings to provide an additional cent of EPS through rounding
the third decimal digit rose dramatically the smaller the absolute size of EPS. For example, as
noted above, for their full sample of firm-quarters, the third decimal digit occurred in the five
through nine range 53% of the time. However, when the sample was pared down to include only
firm-quarters with EPS of $.20 or less, the portion in the five through nine range increased to
54.5%. For firm-quarters with EPS of $.05 or less, the frequency in the five through nine range
rose to 57.5%.
Jorgensen et al. (2014) also replicated the Das and Zhang (2003) study but used annual
data and examined a longer time period (i.e., 1980-2010). EPS was computed and presented
under the rules of Accounting Principles Board Opinion (APBO) No. 15 prior to December 1997
and under the rules of Statement of Financial Accounting Standards (SFAS) No. 128 subsequent
to November 1997. For the APBO No. 15 period, the third decimal digit in calculated EPS
occurred in the five through nine range 53.29% of the time, which was significantly higher than
the expected rate of 50%. For the SFAS No. 128 period, the third decimal digit appeared in the
five through nine range 51.11% of the time, which although lower than what occurred in the
APBO No. 15 era, was still significantly greater than the expected frequency of 50%.
Research Question
Das and Zhang (2003), Miller et al. (2012), and Jorgensen et al. (2014) all provide
evidence indicating income is frequently manipulated to allow EPS to be rounded up so that an
additional cent can be added to the EPS figure reported in the financial statements. The data for
the Das and Zhang (2003) study came entirely from a period (1989-1998) prior to SOX’s
implementation in 2002. As then SEC chairman Arthur Levitt noted in 1999, this period of time
was marked by widespread earnings management in the U.S. (Ketz, 1999). Thus, the findings of
Das and Zhang (2003) come as no real surprise. The two later studies, though, contained samples
with both pre- and post-SOX data. That is, the Miller et al. (2012) sample period was 1995-
2010, while the Jorgensen et al. (2014) sample spanned the 30-year period 1980-2010. Yet,
neither study split its sample into distinct pre- and post-SOX eras so that comparisons could
be made between the two periods relative to management’s propensity to manipulate income to
allow the rounding up of EPS to report an additional cent of EPS.
As noted in the introduction, several studies indicate that at least some forms of earnings
management declined significantly after the implementation of SOX (e.g., Aono & Guan, 2008;
Bartev & Cohen, 2009; Cohen et al. 2008; Jordan & Clark, 2013). This leads to the primary
research question in the current study. In particular, did the propensity to manipulate income so
that EPS can be rounded up by an additional cent change after SOX? To address this question,
tests are performed to detect this form of earnings management in dist inct pre- and post-
SOX samples. Based on prior research (i.e., Das & Zhang, 2003; Jorgensen et al. 2014; Miller
et al. 2012), it is anticipated that, for the pre-SOX period, the third decimal digit of
calculated EPS will occur in the five through nine range significantly more often than would
Page 157
be expected by chance. The real point of this study, though, is to ascertain whether SOX seems
to have constrained this form of earnings management, which would be suggested if the third
decimal digit in the post-SOX era appears in the five through nine range at a significantly lower
rate than what is observed for the pre-SOX period.
METHODOLOGY AND DATA
The methodology used in the current study echoes that of the prior research in this area
(i.e., Das & Zhang, 2003; Jorgensen et al. 2014; Miller et al. 2012). In particular, for listed U.S.
companies with positive income, basic EPS is calculated by dividing income before
extraordinary items by the weighted-average number of common shares outstanding; both
figures are provided in the Fundamental Annuals files in COMPUSTAT and are the same
numbers the companies used to compute their reported year-end EPS figures.
The third decimal digit of calculated EPS represents the key measure being examined. As
Das and Zhang (2003) note, absent manipulation, the third decimal digit should occur in the
numerical range zero through four approximately 50% of the time and likewise should appear in
the range five through nine about 50% of the time as well. If the third decimal digit occurs in the
five through nine range significantly more often than expected (and in the zero through four
range at an abnormally low rate), anecdotal evidence would exist that management
frequently manipulated income upward to ensure that EPS would be rounded up by an additional
cent.
Benford (1938) demonstrated that the numbers zero through nine do not appear
proportionally in the various digital positions of naturally occurring data (i.e., not contrived by
man). In particular, in the two left-most digital positions of a number, low digits (e.g., one
or two) have a much higher probability of appearing than high digits (e.g., eight or nine).
Nigrini (1999) points out that the expected digital frequencies developed by Benford are scale
invariant, which means they hold true even if all numbers in a population are multiplied or
divided by a constant. For example, Benford’s digital rates would apply if financial numbers
denominated in Mexican pesos were divided by the appropriate exchange rate to convert them
to U.S. dollars. Studies have shown that unmanipulated income conforms to Benford’s digital
frequencies (e.g., see Rodriguez, 2004; Jordan et al., 2014).
While net earnings adhere to Benford’s rates, EPS would not be expected to demonstrate
these same digital distributions since EPS figures are not naturally occurring numbers but instead
result from dividing income by a non-constant variable that is unique to each company (i.e.,
number of common shares). Instead, as Das and Zhang (2003), Miller et al. (2012), and
Jorgensen et al. (2014) note, each of the numbers zero through nine should appear in the various
digital positions to the right of the decimal point in EPS at proportional rates (i.e., about 10%
of the time). Hence, 50% of the third decimal digits of calculated EPS should fall in the
numerical range zero through four, while 50% should also occur in the range five through nine.
The frequencies of the numbers appearing in the third decimal digit of calculated EPS are
examined for pre- and post-SOX samples. As Miller et al. (2012) demonstrated, the incentive to
manipulate earnings to ensure that EPS will be rounded up by an additional penny increases
as the size of the EPS figure decreases. For example, adding a penny to EPS of $.10 increases
EPS by 10%, while adding a cent to EPS of $1.10 enhances EPS by less than 1%. For this
reason, the samples in the current study are confined to companies with calculated EPS
between $.01 and $1.00 as these firms would be more inclined to manipulate earnings to add
an additional penny of EPS than would a sample of companies with widely diverging amounts
Page 158
of EPS.
The post-SOX sample comprises all firm-years in COMPUSTAT with calculated EPS in
the range noted above for the years 2003-2012. The pre-SOX sample contains all firm-
years meeting these same parameters but with financial year-ends from December 1997 through
December 2000. The pre-SOX sample period begins in December 1997 because this is when
companies had to start computing and reporting EPS in conformity with SFAS No. 128. Prior to
this date, EPS was calculated under the rules of APBO No. 15, and to cut down on any potential
noise, we wanted to ensure that companies in the pre- and post-SOX samples used the same
reporting standard for EPS (i.e., SFAS No. 128). The years 2001 and 2002 are excluded from
both samples because it was during this period that the egregious financial scandals (e.g., Enron
and WorldCom) that eventually led to the passage of SOX came to light. In this period,
management likely suspected corporate governance legislation would soon be enacted, and these
two years would be considered neither truly pre-SOX nor post-SOX periods, but rather a
mezzanine era producing unnecessary noise if examined.
To ascertain whether the numbers in the ranges zero through four and five through nine
appear in the third decimal digit of calculated EPS at expected rates for the pre- and post-SOX
samples, their actual frequencies are compared to the anticipated rate (50%) using a two-tailed
proportions test. To determine whether SOX’s implementation appears to have affected
management’s propensity to manipulate earnings so that EPS can be rounded up by an
additional cent, a chi-square test is used to compare the pre- and post-SOX samples relative to
the rates of numbers appearing in the zero through four and five through nine ranges.
RESULTS
The pre-SOX sample comprises the 11,440 firm-years in COMPUSTAT’s Annual

Fundamentals file with calculated EPS ranging from $.01 to $1.00 for the period December 1997
through December 2000. The post-SOX sample contains the 20,241 firm-years meeting these
same parameters for the period 2003-2012. The post-SOX sample size exceeds that of the pre-
SOX sample simply because the pre-SOX period was shorter in duration. Even so, the
samples for both periods are sufficiently large enough for statistical testing and are representative
of firms existing during distinct pre- and post-SOX eras.
The main objective of this study is to determine whether a difference exists between
pre- and post-SOX eras in management’s propensity to manipulate earnings to gain an additional
cent of EPS by rounding up the third decimal digit. Das and Zhang (2003) have already
shown that this form of biased reporting occurred during a pre-SOX period (i.e., 1989-1998).
However, their data came largely from firm-quarters where EPS was reported under APBO No.
15, with SFAS No. 128 not becoming effective until late in their sample period. Thus, a question
remains concerning whether this manipulative behavior took place prior to SOX when EPS
was computed and presented under the current reporting standard (i.e., SFAS No. 128). In
addition, no study has tested for this form of earnings management in a distinct post-SOX era.
For the pre-SOX sample, Panel A of Table 1 provides the actual counts and
frequencies at which the third decimal digit of calculated EPS occurred in the numerical ranges
zero through four and five through nine. For example, the third decimal digit fell in the zero
through four range 5,558 times, which represents 48.58% of the total 11,440 firm-years for this
period, while 5,882 (51.42%) of the third decimal digits were in the five through nine range.
The expected frequency for numbers appearing in each of the two ranges was 50%. Thus, the
third decimal digit appeared in the zero though four range less often than expected and in the
Page 159
five through nine range more frequently than anticipated. Panel A also provides the Z statistic
and p-level for the proportions test comparing the actual rates with the expected rate of 50%,
and a key finding for the pre-SOX sample is that the difference between the actual and
expected frequencies is statistically significant (i.e., p-level of .003).
Table 1
THIRD DECIMAL DIGIT FREQUENCIES IN ZERO THROUGH FOUR AND FIVE THROUGH NINE
RANGES
Panel A (Pre-Sox Sample):
N = 11,440 Numerical Range of Third Decimal Digit:
Zero through four Five through nine
Actual count (n) 5,558 5,882
Actual frequency (%) 48.58 51.42
Expected frequency (%) 50.00 50.00
Z statistic -3.021 3.021
p-level .003* .003*
Panel B (Post-Sox Sample):
N = 20,241 Numerical Range of Third Decimal Digit:
Zero through four Five through nine
Actual count (n) 10,144 10,097
Actual frequency (%) 50.12 49.88
Expected frequency (%) 50.00 50.00
Z statistic .323 -.323
p-level .747 .747
*significant at .01 level
Panel B of Table 1 presents the results for the post-SOX sample. Notice that the third
decimal digit fell in the zero through four and five through nine ranges 50.12% and 49.88% of
the time, respectively. The proportions test reveals that these rates do not differ significantly
from the expected frequency of 50% (i.e., p-level of .747). It appears the biased reporting to
generate an additional penny of EPS that existed prior to SOX vanished in the post -SOX period.
Table 1 shows a comparison of the actual rates for the third decimal digits appearing in
the numerical ranges zero through four and five through nine for the pre- and post-SOX eras to
the expected rate of 50%. Table 2, though, presents a chi-square test for a direct comparison
of the rates existing in the pre-SOX period with those occurring in the post-SOX period. The
chi- square test statistic of 6.86 (p-level of .009) shows that the rates differ at a statistically
significant level, thus indicating a change occurred between the periods.
Table 2
CHI-SQUARE TEST COMPARING PRE- AND POST-SOX SAMPLES
Numerical Range of Third Decimal Digit:
Zero through four Five through nine Total
Pre-SOX sample 5,558 (48.58%) 5,882 (51.42%) 11,440 (100%)
Post-SOX sample 10,144 (50.12%) 10,097(49.88%) 20,241 (100%)
Total 15,702 15,979 31,681
Chi-square test statistic = 6.86; p-level = .009
Page 160
One final test involved an analysis of the rates at which each of the numbers zero through
nine appeared in the third decimal digit of calculated EPS for the pre- and post-SOX periods.
As noted previously, each number (zero through nine) should occur in the third decimal digit
approximately 10% of the time. Yet, Das and Zhang (2003) found that for their sample a
significant discontinuity occurred for the number four, with far fewer fours appearing in the third
decimal digit than expected. They surmised this happened because the closer the unmanipulated
third digit came to five (i.e., the traditional minimum cutoff for rounding up), the more
likely earnings were manipulated to ensure the third decimal digit fell in the range for upward
rounding (i.e., five through nine).
For the pre- and post-SOX periods, respectively, Panels A and B of Table 3 provide the
actual counts and rates at which each of the numbers zero through nine appeared in the third
decimal digit of calculated EPS; also provided are Z statistics and p-levels for proportions tests
comparing the actual frequency of each number to the expected rate of 10%. Notice in Panel
A the same discontinuity in the number four as observed by Das and Zhang (2003) occurred in
the pre-SOX sample. That is, fours appeared in the third decimal digit only 9.18% of the time,
which is well below the expected rate of 10%; the discrepancy between the actual and expected
rates of fours is statistically significant (i.e., p-level of .002). Using traditional measures of
statistical significance, the numbers six (p-level of .070) and eight (p-level of .024) appeared in
the third decimal digit for the pre-SOX sample at abnormally high rates. These results are
intuitive given both of these numbers fell in the range (five through nine) that would be rounded
up.
Table 3
THIRD DECIMAL DIGIT FREQUENCIES FOR NUMBERS ZERO THROUGH NINE
Panel A (Pre-Sox Sample)
N = 11,440
Third decimal digit 0 1 2 3 4 5 6 7 8 9
Actual count (n) 1155 1099 1134 1120 1050 1155 1204 1123 1219 1181
Actual frequency (%) 10.10 9.61 9.91 9.79 9.18 10.10 10.52 9.82 10.66 10.32
Expected frequency (%) 10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00
Z statistic .326 -1.412 -.297 -.739 -3.028 .326 1.813 -.644 2.258 1.122
p-level .745 .158 .766 .460 .002* .745 .070*** .519 .024** .262
Panel B (Post-SOX Sample)
N = 20,241
Third decimal digit 0 1 2 3 4 5 6 7 8 9
Actual count (n) 2045 1993 2048 2010 2048 1991 2025 2090 1983 2008
Actual frequency (%) 10.10 9.85 10.12 9.93 10.12 9.84 10.00 10.33 9.80 9.92
Expected frequency (%) 10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00 10.00
Z statistic .476 -.722 .545 -.320 .545 -.769 .009 1.511 -.960 -.367
p-level .634 .470 .585 .749 .585 .442 .993 .131 .337 .714
*,**,*** significant at .01, .05, and .10 levels, respectively
The findings for the post-SOX sample in Panel B of Table 3 stand in stark contrast
to those for the pre-SOX sample in Panel A. More specifically, for the post-SOX era, each
number (zero through nine) appeared in the third decimal digit at approximately its expected
rate, which suggests that during this period earnings were not manipulated to create desired
amounts in this digital position of calculated EPS. Taken together, the results shown in
Page 161
Tables 1, 2 and 3 indicate a definite decline occurred in management’s propensity to

manipulate income so that EPS could be enhanced by garnering an additional cent from
rounding, and this shift toward more neutral financial reporting corresponds with the
implementation of SOX.
DISCUSSION AND CONCLUSION
Prior research (Das & Zhang, 2003; Jorgensen et al. 2014; Miller et al. 2012) showed
that management frequently manipulated income so that EPS could be rounded up by an
additional penny. In particular, the studies provided evidence that the third decimal digit in
calculated EPS occurred in the numerical range five through nine (zero through four)
significantly more (less) often than would be expected for unmanipulated data. Since EPS is
presented in the financial statements only to the cent position (i.e., second decimal digit), having
an inordinate number of firms with the third decimal digit in the five through nine range
meant an abnormally high rate of companies were rounding up EPS to add an additional
cent to reported EPS.
The current study replicates the prior research in this area but does so using distinct
pre- and post-SOX samples to ascertain whether SOX appears to have constrained this biased
form of reporting. Not surprisingly given the results of the previous studies, the pre-SOX
sample possessed an abnormally high (low) rate of third decimal digits in calculated EPS in
the range five through nine (zero through four), suggesting that management during this period
often manipulated earnings so that reported EPS could be rounded up instead of down.
However, for the post-SOX sample, no anomalies appeared in the rates at which the numbers
zero through nine occurred in the third decimal digit. Both individual numbers and groupings of
numbers (i.e., zero through four and five through nine) appeared in the third decimal digit at
their expected rates. This finding provides anecdotal evidence that this form of biased financial
reporting, which existed during the pre-SOX era, disappeared in the post-SOX period.
These results suggest that SOX, with its harsh penalties for fraudulent financial reporting,
played a key role in the elimination of this particular form of earnings management. There is,
of course, a possibility that factors other than SOX led to the more altruistic financial
reporting observed in recent years. For example, Yoon and Miller (2002) found that the
propensity to manage earnings is inversely related to operating performance, with managers
of poor performing firms more prone to manipulative behavior than managers of companies
enjoying strong financial performance. Although the pre- and post-SOX samples in the current
study contained only firm-years with positive EPS ranging from $.01 to $1.00, there exists a
possibility that the profitability of the firms overall differed between the two samples. Thus, the
more altruistic financial reporting in the post-SOX sample observed here could be explained if
the companies in this sample experienced stronger operating performance than those in the pre-
SOX sample. This is not the case, though, as the median returns on investment for the pre- and
post- SOX samples of 3.86% and 3.45%, respectively, show that operating performance differed
very little between the samples.
Gavious and Rosenboim (2013) and Grasso et al. (2009) indicate the egregious financial
scandals in the U.S. made public around the turn of the millennium (i.e., Enron, WorldCom,
HealthSouth, etc.) changed the attitudes of management and auditors about the ethicality of
earnings management. They suggest these changed attitudes played a more dominant role than
SOX in constraining earnings management in the recent past. While the present study sheds little
light on this debate, the project could be replicated in a country that enacted corporate
Page 162
governance legislation similar to SOX but did not experience financial scandals on the
magnitude of those that plagued the U.S. If the results of such a study echo those of the current
research, strong evidence would exist that the corporate governance legislation, and not changed
attitudes by management and auditors, caused the more altruistic financial reporting observed
recently.
Like most studies, the present research has a few limitations that affect the
generalizability of its findings. For example, we confined our analysis to companies with
calculated EPS between $.01 and $1.00 because firms with smaller amounts of EPS have more
incentive than companies with larger amounts of EPS to manipulate earnings to allow the
rounding up of EPS (e.g., see Miller et al. 2012). No inferences should be made from our study
about companies reporting EPS outside this range. In addition, we examined data from
general samples of listed companies and did not perform any industry-specific analyses. Yet,
research (e.g., Datta et al. 2013; He & Yang, 2014) provides evidence indicating that the
propensity to engage in earnings management can vary across industries. Thus, it is possible the
type of biased financial reporting examined in the current study could still exist in particular
industries. Likewise, we did not segregate our samples based of the quality of the companies’
auditors. Prior studies (e.g., Burnett et al. 2012; Chen et al. 2005) have shown that audit
quality (frequently measured as audit firm size or degree of industry specialization of the audit
firm) constrains certain types of earnings management. Separating our samples according to
the quality of the companies’ auditors might have yielded results different from those obtained
(e.g., perhaps earnings manipulation to round up EPS still occurs for companies with lower
quality audit firms). Future research could examine these issues.
REFERENCES
Aono, J. & L. Guan (2008). The impact of Sarbanes-Oxley Act on cosmetic earnings management. Research in
Accounting Regulation, 20, 205-215.
Bartev, E. & D. Cohen (2009). The “numbers game” in the pre- and post-Sarbanes-Oxley eras. Journal of
Accounting, Auditing & Finance, 24(4), 505-534.
Benford, F. (1938). The law of anomalous numbers. Proceedings of the American Philosophical Society, 78(4), 551-
572.
Burnett, B., B. Cripe, G. Martin & B. McAllister (2012). Audit quality and the trade-off between accretive stock
repurchases and accrual-based earnings management. The Accounting Review, 87(6), 1861-1884.
Carslaw, C. (1988). Anomalies in income numbers: Evidence of goal oriented behavior. The Accounting Review,
63(2), 321-327.
Chen, K., K. Lin & J. Zhou (2005). Audit quality and earnings management for Taiwan IPO firms. Managerial
Auditing Journal, 20(1), 86-104.
Cohen, D., A. Dey, & T. Lys (2008). Real and accrual-based earnings management in the pre- and post-Sarbanes-
Oxley periods. The Accounting Review, 83(3), 757-787.
Cullen, J., S. Robb & D. Segal (2008). Revenue manipulation and restatements by loss firms. Auditing: A Journal of
Das, S. & H. Zhang (2003). Rounding-up in reported EPS, behavioral thresholds, and earnings management. Journal
of Accounting & Economics, 35(1), 31-50.
Datta, S., M. Iskander-Datta & V. Singh (2013). Product market power, industry structure, and corporate earnings
management. Journal of Banking & Finance, 37(8), 3273-3285.
Degeorge, F., J. Patel & R. Zeckhauser (1999). Earnings management to exceed thresholds. Journal of Business,
72(1), 1-33.
Gavious, I. & M. Rosenboim (2013). The disciplining role of the market versus government regulation: The case of
Sarbanes-Oxley and the earnings quality of M&A targets. American Law & Economics, 15(1), 300-332.
Page 163
Ghosh, A., A. Marra & D. Moon (2010). Corporate boards, audit committees, and earnings management: Pre- and
post-SOX evidence. Journal of Business Finance and Accounting, 37(9/10), 1145-1176.
Grasso, L., P. Till & R. White (2009). The ethics of earnings management perceptions after Sarbanes-Oxley.
Management Accounting Quarterly, 11(1), 45-69.
Guan, L., D. He & D. Yang (2006). Auditing, integral approach to quarterly reporting, and cosmetic earnings
management. Managerial Auditing Journal, 21(6), 569-581.
He, L. & R. Yang (2014). Does industry regulation matter? New evidence on audit committees and earnings
management. Journal of Business Ethics, 123(4), 573-589.
Jordan, C. & S. Clark (2013). Manipulating sales revenue to user reference points in pre- and post-Sarbanes Oxley
eras. Journal of Forensic & Investigative Accounting, 5(2), 114-130.
Jordan, C., S. Clark & M. Waldron (2014). Cosmetic earnings management before and after corporate governance
legislation in Canada. Accounting and Finance Research, 3(4), 105-114.
Jorgensen, B., Y. Lee & S. Rock (2014). The shapes of scaled earnings histograms are not due to scaling and sample
selection: Evidence from distributions of reported earnings per share. Contemporary Accounting Research,
31(2), 498-521.
Ketz, J. (1999). How goes the SEC’s war against earnings management? Journal of Corporate Accounting and
Finance, 10(3), 41-52.
Kinnunen, J. & M. Koskela (2003). Who is Miss World in cosmetic earnings management? A cross-national
comparison of small upward rounding of net income numbers among eighteen countries. Journal of
International Accounting Research, 2, 39-68.
Lin, F., L. Zhao & L. Guan (2014). Window dressing in reported earnings: A comparison of high -tech and low-tech
companies. Emerging Markets, Finance & Trade, 50(supplement 1), 254-264.
Lin, F. & S. Wu (2014). Comparison of cosmetic earnings management for the developed markets and emerging
markets: Some empirical evidence from the United States and Taiwan. Economic Modelling, 36, 466-473.
Miller, P., G. Martin & P. Bahnson (2012). A penny for your thoughts: Sizing up manipulative EPS rounding.
Strategic Finance, 94(1), 35-39.
Nigrini, M. (1999). I’ve got your number. Journal of Accountancy, 187(5), 79-83.
Payne, J. & S. Robb (2000). Earnings management: The effect of ex ante earnings expectations. Journal of
Accounting, Auditing & Finance, 15(4), 371-392.
Rodriguez, R. (2004). Reducing false alarms in the detection of human influence on data. Journal of Accounting,
Auditing & Finance, 19(2), 141-158.
Schipper, K. (1989). Earnings management. Accounting Horizons, 3(4), 91-102.
Skousen, C., L, Guan & T. Wetzel (2004). Anomalies and unusual patterns in reported earnings: Japanese managers
round earnings. Journal of International Financial Management & Accounting, 15(3), 212-234.
Thomas, J. (1989). Unusual patterns in reported earnings. The Accounting Review, 64(4), 773-787.
Van Caneghem, T. (2002). Earnings management induced by cognitive reference points. British Accounting Review,
34(2), 167-178.
Wilson, T. (2012). Further evidence on the extent of cosmetic earnings management by U.S. firms. Academy of
Accounting and Financial Studies Journal, 16(3), 57-64.
Yoon, S. & C. Miller (2002). Earnings management of seasoned equity offering firms in Korea. International
Journal of Accounting, 37(1), 57-78.
Page 164
Page 165
TIER 1 CAPITAL RATIO S: THE REGULATION OF

RISK BEHAVIOR CON TINUES
Ry a n L a i o l a , Ca l i f o r n i a S t a t e Un i v e r s i t y F u l l e r t o n
ABSTRACT
This study builds on existing research by examining the top bank holding companies
(BHCs) in the US economy. Through the guidelines put forth in Basel I, II, & III requirements,
The Federal Reserve oversees the largest asset size BHCs and determines the appropriate
capital levels for reserves against loan losses. Capital adequacy regulation is aimed at measuring
the individual risk levels of banks and allocating reserves for that amount of risk and expected
potential loan losses. This study develops an original research model and hypothesis using R
square testing to find correlative relationships between the market volatility of returns on equity
and the fluctuations of Tier 1 Capital ratios during the time series data collected. This study adds
to the subject matter of utilizing a book-value measure of risk and a market-value measure of
risk for comparison.
This study examines the effects of regulatory implementation of bank capital adequacy
measures and the market pricing of return on equity. Drawing upon a ten-year data series on the
balance sheets and adjusted stock returns of the top thirty largest bank holding companies in the
United States, this study will use the linear R Square test to provide empirical evidence in
support of the argument that capital supervision has either a negative or positive effect on bank
risk behavior. The results of this study can be used by regulators and management as a capital
planning tool and also add to the discourse on the subject matter.
INTRODUCTION
The recent financial crisis of 2008, which has also been referred to as The Great Recession,
stemmed from a liquidity crisis in the financial economic sector. The cause of the recession has
largely been pinned on US financial institutions’ creations of ever more complex financial
derivatives of mortgage-backed securities driving a valuation bubble that eventually burst.
Liquidity dried up, and banks became insolvent. As real estate values began to fall, borrowers
defaulted on their mortgages and banks began to incur loan losses. This process causes a
reduction in assets on a bank’s balance sheet, which must be covered by the bank’s own capital
reserves (or equity funds). When losses outpace or exhaust the capital reserves, the bank becomes
balance sheet insolvent and is forced to close its doors.
Banks and other financial institutions that are large enough to be significantly integrated
into the economy can cause an economic shock. To further exacerbate this solvency crisis, the
interbank market structure can cause a domino effect of bank-to-bank defaults. The debt side of a
bank’s balance sheet is typically made up of a mix of retail deposits and wholesale debt funding
(Farag, Harland, & Nixon, 2013). This was particularly problematic when pre-recessionary debt-
to-equity ratios were as high as 30.0:1.0, and capital reserves could not withstand write downs of
assets (Fox, 2013).
Regulators had to step in and prop up the financial system through bail-outs known as the
Troubled Asset Relief Program through the passage of the Emergency Economic Stabilization
Act of 2008. This law allowed the US Treasury up to $700B in purchasing authority to help
Page 166
stabilize the US financial system. Of this amount, the Treasury invested $245B into five different
bank programs such as the Asset Guarantee Program, The Capital Assistance Program, and the
Capital Purchase Program. Fortunately, these programs achieved many of their stated goals and
also returned approximately $28B more than initially invested (US Department of the Treasury,
2013).
Subsequent regulation in 2009 known as the Supervisory Capital Assessment Program
(SCAP) allowed for the Federal Reserve Bank to conduct the stress testing. The Federal Reserve
annually audits the largest bank holding companies (BHCs), which are typically any BHC that
has more than $10B in total assets. Under the Comprehensive Capital Analysis and Review
(CCAR) and the Dodd-Frank Act (DFA), the Federal Reserve ensures that certain financial
institutions of systemic influence have adequate capital that account for their own idiosyncratic
risks in order to sustain external market factors and financial stress (Board of Governors of the
Federal Reserve System, 2013).
According to the Federal Reserve Economic Data (FRED), the research division of the
Federal Bank of St. Luis, total net loan charge-offs (NCOTOT) exceeded 3% of total loans
during the Great Recession, while the historical observations ranged from 0.5-1.5% (Federal
Reserve Bank of St. Luis, 2013). Net charge-offs are losses taken on a bank’s loan portfolio, net of
any recoveries associated with the charge-off. With banks incurring twice the amount of
losses in a short period of time, bank failures threatened the US economy and capital reserve
requirements were increased by regulators. The effect of this was to force a more conservative
capital structure with the intent of inducing risk-averse behavior. Further, the US financial sector
has outpaced the growth of the general economy over the past three decades creating some
fundamental changes to the landscape of the financial sector. For example, the past three decades
have seen the appearance of the shadow banking system, a system of non-bank financial firms
that have the same basic functions of a traditional deposit entity (Greenwood & Scharfstein,
2012).
Purpose of the Study
The primary goal of regulatory capital requirements is to force banks to maintain the
necessary funds to weather financial stress, absorb losses, and to continue the going concern. The
regulations and tests are designed to be forward looking to ensure that banks and other financial
institutions have the appropriate capital plan in place to balance the level of credit risk inherent in
the loan portfolio. The problem with these regulatory requirements is that forcing a debt-to- equity
ratio capital structure places downward pressure on ROE and ROA metrics for shareholders. By
forcing banks to increaser their Tier 1 capital, the equity side of the balance sheet expands and
thus dilutes equity values, spreading the earning over a larger portion of equity. Arguably,
capital regulatory oversight can pressure a board of directors into underwriting riskier loans where
they can, in order to boost earnings in other areas when possible.
The purpose of this study is to provide empirical observations and additional discourse to
the effects that regulatory enforcement of more stringent capital structures have on financial
institutions and their shareholders. In answering this question, this study will build on current
research in order to provide bank executives and board directors data to create better strategies
for capital planning and market positioning.
Page 167
Limitation s of the S tudy
The conclusions of this study are limited to the abstract relation between determining a
statistically significant level of correlation on stock market variance metrics and the various
levels of regulatory Tier 1 Capital to risk-adjusted assets ratio. The stock market and global
economy have countless factors that can influence market returns and volatility (VIX) measures.
Further, as regulatory definitions change, investors may not quickly or accurately be able to price in
the estimated effects. Similarly, investor sentiment can change over time, which may deter from
concluding that a correlative relationship exists between the two variables. As an example,
investor sentiment can change from news of the Federal Reserve’s quantitative easing program
being a negative factor (as an indication of an unhealthy economy), to being a positive factor (as
an indication of necessary economic reconstruction). Lastly, the time period of data used in this
study overlays the most volatile periods in recent history which involved unforeseen
macroeconomic factors that had significant influence on the variance of returns in equity markets.
The underlying data used in this analysis has a limited linear relationship due to the nature of
the time period used.
REVIEW OF AVAILABLE LITERATU RE
There has been a great deal of data tracking surrounding bank capital reserves since the
Great Recession and the government bailout of the financial sector. Regulatory agencies monitor
and keep track of data metrics for compliance and progress report purposes. These metrics
include capital requirements, reserve requirements, Tier 1 and Tier 2 minimum capital
requirements, leverage ratios, liquidity ratios, and solvency ratios. Ideally, these regulations are
designed to curb the risk behavior of management and in turn minimize the total risk exposure of
financial institutions. In reviewing how capital reserve requirements affect a bank’s return on
assets and return on equity metrics, management must develop new strategies to boost returns.
The traditional banking business model plays a crucial role in the functioning of the
economy. Not only do they supply payment services for transaction settlement, they also provide
loans, or credit standards, to the household and business sectors. The loan creation process is
what is known as the ‘maturity transformation’ process. This is an essential function of a
financial institution to the economy in that short-term excess funds are allocated to banks, where
they are then turned into long-term loans. Smaller short-term deposits (sources of funding) are
aggregated into larger long-term loans (uses of funding). Here, there is a mismatch in the
maturity timing of a bank’s short-term liabilities (source of funds) and its long-term assets, or
loan portfolio (use of funds). Mismanagement of these accounts can result in a liquidity crisis or a
‘run on banks’. A noted method for mitigating this risk is to create a well-diversified and stable
funding profile which can be sourced through contractual instruments that lock in the invested
funds for a specific period of time (Farag, Harland, & Nixon, 2013).
Microprudential regulation seeks to address any negative externalities that can impact a
specific bank’s risk levels, adjusting their capital reserves to appropriately reflect the risks that
they take. This entails the bank’s own capital structure, loan credit risk, lending market risk, and
liquidity risk. Banks’ main function within the economy is to diversify away lending risk, and
capital regulation aims at assessing those risks and requiring appropriate ‘risk-adjusted’ reserves.
On the other hand, macroprudential regulation seeks to account for systemic risks such as a
build-up in leverage or volatility in the rate of change of the ‘maturity transformation.’
Page 168
In “What We’ve Learned from the Financial Crisis”, Fox notes that banks tend to reduce
their debt levels instead of increase their equity when faced with adverse market conditions or
higher capital requirements. This can have a negative effect on the overall economic activity
level, as banks begin to shrink their debt capital to maintain the appropriate debt-to-equity levels. If
banks were forced to sell common shares, they could react to adverse capital constraints by
expanding their equity. Another idea presented is to require banks to issue convertible debt (debt
instruments that are convertible into equity) in times of crisis (Fox, 2013).
Earlier studies have also noted that bank capital structure can produce conflicting results in
risk behavior. As an example, banks tend to view FDIC insurance as a put option per Black-
Scholes valuation methodology, thus potentially taking on excessive risk with the safety net of
regulatory deposit insurance in the event of default. Option Value Theory generally states that
higher volatility of possible outcomes provide a higher value of the exercisable option itself. The
premiums for deposit insurance do not reflect the underlying risks and more so, banks have been
culprits of moral hazard theory (Marshall & Prescott 2000).
However, in a 2011 study, it was noted that banks with a high degree of ‘franchise value’,
that is, a firm that has high earnings or a significant dividend payout, will hold excess capital
reserves above regulatory requirements (Harding, Liang, & Ross 2011). Further, for larger banks
specifically, it was noted that during the recent financial crisis positive stock returns were related to
stronger capital positions. Prior to the recession, stock returns had no correlation to various capital
levels; which concludes that adequate bank capital levels are more of an asset in an adverse
environment (Demirguc-Kunt, Detragiache, & Merrouche, 2013). Also, Merle found that certain
board member characteristics were related to an increase in Tier 1 capital holdings of the bank,
effectively reducing capital risk. Simple linear regression was used to find positive correlation
between board stock ownership, CEO base pay, and CEO incentive pay and the dependent variable
of Tier 1 capital holdings (Merle, 2013).
Progress of Current Research
Some researchers have recognized that at the onset of the financial crisis, a rare
opportunity to redesign the global financial system presented itself. This is possibly a once-in-a-
generation chance for meaningful reform (Butler, 2009). Capital requirements have been the
main tool that regulators have used over the last several decades to help manage risk. The
underlying logic in this approach is to force financial institutions to have enough reserves set
aside to cover expected and unexpected losses. Thus, the level of financial risk undertaken by
banks is lower. Although this theory is straightforward, it can be quite confusing and complex in
practice (Krawcheck, 2012). There are numerous ways to measure capital and the appropriate
level of capital is debatable. Minimum capital requirements may incorporate a market risk factor
and a firm specific risk factor (or credit risk factor) based on either the bank’s own internal models
or a set of regulatory risk weights (Hirtle, 2003).
One significant impact that the government bailout of banks (TARP) had on the risk
behavior of banks is that certain financial institutions are now considered “too big to fail.” Upon
receiving TARP funds, these corporations have been deemed too important to the general
economy to allow them to fail, which re-introduces the issue of moral hazard theory. The
tendency to take more risk from one party when the majority of potential costs are borne by
outside parties is a moral hazard issue. In other words, if the large, systemic banks are likely to be
bailed out by public taxpayer funds again, they would be willing to take excessive risks,
having less skin in the game.
Page 169
In 2012, Stan and McIntyre conducted research that show that the government’s “too big to
fail” policies bestow more risk-taking benefits upon the larger banks than on their smaller
counterparts. Although it is arguable that simple scales of economies can produce such benefits,
the study used an accounting, rather than a market-based, measure of risk to account for the fact
that the smaller banks are not heavily traded in the public equities markets. The empirical
findings supported the hypothesis that larger banks are riskier than smaller ones by measurement
of variance of returns and by capital to asset ratio measures (Stan & McIntyre, 2012).
As previously stated, the core logic behind regulatory capital requirements is to force
banks to have more of their own funds on the books to allow for a cushion against loan losses. It is
essentially a book value risk mitigant for a potential market value of risk. Bank capital regulation
is largely based on book value accounting. Recently, Bhattacharya conducted research that utilized
book value risk measures and book value returns to present evidence that capital requirements
have actually increased the risk level of the banking industry. When return on equity is
constrained and cannot be properly priced in through efficient market factors, banks are induced to
seek high-return-high-risk assets to satisfy the required rate of return on equity. Return on
assets has doubled and loss on assets increased 2.5 times since capital regulation halved the
loan capacity of banks (Bhattacharya, 2013).
Specific Contributio n of Study
Existing studies do not attempt to draw direct relationships between the Tier 1 capital
requirement as a percentage of risk-weighted assets to real returns to investors. Tier 1 Capital is
book value of accounting; however, a more appropriate measure is to use a ratio of Tier 1 Capital to
risk-adjusted assets. Tier 1 Capital is considered core capital. It consists of common equity or
stock and reserves taken from retained earnings. The loan assets are then adjusted by exposure risk
rating grades and risk parameters to account for potential losses (Federal Reserve Board, 2006).
Regulatory requirements have historically been around 4%. According to the Board of
Governors of the Federal Reserve System in their 2011 “Comprehensive Capital Analysis and
Review: Objectives and Overview”, the CCAR seeks to improve the overall risk management,
capital management, and capital resources of the largest asset size bank holding companies
(Federal Reserve Board, 2011). In other words, banks can take risk, and ultimately are in the
business of risk-taking, but they must be regulated to ensure that they take appropriate measures to
mitigate their risk taking activities.
This study will determine the relationship of how the equity market reacts and interprets
the effectiveness of regulatory capital requirements through the application of simple statistical
models. The study will rely on the Efficient Market Theory to make connections with investor
perception of risk and their required rate of return on bank equity. By adding to the existing
research, this study will apply prior researchers’ conclusions and opinions to statistical outputs
from this study in order to add to the discourse of the effects of regulatory capital requirements.
The results of this study will show that the required rate of return on bank equities should adjust
according to regulatory declaration and enforcement of increased Tier 1 Capital and asset risk
adjustments. Contributory calculations consist of time-series graphical representations of the
largest bank holding companies’ trailing twelve months variance of stock returns and the time
frame of regulatory Tier 1 Capital to risk-adjusted assets ratio requirements.
Page 170
RESEARCH MODE L & HYPOTHESIS
Some trends in current research suggest that capital requirements enhance the safety and
soundness of bank operations and loan loss reserves, while other trends suggest that hard-set
regulatory requirements can create some degree of moral hazard and either increase the risk
behavior of banks, or circumvent natural market pricing occurrences of ROE which puts external
pressure on management to take on additional risk. To test this hypothesis, a closer look at a
traditional measure of market priced return on equity will be used to determine a level of
correlation between aggregate investors’ required rate of return and the ratio of regulatory Tier 1
Capital to risk adjusted assets ratio. The book value measurement of ROE is traditionally shown as
net income divided by shareholder equity. Some researchers have argued that making a book- to-
book comparison is appropriate, while others have argued that market values are more relevant
and accurate (Lee, Lin, & Yu, 2013). This study will utilize historical market returns on equity
since accounting income and capital structure can be manipulated and adjusted, while market
pricing is open to aggregate investor behavior and pricing methodologies. The question to be
answered is whether or not there is a linear correlation between the regulatory decrees on Tier 1
Capital ratios and market return on equity.
Null Hypothesis: The statistical relevance of R2 shows that Tier 1 Ratios have a linear relationship with
stock market ROE standard deviations.
H0 = R2 > 0.5
Alternative Hypothesis: The statistical relevance of R2 shows that Tier 1Ratios do not have a linear relationship with
stock market ROE standard deviations.
H1 = R2 < 0.5
RESEARCH METHODOLOG Y
The research is designed around four steps. First, the subject operating banks and bank
holding companies (BHCs) will be chosen based on size and regulatory supervision levels. The
second step is to collect historical adjusted stock returns on a monthly basis, compute the
monthly returns, then compute the monthly trailing twelve months standard deviation to arrive at a
measure of market risk for the return on equity. The third step is to collect time series data on the
most prominent regulatory requirement – the Tier 1 Capital ratio to risk-adjusted assets. Lastly,
the BHC stock volatility will be measured per time period of regulatory capital adequacy
requirement.
Subjects of Study and Sample Size
This study seeks to analyze the largest bank holding companies in the US domestic
economy with a history data set that has been subject to regulatory scrutiny over the years. An
ideal subject company is large enough to have been subject to the Federal Reserve capital
requirements (which is largely adopted through the International Bank of Settlements’ Basel
policy) and has been in operation long enough to predate the Tier 1 Capital ratio requirements. The
list was taken directly from the Federal Reserve’s CCAR participants, which only applies to banks
and organizations with assets of $10 billion or more. This largely excludes community banks
and regional banks and targets the more complex national commercial banks that are subject to US
regulatory filings.
Page 171
In a press release from November 1st, 2013, the Federal Reserve board of governors
released a revised list of corporations subject to the Comprehensive Capital Analysis and Review
program, which is inclusive of the Tier 1 ratio requirement. Thirty BHCs in total are included on
the list, of which 12 were added to the coming 2014 program review and Basel III requirements.
However, all banks listed have been subject to or participated in Tier 1 ratio measurements. This
list represents a forward looking sample of BHCs since the Federal Reserve’s press release
named these corporations specifically in their future stress testing program (Federal Reserve
Board, 2013).
Table 1
SELECTED SUBJECT BHCs
Previous CCAR participants, also
Participants new to CCAR in 2014
participants in 2014
Ally Financial Inc. BMO Financial Corp.
American Express Company BBVA Compass Bancshares, Inc.
Bank of America Corporation Comerica Inc.
The Bank of New York Mellon Corporation Discover Financial Services
BB&T Corporation HSBC North America Holdings Inc. *
Capital One Financial Corporation Huntington Bancshares Inc.
Citigroup Inc. M&T Bank Corp.
Fifth Third Bancorp Northern Trust Corp.
The Goldman Sachs Group, Inc. RBS Citizens Financial Group, Inc. *
JPMorgan Chase & Co. Santander Holdings USA, Inc. *
KeyCorp UnionBanCal Corp. *
Morgan Stanley Zions Bancorp
The PNC Financial Services Group, Inc.
Regions Financial Corporation
State Street Corporation
SunTrust Banks, Inc.
U.S. Bancorp
Wells Fargo & Company
HSBC North America Holdings Inc., RBS Citizens Financial Group, Santander Holdings
USA, Inc., and UnionBalCal Corporation (denoted *) were excluded from the study due to
private ownership which made it impossible to procure and analyze public market data. This
remaining list was cross-referenced with other institutions leading BHCs tracking lists such as
the FDIC’s “Top 50 BHCs” and with The Banker “Top 1000 World Banks, 2013”. Both of
which measure Tier 1 Capital amounts (Alexander, 2013). The Federal Reserve, by definition of
CCAR, has named these BHCs Systematically Important Financial Institutions (SIFIs) by virtue of
their total asset size and scope of operations. Lastly, the sample size of approximately twenty- six
subjects will allow for statistical relevance of findings from a population.
Data Collection Method
Historical prices were taken from Yahoo! Finance online. Yahoo! Finance’s database
calculates the adjusted closing price of any stock for the entire public life on the respective
market exchange. Yahoo!’s data source is taken from the security’s respective exchange such as
NYSE, NASDAQ, or NYSE MKT, but then is adjusted for dividends and splits. Thus the
Page 172
adjusted close as published with Yahoo! Finance incorporates the effects of ex-dividend date of
distribution and splits and reverse-splits on the stock price which is a more accurate depiction of
true returns to stockholders. Data was then collected from the present day (12/2/13 monthly
adjusted close) to end of year 2003 (Yahoo! Finance, 2013). This time frame of the last ten years
includes a large range of median industry levels of Tier 1 Capital ratios within a range of 9.8% -
13.2%, a new historical high. Then the Tier 1 ratio to risk adjusted assets data was collected on
the above listed BHC’s for the same time period (BankRegData, 2013).
Statistical Techniqu e Used
This study analyzed the ten-year adjusted monthly closing stock price of the most
significant BHCs, those that were categorized by the Federal Reserve as “systematically important
financial institutions” (SIFIs), calculated the monthly returns on each stock, and then calculated
the trailing twelve-month standard deviation of those returns. The resultant time series data
was then graphed and used as a measure of market volatility or risk for the return on equity for
BHC investors. The second variable was the historical Tier 1 Capital ratio, defined as core equity
capital divided by risk-adjusted total assets for the same subject BHCs. Quarterly data figures
were obtained and graphed over the same coinciding time period. Lastly, the R2 statistical test
was performed on both variables to predict future outcomes and determine if the hypothesis is
valid or null. This shall determine the goodness of fit test of the statistical model presented in this
study.
ANALYSIS & RESULTS
Each BHC analysis was conducted on an individual basis to determine a specific linear
relationship between the two variables outlined above. R2 analysis was performed on both the
dependent and independent variable to determine which banks specifically contributed to the
validity of the hypothesis presented here. Each bank has its own market strategy and its own
appetite for risk. As indicated in the data collection of the historical Tier 1 Capital ratios, the
percent of capital allocated to the risk adjusted assets is different for each BHC. Therefore, the
null and alternative hypotheses apply to each individual BHC, not the aggregate list or the
industry as a whole. See appendix excel data for detailed analysis results on each BHC. To
illustrate an example, the analysis results of BMO Financial follow:
Figure 5
Page 173
The total findings resulted in the null hypothesis to be null and the alternative hypothesis to
be valid. Although some of the BHC findings resulted in an R2 value greater than 0.5
statistical relevance, the values only coincided with the Tier 1 Capital ratio variable. The
standard deviation of return on equity did not produce any R2 values above the relevance level of
0.5. The expectation of the this model is to see either a decline in ROE volatility as the Tier 1
Capital ratio increases or to see a linear relationship producing the opposite effect. As each subject
BHC was examined for a positive R2 value greater than 0.5, a summary table of the results
are as follows:
Table 2
SUMMARY OF RESULTS
R Square
BHC StDev StDev R > 0.5 R Square: Tier 1 Tier 1 > 0.5
American Express 0.02419 0.59257 x
Bank of America 0.12786 0.83581 x
BBVA 0.16432 0.15237
BB&T 0.07857 0.60430
BMO 0.00180 0.75629 x
Capital One 0.01053 0.00040
Citigroup 0.07115 0.59499 x
Comerica 0.04483 0.48254
Bank of New York 0.00440 0.83497 x
Discover 0.00223 0.00267
Northern Trust 0.00332 0.61101 x
Fifth Third Bank 0.04062 0.42078
Huntington 0.04062 0.76078 x
JPMorgan 0.02277 0.71072 x
Key 0.08117 0.68019 x
M&T 0.03318 0.68550 x
Morgan Stanley 0.44710 0.71072 x
Regions 0.13788 0.61010 x
State Street 0.02780 0.49515
SunTrust 0.08286 0.73630 x
PNC 0.02919 0.42062
USBancorp 0.00478 0.64541 x
Wells Fargo 0.05548 0.44358
Zions 0.13174 0.77155 x
As the null hypothesis is rejected, the alternative hypothesis is accepted. The hypothesis
expected to show a linear relationship or correlative effect of Tier 1 Capital ratio requirements to
the standard deviation of return on equity. Since the alternative hypothesis is accepted, the
implications are that Tier 1 Capital ratio requirements do not have an effect on the riskiness of
returns on market value of equity of bank holding companies.
Page 174
MANAGERIAL IMPLICATI ONS
One of the primary functions of executives is the fiduciary duty to maximize shareholder
returns. To do this, management needs to invest in future cash flows that add to the present value of
the firm, which takes on a certain level of risk. If banks have to hold more capital in reserves (or
equity) to account for potential loan losses, then management and stockholders may want the firm
to take on higher levels of risk to produce higher returns. A measure of which is the
volatility of stock returns. Therefore, this study was aimed at producing empirical data to provide
stockholders and managers evidence to support their strategic objectives when picking risk-
adjusted assets to invest in.
This data is also relevant to the discourse and objectives of regulators. The main objective
of bank capital supervision is to prevent future financial crises and systemic bank failures. The
implications of their capital reserve requirements are to prevent banks from incurring excessive
loan losses that incur debt-funding default on the part of the bank. The capital reserves
provide equity funding that should be exhausted before debt funding is required in the event of
adverse economic conditions. As prior research has indicated, regulation can have unintended
consequences and may ultimately not achieve the stated objectives of regulatory oversight.
CONCLUSION
There is a significant amount of research regarding the regulatory effects of capital

requirements and this study examines several facets of implications of prior studies. This study
provided a quantitative test to determine whether or not a book measure of risk (Tier 1 Capital
ratio) has a linear relationship on a market measure of risk (standard deviation of return on
equity) on a time series data set over the last 10 years. The results of this study did not indicate a
statistically relevant relationship between the two variables. These findings conclude that the
impact of Tier 1 requirements has not produced any effect on the market risk of such firms. The
expectation was to find that market risk declined as Tier 1 Capital increased or that market risk
increased as Tier 1 Capital increased. This would conclude that either adequate reserves lower
the riskiness of ROE and thus lower the required rate of return on equity, or that adequate capital
reserves increase the market risk of ROE due to the increased pressure from stockholders to
outpace a mandated capital constraint. This presents an argument that capital reserves are not an
efficient use of funds. Many see the new Basel III requirements, as the most valid threat to the
banking business model. Banks have more capital, but management has not adjusted their
operational business model to incorporate the effects of new and increasing regulation of capital
(McKinsey, 2013).
Further research is required to present a clearer relationship between the effects of
enforced balance sheet equity expansion on the overall risk levels of the firm, as measured by
appropriately priced market factors. Arguably, an increase in capital reserves for a bank may or
may not be an indication of risk aversion. A required increase in core capital reserves may be
caused by a an increase in market risk of existing assets, since market conditions can change
after loan capital is deployed. Management and regulators can increase capital reserves in order to
counter balance the perceived market risks. If this were the case, the increases in Tier 1 capital may
signal that the BHC has already taken on additional risk. However the temporal cause and effect
relationship may be less clearly defined as management may also preemptively increase capital
Page 175
reserves prior to investing in riskier future cash flows via a forthcoming change in lending
standards or credit policy. A quantitative analysis may not in fact draw synonymous
conclusions. Additional research is needed to better clarify possible relationships between the
actual measurements of capital adequacy from regulatory agencies.
REFERENCES
Alexander, P. (2013). Top 1000 World Banks 2013. The Banker: Banker Data. Retrieved from:
http://www.thebanker.com/Banker-Data/Banker-Rankings/Top-1000-World-Banks-2013
BankRegData. (2013). “Tier 1 Capital Ratio.” <http://www.bankregdata.com/allHMmet.asp?met=ONE>. Accessed on
Dec. 6, 2013.
Bhattacharya, H., (2013). Capital Regulation and Rising Risk of Banking Industry: A Financial Accounting Perspective.
Academy of Banking Studies Journal, 12(1) 31-59.
Board of Governors of the Federal Reserve System. (2013). 2013 Banking and Consumer Regulatory Policy [Press
release]. Retrieved from http://www.federalreserve.gov/ newsevents/press/bcreg/20131101a.htm
Butler, P. (2009). Learning From Financial Regulation's Mistakes. McKinsey Quarterly, (3), 68-74.
Demirguc-Kunt, A., Detragiache, E., & Merrouche, O. (2013). Bank Capital: Lessons from the Financial Crisis. Journal
of Money, Credit and Banking, 45(6) 1147-1164.
Farag, M., Harland, D., & Nixon, D. (2013). Bank Capital and Liquidity. Bank of England.Quarterly Bulletin, 53(3),
201-215. Retrieved from http://search.proquest.com/docview/1445262398?accountid=9840
Federal Reserve Bank of St. Luis: FRED® Economic Data. (2013). Reports of Condition and Income for All Insured
U.S. Commercial Banks. Loan Loss Reserve / Total Loans for All U.S. Banks (USLLRTL). Retrieved from
http://research.stlouisfed.org/fred2/series/USLLRTL
Federal Reserve Bank of St. Luis: FRED® Economic Data. (2013). Reports of Condition and Income for All Insured
U.S. Commercial Banks. Total Net Loan Charge-offs (NCOTOT). Retrieved from
http://research.stlouisfed.org/fred2/series/ NCOTOT?rid=55&soid=6
Federal Reserve Board. (2011). Comprehensive Capital Analysis and Review: Objectives and Overview [Press release].
Retrieved from http://www.federalreserve.gov/newsevents/press/bcreg/bcreg20110318a1.pdf
Federal Reserve Board. (2006). Notice of Proposed Rulemaking (NPR): Preamble – IV. Calculation of Tier 1 Capital
and Total Qualifying Capital: Basel II Capital Accord. Retrieved from
http://federalreserve.gov/GeneralInfo/Basel2/NPR_ 20060905/NPR/section_4.htm
Fox, J. (2013). What We've Learned from the Financial Crisis. Harvard Business Review, 91(11), 94-101.
Greenwood, R., & Scharfstein, D. S. (2012). How to Make Finance Work. Harvard Business Review, 90(3), 104-110.
Harding, J., Xiaozhong, L., & Ross, S. (2011) Bank Capital Requirements, Capital Structure and Regulation. Journal of
Financial Services Research, 43, pp. 127-148. DOI 10.1007/s10693-011-0127-6
Hirtle, B., (2003). What Market Risk Capital Reporting Tells Us About Bank Risk. Federal Reserve Bank of New York:
Economic Policy Review, 9(3), 37-54.
Krawcheck, S. (2012). Four Ways to Fix Banks. Harvard Business Review, 90(6), 106-111.
Lee, S., Lin, C., & Yu, M. (2013) Book-to-Market Equity, Asset Correlations and the Basel Capital Requirement.
Journal of Business Finance & Accounting, 40(7) & (8), 991-1008. doi: 10.1111/jbfa.12029
Marshall D., & Prescott E., (2000). Bank Capital Regulation with and without State-Contingent Penalties. Federal
Reserve Bank of Chicago. Working Paper WP-00-01
McKinsey & Company. (2012). Why US Banks Need a New Business Model. McKinsey & Company – Insights and
Publications. Retrieved from
http://www.mckinsey.com/insights/financial_services/why_us_banks_need_a_new_business_model
McKinsey & Company. (2013). The Search for a Sustainable Banking Model. McKinsey & Company – Insights and
Publications. Retrieved from
http://www.mckinsey.com/insights/financial_services/the_search_for_a_sustainable_banking_model
Merle, J., (2013). An Examination of the Relationship Between Board Characteristics and Capital Adequacy Risk
Taking at Bank Holding Companies. Academy of Banking Studies Journal, 12(2) 3-11.
Stan, M., & McIntyre, M., (2012). Too Big To Fail? Size and Risk in Banking. Academy of Banking Studies Journal,
11(2) 11-21.
United States Department of the Treasury. (2013). Financial Stability Reports. Troubled Asset Relief Program (TARP)
Tracker. Retrieved from http://www.treasury.gov/ initiatives/financial- stability/reports/Pages/TARP-
Tracker.aspx#Bank
Yahoo!Finance. (2013). “Historical Stock Data”. <http://finance.yahoo.com>. Accessed on Dec. 3rd, 2013.
Page 176
APPENDIX
Figure 2
Page 177
Figure 3
Figure 4
Page 178
Figure 5
Figure 6
Page 179
Figure 7
Figure 8
Page 180
Figure 9
Figure 10
Page 181
Figure 11
Figure 12
Page 182
Figure 13
Figure 14
Page 183
Figure 15
Figure 16
Page 184
Figure 17
Figure 18
Page 185
Figure 19
Figure 20
Page 186
Figure 21
Figure 22
Page 187
Figure 23
Figure 24
Page 188
Page 189
ASSOCIATION BETWEEN PRE IPO EARNINGS

MANAGEMENTS AND POST IPO INSTITUTIONAL
OWNERSHIP DRIFTS
Kyung Joo Lee, University of Maryland-Eastern Shore
John Jongdae Jin, California State University-San Bernardino
Diane Li, University of Maryland-Eastern Shore
ABSTRACT
The purpose of this study is to investigate how earnings management before IPO affects
institutional ownership changes after IPO. It is hypothesized that there is a negative association
between the pre IPO earnings managements and post IPO institutional ownership drifts. Using a
sample of 302 IPO’s, we find that the association between pre IPO earnings managements and
post IPO institutional ownerships get weaker over time so that the association becomes
statistically insignificant by the end of the first year after IPO’s. We also find that the association
between pre IPO earnings managements and post IPO institutional ownership changes is a
statistically significantly negative over the period between the IPO year and 2 years after that.
Both results support the hypothesis. These results hold after controlling for other influencing
factors such as market performance, initial offer price, underwriter reputation, and offer
fraction. The results are robust across different measures of variables and testing methods.
INTRODUCTION
IPO’s may be one of few opportunities where stock issuing firms have significant
information asymmetry over general investment public and hence can take advantage of this
superior information for their own benefits. Some of those benefits stock issuing firms can
entertain through IPO’s could be benefits from attracting more institutional owners in the firms
because institutional investors usually provide diverse benefits to their investee firms ranging
from monitoring/consulting services, better performances, higher spending on R & D, to saving
transaction costs. One way for IPO firms to make themselves attractive to institutional investors
could be window dressing their financials through aggressive earnings managements, which is
working under information asymmetry between IPO firms and general investors including
institutional investors. Sun et al. (forthcoming) addressed this issue and found supporting
empirical evidence: i.e., IPO firms with more aggressive earnings managements tend to have
stronger presence of institutional ownerships than IPO firms without them immediately after
IPO’s. If this is true, a natural extension of this issue would be how this strong presence of
institutional investors inflated by aggressive earnings managements will change after IPO’s as
the information asymmetry between IPO firms and investors fade away.
The purpose of this study is to investigate this issue of institutional ownership changes
after IPO’s in conjunction with earnings managements before IPO’s. It is hypothesized that there
Page 190
is a negative association between the pre IPO earnings managements and post IPO institutional
ownership drifts.
The remainder of this paper is organized as follows. First, a hypothesis is developed
through a review of previous literatures and logical reasoning. Then, sample selection and
measurement of variables are described. The empirical tests and their results are followed. In the
final section, conclusions are addressed.
HYPOTHESIS DEVELOPMENT
It is well documented that institutional investors tend to make diverse value adding
contributions to their investee firms in previous studies. For examples, Clyde (1997) finds that
institutional ownership is directly related to the benefits of policing firms. Stoughton et al. (1998)
& Sun et al. (2008) suggest that institutional investors provide monitoring functions to improve
their performances. Field et al. (forthcoming) find that IPO firms with greater institutional
ownerships outperform those with smaller institutional ownerships. Baysinger et al. (1991) find
that the institutional ownership has a positive impact on corporate R & D spending. Fernando et
al. (2004) suggest that IPO firms with greater institutional ownerships have lower mortality rates
than others with smaller institutional ownerships. Therefore, there seems to be strong incentives
for IPO firms to attract institutional investors at IPO.
Earnings managements can be an effective means of attracting institutional investors at
IPO’s where there are significant information asymmetries between IPO firms and investors. It is
because even intelligent and sophisticate investors like institutional investors could be fooled by
artificially inflated earnings through aggressive earnings managements by IPO firms when
market-determined price information is not available and most financial information about IPO
firms is provided by IPO firms themselves. Previous studies on IPO’s show that IPO firms do
engage in aggressive earnings managements to take advantage of severe information
asymmetries at IPO’s for their economic benefits. For examples, Chaney & Lewis (1995) find
that earnings managements affect firm values with an information asymmetry. Friedlan (1994)
also shows that IPO firms make income increasing discretionary accruals in financial statements
released before IPO’s to increase offer prices. Since inflated earnings by aggressive earnings
managements may increase the initial offer price and be perceived as better performances by
investors at IPO’s, institutional investors who prefer high price stocks and better performers
would buy more of IPO stocks. (See Gompers and Metrick (2001), and Fernando et al. (2004))
However, the increased presence of institutional ownerships by aggressive earnings
managements may disappears as the true earnings become available with diminishing
information asymmetry after IPO’s. Prior studies have shown that pre-IPO aggressive earnings
managements increase initial firm value at IPO’s but decrease subsequent returns to investors
after IPO’s. For example, Ducharme et al. (2001) find a positive relation between pre-IPO
discretionary accruals (a measure of aggressive earnings management) and initial firm value but
a negative relation between initial discretionary accruals and subsequent firm performance.
Similarly, Teoh et al. (1998) report a significant negative relation between discretionary accruals
in the IPO offer year and stock returns over a three-year post-IPO period. Teoh et al. (1998a)
Page 191
find evidence that IPO firms, on average, have high positive issue-year earnings and
discretionary accruals, followed by poor long-run earnings and negative discretionary accruals.
This reversal of institutional ownerships could take long because of transaction costs and
various value adding contributions of institutional investors to IPPO firms after IPO’s.
Therefore, a testable hypothesis derived here from would be
Hypothesis: There is a negative association between the pre IPO earnings managements and post
IPO institutional ownership drifts.
SAMPLE SELECTION AND DATA
Our initial sample of IPO issuers are obtained from the IPO database of Hoovers
Incorporated. The sample period extends from April 1997 to December 2002. Several selection
criteria are applied sequentially. First, financial institutions and utility firms are excluded
because they are in regulated industries and hence usually have different behaviors than
unregulated firms do. Also, the sample excludes ADRs because ADRs are subject not only to US
regulations but also to regulations of foreign country where their base stocks are listed and
traded. Firms with offer price less than one dollar (penny stocks) and firms with offer size less
than one million dollars are excluded. It is because institutional investors, in general, do not
invest in penny stocks and small offers. Finally, relevant data availability in COMPUSTAT data
files over the period of six years surrounding each IPO (i.e., t= [-2, -1, 0, 1, 2, 3]) is required.
These selection criteria yield the initial sample of 302 IPO issuers.
MEASUREMENTS OF VARIABLES
The earnings management is measured by discretionary accruals which are differences

between total accruals and the expected benchmark accruals (nondiscretionary accruals). The
nondiscretionary accruals are industry wide accruals, varying across firm and industry
characteristics, while discretionary accruals are firm specific accruals. Cross-sectional modified
Jones model was used to estimate discretionary accruals of each IPO firm (Jones, 1991; Dechow
et al., 1995; Teoh et al., 1998a). Cross-sectional method is used because a time series approach is
not possible for IPO’s. The cross-sectional approach has an additional advantage in that it
incorporates changes in accruals resulting from changes in economic conditions for the industry
as a whole.
The institutional ownership data are obtained from the 13F filings reported in the
database of Thomson One Banker. We measure institutional ownership by ‘the percentage of
shares owned by all institutional investors’ at the end of first quarter after IPO. We also used ‘the
number of institutional owners’ as an additional measure of institutional ownership. The results
are basically the same.
Other variables that are proven to affect institutional ownerships are market performance,
offer price, offer fraction, and underwriter reputation (see Fernando et. al. (2004), Field & Lowry
(forthcoming), and Lee et. al (forthcoming)). These variables are used in sample description and
regression analyses as control variables. These variables are measured as follows:
Page 192
1. Offer price (OPRC): initial price at which shares were offered at IPO.
2. Market Performance (MPFM): stock returns as measured by the buy-and-hold strategy from IPO date
to Year1 or Year2. Abnormal returns adjusted by market returns (equally- or value-weighted) were
also used.
3. Offer fraction (OFRC): the number of shares offered as a fraction of total number of shares
outstanding.
4. Underwriter Reputation (UWRP): underwriter reputation based on the rankings of Carter and Manaster
(1990), and updated according to the information in Jay Ritter’s website.
Descriptive statistics of the above variables are presented in Table 1. On average, the IPO
firms in the sample have about $879 million in market value after IPO’s. Mean (median) value of
offer price is $14.77 ($14.00), while mean (median) value of institutional ownerships after IPO’s
is 25.60% (21.00%). Mean (median) of offer fraction is 29.82% (median of 24.35%). The sample
firms appear to choose highly reputed underwriters with mean (median) rank of 8.15 (9.10) out
of 10 point scale.
Table 1
DESCRIPTIVE STATISTICS FOR SELECTED VARIABLES
Quartiles
Variables Mean Standard
Deviation
25% 50% 75%
Discretionary accruals (DAC) -0.128 0.321 -0.251 -0.057 0.056

Offer price (OPRC): ($) 14.77 7.37 11.00 14.00 17.50
Offer fraction (OFRC): (%) 29.82 20.49 17.62 24.35 33.33
Underwriter Reputation (UWRP) 8.15 1.51 8.10 9.10 9.10
Institutional Ownership (INOS): (%) 25.60 18.50 13.00 21.00 32.00
EMPIRICAL TESTS AND RESULTS
Two testing methods are used to investigate the hypothesis in this study. The first method
is testing on changes in the association between earnings managements and institutional
ownerships over time. If the association gets weaker over time after IPO’s, the effect of earnings
managements on institutional ownerships diminish over time, supporting the hypothesis. The
second method is testing on associations between earnings managements and changes in
institutional ownerships. If the association turns out to be significantly negative after IPO’s,
earnings managements have negative impacts on institutional ownership changes. It may indicate
that aggressive earnings managements before IPO’s that increase institutional ownerships at
IPO’s induce bigger drops in institutional ownerships after IPO’s.
Page 193
Tests on Changes in Association between Earnings Managements and Institutional

Ownerships
In order to examine the change in association between institutional ownerships and

earnings managements over time, the following simple regression is estimated at four different
points in time: i.e., at the end of the first quarter after IPO (Quarter 1), at the end of year of IPO
(Year 0), at the end of the first year after IPO (year 1), and at the end of the second year after
IPO (Year 2).
INOSi = β0 + β1EMGTi + ε (1)
Where: INOSi = institutional ownership, defined as the percentage of shares

owned by all institutions,
EMGTi = discretionary accruals in year t-1 (one year before IPO).
Table 2 presents the results from regression model (1) for four different points in time.
The regression coefficients of EMGT for Quarter 1, Year 0, Year 2, and Year 2 are 0.085, 0.068,
0.070, and 0.020, respectively. The regression coefficient of EMGT for Quarter 1 of 0.085 is
statistically significant at 1% significant level, while that for Year 0 of 0.068 is statistically
significant at 5% significant level. The other two regression coefficients are not significant at any
meaningful level of significance.
Table 2
EFFECT OF PRE-IPO EARNINGS MANAGEMENT ON POST-IPO INSTITUTIONAL
OWNERSHIP: SIMPLE REGRESSION ANALYSIS
INOSi = β0 + β1EMGTi + ε
Independent Expected Quarter 1 Year0 Year1 Year2

Variables Signs
Coefficients Coefficients Coefficients Coefficients
(t-value) (t-value) (t-value) (t-value)
Intercepts 0.267 0.330 0.395 0.436
(23.50)*** (25.64)*** (24.70)*** (25.47)***
EMGT + 0.085 0.068 0.070 0.024
(2.57)** (1.84)* (1.51) (0.49)
Adj. R2 (%) 1.84 0.78 0.42 0.00
F-value 6.63 3.37 2.27 0.24

(p-value) (0.011)** (0.067)* (0.133) (0.627)
***: Significant at <0.01; **: significant at <0.05; *: significant at <0.10;
These results show that the effect of earnings managements on institutional ownerships
diminishes from very significant (Quarter 1), significant (Year 0), to insignificant (after Year 0).
In other words, the association between earnings managements and institutional ownerships gets
weaker over time, which supports the hypothesis stating a negative association between pre IPO
earnings managements and post IPO institutional ownership drifts.
Page 194
As an attempt to investigate if this result holds after controlling for the other influencing
variables mentioned above, the following multiple regression model is estimated:
INOSi = β0 + β1EMGTi + β2OPRCi + β3UWRPi + β4OFRCi + ε (2)
Where: EMGTi = discretionary accruals in year t-1 (one year before IPO),
OPRCi = initial offer price,
UWRPi = underwriter reputation for ith firm, measured by the rankings of
Carter and Manaster (1990), and updated according to the information in
Jay Ritter’s website,
OFRCi = offer fraction, defined as the number of shares offered divided by
total number of shares outstanding after IPO.
Results from the regression model (2) are presented in Table 3. Results from the multiple
regression model (2) are essentially the same as those from the regression model (1). The
regression coefficient of EMGT for Quarter 1 of 0.087 is statistically significant at 1%
significant level, while that for Year 0 of 0.068 is statistically significant at 5% significant level.
The other two regression coefficients for Year 1 and Year 2 (1.54 and 0.52, respectively) are not
significant. Even after controlling for the other influencing variables, we still have weakening
associations between pre IPO earnings managements and post IPO institutional ownerships over
time, which is consistent with the hypothesis.
Table 3
EFFECT OF PRE-IPO EARNINGS MANAGEMENT ON POST-IPO INSTITUTIONAL
OWNERSHIP: MULTIPLE REGRESSION ANALYSIS
INOSi = β0 + β1EMGTi + β2OPRCi + β3UWRPi + β4OFRCi + ε
Independent Expected Quarter 1 Year0 Year1 Year2

Variables Signs
Intercepts ? -0.089 -0.113 -0.156 -0.198
(1.14) (1.28) (1.43) (1.70)*
EMGT + 0.087 0.068 0.070 0.025
(2.70)*** (1.87)* (1.54) (0.52)
OPRC + 0.056 0.089 0.111 0.125
(1.91)* (2.66)*** (2.70)*** (2.85)***
UWRP + 0.019 0.021 0.028 0.034
(2.49)** (2.40)** (2.54)** (2.94)***
OFRC + 0.173 0.134 0.115 0.101
(3.44)*** (2.34)** (1.61) (1.34)
Adj. R2 (%) 8.82 8.00 7.61 8.44
F-value 8.28 7.54 7.20 7.94

(p-value) (0.000)*** (0.000)*** (0.000)*** (0.000)***
Page 195
Tests on Association between Earnings Managements and Changes in Institutional

Ownerships
As a more direct attempt to examine the hypothetical relationship between pre IPO
earnings managements and post IPO institutional ownership drifts, the following simple
regression model is estimated.
∆INOSi = β0 + β1EMGTi + β2MPFMi + ε (3)
Where: ∆INOSi = changes in institutional ownerships.

EMGTi = discretionary accruals in year t-1 (one year before IPO).
MPFMi = stock returns as measured by the buy-and-hold strategy from IPO dates
to Year 1 or Year 2.
MPFM, a market performance measure, is included in the regression model for a control
purpose because the market performance of stocks is supposed to be very important determinant
of investment decisions. Results from regression model (3) are presented in Table 4.
Table 4
EFFECT OF PRE-IPO EARNINGS MANAGEMENT ON THE CHANGES IN POST-IPO
INSTITUTIONAL OWNERSHIP: AFTER CONTROLLING FOR MARKET PERFORMANCE
∆INOSi = β0 + β1MPFMi + β2EMGTi +ε
Number of Institutional Owners Percentage of Shares Owned by

All Institutions
Independent Expected Year1 - Year0 Year2 – Year0 Year1 - Year0 Year2 – Year0
Variables Signs
Intercepts ? 0.324 0.606 0.062 0.085
(6.08)*** (4.39)*** (6.52)*** (6.75)***
MPFM + 0.337 0.648 0.024 0.036
(9.12)*** (9.68)*** (3.66)*** (5.88)***
EMGT - 0.013 -1.235 -0.013 -0.096
(0.08) (3.10)*** (0.47) (2.66)***
Adj. R2 (%) 21.46 28.06 3.71 12.96
F-value 47.71 49.17 6.76 19.46

(p-value) (0.0001)*** (0.001)*** (0.0013)*** (0.0001)***
∆INOSi are measured over two different time periods using two different methods.
Changes in institutional ownerships between Year 0 and Year 1and those between Year 0 and
Year 2 are used. Over the two time periods mentioned, changes in institutional ownerships are
measured by the changes in number of institutional ownerships or by the changes in the
Page 196
percentage of institutional ownerships in total shares outstanding. The regression coefficients of

EMGT for the period between Year 0 and Year 1 are statistically insignificant across the
different measures of ∆INOSi, which means no meaningful association between pre IPO earnings
managements and institutional ownership changes over one year period between Year 0 and year
1.
On the other hand, the regression coefficients of EMGT for the period between Year 0 an
Year 2 are -1.235 and -0.096 for changes in number of institutional owners and changes in
percentage of institutional ownerships, respectively. Both regression coefficients are statistically
significant at 1% significance level, which means a negative association between pre IPO
earnings managements and post IPO institutional ownership changes over the 2 year period
between Year 0 and year 2. This is supporting the hypothesis.
In order to investigate if the above results presented in Table 3 hold even after controlling
the other influencing variables such as initial offer price, underwriter reputation, offer fraction,
the following multiple regression model is estimated, again.
∆INOSi = β0 + β1EMGTi + β2OPRCi + β3UWRPi + β4OFRCi + β2MPFMi + ε (4)
Where: EMGTi = discretionary accruals in year t-1 (one year before IPO),
OPRCi = initial offer price,
UWRPi = underwriter reputation for ith firm, measured by the rankings of
Carter and Manaster (1990), and updated according to the information in
Jay Ritter’s website,
OFRCi = offer fraction, defined as the number of shares offered divided by
total number of shares outstanding after IPO.
Results from the multiple regression model (4) presented in Table 5 are consistent with
those in Table 3. The regression coefficients of EMGT for the period between Year 0 and Year 1
are statistically insignificant for both measures of institutional ownership changes, indicating no
meaningful relationship between pre IPO earnings managements and post IPO institutional
ownership changes during one year period between Year 0 and year 1.
On the other hand, the regression coefficients of EMGT for the period between Year 0
and year 2 are -1.235 and -0.096 for changes in number of institutional owners and changes in
percentage of institutional ownerships, respectively. Both regression coefficients are statistically
significant at 1% significance level, which means a negative association between pre IPO
earnings managements and post IPO institutional ownership changes over the 2 year period
between Year 0 and year 2. This is supporting the hypothesis that there is a negative association
between pre IPO earnings managements and post IPO institutional ownership drifts.
Another interesting observation presented in Tables 4 and 5 is the regression coefficient
of the market performance measure, MPFM. The regression coefficient of MPFM is statistically
significantly positive across all time periods and different institutional ownership change
measures, which confirms a common belief that investors in general including institutional
investors invest more in the better performing stocks.
In sum, the results from tests on changes in associations between pre IPO earnings
managements and post IPO institutional ownership presented in Tables 2 and 3 suggest that the
association between pre IPO earnings managements and post IPO institutional ownerships get
weaker over time so that the association becomes statistically insignificant by the end of the first
Page 197
year after IPO’s. The results from tests on associations between pre IPO earnings managements
and changes in post IPO institutional ownerships presented in Tables 4 and 5 suggest that the
association is a statistically significantly negative over the 2 year period between the IPO year
and 2 year after that. Thus, all empirical results support the hypothesis stating a negative
association between pre IPO earnings managements and post IPO institutional ownership drifts.
Table 5
EFFECT OF PRE-IPO EARNINGS MANAGEMENT ON THE CHANGES IN POST-IPO
INSTITUTIONAL OWNERSHIP: AFTER CONTROLLING FOR MARKET PERFORMANCE
AND OTHER IPO-RELATED VARIABLES
∆INOSi = β0 + β1MPFMi + β2EMGTi + β3OPRCi + β4UWRPi + β5OFRCi + ε
Number of Institutional Owners Percentage of Shares Owned by

All Institutions
Independent Expected Year1 - Year0 Year2 – Year0 Year1 - Year0 Year2 – Year0
Variables Signs
Intercepts ? 0.324 0.606 0.062 0.085
(6.08)*** (4.39)*** (6.52)*** (6.75)***
MPFM + 0.337 0.648 0.024 0.036
(9.12)*** (9.68)*** (3.66)*** (5.88)***
EMGT - 0.013 -1.235 -0.013 -0.096
(0.08) (3.10)*** (0.47) (2.66)***
OPRC ? -0.007 -0.005 -0.000 0.002
(1.03) (0.26) (0.07) (1.23)
UWRP ? -0.008 0.025 0.007 0.012
(0.24) (0.28) (1.18) (1.50)
OFRC ? 0.024 -1.093 -0.015 -0.106
(0.10) (1.70)* (0.35) (1.83)*
Adj. R2 (%) 21.05 28.12 3.32 15.58
F-value 16.89 20.33 3.05 10.15

(p-value) (0.0001)*** (0.001)*** (0.0106)** (0.0001)***
CONCLUSION
The purpose of this study is to investigate how aggressive earnings management before
IPO affects institutional ownership changes after IPO. It is hypothesized that there is a negative
association between the aggressiveness of earnings management before IPO’s and institutional
ownership drifts after IPO’s.
Using a sample of 302 IPO firms between 1997 and 2002, we find empirical results
supporting our hypothesis. The results show that the association between pre IPO earnings
managements and post IPO institutional ownerships get weaker over time so that the association
Page 198
becomes statistically insignificant by the end of the first year after IPO’s. The results also show
that the association between pre IPO earnings managements and post IPO institutional ownership
changes is a statistically significantly negative over the 2 year period between the IPO year and 2
years after that. Thus, all empirical results support the hypothesis stating a negative association
between pre IPO earnings managements and post IPO institutional ownership drifts. Inflated
presence of institutional investors surrounding IPO’s due to information asymmetry and
aggressive earnings managements gradually fade away after IPO’s.
These results hold even after controlling for the other influencing variables on post-IPO
institutional ownership changes such as market performance, initial offer price, underwriter
reputation, and offer fraction. These results are robust across different measures of variables and
testing methods.
REFERENCES
Armstrong, C., G. Foster & D. Taylor (2008). Earnings Management around Initial Public Offerings: A Re-
examination. Unpublished working paper.
Baysinger, B., R. Kosnik & T. Turk (1991). Effects of Board and Ownership Structure on Corporate R&D Strategy.
The Academy of Management Journal, 34, 205-214.
Carter, R. & S. Manaster (1990). Initial Public Offerings and Underwriter Reputation. Journal of Finance, 45, 1045-
1067.
Chaney, P. & C. Lewis (1995). Earnings Management and Firm Valuation under Asymmetric Information. Journal
of Corporate Finance, 1, 319-345.
Clyde, P. (1997). Do Institutional Shareholders Police Management? Managerial and Decision Economics, 18, 1-10.
Dechow, P., R. Sloan & A. Sweeney (1995). “Detecting Earnings Management.” The Accounting Review 70: 193-
225.
Ducharme, L., P. Malatesta & S. Sefcik (2001). Earnings Management: IPO Valuation and Subsequent Performance.
Journal of Accounting, Auditing and Finance, 16, 369-96.
Fernando, C., S. Krishnamurthy & P. Spindt (2004). Are Share Price Levels Informative? Evidence from the
Ownership, Pricing, Turnover and Performance of IPO firms. Journal of Financial Markets, 7, 377-403.
Field, L. & M. Lowry. Forthcoming. “Institutional Versus Individual Investment in IPO’s: The Importance of Firm
Fundamentals”. Journal of Financial and Quantitative Analysis.
Friedman, John. 1994. “Accounting Choice of Issuers of Initial Public Offerings”. Contemporary Accounting
Research 11, No. 1: 1-31.
Gompers, P. and A. Metrick. 2001. “Institutional Investors and Equity Prices’. Quarterly Journal of Economics. Vol
116, No. 1: 229-259.
Jo, H., Y. Kim and M. Park 2007. “Underwriter Choice and Earnings Management: Evidence from Seasoned Equity
Offerings.” Review of Accounting Studies 12: 23-59.
Jones, J. 1991. “Earnings Management during Import Relief Investigations.” Journal of Accounting Research 29:
193-228.
Kim, Yongtae and Myung S. Park. 2005. “Pricing of Seasoned Equity Offers and Earnings Management”. Journal
of Financial and Quantitative Analysis 40 No. 2. 435-463.
Ke Bin and Santhosh Ramalingegowda. 2005. “Do Institutional Investors Exploit The Post-earnings Announcement
Drift?” Journal of Accounting and Economics 39: 25-53.
Schipper, K. 1989. “Commentary on Earnings Management.” Accounting Horizons 3: 91-102.
Stoughton, N. and J. Zechner. 1998. “IPO-mechanisms, Monitoring and Ownership Structure.” Journal of Financial
Economics 49: 45-77.
Teoh, Siew Hong, I. Welch, and T. J. Wong. 1998. “Earnings Management and the Long-term Market Performance
of Initial Public Offerings.” Journal of Finance 53: 1935-1974.
Page 199
Teoh, Siew Hong, T. J. Wong, and Gita R. Rao. 1998. “Are Accruals during Initial Public Offerings Opportunistic?”
Review of Accounting Studies 3, 175-208.
Sun, Yong, K. Lee, D. Li, and J. Jin. 2008. “The Effect of Underwriter Reputation on Pre-IPO Earnings
management and Post-IPO Operating Performance”. Academy of Accounting and Financial Studies
Journal.
Sun, Yong, K. Lee, J. Jin, & D. Li. Forthcoming. ‘The Short-Term Effect of Pre-IPO Earnings Management on Post-
IPO Ownership Structure’, Academy of Accounting and Financial Studies Journal.
Page 200
Page 201
THE ASSOCIATION BETWEEN FINANCIAL

LITERACY AND TRUST IN FINANCIAL MARKETS
AMONG NOVICE NONPROFESSIONAL INVESTORS
Lee M. Kersting, Northern Kentucky University
Robert N. Marley, University of Tampa
Mark J. Mellon, University of South Florida
ABSTRACT
This study contributes to the capital markets literature by examining the association
between novice nonprofessional investors’ financial literacy and their trust in financial markets
and individuals who operate those markets. We apply psychological contract theory to a
financial market setting to predict that financial literacy will be negatively associated with trust
in financial markets. In an experimental setting, we provide novice nonprofessional investors
with an instrument used to assess financial literacy and their trust in financial markets.
Controlling for investing experience, age, gender, and propensity for risk, we find a negative
association between novice nonprofessional investors’ financial literacy and their trust in
financial markets. Our findings should be of interest to policy makers and financial market
practitioners, potentially prompting future research to examine the effect of novice
nonprofessional investors’ financial literacy on their financial market participation.
INTRODUCTION
A growing number of individuals are required to make their own investment decisions as
private enterprise shifts from offering employees defined-benefit retirement plans in favor of
providing defined-contribution retirement plans (Franklin, 2011). This shift requires individuals
to take a more direct role in making long-term investment decisions, forcing many individuals to
assume a new and unfamiliar role as a novice nonprofessional investor (Campbell, Jackson,
Madrian & Tufano, 2011; Dorfman, 2013). Given that all investment decisions require an act of
faith (i.e., trust) on the part of the investor, research that examines the effects of trust on
investors’ financial decisions is important to academia, policy makers, and financial market
practitioners (Guiso, Sapienza & Zingales, 2008). Accordingly, the specific purpose of this study
is to apply a theoretical basis to explain how novice nonprofessional investors’ financial literacy
(an antecedent of trust) affects their trust in the financial markets and in the individuals who
operate the financial markets. This study defines financial literacy as the knowledge one has of
concepts inherent to investing and to the operation of financial markets in general (Mandell,
2006).1
Recognizing the literature is replete with various definitions of trust, this study follows
the definition provided by Sapienza and Zingales (2012): “the expectation that another person or
institution will perform actions that are beneficial, or at least not detrimental, to us regardless of
our capacity to monitor those actions.” Accordingly, an individual’s decision to commit funds to
a market-traded security requires that the individual trust the financial markets and the
individuals involved in operating financial markets to function as intended (Changwony,
Campbell & Tabner, 2014).
Page 202
While a robust stream of literature demonstrates that trust affects investors’ participation
in financial markets (Van Rooij, Lusardi & Alessie, 2011; Mayer, 2008; Guiso et al. 2008;
Guiso, 2010; Georgarakos & Pasini, 2011), a paucity of research examines variables that affect
trust itself. Given that many individuals are increasingly required to make their own investment
decisions, an examination of variables that affect novice nonprofessional investors’ trust is
needed to establish a greater understanding of how trust can be fostered. Policy makers, in
particular, should be especially interested in research that examines variables affecting novice
nonprofessional investors’ trust because: (1) novice nonprofessional investors’ financial market
participation rates are low, (2) their participation has declined significantly since the financial
crisis, and (3) those novice nonprofessional investors who choose participate in financial markets
remain chronically underinvested (Guiso, Haliassos & Jappelli, 2002; Changwony et al. 2014).
By more thoroughly understanding the factors affecting trust, policy makers can pursue
empirically sound policies to improve novice nonprofessional investors’ trust, ultimately
encouraging greater participation in financial markets.2
Psychological contract theory concerns the mutual expectations regarding inputs and
outputs owed to individuals which arise from the relationship existing between an individual and
another party (Rousseau, 1989; Morrison & Robinson, 1997). Extending psychological contract
theory from a labor market setting to a financial market setting, we predict novice
nonprofessional investors’ financial literacy will have a negative association with novice
nonprofessional investors’ trust in financial markets and the individuals who operate them.
Research that applies psychological contract theory in labor market contexts demonstrates that as
an employee become increasingly mature, that is, as an employee gains greater knowledge, the
employee requires more of his expectations to be written into a formal labor contract (Shruthi &
Hermanth, 2012). Less knowledgeable employees permit more of their expectations to exist in a
non-written “psychological” contract (Shruthi & Hermanth, 2012). Thus, knowledgeable
employees are less trusting of their employer than less knowledgeable employees. Extending this
research to a financial market setting, we predict that more knowledgeable investors will be less
trusting of financial markets and in the individuals who operate the financial markets.
We conducted an experiment in which novice nonprofessional investors completed two
questionnaires; one questionnaire assessed investors’ financial literacy and their propensity for
risk while the other assessed investors’ degree of trust in the financial markets. As predicted, we
find a negative association between investors’ financial literacy and their trust in financial
markets and the individuals involved with operating financial markets, thereby suggesting the
more knowledgeable investors are, the less trusting they are in the financial markets.
Our findings contribute to the academic literature by providing a theoretical basis to
explain empirical evidence that suggests investors’ financial literacy negatively affects their trust
in financial markets. Consequently, our results should also be of interest to policy makers
seeking both to influence the public’s trust in the financial markets and to encourage greater
financial market participation.
The remainder of the study is organized as follows. The next section provides a review of
the trust and financial literacy literature, discusses the main research questions addressed by this
literature, and develops our theory-based hypothesis. Then, details of our participants and the
method we used to test our hypothesis are discussed. The fourth section presents our empirical
results and analysis. Finally, we summarize our results, discuss this study’s limitations, and
provide suggestions for future research.
Page 203
LITERATURE REVIEW
Trust
Trust is an important element in free market transactions and is an essential precondition

for financial market participation because investing requires individuals to part with their money
in exchange for promises of capital appreciation or dividends (Arrow, 1972). Sapienza and
Zingales (2012) go so far as to suggest that investments would cease to exist in the absence of
trust. Until recently, empirical work largely ignored the role of trust in explaining stock market
participation. Although there is scant research assessing the impact of trust on financial market
participation, there are a few studies that have analyzed trust and its impact on financial decision-
making by focusing broadly on either public trust, generalized trust, or personalized trust.3 These
areas of trust research are described below.
Public trust can be broadly defined as a society’s level of trust which arises from
environmental, societal, and institutional differences between countries (Zak & Knack, 2001).
The extant literature suggests that public trust has a direct effect on decision-making and thus an
indirect effect on financial markets. For instance, public trust has been shown to be positively
correlated with economic growth (Carlin, Dorobantu & Viswanathan, 2009). In addition, public
trust is also positively correlated with prospective investors’ participation in the stock market
(Carlin et al. 2009). The results support the idea that trust is paramount to investment decisions
and that without trust the financial markets may not exist (Arrow, 1972; Sapienza & Zingales,
2012).
Generalized trust refers to whether individuals feel that people are generally trustworthy
or whether individuals must be careful when dealing with others (Guiso et al. 2008).4 Therefore,
generalized trust is not directed toward a specific party or entity. The extant literature
demonstrates that generalized trust partially explains households’ decisions to invest in the stock
market (Guiso et al. 2008; Sapienza & Zingales, 2012). Specifically, Guiso et al. (2008) show
that individuals who exhibit higher levels of generalized trust are approximately 50 percent more
likely to participate in financial markets. Sapienza and Zingales (2012) find that generalized trust
affects the level of financial market participation. For example, households that reported the
lowest levels of generalized trust stated an intention to divest from the financial market in the
future while households that reported the highest level of generalized trust stated an intention to
increase stock ownership in the future. Therefore, Guiso et al. (2008) and Sapienza and Zingales
(2012) illustrate that generalized trust impacts decisions to participate in financial markets.
In addition to public trust and generalized trust, the trust literature also mentions how
personalized trust impacts individuals’ decisions to participate in financial markets.5
Personalized trust assesses an individual’s trust in specific individuals or institutions and is
typically measured by asking individuals how much they trust the individuals who serve as
intermediaries to the financial markets. For example, Monticone (2010) assesses the level of trust
that Italian banking customers had in their bank advisor or financial promoter concerning
financial investments. Additionally, Sapienza and Zingales (2012) assess the level of trust
individuals have in the stock market. They find that individuals’ responses were 2.13 on a 5-
point scale.6 This indicates that individuals, in general, exhibit a low level of trust in the stock
market.
In addition to illustrating individuals’ personalized level of trust, the research has also
illustrated how personalized trust can impact financial decisions. For instance, research has
shown that personalized trust in a financial intermediary impacts investors’ financial decisions.
Page 204
Specifically, the more trust an individual has in a financial institution or in a financial advisor,
the greater the probability that the individual will hold risky assets (Guiso et al. 2008;
Monticone, 2010). Additionally, personalized trust, which in this case relates to a household’s
trust toward the bank as a broker, partially explains households’ investments in the stock market
(Guiso et al. 2008). Specifically, this measure of personalized trust has a positive and significant
effect on stock market participation as well as the decisions on which specific stocks to invest in.
We next extend prior theoretical research to predict how financial literacy should affect the level
of trust that individuals have in the stock market and in the individuals who operate the stock
market.
Psychological Contract Theory and the Effect of Financial Literacy on Trust
Financial literacy has been found to affect an individual’s decision to participate in

financial markets (Van Rooij et al. 2011). However, the link between financial literacy and the
level of trust that an individual has in the financial markets and in those who operate the
financial markets has not been examined.
Psychological contract theory concerns the mutual expectations regarding inputs and
outputs owed to individuals which arise from the relationship existing between an individual and
another party (Rousseau, 1989; Morrison & Robinson, 1997). Primarily examined in labor
market settings, the “psychological contract” is frequently applied to explain an employee’s
expectations with regard to what her employer owes her in return for her continued work
performance and loyalty (Rousseau & McLean Parks, 1993). However, contrary to legal
connotations which can be associated with the word “contract,” the psychological contract is not
akin to an explicit contract or express document; rather it represents notions of trust, or at least
an implied mutual understanding of expectations which might be different from one individual to
another (Rousseau, 1989).
Psychological contracts can emerge from a number of different contexts. For example,
when one party believes that a promise of future return has been made (e.g. pay for performance)
or when a contribution has been given (e.g. an exchange of some sort) and thus an obligation is
owed (Robinson & Rousseau, 1994). While psychological contract research has been primarily
conducted in labor market studies examining contracts negotiated between employers and
employees, we extend the theory to a financial market context in which investors exchange
money expecting future capital returns (Robinson & Rousseau, 1994). Psychological contract
theory is contingent upon the party’s belief about the contract, thus the investors’ belief that their
investment will lead to a return forms the basis of a psychological contract.7
Within the context of this study, we extend the psychological contract to a financial
market setting to refer to the mutual expectations of inputs and outputs due to the relationship
arising among investors, financial markets, and the individuals involved with operating the
markets. That is, we extend the psychological contract to represent the obligations, rights, and
rewards an investor believes he is owed by the financial markets and those who operate the
markets in return for the investor’s capital. Specifically, we hold that financial markets expect
investors to commit capital to risky ventures, while investors expect financial markets to operate
in an unbiased (i.e., fair) manner.
Psychological contract research from the labor market domain demonstrates that more
“mature” (i.e., more knowledgeable) employees demand greater portions of their expectations to
be written into a formal employment contract as compared to less mature employees who permit
more of their expectations to exist in a non-written psychological contract (Shruthi & Hermanth,
Page 205
2012). In other words, more knowledgeable employees are less trusting of their employer
compared to less knowledgeable employees. Extending this logic in an analogous manner to a
financial market setting, an argument predicting that more knowledgeable investors should be
less trusting of financial markets can be advanced. That is, an argument can be made that
individuals with higher levels of financial literacy will have lower levels of trust in financial
markets. For instance, Guiso (2010) suggests that the financial crisis has had a significantly
negative impact on individuals’ perceived trustworthiness of financial intermediaries and their
managers. An extension to this argument would then suggest that individuals with higher
financial literacy are more likely to understand incentive structures applicable to the various
market participants and thereby have a better understanding of the potential for those
participants’ self-interested actions to negatively impact their wealth. It could be argued that
individuals with higher financial literacy understand how financial markets are intended to
operate and thus are more able to perceive scenarios under which the market would not operate
as intended. Those investors with higher financial literacy could perceive the fallibility of
gatekeepers (i.e., financial intermediaries, financial advisors) and therefore have lower levels of
trust in the financial market and in those individuals associated with the financial market.
On the other hand, it could be argued that individuals with lower levels of financial
literacy are less likely to trust that financial markets and the individuals associated with those
financial markets will operate as intended. By definition individuals with low financial literacy
do not fully understand how financial markets operate and, it could be argued, do not understand
what functions individuals involved in the financial markets perform. Therefore, individuals with
low financial literacy would be likely to have lower levels of trust in the financial market and the
individuals who operate those markets because they lack the general understanding of how the
markets operate and what the individuals associated with the financial market are responsible for.
The impact on trust would therefore result from a lack of knowledge, whereby individuals do not
place trust in environments they do not fully understand.
Extending psychological contract theory research findings to a financial market context
involving novice nonprofessional investors, we predict a negative association between
nonprofessional investors’ financial literacy and their trust such that:
H1 The financial literacy of novice nonprofessional investors is negatively associated with their trust
in the financial markets and those individual who operate those markets.
Control Variables
Both intuition and prior literature suggest various factors in addition to financial literacy
that may cause an investor’s trust in the financial markets and those individuals who operate
those markets to vary. Intuition suggests that an investor’s trust will be affected both positively
and negatively by his or her investment experience. Prior literature suggests that one’s age,
gender and propensity for risk all have the potential to affect one’s trust. For example, Castle et
al. (2012) identify an age effect such that older individuals are more trusting than younger
individuals. Additionally, Jacobsen, Lee, and Marquering (2008) have identified a gender
difference such that men are significantly more optimistic than women with respect to economic
and financial indicators. Further, trust involves an element of risk (Lewis & Weigert, 1985).
Consequently, drawing upon intuition and prior research, we therefore control for the investor’s
investment experience, age, gender and propensity for risk in robustness tests of the relation
between financial literacy and trust.
Page 206
METHODOLOGY
Experimental Task
Experimental sessions were conducted with 88 undergraduate business students. Each

session lasted approximately 20 minutes. During each session, participants completed two
questionnaires. The first questionnaire was used to obtain measures of each participant’s
financial literacy and his or her propensity for risk.8 After completing the questionnaire
participants completed a second questionnaire on which they provided demographic information,
such as gender and age, as well as information about their investing experience and answered a
question relating to their trust in the financial markets and those who operate the financial
markets.
Dependent Variable Measure
The dependent variable was measured as each participant’s response to the following
question: “I trust the financial markets and those individuals involved with them to operate as
intended.” The responses were made on a 9-point Likert scale with end-points labeled “Strongly
Disagree” (1) and “Strongly Agree” (9).
Independent Variable Measure
The first ten questions on the first questionnaire measured financial literacy. These ten
questions had been used previously by Agnew and Szykman (2005) and were similar to those
used in other studies to measure financial literacy (Dwyer, Gilkeson & List, 2002; Wilcox,
2003). In accordance with prior literature, each participant’s financial literacy was measured as
the number of questions answered correctly (Dwyer et al. 2002; Wilcox, 2003; Agnew &
Szykman, 2005; Marley & Mellon, 2013).
Control Variables Measures
Each individual’s investment experience, age, gender and propensity for risk were
measured. Investment experience was measured using the investor’s response to the following
statement on the second questionnaire: “Please circle your level of investment experience.” The
responses were made on a 9-point investment experience scale (1 to 9, with endpoints labeled
“no investment experience” and “a great deal of investment experience,” respectively).
The individual’s age and gender were obtained by asking individuals to self-report. The
individual’s propensity for risk measure was obtained by asking participants to select which of
five distinct gambles would be their preference to play. This approach is similar to that used in
prior literature to measure individuals’ propensity for risk (Eckel & Grossman, 2002). The
gambles were constructed such that the number of the gamble was correlated with the level of
risk associated with it. Therefore, a risk-averse individual would most likely select Gamble 1 or
2 while a risk-seeking individual would most likely select Gamble 4 or 5. Consistent with prior
literature each participant’s propensity for risk was scored as the number of the gamble that he or
she selected (Eckel & Grossman, 2002).
Page 207
RESULTS
The descriptive statistics regarding participants’ trust, financial literacy, investing

experience, age, gender and propensity for risk are presented in Table 1. The participants’ mean
trust score was 4.63 (2.19 Std. Dev.) with scores ranging from 1 to 9. Therefore participants
exhibited a moderate amount of trust in the financial markets and in those who operate the
financial markets. The mean financial literacy of the participants was 4.93 (1.74 Std. Dev.) with
a range from 2 to 10. The participants’ self-reported investment experience ranged from 1 to 8 on
a 9-point investment experience scale (1 to 9, with endpoints labeled “no investment experience”
and “a great deal of investment experience”, respectively. The average reported investment
experience was 2.05 (1.59 Std. Dev). The age of the participants ranged from 20 to 57 years old,
with a mean age of 23 years. Forty-nine (55.7%) of the participants were female and thirty-nine
(44.3%) were male.9 Table 1 also illustrates that participants’ average propensity for risk was
2.22 (1.28 Std. Dev). The lower propensity for risk score indicates that on average the
participants are somewhat risk-averse.
Table 1
DESCRIPTIVE STATISTICS OF NOVICE NONPROFESSIONAL INVESTORS
Standard 25th 50th 75th

Mean Deviation Min Max Percentile Percentile Percentile
a
Trust 4.63 2.19 1 9 3.00 5.00 6.00
Financial Literacyb 4.93 1.74 2 10 4.00 5.00 6.00
Investment Experiencec 2.05 1.59 1 8 1.00 1.00 3.00
Aged 22.99 5.32 20 57 21.00 22.00 22.75
Gendere 0.44 0.50 0 1 0.00 0.00 1.00
Propensity for Riskf 2.22 1.28 1 5 1.00 2.00 3.00
a
Trust in the Financial Markets: The response to the following statement in the exit questionnaire “I trust the financial markets
and those individuals involved with them to operate as intended.” The responses were made on a 9-point scale (1 to 9, with
endpoints labeled “strongly disagree” and “strongly agree,” respectively).
b
Financial Literacy: The degree to which an individual is financially literate, measured by the number of questions answered
correctly on the financial literacy scale developed by Agnew and Szykman (2005).
c
Investment Experience: The response to the following statement in the exit questionnaire: “Please circle your level of
investment experience.” The responses were made on a 9-point investment experience scale (1 to 9, with endpoints labeled “no
investment experience” and “a great deal of investment experience,” respectively).
d
Age: The individuals’ self-reported age.
e
Gender: Coded as 1 for males and 0 for females.
f
Propensity for Risk: The individual’s propensity for risk, measured on the financial risk scale developed by Eckel and Grossman
(2002).
Page 208
Correlation Analysis
Table 2 presents bivariate correlations for variables to be used in our regression analysis
which is presented in Table 3. We predict that financial literacy is negatively associated with
trust in the financial markets. As per Table 2 and consistent with our hypothesis there is a
significantly negative correlation between trust and financial literacy (-0.19, p = 0.08). The
correlations between trust and investment experience, age, gender and propensity for risk are
also presented in Table 2; none of our control variables are significantly correlated with trust.
These results, however do not take into account effects present with all the variables in the
model.
There are significant correlations between our independent variable and our control
variables. For example, investment experience is significantly correlated with financial literacy
(0.21, p = 0.06), age (0.20, p = 0.06) and gender (0.24, p = 0.03), while financial literacy is
significantly correlated with gender (0.34, p = 0.01). To address concerns that multicollinearity
issues may bias our findings, we considered whether any of the variance inflation factors (VIF)
for the independent variable and control variables were greater than ten or whether the average
VIF was significantly greater than one. Given that the largest VIF was well below ten (1.18) and
that the average VIF (1.11) was not significantly greater than one, we have no concerns that
multicollinearity influenced our results.
Table 2
BIVARIATE CORRELATIONS FOR VARIABLES TO BE USED IN REGRESSION ANALYSIS
Financial Investment Propensity

Trust Literacy Experience Age Gender for Risk
-0.19 0.07 -0.06 -0.09 0.09
Trusta 1 (0.08) (0.49) (0.55) (0.42) (0.40)
-0.17 0.21 0.15 0.34 0.07
Financial Literacyb (0.11) 1 (0.06) (0.17) (0.01) (0.53)
Investment 0.04 0.08 0.20 0.24 -0.02

(0.70) (0.45) 1 (0.06) (0.03) (0.88)
Experiencec
-0.11 -0.05 0.19 0.02 -0.07
Aged (0.32) (0.65) (0.07) 1 (0.89) (0.53)
-0.08 0.34 0.20 -0.02 0.10
Gendere (0.49) (0.01) (0.07) (0.85) 1 (0.35)
0.04 0.08 -0.03 0.19 0.20
Propensity for Riskf (0.70) (0.45) (0.80) (0.07) (0.07) 1
*Pearson correlation statistics are reported above the diagonal and nonparametric Spearman correlation statistics are
reported below the diagonal. Two-tailed p-values are in parentheses.
a
Trust in the Financial Markets: The response to the following statement in the exit questionnaire: “I trust the
financial markets and those individuals involved with them to operate as intended.” The responses were made on a
9-point scale (1 to 9, with endpoints labeled “strongly disagree” and “strongly agree,” respectively).
b
Financial Literacy: The degree to which an individual is financially literate: measured by the number of questions
answered correctly on the financial literacy scale developed by Agnew and Szykman (2005).
Page 209
c
Investment Experience: The response to the following statement in the exit questionnaire: “Please circle your level
of investment experience.” The responses were made on a 9-point investment experience scale (1 to 9, with
endpoints labeled “no investment experience” and “a great deal of investment experience,” respectively).
d
e
Gender: Coded as 1 for males and 0 for females.
f
Propensity for Risk: The individual’s propensity for risk: measured on the financial risk scale developed by Eckel
and Grossman (2002).
Regression Analysis
To further investigate the robustness of the negative relation between financial literacy
and trust we conducted additional regression analyses. Table 3 allows us to perform a more
thorough analysis of our hypothesis which predicts that a novice nonprofessional investor’s
financial literacy will be negatively associated with his or her trust in the financial markets and in
those individuals responsible for operating the financial markets.
Table 3 displays the results of regression analysis performed to examine the relation
between an individual’s financial literacy and his or her trust in the financial markets. In the first
model, investors’ trust is regressed onto their financial literacy. In the second model, investors’
trust is regressed onto their financial literacy while controlling for their prior investment
experience. In the third model, the investors’ trust is regressed onto their financial literacy while
controlling for their prior investment experience and a number of individual character traits,
more specifically the investor’s age, gender and propensity for risk.
In regards to the first model, the coefficient on financial literacy is negative and
significant (-0.19, t = -1.80, p = 0.038). This finding is unaffected by the inclusion of the control
variables related to the investor’s investment experience and/or character traits as the coefficient
on financial literacy in the second model (-0.22, t = -1.99, p = 0.025) and the third model (-0.20,
t = -1.71, p = 0.046) continues to be negative and significant.10 Consequently, the regression
analysis provides additional support for the existence of a negative association between financial
literacy and trust in the financial markets. Specifically, regression results support the
psychological contract theory argument that more knowledgeable investors have a better idea of
how markets operate and what expectations they should have when investing in a market.
Therefore, because they have this knowledge, they are less trusting of financial markets and
those who operate them because they understand more about the markets themselves. The
evidence thereby demonstrates an association between a novice nonprofessional investor’s
financial literacy and trust, such that the higher the investor’s financial literacy, the lower his or
her trust in the financial markets and in those individuals responsible for operating the financial
markets.
Page 210
Table 3
REGRESSION MODELS OF AN INDIVIDUAL’S TRUST IN THE FINANCIAL MARKETSa
Full Model: Trust in the Financial Markets = α0 + α1Financial Literacy + α2Investment Experience + α3Age +
α4Gender + α5Propensity for Risk + ε
Variable Model 1 Model 2 Model 3

Intercept 5.81 *** 5.63 *** 5.69 ***
Financial Literacyb -0.19 ** -0.22 ** -0.20 **
Investment Experiencec 0.12 0.14
Character Traits
Aged -0.06
Gendere -0.07
Propensity for Riskf 0.11
Adjusted R2 0.03 0.03 0.01

Sample Size 88 88 88
*, **, *** One-tailed significance at the 0.10, 0.05, and 0.01 levels, respectively, for predicted effects.
a
Trust in the Financial Markets: The response to the following statement in the exit questionnaire: “I trust the
financial markets and those individuals involved with them to operate as intended.” The responses were made on a
9-point scale (1 to 9, with endpoints labeled “strongly disagree” and “strongly agree,” respectively).
b
Financial Literacy: The degree to which an individual is financially literate, measured by the number of questions
answered correctly on the financial literacy scale developed by Agnew and Szykman (2005).
c
Investment Experience: The response to the following statement in the exit questionnaire: “Please circle your level
of investment experience.” The responses were made on a 9-point investment experience scale (1 to 9, with
endpoints labeled “no investment experience” and “a great deal of investment experience,” respectively).
d
e
Gender: Coded as 0 for females and 1 for males.
f
Propensity for Risk: The individual’s propensity for risk: measured on the financial risk scale developed by Eckel
and Grossman (2002).
CONCLUSIONS AND AREAS FOR FUTURE RESEARCH
Prior research demonstrates that investors’ participation in financial markets is positively

related to trust. This study extends the capital markets literature by drawing upon psychological
contract theory to explain the association between novice nonprofessional investors’ financial
literacy and their trust in financial markets. Our results indicate that there is a negative
association between novice nonprofessional investors’ financial literacy and their trust that
financial markets and the individuals involved with the markets operate as intended. Further, if
broader participation in financial markets is a goal of policy makers, this paper has two
additional implications: First, it is important to directly examine how an investor’s financial
literacy ultimately affects his or her participation in financial markets. Second, our findings
Page 211
highlight the need for researchers to answer the call for research by Guiso et al. (2008) which
examines factors affecting investors' perception of market trustworthiness.
Our results stem from a one-item measure of trust and a measure of financial experience
that does not capture detailed information related to individuals’ returns from prior investments.
As a result, future research may find it fruitful to extend this study by examining if our findings
are robust to a more comprehensive measure of trust and investment experience. Lastly, we
examine the association between financial literacy and trust, thus we cannot infer causality.
ENDNOTES
1 Our definition of financial literacy does not relate to the quality of the actual financial decisions made by
investors. Rather, financial literacy serves as a measure of an individual’s knowledge of investing concepts
and how financial markets operate. Accordingly, this paper does not examine whether financial literacy
leads investors to make “better” or more optimal financial decisions.
2 Throughout the remainder of this paper, our use of the word “investor” denotes a novice nonprofessional
investor.
3 Another line of research analyzes cross-cultural differences in trust and finds significant differences (e.g.,
Guiso et al., 2008). This study, however, does not examine cross-cultural differences in trust.
4 For example, Guiso et al. (2008) asked participants in their survey, “Generally speaking, would you say
that most people can be trusted or that you have to be very careful in dealing with people?” (p. 2558).
5 Monticone (2010) defines personalized trust as the trust an individual has in his or her own bank or
financial advisor.
6 The end points on the 5-point scale were (1) I do not trust at all and (5) I trust completely.
7 Psychological contract theory does not require that both parties have the same understanding of the contract
(Robinson & Rousseau, 1994).
8 See the Appendix for the questionnaire.
9 These demographic measures are consistent with our a priori expectations and with demographic measures
reported in a number of studies that used undergraduate students as proxies for nonprofessional investors
(e.g., Chewning, Coller & Tuttle, 2004; Liyanarachchi & Milne, 2005).
10 Untabulated results show that our findings are unaffected by using the investors’ years of investing
experience as the measure of their investment experience rather than their self-reported level of investment
experience.
Page 212
REFERENCES
Agnew, J. R. & L. R. Szykman (2005). Asset allocation and information overload: The influence of information
display, asset choice, and investor experience. The Journal of Behavioral Finance, 6(2), 57-70.
Arrow, K. J. (1972). Gifts and exchanges. Philosophy & Public Affairs, 1(4), 343-362.
Campbell, J. Y., H. E. Jackson, B. C. Madrian & P. Tufano (2011). Consumer financial protection. Journal of
Economic Perspectives, 25(1), 91-114.
Carlin, B. I., F. Dorobantu & S. Viswanathan (2009). Public trust, the law, and financial investment. Journal of
Financial Economics, 92(3), 321-341.
Castle, E., N. I. Eisenberger, T. E. Seeman, W. G. Moons, I. A. Boggero, M. S. Grinblatt & S. E. Taylor (2012).
Neural and behavioral bases of age differences in perceptions of trust. Proceedings of the National
Academy of Science of the United States of America. doi:10.1073/pnas.1218518109
Changwony, F. K., K. Campbell & I. T. Tabner (2014). Social engagement and stock market participation. Review
of Finance, forthcoming.
Chewning Jr, E.G., M. Coller & B. Tuttle (2004). Do market prices reveal the decision models of sophisticated
investors?: Evidence from the laboratory. Accounting, Organizations and Society, 29(8), 739-758.
Dorfman, J. (2013, July 25). Detroit's bankruptcy should be a warning to every worker expecting a pension, or social
security. Forbes. Retrieved from http://www.forbes.com/sites/jeffreydorfman/2013/07/25/detroits
-bankruptcy-should-be-a-warning-to-every-worker-expecting-a-pension-or-social-security/
Dwyer, P.D., J. H. Gilkeson & J. A. List (2002). Gender differences in revealed risk taking: evidence from mutual
fund investors. Economics Letters, 76(2), 151-158.
Eckel, C.C. & P. J. Grossman (2002). Sex differences and statistical stereotyping in attitudes toward financial risk.
Evolution and Human Behavior, 23(4), 281-295.
Franklin, M.B. (2011, November). What killed pensions. Kiplinger's Personal Finance. Retrieved from
http://www.kiplinger.com/article/retirement/T047-C022-S001-what-killed-pensions.html
Georgarakos, D. & G. Pasini (2011). Trust, sociability, and stock market participation. Review of Finance, 15(4),
693-725.
Guiso, L., M. Haliassos & T. Jappelli. (2002). Household Portfolios. Cambridge, MA: MIT Press.
Guiso, L., P. Sapienza & L. Zingales (2008). Trusting the stock market. The Journal of Finance, 63(6), 2557-2600.
Guiso, L. (2010). A trust-driven financial crisis. Implications for the future of financial markets. (EUI Working
Paper ECO2010/07). European University Institute. Retrieved December 14, 2014, from
http://cadmus.eui.eu/handle/1814/13657
Jacobsen, B., J. Lee & W. Marquering. (2008). Are men more optimistic? (SSRN Working Paper 1030478). Social
Science Research Network. Retrieved December 14, 2014, from http://papers.ssrn.com/sol3/papers.cfm
?abstract_id=1030478
Lewis, J. D. & A. Weigert (1985). Trust as a social reality. Social Forces, 63, 967-985.
Liyanarachchi, G. A. & M. J. Milne (2005). Comparing the investment decisions of accounting practitioners and
students: An empirical study on the adequacy of student surrogates. Accounting Forum, 29, 121-135.
Mandell, L. (2006). If it's so important, why isn't it improving? (SSRN Working Paper 923557). Social Science
Research Network. Retrieved December 14, 2014, from http://ssrn.com/abstract=923557
Marley, R. & M. J. Mellon. (2015). The effects of current and expanded analyst ownership disclosure on
nonprofessional investors’ judgments and decision-making. (SSRN Working Paper 2549872). Social
Science Research Network. doi:10.2139/ssrn.2549872
Mayer, C. (2008). Trust in financial markets. European Financial Management, 14(4), 617-632.
Monticone, C. (2010). Financial literacy, trust and financial advice. (Research Gate Working Paper). Retrieved
December 14, 2015, from http://www.researchgate.net/publication/228432162_Financial_Literacy_
Trust_and_Financial_Advice
Morrison, E. W. & S. L. Robinson (1997). When employees feel betrayed: A model of how psychological contract
violation develops. Academy of Management Review, 22(1), 226-256.
Robinson, S. L. & D. M. Rousseau (1994). Violating the psychological contract: Not the exception but the norm.
Journal of Organizational Behavior, 15(3), 245-259.
Rousseau, D. M. (1989). Psychological and implied contracts in organizations. Employee Responsibilities and
Rights Journal, 2(2), 121-139.
Rousseau, D. M. & J. McLean Parks (1993). The contracts of individuals and organizations. In L. L. Cummings &
B. M. Stow (Eds.), Research in Organizational Behavior, 15, 1-47. Greenwich, CT: JAI Press.
Sapienza, P. & L. Zingales (2012). A trust crisis. International Review of Finance, 12(2), 123-131.
Page 213
Shruthi, V. K. & K. P. Hemanth (2012). Influence of psychological contract on employee-employer relationship.

Journal of Exclusive Management Science, 1(7), 1-10.
Van Rooij, M., A. Lusardi & R. Alessie (2011). Financial literacy and stock market participation. Journal of
Financial Economics, 101(2), 449-472.
Wilcox, R.T. (2003). Bargain hunting or star gazing? Investors’ preferences for stock mutual funds. The Journal of
Business, 76(4), 645-663.
Zak, P. J. & S. Knack. (2001). Trust and growth. The Economic Journal, 111(April), 295-321.
Page 214
APPENDIX
Pre-Experiment Questionnaire
For each of the following 11 questions please identify the choice that best completes the statement
or answers the question.
1. Which of the following types of investments are typically found in a money market fund?
a. Stocks
b. Bonds
c. Short Term Securities
2. When is the best time to transfer money into a long-term bond fund? When interest rates are
expected to ________ .
a. Increase
b. Remain stable
c. Decrease
d. Interest rate doesn’t matter
e. Don’t know
3. If you were to invest $1,000 in a STOCK FUND, would it be possible to have less than $1,000
when you decide to withdraw or move it to another fund?
True False
4. If you were to invest $1,000 in a BOND FUND, would it be possible to have less than $1,000
when you decide to withdraw or move it to another fund?
True False
5. If you were to invest $1,000 in a MONEY MARKET FUND, would it be possible to have less
than $1,000 when you decide to withdraw or move it to another fund?
True False
6. A stock fund’s beta rating can best be described as __________ .

a. A measure of relative volatility of the fund versus the S&P 500 index
b. A measure of relative growth versus the S&P 500 index
c. A measure of the relative capital outflow of the fund versus the S&P 500 index
7. A money market mutual fund is guaranteed by the U.S. government against principal loss.
True False
8. High yield bond funds are invested in bonds with strong credit ratings.
True False
9. If you invest in a bond mutual fund with an average maturity of five years, this means that you
cannot withdraw your money from the fund within a five-year period without incurring a penalty.
Page 215
True False
10. A stock market index fund is actively managed by a fund portfolio manager.
True False
11. We would like you to seriously consider the following hypothetical gambles. Which of the
following five gambles would you prefer if you could only choose one? Please indicate your
choice __________.
Gamble Probability Payoff

Choice Event (%) ($)
1 A 50 16
B 50 16
2 A 50 24
B 50 12
3 A 50 32
B 50 8
4 A 50 40
B 50 4
5 A 50 48
B 50 0
Page 216
Page 217
THE EFFECTS OF COMPENSATION SCHEME,

SOURCE CREDIBILITY, AND RECEIVER
INVOLVEMENT ON THE ORGANIZATIONAL
BUDGETING PROCESS
Gloria J. McVay, Winona State University
ABSTRACT
The results of this study provide several valuable insights into the budgeting process. The
Elaboration Likelihood Model predicts that subjects will elaborate on a message when receiver
involvement is high, but could bias the integration of mixed messages to support the initial
positions favored by the subjects. Incentive compensation formulas were used to establish initial
attitudes toward messages in the high involvement subjects. Pro-attitudinal messages were
expected to receive more weight than counter-attitudinal messages. Experimental results
revealed parameter estimates that were consistently in the right direction, but statistical tests
were not significant. Lack of power may have caused the interaction of initial attitude
(manipulated through compensation scheme) and message direction to go undetected.
In contrast, low involvement receivers were predicted to use heuristics, such as relying
on peripheral cues, when evaluating messages. In this study, source credibility acted as the
peripheral cue. A qualitative analysis of receiver involvement presented evidence of more
decision-related thoughts associated with source credibility under conditions of low receiver
involvement than under conditions of high receiver involvement. This supports the ELM
prediction that low receiver involvement leads to more reliance on peripheral cues such as
source credibility.
The ELM predicts that high involvement subjects will use source credibility as evidence
to support their initial attitude. Content analysis also supported this prediction. Subjects in the
high receiver involvement groups were more likely to present message source arguments when
the direction of the message (sales-increasing versus sales-decreasing) favored their initial
attitude.
Under conditions of high receiver involvement, source credibility was hypothesized to
interact with compensation scheme and message direction. In interpreting the results of this
three-way interaction, budget-adjustments were broken down into two categories: the first
budget adjustment (the change in the sales forecast after receiving the first message) and the
second budget adjustment. The three-way interaction term for source credibility, message
direction, and incentive compensation group (high goal versus low goal) was statistically
significant when the differences between means were segregated into two groups.
For first budget adjustments, high goal subjects placed more weight on low credibility,
increasing messages than did low goal subjects. Weights for high credibility sources and sales-
decreasing messages were not significantly different between groups. For second budget
adjustments, high goal subjects placed more weight on increasing messages from high credibility
sources than did low goal subjects. Weights for sales-decreasing messages were not significantly
different between groups. And finally, for all groups and conditions, sales-decreasing messages
were weighted more heavily than sales-increasing messages. These findings suggest a pro-
attitudinal bias toward message sources that supported the initial attitude of the decision-maker.
Page 218
In both cases, the direction of credibility’s effects was as predicted. The unexpected finding was
that the effect only occurred for sales-increasing messages.
A possible interpretation of this finding is that sales-decreasing messages – messages
that suggest a downward revision of a sales forecast – are less subject to bias because of an
overall conservative bias in the budgeting process. Decision-makers might bring skepticism to
the judgment and decision-making process because the initial forecast is the result of an
aggregation of sales managers’ opinions. The uncertainty concerning the starting point might
lead to a greater saliency of decreasing messages. Decreasing messages might seem less
refutable. Any biases, therefore, could tend to enter the decision process through sales-
increasing messages where the decision-maker is more discriminating about the interpretation of
the message content and message source. If this scenario holds in real-world budgeting contexts,
a sub-optimal sales forecast could result. This bias and the resulting implications for firm
profitability need further investigation
In summary, the predictions of the ELM as tested in this study were generally supported.
The consistency of the results with the theory suggests that insignificant test results may be due
to lack of power. Differences in parameter estimates among cells consistently had the correct
sign. A larger number of subjects might be needed to provide sufficient power to adequately test
all the hypotheses under consideration in this study.
INTRODUCTION
The budgeting process is central to the functions of corporate accounting managers. As

the guardians of corporate assets, managerial accountants are charged with the responsibility of
overseeing the formulation of the annual budget. Ideally, managers strive to make budgets
accurate and free from bias, but the assumption that the budgeting process is actually free from
bias is unwarranted. Many biases, including the way managers are compensated (rewards based
on whether or not expenses come in under budget or sales exceed budgeted goals), the effect of
external sources of budget-relevant information, and the credibility of the sources of this
information all affect the budget process.
Annual budgets not only summarize expected inflows of resources, how assets will be
expended based on those expectations, and the consequent return on investment and financial
position, but also provide benchmarks for the evaluation of managerial performance and the
resultant rewards and penalties. The optimum allocation of resources implies that the budgeting
process is free from bias. Biases in the budget process leading to an under-utilization of firm
resources will impair the firm’s competitive advantage, stability, and long-term profitability. The
success of employees, owners, and managers of an organization, as well as the health of an
organization itself are thus affected by these potential biases.
Prior research has focused on antecedents and outcomes of the budgeting process, but this
has yielded little information about the types of information that affect budget decisions. This
study evaluates potential biases that may enter into a participatory budgeting process and thereby
adversely affect budget estimates. The Elaboration Likelihood Model (Petty & Cacioppo 1986a,
1986b, 1990; Petty, Cacioppo, Strathman, & Priester, 2005) provides the basis for identifying
potential biasing factors within the communication that occurs during the budgeting process. The
importance of the information to the decision-maker (the receiver), the credibility of the
information source (the sender), and receiver involvement as manifested in the personal goals of
management (how the decision-maker is compensated) are manipulated within an experimental
setting. In addition to providing new and important insights into the impact of communication
Page 219
and cognition variables on the organizational process of budgeting, this study represents the only
application of the Elaboration Likelihood Model in this field.
The Elaboration Likelihood Model
The key focus of this study is how various factors related to the communication of budget
information from one person to another might bias the decision-maker in determining the
outcome of the budgeting process. According to Petty and Cacioppo (1986a, 1986b, 1990), a key
construct of the Elaboration Likelihood Model (ELM) is the elaboration likelihood continuum,
which provides a framework for predicting cognitive effort based on factors related to motivation
and ability to evaluate the message. When motivation and ability factors are high, the ELM
predicts the “central route” of message processing – more cognitive effort and relatively high
elaboration of message content. Conversely, when either motivation or ability is low, the ELM
predicts a “peripheral route” of processing, with less elaboration of the message content, less
substantive thought, and more reliance on peripheral cues such as source credibility or
attractiveness. The “peripheral route” is a heuristic, or cognitive shortcut, to forming an attitude
or belief.
The most robust findings of this model have related to two factors. The first factor –
receiver involvement – is one of several receiver motivation factors within the ELM framework.
When receiver involvement is low due to low personal relevance of the message to the receiver,
subjects exhibit a greater reliance on peripheral cues. The second factor is a message source
factor – credibility of the sender of the message. According to the ELM, source credibility acts as
a peripheral cue when elaboration likelihood is low. However, it can also act as a piece of
evidence when elaboration likelihood is relatively high. In the case of high elaboration, the
critical factor determining the direction of credibility’s effects appears to be the nature of the
position advocated by the message – specifically, whether the message advocates a position
initially favored or opposed by the receiver. Both of these factors – receiver involvement and
source credibility – have consistently shown significance in prior studies (Hornikx & O’Keefe,
2009; McComas, 2008; Hallahan, 2008; MacGeorge, 2008; Noar, Palmgreen, Zimmerman,
Lustria, & Lu, 2010).
When budgeting is viewed as a decision process that is dependent upon communication
between interactants, the potential biasing effects of both receiver involvement and sender
credibility become evident. The budgeting decision-maker typically relies upon information
provided by other sources (senders). For example, the sales manager might provide a sales
forecast as a starting point in the preparation of the annual budget. Hence, the sales manager (the
sender) and the budget decision-maker (the receiver) interact, at least in the initial phase of the
budgeting process. Subsequently, the decision-maker might seek out corroborating evidence
from parties both within and outside the company to substantiate the sales forecast. Other factors
- such as forecasted economic conditions or steps taken by competing companies – can temper
the original sales forecast estimate. Normally, the level of budgeted sales is important (has
personal relevance) to the receiver because the decision-maker has some accountability for
results and may be compensated based upon achieving specific goals.
The ELM, however, is sensitive to degrees of involvement. High (as opposed to moderate
or low) levels of involvement lead to a relatively higher likelihood of message elaboration and
less attention to peripheral cues such as source credibility (Liu, 2008). When elaboration
likelihood is lower, the receiver’s perception of the credibility of the sender may be used
heuristically in the weighting of information. Source credibility might also interact with receiver
Page 220
involvement in another way; source credibility can be used to support the initial attitude of the
message receiver (Petty & Cacioppo 1983a, 1983b, 1990).
According to the ELM model, a person selects either the “central route” or the
“peripheral route” as a method to change or form attitudes or beliefs in response to a persuasive
message. The central route prevails under conditions leading to high elaboration, while the
peripheral route dominates under conditions of low elaboration. According to O’Keefe (2009),
when elaboration is relatively high, the directional effects of a message will be influenced by the
receiver’s initial attitude and the message’s advocated position, considered jointly. Conversely,
given relatively low elaboration, the persuasive effect will be influenced much more by the use
of a simple decision rule, such as the credibility of the source.
Management Compensation and Budgeting Decisions
The potential conflict of goals resulting from the financial incentives imbedded in
management compensation schemes is well documented in the literature (Chow, 1983;
Frederickson, 1992; Shields & Waller, 1988; Waller & Chow, 1985; Young et al. 1993).
Untangling the effects of incentives is relevant to the budgeting process (Libby & Lipe, 1992)
because conflicts between the goals of management and investors bias the allocation of resources
entrusted to management. As a result, incentive structures are a potentially important dimension
in the decision-making environment (Hogarth, 1993) and hence the budgeting process. Linkage
between incentive structures and the attainment of targeted sales or profitability goals creates an
environment of high receiver involvement within the decision-maker due to the personal
relevance attached to the outcomes.
The tenets of the Elaboration Likelihood Model, as applied to the budgeting process,
provide several testable hypotheses, given the conditions of both varied compensation schemes
and differing levels of source credibility. The following hypotheses are tested within the
experimental setting of this study.
H1: High involvement message receivers will bias the weighting of mixed messages to support their
initial attitude.
Pro-attitudinal messages will be weighted more heavily than counter-attitudinal

messages.
H2: When receiver involvement is low, source credibility will act as a peripheral cue in the budgeting
decision process.
Subjects will present evidence of more decision-related thoughts associated with source
credibility under conditions of low receiver involvement than under conditions of high receiver
involvement.
H3: When receiver involvement is high, the direction of credibility’s effects will depend on whether the
message advocates the position initially opposed or favored by the receiver.
The receiver’s compensation scheme, the source’s credibility, and the direction of the
message, will jointly influence the weighting of the evidence contained within the message.
Page 221
This study examines decision-making within the budget process by manipulating

compensation scheme and two important communication variables – sender credibility and
receiver involvement. The Elaboration Likelihood Model (Petty & Cacioppo, 1986) is used to
predict cognitive effort and issue-relevant elaboration of messages. High-involvement message
receivers are predicted to exhibit more attention to the message (the central route of cognition)
and less attention to the credibility of the receiver (the peripheral route of cognition). Source
credibility is also predicted to interact with initial attitude with high credibility sources given
relatively more weight in pro-attitudinal arguments and vice versa.
METHODS
Experimental Design
The design for this proposed study is a 3 (compensation scheme) x 2 (high or low level of
involvement) x 2 (sales increasing vs. decreasing message) x 2 (high or low credibility source)
mixed design. There are three treatment groups in this experiment. Group 1 was a low
involvement group where compensation is fixed; Groups 2 and 3 were high involvement groups
where compensation consists of both extra credit points and an incentive linked to the final
budget forecast respondents provided in the experimental task. Receiver involvement was
operationalized in this study in two ways – first, through the language and role assumption
embedded within the case materials (both Groups 2 and 3) and second, through the manipulation
of compensation schemes (Group 2 was given greater compensation for meeting a high goal,
while Group 3 was compensated on the basis of meeting a low goal). In real life, if an
executive’s pay scheme is tied to actual sales, the tendency would be to set a “high goal” for
budgeted sales. A higher sales forecast would result in the availability of more planned
production (lowering the possibility of a stock out) and a higher target for sales personnel to
meet. On the other hand, if an executive’s incentive compensation is tied to attaining the
budgeted level of sales, the tendency would be to set a “low goal” to make it easier to achieve.
Subjects were randomly assigned to one of the three groups.
Sample
The subjects were 107 MBA students from a large Southern university1. MBA students
were selected in order to obtain a broader base of business knowledge and experience (report
sample characteristics measured: age, gender, years of work experience, budgeting experience,
accounting coursework completed, current position level, and undergraduate degree). Statistical
tests for differences of mean values among the three experimental groups on these demographic
characteristics indicate the successful random assignment of subjects across groups. None of the
tests for differences in demographic variables resulted in a test statistic with a significant p-
value.
Materials and Measures
The task and case materials were developed on the basis of information widely available
in classic cost accounting textbooks dating back to Horngren (1973). Horngren states that “the
sales prediction is the foundation for the quantification of the entire business plan . . .the chief
sales officer has direct responsibility for the preparation of the sales budget” (p. 186). In
Page 222
discussing sales forecasting procedures, Horngren continues: “If possible, the budget data should
flow from individual salesmen or district sales managers upward to the chief sales officer. A
valuable benefit from the budgeting process is the holding of discussions, which generally result
in adjustments and which tend to broaden participants’ thinking … Previous sales volumes are
usually the springboard for sales predictions” (p. 187).
Horngren provides a list of factors that are considered in arriving at the sales forecast. He
states that the sales forecast is made after consideration of the following: 1) past sales volume; 2)
general economic and industry conditions; 3) the relationship of sales to economic indicators
such as gross national product, personal income, employment, prices, and industrial production;
4) relative product profitability; 5) market research studies; 6) pricing policies; 7) advertising and
other promotion; 8) quality of sales force; 9) competition; 10) seasonal variation; 11) production
capacity; and 12) long-term sales trends for various products.
The case materials in this study focused on the sales forecast as the budget variable under
consideration. Previous sales volumes and a sales forecast developed by the fictitious company’s
district sales managers provided the starting point for setting the sales forecast for the upcoming
year. Subjects were provided with two additional pieces of information (from two different
sources) that are relevant to evaluating and adjusting the sales forecast. One message contained
information about the potential favorable impact on sales due to a trend in foreign currency rates
(the dollar versus the yen) and the other message contained information about a competitor’s
acquisition of a downstream supplier common to both companies (leading to a cost advantage for
the competitor). The sources of the two messages were varied to provide differences in the level
of source credibility (high credibility versus low credibility).
Two additional pieces of information were then provided within messages received from
two different sources. The first message respondents received contained information that, if the
process is free of bias, should lead to a budget that forecasts increasing sales, while the second
message respondents received contained sales-decreasing information. The order of the pieces of
information was randomized across subjects and cells. Based upon the compensation scheme for
the subjects, the messages are designated as pro- or contra-attitudinal. In the “Low Goal”
manipulation, a message that suggested that sales would decrease was considered a pro-
attitudinal message. The converse was true for the “High Goal” manipulation. In other words, the
initial attitude of the subject, manipulated through compensation scheme, determined whether
each message was pro- or contra-attitudinal. Subjects were also asked to document their starting
point in the decision process (a manipulation check to verify “district sales managers’ forecast”
as the anchor).
Respondents were also asked to complete an 18-item measure of Need for Cognition
(Cacioppo, Petty & Kao, 1984). Need for Cognition was used as a covariate in the analyses of
the data because we thought that it might be possible that participants’ responses to case
materials might vary by the level of this trait.
Procedures
Subjects were provided with case materials, including the district sales managers’
aggregated sales forecast as a starting point in the decision process. Subjects were asked to
update their estimate of next year’s sales forecast after receiving each individual piece of
information. In all three groups, subjects were asked to provide a sales forecast for the upcoming
year based on information provided in the case. When combined with the sales forecast provided
by the district sales managers, the messages suggested either an upward or downward revision in
Page 223
the forecast. Participants were also asked to provide a numerical evaluation of the strength of
each message.
At the conclusion of the experimental task, subjects were asked to complete a
questionnaire that captured demographic information (age, gender, work experience, budgeting
experience, accounting coursework, level of current position, undergraduate field of study.
Subjects also completed a ten-item semantic differential credibility scale for each of the message
sources contained in their case material. This scale was developed by Ohanian (1990) and
measures the two components of source credibility – trustworthiness and expertise (a total
credibility score was calculated for each message source and serves as a manipulation check).
The final step in the experiment was the calculation of subjects’ compensation for participation.
ANALYSIS
Adapting the work of Bamber et al. (1997), the regression model used in this study
permitted estimation of the two message-sensitivity parameters ( and ) when the anchor, Sk1
(given as the district sales managers’ sales forecast), amount of budget revision, S k  Sk1
(collected as data from subjects) and subjective evaluation of evidence, s(xk) (collected as data
from subjects) are known. The regression model used to estimate  and  is specified using the
following regression equation):
Sk  Sk1 =  +  (COGN)
+ 1 (D)( Sk-1) s(xk) + 1 (1 - D)(1 - Sk-1) s(xk)
+ 2 (CREDj)(D)(Sk-1) s(xk) + 2 (CREDj )(1 - D)(1 - Sk-1) s(xk)
+ 3 (LOWj)(D)(Sk-1) s(xk) + 3 (LOWj )(1 - D)(1 - Sk-1) s(xk)
+ 4 (HIGHj)(D)(Sk-1) s(xk) + 4 (HIGHj )(1 - D)(1 - Sk-1) s(xk)
+ 5 (LOWj)(CREDj)(D)(Sk-1) s(xk) + 5 (LOWj)(CREDj)(1 - D)(1 - Sk-1) s(xk)
+ 6 (HIGHj)(CREDj)(D)(Sk-1) s(xk) + 6 (HIGHj) CREDj)(1 - D)(1 - Sk-1) s(xk)
+
where, Sk: adjusted budget after k pieces of evidence,

Sk-1: the anchor or prior budget
s(xk): subjective evaluation of the kth piece of evidence
: sensitivity to sales-decreasing messages, 0   1, and
: sensitivity to sales-increasing messages, 0  :1
COGN: Need for Cognition, a covariate
LOW: Low Goal incentive, and HIGH = High Goal incentive
CRED: Source Credibility (Contrast coded 0 (1) for low (high)credibility),
D: Message direction (D=0 for sales-increasing; D=1 for sales-decreasing),
,  are intercept and error terms, respectively.
Regression analysis was used to calculate coefficients for the terms in the regression
equation. As noted in Cohen and Cohen (1983), in the presence of significant interaction terms,
the main effects should be interpreted with caution and in light of the interaction effects. Prior to
determining the significance of the independent variables of interest, the error associated with
COGN (if significant) was partialed out from the total model error. Table 1 presents the
regression coefficients for  and  that correspond to the cells within the experimental design.
Page 224
RESULTS
This study used regression analysis to analyze the data and test the hypotheses under
consideration. The analysis of data began with tests of the assumptions underlying the regression
technique – normality, homoscedasticity (common variance of residuals), and outlier and
leverage points. Scatter plots showing the studentized residuals of the predictor variables to
predicted values of budget adjustment (Sk  Sk1) were developed and reviewed. The scatter plots
showed no particular pattern, indicating no violation of homoscedasticity.
Based on a review of the studentized residuals, four data points were identified as
potential outliers. Further analysis of these four data points was conducted, resulting in the
removal of two points from the data set. The two data points were removed because the subject,
when evaluating the first message, adjusted the budget upwards but marked the message on the
subjective evaluation scale (sx) as a sales-decreasing message. Then, when evaluating the second
message, the subject adjusted the budget down and rated the message on the sales-increasing
scale. The other two outliers were examined for coding problems; the data appeared to be
correctly coded and were included in the data set.
The regression model provided estimates for the dependent variable – sensitivity to sales-
increasing versus sales-decreasing messages. A significant difference in the weighting of sales-
increasing versus sales-decreasing messages provided evidence of bias within the budget task.
The alpha () terms in the model represent subjects’ sensitivity toward sales-decreasing
messages; the beta () terms in the model represent subjects’ sensitivity toward sales-increasing
messages.
The regression model designed to test the hypotheses yielded statistically significant
results (F(12,199) = 49.17, p<0.0001) with an adjusted R2 of 0.73. Table 1 summarizes the alpha
and beta values by group, source credibility, and message direction. These regression estimates
were used to develop statistical tests for the significance of main effects and interaction terms.
The actual statistical tests are available upon request from the first author.
Table 1
MESSAGE SENSITIVITY ESTIMATES
Low Involvement High Involvement
Control High Goal Low Goal Message Means
Low Credibility
Increasing Sales 0.15269 0.19106 0.14534 0.16303
Decreasing Sales 0.23789 0.23009 0.28757 0.25185
High Credibility
Group Means 0.20187 0.21737 0.21041
a
where terms are as previously defined.
b
t-statistics is significantly different from zero at p<0.05.
The regression model was also run two additional times, segregating the first budget
adjustments (F(12,93) = 40.28; p<.0001, adjusted R2 = 0.82) from the second budget adjustments
(F(12,93) = 18.79; p<.0001, adjusted R2 = 0.67). Table 2, panel A, presents the alpha and beta
values by group, source credibility, and message direction for budget adjustments made after the
first message, while Table 2, panel B, presents the results of budget adjustments after the second
Page 225
message. Statistically significant differences in group sensitivities toward messages were

obtained when the data was segregated. The lack of significant differences when the data is
pooled might point to a lack of independence between the two budget adjustments. This
phenomenon was also observed in the test results for hypothesis three, which points to a need to
segregate budget-adjustments and conduct independent tests of the data.
Table 2
MEAN SENSITIVITY SCORES (BUDGET ADJUSTMENT) BY GROUP AND MESSAGE TYPE
Panel A: First Belief Adjustment
Low Credibility
High Credibility
Group Means 0.20938 0.24185 0.20139
Panel B: Second Belief Adjustment
Low Credibility
Increasing Sales 0.07869 -0.02757 0.20778 0.08630
High Credibility
Group Means 0.18133 0.16548 0.22159
Table 3 presents the results of tests for significance of main effects for message direction,
compensation scheme, receiver involvement, and source credibility. Tests for two- and three-way
interactions are also presented. Significant interaction terms were further analyzed by testing for
significant differences in mean sensitivities for sales-increasing and sales-decreasing messages.
The first hypothesis states that high involvement message receivers will bias the
weighting of mixed messages to support their initial attitude. It was expected that pro-attitudinal
messages would be weighted more heavily than counter-attitudinal messages. In other words, the
receiver’s initial attitude and the message’s advocated position, considered jointly, were
anticipated to influence the direction of elaboration. Experimental results do not support
hypothesis one. The interaction of compensation scheme (testing for groups two and three only)
and message direction is not significant (F (1,199) = 0.07; p =.7881). The data was also analyzed
by segregating reactions to the first message from reactions to the second to determine if
combining the data masked a significant interaction between compensation group and message
direction. The interaction term remained insignificant.
The second hypothesis states that when receiver involvement is low, source credibility
will act as a peripheral cue. It was anticipated that subjects would present evidence of more
involvement than under conditions of high receiver involvement. In this study, source credibility
acted as the peripheral cue. We expected that when receiver involvement was low, source
Page 226
credibility would act as a peripheral cue. This hypothesis was tested using both qualitative and
quantitative methods.
Table 3
TESTS OF MAIN EFFECTS AND INTERACTIONS - F VALUES AND (P-VALUES)
Groups Combined First Second

Adjustment Adjustment
Main – Message Direction All 6.76 1.61 5.71
(.0100)*** (.2073) (.0189)**
Main - Compensation All 0.17 1.01 0.97
(.8466) (.3685) (.3841)
Main – Receiver Involvement All 0.29 0.22 0.11
(.5936) (.6421) (.7459)
Main – Source Credibility All 0.05 0.84 0.67
(.8189) (.3609) (.4160)
Credibility*Involvement 1 & 2/3 0.01 0.94 0.28
(.9159) (.3338) (.5961)
Direction*Compensation 2&3 0.07 1.83 1.58
(.7881) (.1799) (.2120)
Direction*Credibility 2&3 0.00 0.40 0.51
(.9770) (.5287) (.4764)
Direction*Comp*Credibility 2&3 0.25 2.83 7.66
(.6207) (.0960)* (.0068)***
Direction*Comp*Involvement 1 & 2/3 0.08 0.04 0.01
(.7827) (.8352) (.9149)
A content analysis of open-ended questions supports the receiver involvement predictions

of hypothesis two. As part of the experimental task, subjects were asked to list their thoughts
concerning the evaluation of each message to determine if source credibility was salient to the
subjects. An analysis of this thought-listing task was conducted by having two trained,
independent raters read the thought-listing text in order to note the presence or absence of
statements concerning message content and source credibility. Interrater reliability was assessed
by calculating a kappa measure of agreement. A kappa value of .84 was obtained, indicating
excellent agreement between raters2. Because of the high degree of agreement, the counts for
both argument types were averaged between the two raters. Table 4 summarizes the findings of
this analysis.
Table 4
SOURCE AND CONTENTS ARGUMENTS
---------Source Arguments---------
Group Content Percent Increasing Decreasing Percent
Arguments Usage Messages Messages Usage
Low Involvement
Control group (72) 51 70.8% 22 20 58.3%
High Involvement
Low goal (70) 52 72.2% 12 16 37.1%
High goal (72) 51.5 73.6% 19 8 37.5%
Page 227
Under the low receiver involvement condition, message source arguments were more
likely to be present in the thought-listing task (58.3% of subjects gave a source argument in the
low involvement control group versus 37.1% and 37.5% in the two high involvement groups).
This supports the prediction of hypothesis two that low involvement receivers will place more
reliance on peripheral cues such as source credibility. Message content arguments appeared to be
equally present at both levels of receiver involvement. This could be due to a fairly high amount
of involvement for both levels of the involvement variable. However, on a relative basis, the
control group showed indications of more reliance on source credibility (58.3% gave source
arguments versus 37.1% and 37.5%) – indicative of a relatively lower level of receiver
involvement. Because this experiment did not specifically measure cognitive effort, statements
concerning receiver involvement, cognitive effort, and reliance on peripheral cues can only be
inferred from the results obtained.
A quantitative analysis of receiver involvement was conducted using correlation and
regression analysis techniques. Correlation analysis provided some evidence for an association
between subjects’ ratings of credibility of a source and their subjective evaluation of the message
strength. Under conditions of low receiver involvement, a higher correlation between the two
scores was anticipated than under conditions of high receiver involvement. A significant
correlation was found, but only when the source was highly credible and the message was the
first message received (r = 0.84100; p = 0.0045). However, regression analysis failed to reject
the null hypothesis for a receiver involvement main effect (F(1,199) = 0.29; p = 0.5936) or an
interaction between receiver involvement and source credibility (F(1,199) = 0.01; p = .9159). In
summary, hypothesis two is supported by qualitative analysis (content analysis of a thought-
listing task), received some support through correlation analysis, but is not supported by
regression analysis.
The third hypothesis stated that when receiver involvement was high, the direction of the
effects of credibility would depend on whether the message advocated the position initially
opposed or favored by the receiver. It was expected that the receiver’s compensation scheme, the
source’s credibility, and the direction of the message would jointly influence the weighting of the
evidence contained within the message. As noted in the preceding section, when receiver
involvement and elaboration likelihood are expected to be low, source credibility should act as a
peripheral cue. This would be evidenced by more decision-related thoughts associated with the
credibility of the message source. Source credibility performed the role of a heuristic, or “mental
shortcut.” Conversely, when receiver involvement and elaboration likelihood are expected to be
high, source credibility is anticipated to interact with the subjects’ initial attitude and act as a
piece of evidence. In the case of high elaboration, the critical factor determining the direction of
credibility’s effects appears to be the nature of the position advocated by the message -
specifically, whether the message advocates a position initially opposed or favored by the
receiver (Petty &Wegener, 1998).
Hypothesis three posits a significant three-way interaction of incentive compensation
scheme, source credibility, and message direction. The direction of credibility’s effects would
depend on whether the message advocates the position initially opposed or favored by the
receiver. The receiver’s compensation scheme, the source’s credibility, and the direction of the
message, would jointly influence the weighting of the evidence contained within the message.
The test for a three-way interaction of compensation scheme, source credibility, and message
direction was not significant when first and second budget adjustments were tested as combined
data. However, a separation of the two budget adjustments revealed a significant interaction
Page 228
among these variables. The full regression model was run two more times, once using the data
from the first budget adjustments and the second time using the data from the second budget
adjustments. Both models showed evidence of a three-way interaction of compensation scheme,
source credibility, and message direction. The interaction term using the first budget adjustments
data resulted in a test result of F(1,93) = 2.83, p = .0960. The interaction term using the second
budget adjustments data resulted in a test result of F(1,93) = 7.66, p = .0068.
Interestingly, in the high receiver involvement groups, credibility arguments were more
prevalent in the “high goal” group when the messages were from high credibility sources and
were sales-increasing messages. While the count differences are not as large, receivers in the
“low goal” groups used low source credibility to support their attitude toward sales-decreasing
messages. This is in line with ELM theory which predicts that high involvement receivers will
use peripheral cues as evidence to support their initial attitude.
Table 5 summarizes the F-tests for equality of mean alpha and beta terms between groups
two and three (the incentive compensation groups). A significant p-value indicates that there was
a difference between the mean sensitivities toward sales-increasing (beta values) and sales-
decreasing (alpha values) messages in the groups tested.
Table 5
INTERACTION OF COMPENSATION SCHEME, SOURCE CREDIBILITY, AND
MESSAGE DIRECTION
Panel A – First Belief Adjustment
High Goal Low Goal F p
Low Credibility
Increasing Sales 0.27572 0.11249 7.94 0.0059***
High Credibility
Panel B – Second Belief Adjustment

High Goal Low Goal F p
Low Credibility
Increasing Sales -0.02757 0.20778 3.90 0.0514
High Credibility
Increasing Sales 0.26242 0.11253 5.65 0.0195**
* p<0.10; ** p<0.05; *** p<0.01
Reviewing the results of the first budget adjustments, subjects in the High Goal group
placed more weight on low credibility, sales-increasing messages than subjects in the Low Goal
group. A comparison of betas showed a significant difference (p = .0059) between the two
incentive compensation groups (0.27572 for the high goal group versus 0.11249 for the low goal
group). Subjects in the Low Goal group showed evidence of using the low credibility cue to
lower the weighting of a sales-increasing message. This supports hypothesis three.
After the second message, respondents in the High Goal group placed more weight on
high credibility, sales-increasing messages. In this case, a comparison of betas showed a
significant difference (p=0.0195) between the two incentive compensation groups (0.26242 for
Page 229
the high goal group versus 0.11253 for the low goal group). Across both groups, biases entered
the decision process through sales-increasing messages only. Low credibility sources were
weighted more heavily by the “Low Goal” group, while high credibility sources were weighted
more heavily by the “High Goal” group. Decreasing messages from both low and high credibility
sources did not result in different sensitivities between the two groups.
The content analysis presented in Table 4 also supports this phenomenon. Subjects in the
attitude. This was particularly true in the case of high credibility, sales-increasing messages
received by the “High Goal” group. Thus, these findings provide support for hypothesis three.
Table 6 summarizes the overall findings and implications of this study.
Table 6
RESULTS SUMMARY
Panel A: ELM Predictions Findings Implications
Not statistically significant. The While not significant, the

High involvement receiversa
integration of mixed evidence did not direction of the findings are
biased by message direction
produce significant directional bias consistent with prior agency-
between the two incentive compensation based research that finds
groups, but all  and  values were in compensation-based bias
the right direction. (slack-inducing behavior).
Low involvement receivers use Mixed findings. Content analysis Low involvement receivers
source credibility as peripheral supports the credibility heuristic; may use heuristics that lead to
cue quantitative support is weak. sub-optimal decisions.
High involvement receivers Supported. Support is found, but for Incentive compensation sets
(incentive compensation) use increasing messages only. up an initial attitude toward
source credibility as pro- evidence that biases judgment
attitudinal evidence based on source credibility.
Panel B: HE Model Findings Implications
Predictions
Recency order effects Not supported. Weak order effects. Initial attitude may alter the
Some evidence of primacy for increasing order effects predicted by the
messages; recency for decreasing HE model. Further research
messages. needed.
Contrast effect Supported. Second message induces Biases may enter the budget
more belief change when processed after decision process through the
evidence that is opposite in sign. contrast effect.
a
Initial attitude toward evidence is manipulated in high involvement receivers through the incentive
compensation manipulation.
Page 230
DISCUSSION
The results of this study provide several valuable insights into the budgeting process.
The Elaboration Likelihood Model predicts that subjects will elaborate on a message when
receiver involvement is high, but could bias the integration of mixed messages to support the
initial positions favored by the subjects. Incentive compensation formulas were used to establish
initial attitudes toward messages in the high involvement subjects. Pro-attitudinal messages were
expected to receive more weight than counter-attitudinal messages. Experimental results
revealed parameter estimates that were consistently in the right direction, but statistical tests
were not significant. Lack of power may have caused the interaction of initial attitude
(manipulated through compensation scheme) and message direction to go undetected.
In contrast, low involvement receivers were predicted to use heuristics, such as relying on
peripheral cues, when evaluating messages. In this study, source credibility acted as the
peripheral cue. A qualitative analysis of receiver involvement presented evidence of more
involvement than under conditions of high receiver involvement. Subjects provided written
comments concerning their thoughts in arriving at budget decisions. Content analysis was used to
determine the presence or absence of both message content and message source arguments.
Under conditions of low (high) receiver involvement, message source arguments were more
(less) likely to be present in the thought-listing task. This supports the ELM prediction that low
receiver involvement leads to more reliance on peripheral cues such as source credibility.
The ELM predicts that high involvement subjects will use source credibility as evidence
to support their initial attitude. Content analysis also supported this prediction. Subjects in the
attitude. Regression analysis failed to support a statistically significant interaction between
receiver involvement and source credibility.
Under conditions of high receiver involvement, source credibility was hypothesized to
interact with compensation scheme and message direction. In interpreting the results of this
three-way interaction, budget-adjustments were broken down into two categories: the first budget
adjustment (the change in the sales forecast after receiving the first message) and the second
budget adjustment. The three-way interaction term for source credibility, message direction, and
incentive compensation group (high goal versus low goal) was statistically significant when the
differences between means were segregated into two groups.
For first budget adjustments, high goal subjects placed more weight on low credibility,
increasing messages than did low goal subjects. Weights for high credibility sources and sales-
decreasing messages were not significantly different between groups. For second budget
adjustments, high goal subjects placed more weight on increasing messages from high credibility
sources than did low goal subjects. Weights for sales-decreasing messages were not significantly
different between groups. And finally, for all groups and conditions, sales-decreasing messages
were weighted more heavily than sales-increasing messages. These findings suggest a pro-
attitudinal bias toward message sources that supported the initial attitude of the decision-maker.
In both cases, the direction of credibility’s effects was as predicted. The unexpected finding was
that the effect only occurred for sales-increasing messages.
A possible interpretation of this finding is that sales-decreasing messages – messages that
suggest a downward revision of a sales forecast – are less subject to bias because of an overall
conservative bias in the budgeting process. Decision-makers might bring skepticism to the
Page 231
judgment and decision-making process because the initial forecast is the result of an aggregation
of sales managers’ opinions. The uncertainty concerning the starting point might lead to a greater
saliency of decreasing messages. Decreasing messages might seem less refutable. Any biases,
therefore, could tend to enter the decision process through sales-increasing messages where the
decision-maker is more discriminating about the interpretation of the message content and
message source. If this scenario holds in real-world budgeting contexts, a sub-optimal sales
forecast could result. This bias and the resulting implications for firm profitability need further
investigation.
In summary, the predictions of the ELM as tested in this study were generally supported.
The consistency of the results with the theory suggests that insignificant test results may be due
to lack of power. Differences in parameter estimates among cells consistently had the correct
sign. A larger number of subjects might be needed to provide sufficient power to adequately test
all the hypotheses under consideration in this study.
Limitations
This study had several limitations related to the nature of the laboratory experiment
methodology. The first limitation to generalizability was the use of MBA students as subjects.
External validity is weakened to the extent that subjects do not react similarly to a more
experienced population of decision-makers. However, past budgeting research (Young, 1985;
Chow et al. 1991; Fisher et al. 1996, 2000) has consistently used student subjects to study
people’s reactions to different contexts. The use of MBA students helped control for individual
(i.e., age and gender) and organizational variables (i.e., industry, corporate culture, and
organizational structure) that may not be consistent across firms. The use of MBA students might
also mitigate some of the problems associated with using undergraduate students, but would not
overcome all external validity problems.
A second limitation related to the use of MBA students is that the consequences of biased
decision-making are much greater for practitioners. Setting a budget too high or too low results
in consequences that are difficult to duplicate in an experimental setting. The subjects in this
experiment did not face the magnitude of effects from over- or under-budgeting sales for the
upcoming year. Because the sales forecast is used to plan production for the next year, mis-
specified sales forecasts can lead to lost sales from under-production or surplus inventory from
over-production. Both impact company profits, and can lead to reputation effects for the
decision-maker in a multiple-period setting. This experiment was single-period, and did not
consider reputation effects.
This study was also limited in its scope because the experiment considered just one
budget variable – the sales forecast. The sales forecast is the starting point in developing an
annual operating budget, and is therefore a very critical piece to consider. However, it is unclear
whether a revenue decision varies from an expense decision. Because this study only tested a
revenue decision, the findings might not be relevant in an expense decision context. The biases
and cognitive functions associated with an expense versus a revenue context can be different. It
is also possible that an overriding profitability (revenues less expenses) parameter might provide
additional insights. Finally, validity was also threatened to the extent that the parameters of the
subjects’ compensation schemes are not reflective of real world conditions.
Page 232
CONCLUSION
This study remains an important addition to the research in management communication

because of its potential to shed light on a key communication function within an organization —
managing the budgeting process. This study expands the prior literature in several important
ways. First, it proposes the examination of factors affecting the decision-making process of
managers within the budgeting context. Previous studies have examined antecedent and
consequent factors related to participative budgeting, while ignoring the process (this literature
stream is summarized in Kren, 1997). More recent studies have viewed the budgeting process as
a negotiation between subordinate and supervisor (Fisher, Frederickson, & Peffer, 2000), but
have failed to consider any other communication variables that are inherently part of such a
process (as discussed in McVay, Sauers & Clark, 2008). This study incorporates two
communication factors that could bias the budget judgment and decision-making process.
In addition, this research extends current research using the Elaboration Likelihood
model by 1) focusing on empirical data in 2) an applied context other than advertising or political
campaigns. Earlier studies have been much more theoretically-oriented in identifying cognitive
processes related to information processing and persuasion (see discussion in Bloemer, Brijs, &
Kasper, 2009; also Stephenson, Benoit, & Tschida, 2001). Also, much of the research conducted
using the Elaboration Likelihood Model (ELM) of persuasion has focused on a small number of
message topics. ELM research has repeatedly employed the topic of senior comprehensive
exams (for a review of the studies that have examined this and other issues using the ELM, see
O’Keefe, 1998, 2009; Petty & Wegener, 1998, 1999.)
According to O’Keefe, there are approximately twice as many studies on two particular
topics (senior comprehensive exams and tuition changes) than on all other topics combined. By
examining persuasive arguments in another domain, the generalizability of findings from prior
studies is enhanced. As O’Keefe (1990) noted, “there is much to be gained by diversifying the
ELM’s evidentiary base” (p. 110). To date, there has been little experimental work that utilizes
the ELM in a professional decision-making context (Petty & Wegener, 1998); another limitation
of previous studies in this area, according to O’Keefe (1998), is a “failure to specify the
properties that make specific arguments relatively more or less persuasive” (p. 73)
The results of this study suggest a number of opportunities for future research. The
findings related to source credibility explore the impact that message source has on the
integration of information and belief updating. Further studies should examine the relative
importance of the sub-components of source credibility. The weighting of the various attributes
of source credibility might differ, providing insight into those factors that are most influential in
biasing the judgment of a decision-maker.
Source credibility factors might also be moderated due to the relationship of the source to
the firm – an external versus an internal source of information. The consistency of evidence from
different sources within or external to the firm could be examined to determine differences in the
persuasiveness of a particular message. The interaction of evidence consistency, initial attitude of
the decision maker, and source credibility warrants further exploration of these complex
relationships.
Page 233
END NOTES
1
A power calculation was performed to determine the number of subjects needed to provide a power level of .80,
with alpha set at .05. Because effect size is not known (although a meta-analysis of receiver involvement conducted
by Johnson & Eagley [1989] suggests an effect size of approximately .20), number of subjects (n*) was calculated at
three different effect (f2) sizes. When f2 = .10, n* = 192; when f2 = .15, n* = 133; when f2 = .20, n* = 103. (Cohen &
Cohen, 1983).
2
A kappa exceeding 0.75 represents excellent agreement (Fleiss 1981; Fleiss and Cohen 1973).
REFERENCES
Bamber, E. M., R. J. Ramsay, and R. M. Tubbs. (1997). An examination of the descriptive validity of the belief-
adjustment model and alternative attitudes to evidence in accounting. Accounting, Organizations, and Society
22 (3/4), 249-268.
Bloemer, J., Brijs, K., & Kasper, H. (2009). The CoO-ELM model: A theoretical framework for the cognitive
processes underlying country of origin-effects. European Journal of Marketing, 43 (1/2), 62-89.
Cacioppo, J. T., R. E. Petty, and C. F. Kao. (1984). The efficient assessment of need for cognition. Journal of
Personality Assessment 48 (3), 306-308.
Cacioppo, J. T., R. E. Petty, C. F. Kao, and R. Rodriguez. (1986). Central and peripheral routes to persuasion: An
individual difference perspective. Journal of Personality and Social Psychology 51, 1032-1043.
Cacioppo, J. T., R. E. Petty, and K. J. Morris. (1983). Effects of need for cognition on message evaluation, recall,
and persuasion. Journal of Personality and Social Psychology 45, 805-818.
Chow, C. W. (1983). The effects of job standard tightness and compensation scheme on performance: An
exploration of linkages. The Accounting Review (January) 58, 667-685.
Chow, C. W, J. Cooper and K. Haddad. (1991). The effects of pay schemes and ratchets on budgetary slack and
performance: A multiperiod experiment. Accounting, Organizations, and Society 16 (1), 47-60.
Fisher, J., J. Frederickson, and S. Peffer. (2000). Budgeting: An empirical investigation of the effects of negotiation.
The Accounting Review 75 (1), 93-114..
Fisher, J., J. Frederickson, and S. Peffer. (1999b). The effects of information asymmetry and negotiation type on
budgeting. Working paper
Frederickson, J. (1992). Relative performance information: The effects of common uncertainty and contract type of
agent effort. The Accounting Review 7 (4), 647-669.
Hallahan, K. (2008). Need for cognition as motivation to process publicity and advertising. Journal of Promotion
Management 14 (3/4), 169-194.
Hogarth, R. M. (1993). Accounting for decision and decisions for accounting. Accounting, Organizations, and
Society, 407-424.
Horngren, C. Cost Accounting. A Mangerial Emphasis, Third Edition, Prentice Hall, 1973.
Hornikx, J., & O’Keefe, D. J. (2009). Adapting consumer advertising appeals to cultural values: A meta-analytic
review of effects on persuasiveness and ad liking. Communication Yearbook 33, 39-71.
Libby, R. & Lipe, M. G. (1992). Incentive effects and the cognitive processes involved in accounting-related
judgments. Manuscript, Cornell University.
Liu, X. (2008). Moderation of media issue salience: Retesting the agenda setting effect within the elaboration
likelihood model. Conference Papers -- International Communication Association 1-30. Communication &
Mass Media Complete, EBSCOhost (accessed July 16, 2010).
MacGeorge, E. (2008). Message and source factors as influences on advice evaluation and outcomes. Conference
Papers – International Communication Association, 1-37. Communication & Mass Media Complete,
EBSCOhost (accessed July 17, 2010).
McComas, K. (2008). Information processing, trust, and perception of environmental cancer risk. Conference Papers
– International Communication Association, 1-25. Communication & Mass Media Complete, EBSCOhost
(accessed July 17, 2010).
McVay, G., Sauers, D., & Clark, M. (2008). Belief adjustment in the budgeting process: Examining the contextual
validity of the belief-adjustment model. Academy of Accounting and Financial Studies Journal 12 (2), 131-
139.
Page 234
Noar, S.M., Palmgreen, P., Zimmerman, R.S., Lustria, M.L., & Lu, H. (2010). Assessing the relationship between
perceived message sensation value and perceived message effectiveness: Analysis of PSAs from an effective
campaign. Communication Studies 61(1), 21-45.
Ohanian, R. (1990). Construction and validation of a scale to measure celebrity endorsers’ perceived expertise,
trustworthiness, and attractiveness. Journal of Advertising 19(3), 39-52.
O’Keefe, D. (1990). Persuasion theory and research. Newbury Park, CA: Sage Publications.
O’Keefe, D. (1998). Justification explicitness and persuasive effect: A meta-analytic review of the effects of varying
support articulation in persuasive messages. Argumentation and Advocacy 35(2), 61-76.
O’Keefe, D. J. (2009). Theories of persuasion. In R. Nabi & M. B. Oliver (Eds.), Handbook of media effects: 269-
282. Thousand Oaks, CA: Sage.
Petty, R. E., and J. T. Cacioppo. (1986a). Communication and persuasion: Central and peripheral routes to attitude
change. New York: Springer-Verlag.
Petty, R. E., and J. T. Cacioppo. (1986b). The elaboration likelihood model of persuasion. Advances in Experimental
Psychology 19, 123-205.
Petty, R. E., and J. T. Cacioppo. (1990). Involvement and persuasion: Tradition versus integration. Psychological
Bulletin 107 (3), 367-374.
Petty, R. E., and & Wegener. (1998). Attitude change: Multiple roles for persuasion variables. In D. Gilbert, S.
Fiske, and G. Lindzey (Eds.), The Handbook of Social Psychology (Vol. II): 323-370. New York: Oxford
University Press.
Petty, R. E., and & Wegener. (1999). The Elaboration Likelihood model: Current status and controversies. In S.
Chaiken and Y. Trope (Eds.), Dual-process theories in social psychology: 41-72. New York: Guilford Press.
Petty, R. E., Cacioppo, J. T., Strathman, A. J., & Priester, J. R. (2005). To think or not to think: Exploring two routes
to persuasion. In T. C. Brock & M. C. Green (Eds.). Persuasion: Psychological insights and perspectives
(2nd ed.): 81-116. Thousand Oaks, CA: Sage.
Shields, M. D., and W. S. Waller. (1988). A behavioral study of accounting variables in performance-incentive
contracts. Accounting, Organizations, and Society: 13 (6), 581-594.
Stephenson, M.T., Benoit, W.L., & Tschida, D. A. (2001). Testing the mediating role of cognitive responses in the
elaboration likelihood model. Communication Studies 52 (4), 324-337.
Tubbs, R. M., W. F. Messier, and W. R. Knechel (1990). Recency effects in the auditors’ belief revision process.
The Accounting Review 65 (April), 452-460.
Waller, W. S. (1988). Slack in participative budgeting: The joint effect of a truth-inducing pay scheme and risk
preferences. Accounting, Organizations, and Society 1, 87-98.
Waller, W. S., and R. A. Bishop. (1990). An experimental study of incentive pay schemes, communication, and
intrafirm resource allocation. The Accounting Review 65 (April), 812-836.
Waller, W. S., and C. W. Chow. (1985). The self-selection and effort effects of standard-based employment
contracts: A framework and some empirical evidence. The Accounting Review 60 (July), 458-476.
Young, M. S. (1985). Participative budgeting: the effects of risk aversion and asymmetric information on budgetary
slack. Journal of Accounting Research (Autumn), 829-842.
Young, M. S., J. Fisher and T. M. Lindquist. (1993). The effects of intergroup competition and intragroup
cooperation on slack and output in a manufacturing setting. The Accounting Review 68 (July), 466-481.
Page 235
OWNERSHIP STRUCTURE AND INFORMATION

DISCLOSURE: AN APPROACH AT FIRM LEVEL IN
VIETNAM
Pham T. B. Ngoc, Hoa Sen University
Quach M. Hung, University of Economics HCMC
ABSTRACT
This paper aims at investigating the impact of firm ownership concentration on firm-
specific information disclosure for the case of an emerging country. By analyzing an unbalanced
panel data including 195 Vietnamese firms on the Hochiminh stock market from 2006 to 2011
since Vietnam strongly opened its economy by completing the ASEAN free trade area in 2006
and joining the WTO in 2007, we find that the largest owner tends to disclose more information
to the market if his asset, tied with the stock price, is large, fearing the market could punish his
bad behavior by discounting the stock price. Moreover, as the Vietnamese stock market operates
in a market oriented economy originating from a central planned economy, we further
investigate origins of the largest owners, private vs. state ownership. Accordingly, the impact of
ownership structure on information disclosure is only engaged with wholly private firms but not
significant when the largest owner is related to the state.
Keywords: ownership concentration, state joint stock firms, information disclosure.
JEL Classification: G14, G32, G34
INTRODUCTION
The Hochiminh stock exchange (HOSE), which occupies 85% of the Vietnamese stock
market capitalization, has developed to have 302 listed stocks, 2 funds and 28 bonds until 2013
since the first trading day at 28/07/2000 with only two listed stocks. This fresh market accounts
27% of the GDP with the capitalizing value about USD40 billion for the year 2013. After a
period of fast development, the VN Index as denoted the HOSE market index has confronted a
decreasing trend from the range 950-1000 in the years 2007-2008 to be sustainably above 500 in
recent years, vis-à-vis the world crisis. This fact possibly reflects a stagnancy in the Vietnamese
economy since the stock market is always considered as one of the best ways to forecast the
economic performance as it shows much economic information through its stock pricing system
according to Roll (1988). A research question is posed that how to improve the informativeness
environment focusing on corporate information disclosure practices in order to promote the
investment climate attractive enough to both foreign and domestic investors.
Corporate disclosure is considered as the communication between the firm managers,
controlling owners and the outside investors about firms’ performance, financial situation,
potential development or even risks such as cost of capital, stock performance and bid-ask
spreads via regulated financial, annual reports, press release or firm online news (Healy and
Palepu, 2001). According to Roll (1988), not only macro events and industry movements but
also firms’ specific information are captured in stock price fluctuations. However, the reflection
Page 236
of those kinds of information into the stock price variation is not always good because of the
corporate disclosure problem. A bad disclosure can block the information to be captured in the
stock price and create the asymmetric information between the insiders and uninformed
investors. Due to this market failure, the cost for acquiring information becomes higher for the
outside investors (Jiang, et al. 2011). The more severe the situation, the less attractive an
investment environment. Thus, a fund attraction progress for the capital market in order to lower
the domestic cost of capital could be in vain if the corporate disclosure problem is not resolved
(Jiang, et al. 2011; Lawrence, 2013).
In this study, the impact of the ownership structure on disclosure activities is the main
research objective. The first argument behind this suggestion is based on the agency problem
mentioned by Jensen and Meckling (1976) that controlling shareholders would hide corporate
information from outside investors for their private benefits, such as covering their corruption,
private contracts or inside trading. Perhaps, ownership concentration especially that associated
with state ownership in Vietnam doesn’t favor the information disclosure system. Hence, the
ownership level would have a negative relationship with the corporate disclosure level. However,
there could be another argument developed by Healy and Palepu (2001) that a high holding rate
of a firm stock is a strong commitment for the owner’s honesty. Investors could easily punish
his bad behavior by discounting his asset easily and directly on the stock market. Thus, the effect
of ownership concentration on the corporate disclosure is important but with mixed results.
To the best of our knowledge, this paper is the first one to shed a light on the
transparency of Vietnam stock market by analyzing the firm-specific return variation and its
relationship with ownership concentration. By using the OLS, fixed effects, and random effects
estimations together with robustness checks for a panel data including 195 listed firms on the
Hochiminh Stock Exchange from 2006 to 2011, we find a robustly positive impact of ownership
concentration on information disclosure in the context that firms try their effort to send
transparent signals to investors. Interestingly, this affect is only significant for wholly private
firms, but not for the state-related firms. There could be the fact that state-related firms want to
protect information as state ownership plays a major role in transition economies, from former
central planned economy, for the case of Vietnam.
The following section is literature review. The third and fourth sections present research
methodology and estimation strategy, and data description. Next, the fifth section provides the
empirical results and analysis. Finally, we come up some conclusions and policy implications in
the final section.
LITERATURE REVIEW
According to Morck, Yeung, and Yu (2000), low level of informativeness is a common

phenomenon in emerging financial markets resulting from many serious corporate and institution
structure problems. According to Ball (2001) and Chan and Hameed (2006), the bad enforcement
of disclosure regulations in emerging markets makes the insiders and managers have less
incentive to release more information for the outside investors. In addition, corporate structure,
for instance ownership concentration is considered as a main cause for the problem in emerging
economies. Ownership concentration is the situation when one or a few shareholders hold a large
percentage of stock of a firm. When the stock holding is equivalent to the voting right, the result
is the power concentration (Thomsen and Pedersen, 2000). In developing countries, the
ownership usually concentrates in hands of the entrepreneurs and their relatives. Meanwhile, in
Page 237
the case of transition economies, the largest shareholders are related to the state. A high
ownership concentration rate certainly has a large influence on firms’ disclosure policy.
There could be a controversial impact of ownership structure on the disclosure activities.
On the one hand, Jensen and Meckling (1976) state that the controlling shareholders could access
to the inside information sooner than the outside investors. Moreover, Fan and Wong (2005),
Kim and Yi (2006) and Gul, Kim and Qiu (2010) argue that the controlling owners, by exploiting
their concentrated power, have incentives to hide those information or disclose them in a
selective way in order to benefit their self-serving activities, such as hiding the private contracts,
corruption or inside trading. Hence, in this case, the ownership concentration would have a
negative relationship with the corporate disclosure level.
On the other hand, Healy and Palepu (2001) argue that the high holding rate of a firm is a
strong commitment for the owner honesty. As a result, investors could easily punish his bad
behavior by discounting his asset easily and directly on the stock market. That certainly could
encourage the owners to release firm information. Jalila and Devi (2012) argue that strict and
close supervision of large owners forces the manager working for the common good of the
shareholders. They also note that large investors such as institutions, international investors and
professional funds tend to make the manager to disclose the information better.
Jalila and Devi (2012) summarize these two opposite impacts through channels of the
entrenchment effect and the alignment effect. The former effect derives from the conflict
between larger owners who could be founders or managers at the same time and minority outside
investors. The owners with the inside information behave like a predators to minority
shareholders. Meanwhile, the latter shows the conflict between owners and corporate managers
that the effect can rise when the manager acts for his own benefit but not for the owners. In this
case, the large owner will behave in the common interest with the shareholders by supervising
closely the moral hazard activities of the managers.
Previous empirical studies provide ambiguous results. In Jiang, et al. (2011), the stock
price reflects the asymmetric information, which comes from the signal of the bid-ask spread,
depending on ownership concentration for the case of New Zealand. This study is applied for
103 firms with 390 firm-year observations in order to show that the ownership concentration
rises the asymmetric information. Moreover, Chau and Gray (2010), by using a voluntary
disclosure index, find a positive relationship between the family ownership and voluntary
disclosure in Hong Kong. However, Gul, Kim and Qiu (2010) show the negative impact of
ownership on corporate disclosure for the China with 1,142 firms from 1996 to 2003. Their
results support the entrenchment effect with which the largest shareholders has incentive to hide
the information from outside investors. Meanwhile, Jung & Kwon (2002) analyzes the return of
the informativeness earning per share with the ownership concentration, institution and block
holding in Korea. Their study provides evidence that ownership concentration, institution and
blockholdings supply more information to the stock return but not for the case of the cheabol
group. Due to its complicated ownership and activities, cheabol managers and owners don’t want
to disclose their secret. The same situation happens to the state joint stock firms that the
government ownership does not favor the information disclosure according to Gul, Kim and Qiu
(2010).
Page 238
RESEARCH MODEL AND METHODOLOGY
In this research, we apply the approach of Fernandes and Ferreira (2008) and Gul, Kim
and Qiu (2010) which were developed from the foundation of Roll (1988). Accordingly, two
main steps will be conducted as follows:
Step 1: Measuring corporate disclosure
According to Roll (1988), the stock price fluctuation of a firm is composed by 3

components: systematic economic influences, industrial factors and firm specific information.
An investor’s paying decision, which is revealed in stock variation, will be based on the macro
information and industrial factors and firm specific information. Thus, after regressing stock
variation to market and industry returns, which represent the public information, the unexplained
part of the stock price variation is identified as the firm specific information. Then we can use it
as a proxy to capture the contribution level of disclosing information.
Where:
: daily return of stock i at the end of day t in a trading year. Being different from
Gul, Kim and Qiu (2010), Fernandes and Ferreira (2008) and Morck, Yeung and Yu (2000)
which use the trading year from 1st January to 31st December, we apply a trading year from 1st
April to 31st March for the reason to avoid the overlap counting problem of the new Board of
Directors (BoD) and Board of Management (BoM), who are elected in the shareholders annual
meeting at the end of the first year quarter, and the old BoD and BoM.
: market return variation (VNIndex) at the end of day t in a trading year.
: industry return variation at the end of day t in a trading year. This indicator is
constructed by making a stock index of all the firms in the same industry, excluding the
estimated firm. The industrial classification is taken from the HOSE website (hsx.vn) and a
chosen industry is required to have at least 4 firms and in a trading year. The index formula is as
follow:
∑
INDRETit =
∑
where n is the number of firm in the industry. The SIZE is identified as the total share
listed of each firm at the beginning of the year.
The lagged components of MKRET and INDRET are included in the equation (1) for
capturing the impact of the information from the last trading day on the actual stock trading
variation.
2
After running the regression for every firm, we got the value for each firm-year
regression which captures the percentage of market and industry return variation over the total
Page 239
return variation (where i: identifies firm i, and y: represents the estimated year y). The value of
2
specific firm information captured in the stock price of firm i in year y is: 1 -
2
Because the 1 - value only varies between [0,1], a transformation is required in order
to convert it to a real number. Hence, we acquire the measurement variable for the corporate
disclosure or the informativeness of the firm i in year y:
Step 2: Measuring the impact of ownership concentration on corporate disclosure
∑ ሺʹሻ
where:
TOPHOLDiy : is the percentage of share of the largest holder of firm i at the beginning
of the trading year y. In addition, the share of the largest holder is the ownership combination of
family members, related people or company. Whereas, if the shareholder is government related,
even if the ownership is discarded into several persons, institutions or public firms, the
ownership will be also combined and counted for the state. The data is collected manually from
the annual report of each firm which is released at the first quarter of the year.
CONTROL variables
VOLiy: daily average trading volume of firm i at year y. Chan and Hameed (2006)
suggest that a high trading volume could mean the high speed of price adjustment which helps it
keeping up with the market movement. Hence, it is believed that the more highly traded stock,
the more its price is synchronized with the market return. Thus, it is expected to have a negative
relationship with the corporate disclosure. However, Gul, Kim and Qiu (2010) found an opposite
effect as high trading volume could be an effective signal for liquidity when the firm specific
information capitalized in the stock price. The difference in their results could be the
consequence of using different proxy for measuring the trading volume.
SIZEiy: denoted by the total shares of firm i at the beginning of the year. According to
Chan and Hameed (2006), Gul, Kim and Qiu (2010) the SIZE control variable is used in order to
capture the influence of the large firms on the market and industrial index which is formed by
weighting each stock variation by its size. Hence, the SIZE is expected to have a negative
relation with the dependent variable due to its positive theoretical correlation with the market and
industrial index.
INDSIZEiy: the total SIZE of all firms in the same industry of firm i. When the industry
is too large or too small, the interaction of the firm stock variation with the market index or the
industry index also varies. Hence, this industry variable is included in the model to control the
industrial level variations.
Page 240
An extension
The equation (2) will be applied for wholly private joint stock firms and state- related
joint stock firms separately in order to estimate the role of private vs. state ownership structure
on the information disclosure. Gul, Kim and Qiu (2010) use only a dummy variable to capture
the impact of government ownership on the corporate disclosure level. We consider this method
is not sophisticated enough for emphasizing that effect. Hence, in this paper we divide the
sample in two groups: the state-related joint stock firms, which the largest shareholder is related
to the state, and the private joint stock firms.
Robustness check
We are checking if the impact of ownership concentration on information disclosure is

still consistent when applying different measurements for this regressor.
Firstly, a simpler additional ownership index HSUM is suggested. HSUM is only the
sum of top 3 largest shareholders‘ ownership.
TOPHOLDkiy is the percentage of share which is hold by the top three shareholders of
firm i at the beginning of the year y respectively.
Next, according to Jiang, et al (2011), a different proxy for measuring the ownership
concentration is applied in order to put more weight on the larger shareholding:
where:
Hiy is the ownership concentration index of firm i at the begining of year y.
Estimation methods
Fernandes and Ferreira (2008) and Gul, Kim and Qiu (2010) employ the OLS estimation
and follow the Petersen (2009) by using robust standard errors corrected for analyzing financial
panel firm-level data. This method is crucial for correcting the potential serial dependency
problems. However, Petersen (2009) also suggests that while analyzing panel data with potential
fixed effects, the fixed effects (FE) or the random effects (RE) estimations will be more efficient.
F-test and LM-test will be applied to examine the appropriateness of the FE and RE
model correspondingly. If both models are qualified, they will be tested by the Hausman test for
choosing the most appropriated model. In addition, time dummies will be included to control the
potential time effects. Industry dummies are employed in the OLS and RE model for controlling
the industrial effect. Moreover, the clustered standard errors are used for correcting the
heteroskedasticity and within panel serial correlation problem. VIF test for multicollinearity is
also applied for finding the correlation level among independent variables.
Page 241
DATA DESCRIPTION
Data sample
The data covers 729 firm-year observations of 195 listed firms on the Hochiminh Stock
Exchange from 2006 to 2011. The stock market return with the trading volume is collected from
cophieu68.com website. Meanwhile, the data about the ownership, firm size is gathered
manually from annual and financial report taken from the firm website and the official website of
Hochiminh Stock Exchange (hsx.vn). Every firm is classified following the industry category of
HOSE and must have at least 200 trading days per year. Financial firms and outliers including
firms with more than 100 million stocks or firms with only one year data are excluded.
Summary Statistics
As can be seen from Table 1, the value R2, which represents the level of the
capitalization of market and industrial information in the stock price variation, has the mean
0.3045, which is much lower than the mean 0.454 for the case of China in Gul, Kim and Qiu
(2010), but much higher than the value 0.193 for the US in Piotroski and Roulstone (2004).
The amount 1 – R2 indicating firm specific information level captured in the stock price
is 0.6955. As compared with the result in Fernandes and Ferreira (2008), Vietnamese corporate
disclosure level belong to the bottom third of 24 emerging markets. However, among the
developed markets, Vietnamese disclosure level can only outperform slightly the case of Greece.
After transforming it by taking log, we get the corporate disclosure measurement INFO.
Table
Table 11
SUMMARY
SummarySTATISTICS
Statistics
Number of Standard
Variables Mean Min Max
observation Deviation
Dependent variable
1 - R2 729 0.6967 0.1699 0.0257 0.7349
INFO 729 1.0283 1.0032 -1.0196 3.6352
Independent variables
Ownership concentration
TOPHOLD 723 0.3681 0.1918 0.0500 0.8370
Control variables
VOL 729 91,376 142,699 478 1,174,396
SIZE 729 21,000,000 16,900,000 1,138,501 82,700,000
INDSIZE 729 808,000,000 639,000,000 3,000,000 1,970,000,000
Also from Table 1, the first main independent variable TOPHOLD which denotes
ownership concentration presents that the largest shareholder holds 36.81% of the stock on the
average. However, this value varies a lot from 5% to 83.7%. Although the mean value is lower
than that of 42.8% for the case of China (Gul, Kim and Qiu, 2010), it is very high compared to
the case of the US. Jiang et al (2011) report that 20 top shareholders in US hold only 37.66% of
the total amount of shares.
Meanwhile, Figure 1 shows that the disclosure situation of Vietnamese listed firms is
positively and significantly enhancing through time. A simple explanation for this phenomenon
Page 242
could be that the Vietnamese policy makers and regulators have success in making a good
environment for better disclosure activities. It could be the result of strict law enforcement or
more experienced market supervisors and investors. However, it is only the first intuitive
observation. The detailed results will only be acquired after finishing the regressing phase.
Figure 1: Informativeness level by year
A positive relationship between corporate disclosure and ownership concentration could

be seen in Figure 2. Although many previous researches only concentrate on the linear trend,
Gul, Kim and Qiu (2010) suggest a concave relation. In our study, we do not analyze the
quadratic relationship.
When taking ownership structure into account, the number of state-related firms is 114
among 195 firms as a whole. The state controls on average 39.47% of the shares of these firms.
However, there is a strong firm concentration around the level 50% as many Vietnamese state-
related firms protect the state’s power by keeping the voting share more or less than 51% in the
privatization progress.
Page 243
Figure 2: The relationship between corporate disclosure and ownership concentration of

state joint stock firms
EMPIRICAL RESULTS AND ANALYSIS
The relationship between ownership concentration and corporate disclosure
The OLS, FE, and RE methods are applied for the equation (2). Accordingly, the
estimation results for all firms, state-related and wholly private firms are introduced in Table 2.
The F-test, LM test and Hausman test have been performed for each group. All groups had
passed the F-test and LM test with the significance level at 1%. Therefore, the OLS model is
proven to be less appropriated than the FE and RE model and would not be applied. Moreover,
the Hausman test indicated that the FE method is more appropriate for the group of all firms and
state- firms at 5% significance level. Meanwhile, the RE method is suitable for the group of
private joint stock. The selected results will be displayed in bold.
Column 1 provides the positive coefficient 1.06 which is significant at 5% significance
level. As predicted in the descriptive statistic section, the slope is pretty small. We find the
results that the more ownership is concentrated, the better corporate disclosure is in the
Vietnam’s stock market. In other words, the more stock the largest shareholder holds, the more
he discloses firm specific information to the public. Then, the information could be well
capitalized in the stock price variation. However, the result is opposite with the negative result in
Gul, Kim and Qiu (2010) so we need a further robustness check.
Column 3 shows insignificant coefficients for the state-related firms which is in line with
the finding of Jung and Kwon (2002) about Korean cheabol firms. Their explanation is that the
positive effect of the ownership concentration is offset by the entrenchment effect in the cheabol
firms. Their largest shareholders, with the complicated system of possession, have large
incentive to hide their activities or to exploit the minor investors. However, the state joint stock
firms in Vietnam may have not incentive to exploit the minor shareholders. Their low level of
disclosure could be explained by inefficient governance which reveals their passive role in the
Page 244
disclosure activities. The finding is also consistent with Gul, Kim and Qiu (2010). In their
research, the included dummy variable for the state related top shareholder is negatively
correlated with the corporate disclosure.
Table 2
Table 2
THE IMPACT OF OWNERSHIP CONCENTRATION ON CORPORATE DISCLOSURE
The Impact of Ownership Concentration on Corporate Disclosure
Dependent variable: INFO
Column 1 2 3 4 5 6
All firms State JS firms Private JS firms
GROUP FE RE FE RE FE RE
1.0570 0.4312 1.5935 0.2171 1.4114 0.9005
TOPHOLD
(2.28)** (2.02)** (0.94) (0.67 ) (2.65)*** (3.17)***
Control variables
-1.39E-08 -1.11E-08 -1.33E-08 -9.22E-09 -7.66E-09 -1.33E-08
SIZE
(-2.25)** (-3.88)*** (-1.48) (-2.26)** (-0.98) (-3.64)***
-4.64E-11 2.12E-11 5.11E-11 1.40E-10 -3.65E-10 -2.77E-10
INDSIZE
(-0.25) (0.12) (0.21) (0.63) (-0.99 ) (-0.89)
-1.46E-06 -2.02E-06 -1.41E-06 -2.21E-06 -1.46E-06 -1.91E-06
VOL
(-4.40)*** (-6.83)*** (-3.08 )*** (-4.15 )*** (-3.75)*** (-6.28 )***
Year dummies
0.0648 0.0432 -0.0610 -0.0415 0.4867 0.3565
2007
(0.59) (-0.42) (-0.46) (-0.32) (2.19)** (2.00)**
0.2104 0.2324 0.0613 0.1147 0.7042 0.6469
2008
(1.51) (-1.79)* (0.35) (0.71) (2.52)** (2.86)***
0.6357 0.6258 0.5262 0.5097 1.1856 1.1058
2009
(3.25)*** (3.47)*** (2.22 )** (2.03)*** (2.82)*** (3.28)***
1.3711 1.3202 1.1140 1.0419 2.1360 2.0562
2010
(5.60)*** (6.01)*** (3.85)*** (3.89)*** (3.94)*** (4.83)***
1.4176 1.3766 1.2941 1.2288 2.0230 1.9826
2011
(5.27)*** (5.69)*** (4.12)*** (4.18)*** (3.32)*** (4.14)***
0.2366 0.3947 -0.0115 0.2619 -0.1681 -0.0369
constant
(1.12) (1.92)* (-0.02) (-1.07) (-0.56 ) (-0.11)
Dummies Industry Industry Industry
F-test (3.58)*** (3.33)*** (3.19)***
LM test (98.5)*** ( 23.22)*** (41.69)***
Hausman test (12.65)** (14.37)** (14.37)** (7.96) (7.96)
Obs 723 723 417 417 306 306

Note: The robust results are with bold columns 1,3 and 6. Industry dummies are used in the RE regressions.
Standard errors are in parentheses. Statistics significance at: 10% level (*), 5% level (**), 1 % level (***).
In contrast, the wholly private joint stock group has significant results in columns 5-6.
The magnitude of the coefficient in the FE model is also higher than the overall group (1.4114 >
1.057) and its level of significance rises as well (2.65 > 2.28). The answer for this finding is the
Page 245
credit of a firm is tied closely to the credit of its founders, owners or managers. In addition,
social capital plays important role in Vietnamese economy. A good credit, commitment of the
firm or the owner could be a collateral for borrowing or issuing new stock. This is also a vital
purpose of each firm for being listed. Hence, maintaining a good reputation is crucial for not
only a firm but also the owner especially in the period of the financial crisis.
Besides, when holding a large share ratio, a bad behavior once discovered could be
catastrophic for the owner. Fan and Wong (2005) mentions that, the outside investors will defend
themselves from the information disadvantage by adjusting the stock price. The more
asymmetric information is, the larger the discount ratio is. Thus, the largest shareholder, while
caring for their asset, will choose to behave properly. This also explains the insignificant result
for state-related firms. Since the managers only own a little amount of share, they don’t have
incentive to disclose much information about the firm activities. The agency problem between
the owners and the managers is more severe in the public firm where the ownership belongs to
the state and the risk of corruption is commonly high.
The year dummies included in every model has shown significant positive increase in
corporate disclosure each year. Furthermore, the significance level also rises year by year. In
short, the corporate disclosure of Vietnamese firms becomes better over time. This is a good
signal for the progress of privatization and transparency of Vietnam.
Robustness check
Robustness check is applied in order to gain more confidence in the main relationship
between the ownership concentration and the corporate disclosure (Table 3). The regression
progress for the equation (2) will be replicated with 2 different new independent variables
HSUM and H with 3 data samples: all firms, state-related and wholly private joint stock firms.
The private group is expected to gain a positive relationship between ownership concentration
and information disclosure and the state-related group should have a neutral impact. FE and RE
methods are used with the F-test’s and LM test’s significance level at 1%. Therefore, the OLS
model is proven to be less appropriated than the FE and RE model and would not be applied. The
Hausman tests show suitable results in Column 2 (state-related firms/FE) and Columns 4 and 6
(all firms and wholly private firms/RE).
As predicted, the relationship of ownership concentration and corporate disclosure is
significantly positive in both case HSUM and H. The results are consistent with all the previous
findings. Ownership concentration contributes significantly and positively to corporate
disclosure in every models and independent variables. Moreover, this effect happens stronger in
the private sector than in the public sector. In the case of state related firms, the effect is offset by
the negative entrenchment incentive and become insignificant.
More specifically, in the case of HSUM independent variable, the coefficient of the
private group is also significantly positive (0.8663). Meanwhile, the coefficients of state joint
stock group are not significant in both models despite its positive values.
In addition, the results from the H proxy test (Table 4) prove that the coefficient of the
private group is much greater than the state group in both fixed effects and random effects
models (see the results in columns 3 and 6 as compared to those in 2 and 5). Furthermore, the
coefficients of the state group are not significant in both models, confirming again that the state
ownership has no influence on corporate disclosure.
Page 246
Table 3
THE IMPACT OF OWNERSHIP CONCENTRATION Table 3 ON CORPORATE DISCLOSURE –
THE CASE
The Impact of Ownership Concentration OF HSUMDisclosure – The case of HSUM.
on Corporate
FIXED EFFECT RANDOM EFFECT
Column 1 2 3 4 5 6
GROUP Total State Private Total State Private
0.6580 1.1071 0.8987 0.4972 0.2432 0.8663
HSUM
( 2.07 )** (1.10) (1.99)** ( 2.44)** (0.77) (2.89)***
Control variables
-1.4E-08 -1.4E-08 -8.7E-09 -1.1E-08 -9.3E-09 -1.3E-08
SIZE
(-2.30 )** (-1.53 ) (-1.08 ) (-3.97)*** (-2.27) (-3.53)***
-4.0E-11 5.6E-11 -3.4E-10 1.3E-11 1.4E-10 -2.7E-10
INDSIZE
(-0.21) (0.23) (-0.90) (0.08) (0.63) (-0.85)
-1.5E-06 -1.4E-06 -1.4E-06 -2.0E-06 -2.2E-06 -1.9E-06
VOL
(-4.40 )*** (-3.09)*** (-3.64)*** (-6.75 )*** (-4.15)*** (-5.85)***
Year dummies
0.065 -0.076 0.456 0.053 -0.041 0.390
2007
-0.580 (-0.57 ) (1.85)* (0.51) (-0.32) (1.98)**
0.209 0.045 0.686 0.238 0.115 0.671
2008
-1.510 -0.260 (2.21)** (1.83)* (0.71) (2.74)***
0.632 0.509 1.161 0.632 0.511 1.120
2009
(3.21)*** (2.17 )** (2.56)** (3.5)*** (2.3 )** (3.16)***
1.368 1.096 2.115 1.334 1.044 2.080
2010
(5.55)*** (3.8)*** (3.65 )*** (6.07)*** (3.9)*** (4.7 )***
1.408 1.277 1.998 1.385 1.230 1.991
2011
(5.20)*** (4.07)*** (3.09)*** (5.71)*** (4.19)*** (4.02)***
0.359 0.196 -0.070 0.343 0.255 -0.144
Constant
(2.01 )** (0.44) (-0.20) (1.65)* (1.05) (-0.37)
F-test (3.54 )*** (3.32 )*** (3.08)***
LM-test (97.76)*** (40.88)*** (38.47)***
Hausman test (10.37) (11.88)* (5.89) (10.37) (11.88)* (5.89)
Obs 723 417 306 723 417 306
Note: The robust results are with bold columns 2, 4 and 6. Standard errors are in parentheses. Statistics significance
at: 10% level (*), 5% level (**), 1 % level (***).
Page 247
Table 4
Table 4
THE IMPACT OF OWNERSHIP
The Impact CONCENTERATION
of Ownership Concentration ONDisclosure
on Corporate CORPORATE DISCLOSURE
– The case of H. –
FIXED EFFECTTHE CASE OF H RANDOM EFFECT
Column 1 2 3 4 5 6
GROUP Total Public Private Total Public Private
1.1145 1.0926 1.2645 0.5590 0.2425 1.0589
H
(2.57)** (0.58 ) (2.81)*** ( 2.15)** (0.62 ) (3.08)***
Control variables
-1.4E-08 -1.4E-08 -8.1E-09 -1.1E-08 -9.2E-09 -1.3E-08
SIZE
(-2.26) ** (-1.53) (-1.04) (-3.89)*** (-2.25)** (-3.55 )***
-3.7E-11 5.7E-11 -3.4E-10 2.3E-11 1.4E-10 -2.5E-10
INDSIZE
(-0.20 ) (0.24) (-0.91) (0.14) (0.64) (-0.82)
-1.5E-06 -1.4E-06 -1.5E-06 -2.0E-06 -2.2E-06 -1.9E-06
VOL
(-4.40)*** (-3.12)*** (-3.71)*** (-6.81 )*** (-4.15)*** (-6.29)***
Year dummies
0.064 -0.077 0.444 0.045 -0.042 0.352
2007
(0.59) ( -0.59) (2.03)** (0.44) (-0.32) (2.00)**
0.210 0.044 0.668 0.234 0.115 0.640
2008
(1.52) (0.26) (2.38)** (1.81)* (0.72) (2.85)***
0.627 0.501 1.133 0.624 0.509 1.081
2009
(3.32)*** (2.14)** ( 2.70 )*** (3.46)*** (2.3 )** (3.25)***
1.362 1.087 2.081 1.319 1.041 2.030
2010
(5.61)*** (3.77 )*** (3.86)*** (6.02)*** (3.88)*** (4.84 )***
1.407 1.268 1.968 1.375 1.228 1.953
2011
(5.27)*** (4.05)*** (3.24)*** (5.69 )*** (4.18)*** (4.13)***
0.427 0.436 0.122 0.450 0.294 0.097
Constant
( 3.11)*** ( 1.07 ) (0.55) (2.31)** (1.23) (0.3)
F-test (3.57 )*** (3.31)*** (3.15 )***
LM-test (98.79)*** (40.67)*** (41.37)***
Hausman test (12.00)* (11.59)* (6.95) (12.00)* (11.59)* (6.95)
Obs 723 417 306 723 417 306
Note: The robust results are with bold columns 1,2 and 6. Standard errors are in parentheses. Statistics significance
at: 10% level (*), 5% level (**), 1 % level (***).
CONCLUSION
In this study, we aim to shed a light into the relationship between the ownership
concentration and corporate disclosure for the whole, and separated private vs. state joint stock
firms by applying a data set of 195 firms listed on the Hochiminh stock exchange from 2006-
2011.
Firstly, the corporate disclosure issue of Vietnamese firms has been carefully analyzed. In
an emerging country like Vietnam, the level of corporate disclosure of Vietnamese firms is pretty
Page 248
low. According to our calculation, Vietnamese corporate disclosure level belong to the bottom
third of 24 emerging markets of the Fernandes and Ferreira (2008) study.
Secondly, we find evidence that ownership concentration favors corporate disclosure in
support for the alignment effect as being mentioned in Jalila and Devi (2012). When
investigating further types of ownership, the positive relationship between ownership
concentration and corporate disclosure is associated with the private sector but not for the private
sector.
Many previous studies suggested that the high ownership concentration in emerging
countries could be related to the bad disclosure performance. For the case of Vietnam which is
originating from a central planned economy to a market oriented economy, ownership structure
of state joint stock firms can be a major obstacle for the disclosure process. This may result from
the fact that state ownership representatives have less incentive to disclose information possibly
due to agency problem. On the contrary, the owner of the private firm care more about disclosure
activities. The explications could be that: (1) the owner fears the lack of disclosure could cause
rumors or doubt among outside investors which could harm their credits; (2) not only the
reputation could be harm, but also the stock price into which the owner asset is tied; (3) more
disclosure will send a signal of a good credit for both the firm and its owner which could be a
great help in gaining for capital or loan.
In sum, it is noticed that a strong trend of enhancing disclosure situation over time is
crucial to express a good signal from the market to call for more investors. Acknowledging the
importance of disclosing information on the financial market and the economy, policy makers
should improve the law enforcement on the listed and even non listed firms to enhance the
information transparency. Better information disclosure regulations can protect the minor
investors. So privatization could be the most effective way to set the situation go straight.
However, the effort could be still in vain if most of firms’ shares remain in hands of ineffective
state agents.
REFERENCES
Ball, R. (2001). Infrastructure requirements for an economically efficient system of public financial reporting and
disclosure. Brookings-Wharton papers on financial services, 2001(1), 127-169. DOI:
10.1353/pfs.2001.0002
Cameron, A. C., & Trivedi, P. K. (2009). Microeconometrics using stata (Vol. 5). College Station, TX: Stata Press.
Chan, K., & Hameed, A. (2006). Stock price synchronicity and analyst coverage in emerging markets. Journal of
Financial Economics, 80(1), 115-147. http://dx.doi.org/10.1016/j.jfineco.2005.03.010
Chau, G., & Gray, S. J. (2010). Family ownership, board independence and voluntary disclosure: Evidence from
Hong Kong. Journal of International Accounting, Auditing and Taxation, 19(2), 93-109.
http://dx.doi.org/10.1016/j.intaccaudtax.2010.07.002
Claessens, S., Djankov, S., Fan, J., & Lang, L. (2002). Disentangling the Incentive and Entrenchment Effects of
Large Shareholdings. Journal of Finance, 57 (6), 2741-2771. DOI: 10.1111/1540-6261.00511
Fan, J. P., & Wong, T. J. (2005). Do external auditors perform a corporate governance role in emerging markets?
Evidence from East Asia. Journal of accounting research, 43(1), 35-72. DOI: 10.1111/j.1475-
679x.2004.00162.x
Fernandes, N., & Ferreira, M. A. (2008). Does international cross-listing improve the information environment.
Journal of Financial Economics, 88(2), 216-244. http://dx.doi.org/10.1016/j.jfineco.2007.06.002
Gomes, A. (2000). Going public without governance: managerial reputation effects. The Journal of Finance, 55(2),
615-646. DOI: 10.1111/0022-1082.00221
Greene, William H. (2003). Econometric analysis, 5th. Ed.. Upper Saddle River, NJ.
Gujarati, Damodar N. (2003). Basic Econometrics. 4th: New York: McGraw-Hill.
Page 249
Gul, F. A., Kim, J. B., & Qiu, A. A. (2010). Ownership concentration, foreign shareholding, audit quality, and stock
price synchronicity: Evidence from China. Journal of Financial Economics, 95(3), 425-442.
http://dx.doi.org/10.1016/j.jfineco.2009.11.005
Healy, P. M., & Palepu, K. G. (2001). Information asymmetry, corporate disclosure, and the capital markets: A
review of the empirical disclosure literature. Journal of accounting and economics, 31(1), 405-440.
http://dx.doi.org/10.1016/S0165-4101(01)00018-0
Jalila, S., & Devi, J. (2012). Ownership Structure Effect on the Extent of Segment Disclosure: Evidence from
Malaysia. Procedia Economics and Finance, 2, 247-256. http://dx.doi.org/10.1016/S2212-5671(12)00085-
8
Jensen, M. C., & Meckling, W. H. (1976). Theory of the firm: Managerial behavior, agency costs and ownership
structure. Journal of financial economics, 3(4), 305-360. http://dx.doi.org/10.1016/0304-405X(76)90026-X
Jensen, M. (1986). Agency Costs of Free Cash Flow, Corporate Finance, and Takeover. American Economic
Association Papers and Proceedings, 76, 323-29
Jiang, H., Habib, A., & Hu, B. (2011). Ownership concentration, voluntary disclosures and information asymmetry
in New Zealand. The British Accounting Review, 43(1), 39-53. http://dx.doi.org/10.1016/j.bar.2010.10.005
Jung, K., & Kwon, S. Y. (2002). Ownership structure and earnings informativeness: Evidence from Korea. The
International Journal of Accounting, 37(3), 301-325. http://dx.doi.org/10.1016/S0020-7063(02)00173-5
Kim, J. B., & Yi, C. H. (2006). Ownership structure, business group affiliation, listing status, and earnings
management: Evidence from korea*. Contemporary Accounting Research, 23(2), 427-464. DOI:
10.1506/7T5B-72FV-MHJV-E697
Lawrence, A. (2013). Individual investors and financial disclosure. Journal of Accounting and Economics.
http://dx.doi.org/10.1016/j.jacceco.2013.05.001
Morck, R., Yeung, B., & Yu, W. (2000). The information content of stock markets: why do emerging markets have
synchronous stock price movements?. Journal of financial economics, 58(1), 215-260.
http://dx.doi.org/10.1016/S0304-405X(00)00071-4
Ngo, M. H., & Rieger, M. O. (2013). Ambiguity Aversion and the Equity Premium Puzzle in Stock Markets
Worldwide.
Petersen, M. A. (2009). Estimating standard errors in finance panel data sets: Comparing approaches. Review of
financial studies, 22(1), 435-480. doi: 10.1093/rfs/hhn053
Piotroski, J. D., & Roulstone, D. T. (2004). The influence of analysts, institutional investors, and insiders on the
incorporation of market, industry, and firm-specific information into stock prices. The Accounting Review,
79(4), 1119-1151. http://dx.doi.org/10.2308/accr.2004.79.4.1119
Roll, R. (1988). R2. The Journal of Finance, 25, 545-566.
Thomsen, S. and Pedersen, T. (2000). Ownership structure and economic performance in the largest European
companies. Strategic Management Journal, 21 (6), 689-705 DOI: 10.1002/(SICI)1097-
0266(200006)21:6<689::AID-SMJ115>3.0.CO;2-Y
Page 250
APPENDIX
Table 5
SAMPLE
Table DISTRIBUITION BY INDUSTRY
5 - Sample distribution by industry
No Industry Number of firms Number of observations
1 Agriculture and forestry 7 28
2 Construction 24 76
3 Information and technology 4 12
4 Manufacturing 80 323
5 Mining 8 28
6 Power and gas 9 34
7 Real estates 20 62
8 Trade 28 111
9 Transporting 15 55
Total 195 729
Table 6
EMPIRICAL RESEARCHES ABOUT OWNERSHIP CONCENTRATION AND CORPORATE
Table 6 - Empirical Researches aboutDISCLOSURE
Ownership Concentration and Corporate Disclosure
Ownership
Period & Disclosure
Authors Methodology concentration Results
Sample measurement
measurement
1996 - 2003
Gul, Kim
1142 firms Multivariate Stock Percentage of Negative relationship
and Qiu
6129 obs regression Synchronization topshareholder with concave function
(2010)
from China
2002 Family ownership
Chau & Voluntary Percentage of
273 firms OLS disclose more
Gray (2010) disclosure index family member
Hong Kong information
Dummy variable Positive relationship
Jung and
Return Equal 1 if between ownership
Kwon Korea OLS
informativeness ownership above concentration and return
(2002)
mean value. informativeness
103 firms
Panel corrected Ownership
Jiang et al 390 firm-year
standard error Bid - Ask spread Ownership index concentration blocks the
(2011) obs
(PCSE) information
New Zealand
Page 251
Table 7
THE IMPACT
Table 7 -TheOF OWNERSHIP
Impact CONCENTRATION
of Ownership ON CORPORATE
Concentration on Corporate Disclosure –DISCLOSURE
OLS results. –
OLS
OLS RESULTS
RESULTS
Column 1 2 3 4 5 6
INFO wINFO
GROUP Total State Private Total State Private
0.2511 -0.0055 0.7634
TOPHOLD
(1.21) (-0.02) (2.55)**
0.3279875 0.0904937 0.886635
wTOPHOLD
(1.3) (0.31) (2.42)**
Control variables
-7.19E-09 -3.05E-09 -1.06E-08 4.8E-05 7.2E-05 1.8E-05
SIZE
(-2.88)*** (-0.87 ) (-2.98)*** (2.21)** (2.74)*** (0.58)
3.71E-11 1.66E-10 -1.85E-10 -7.0E-08 1.7E-07 -1.0E-06
INDSIZE
(0.22) (0.72) (-0.65) (-0.09) (0.16) (-0.76)
-2.56E-06 -3.05E-06 -2.28E-06 -1.4E-02 -1.7E-02 -1.2E-02
VOL
(-7.11)*** (-3.90 )*** (-6.6)*** (-7.57)*** (-4.67)*** (-6.18)***
Year dummies
0.0300 0.0308 0.1726 -36.15 206.77 74.18
2007
(0.32) (0.24) ( 1.05) (-0.1) (0.45) (0.11)
0.2583 0.2470 0.4989 946.97 1211.73 1547.25
2008
(2.6)*** (1.5) (2.41)** (1.84)* (1.88)* (1.83)*
0.6116 0.6015 0.9524 2407.94 2258.69 3545.82
2009
(6.04)*** (2.78)*** (3.23 )*** (3.35)*** (2.54)** (2.79)***
1.2247 1.0818 1.8908 5342.47 4203.58 7819.51
2010
(10.45)*** (4.12)*** (5.17 )*** (5.77)*** (3.67)*** (4.55)***
1.2699 1.2784 1.8139 5597.67 5038.63 7443.76
2011
(10.18)*** (4.4)*** (4.38 )*** (5.44)*** (4.01)*** (3.91)***
0.3851 0.1952 0.0616 188.43 -622.02 -662.71
Constant
(3.8)*** (0.87) ( 0.20 ) (0.22) (0.59) (-0.52)
Dummies Industry Industry Industry Industry Industry Industry
Obs 723 417 306 723 417 306
Note: Standard errors are in parentheses. Statistics significance at: 10% level (*), 5% level (**), 1 % level (***).
Table 8
Table 8 - The Correlation
THE CORRELATION MATRIXMatrix and Test
AND TEST FORfor Multicollinearity
MULTICOLLINEARITY
INFO TOPHOLD SIZE INDSIZE VOL Variable VIF 1/VIF
INFO 1.0000 SIZE 3.13 0.319983
TOPHOLD 0.0888 1.0000 TOPHOLD 2.97 0.336889
SIZE -0.0769 0.2396 1.0000 INDSIZE 2.08 0.481463
INDSIZE 0.3363 0.0263 0.0508 1.0000 VOL 1.58 0.632529
VOL -0.4339 -0.1216 0.3458 -0.0584 1.0000 Mean VIF 2.44
Page 252
Page 253
DAILY PRICE CHANGES IN THE REAL ESTATE

MARKET
David Rodriguez, Henderson State University
ABSTRACT
Daily price changes are generally missing from data used to investigate the real estate
market, leading to an incomplete examination of the market. Given that some properties undergo
numerous changes before reaching a sale, this paper makes an effort to include all available
data. This paper studies the daily price changes of over 11,000 properties from 2004 to 2011,
providing nearly 13,000 price changes. The average price change is a downward movement
ranging from 6% to 8%. The results suggest a marginal benefit in timing a price change as
gauged by time on market and discount from the last list price. While observational analyses
have frequently made a case for Friday being a superior day to list, the current findings indicate
that Wednesday may be the optimal day to initiate a change. Wednesday price changes reduce
the discount by approximately 1.9%, allowing sales marginally closer to the asking price, and
reduce time on market by approximately 5 days. Both reductions are significant at the 10% level.
INTRODUCTION
A quick search on any search engine will return a wealth of media coverage and research
on housing price trends, the average time on market, and even how particular factors are
expected to be reflected in single-family housing prices. Similarly, real estate investment trusts
(REITs) have received a fair amount of attention possibly due not only to the size of the market
but also to how easily data can be retrieved. However, the real estate market is more than just
single-family houses and REITs. Real estate, as an asset class for the active investor, has been all
but ignored by both the media and the academia. In fact, the United States Census Bureau has
published results in 2013 showing that equity in a primary home, rental property, or other real
estate makes up the majority of the net worth of Americans across all demographics (e.g., race,
age, and education). The primary goal of the current research is to increase the attention given to
this seemingly significant asset class by examining a more complete data sample—that is, one
that incorporates daily price changes. Real estate as an investment class is worthy of attention,
and discovering how to more accurately price a parcel will improve the efficiency of this market.
Active investors in the real estate market are accustomed to its lack of efficiency. Unlike
investing in stocks or REITs, investing directly in real estate forces participants to trade in a
fragmented, non-centralized, non-standardized market. This simple fact alone forces participants
to specialize in a particular area to leverage their individual experiences and local knowledge.
Unfortunately, since parcels are unique and are rarely traded, determining a market price for
them can be difficult for even seasoned investors or brokers. To determine the value of a
property, participants often rely on rule-of-thumb valuations or simple comparative market
analyses. These methods are anything but precise, as exemplified by the frequent price changes
in the market and the lack of uniformity in estimating list prices. In fact, even though multiple
listing services (MLSs) provide daily data on real estate, this data is not examined. This is not to
say that multiple-listing data is unavailable or has not been used. However, attention has been
drawn only to the original list price, final list price, sales price, and time on market. The simple
Page 254
reason is that these price points are the most easily collected. In view of this, this paper utilizes a
complete proprietary MLS dataset and examines whether day-of-the-week advantages exist in
the real estate market. Specifically, this research investigates the day-of-the-week differences of
listings as well as all price changes undertaken within the small investment property space.
This study focuses on small multi-unit properties to examine a more investor-oriented
class of buyers, sellers, and brokers. While there is no guarantee that every broker working with
investment properties is knowledgeable or that every buyer and seller in this market is concerned
only with the investment potential, this segment provides a theoretically “clean” sample to work
with since participants in the single-family home market can become emotionally involved with
the purchase or sale of a home. This emotional involvement alone can reduce the efficiency of a
market. In contrast, an investor may be less inclined to bring emotion into a transaction and more
concerned with the investment potential of a parcel. This makes the market more structured and
ultimately more efficient.
This study adds to the discussion on whether it truly matters when a parcel is listed, and
introduces an examination of daily price changes. Determining the best day or month to list one’s
property has been a topic of discussion in several white papers developed by market participants,
but the analysis has been limited. For example, Redfin has published several white papers
examining the real estate market. In 2013, after surveying 17 markets, the company created a
top-five list of home-selling tips (Ellis, 2013). Ellis reported that Friday is the best day to list a
home since doing so results in a faster sale and a sale price closer to the asking price. Another
finding is that April is the best month to list a home, but any month prior to June is ideal. While
these are interesting findings, the study focused only on single-family houses and looked at only
the original list date and price rather than including the sale date and final price. This fact ignores
all of the time and potential changes during the listing period. Follow-up examinations, such as
those discussed by Fischler (2012), have observed other markets and found that Wednesdays are
optimal, using the same methods of observational analysis. However, in those instances,
empirical analysis was lacking, and while the day of the week a parcel is listed may be
considered significant, the attention it has received seems to dwarf the focus on price changes.
This distinction is noteworthy since both the original listing and each price change lead to a top
listing position in MLSs, which position a listing on top when it is new or when a price change is
made. This positioning brings a listing in front of investors’ eyes and presumably increases
attention to it in a similar way to that of a new listing. The MLS from which the data was
gathered treats a “new” listing similar to a “PCHG.” In both cases, the listing is top-listed for 5
days. The argument presented here is that since the goal is to draw attention to a parcel,
practitioners should have a firm understanding of both the optimal day to list and the optimal day
to initiate price changes to garner the most attention from prospects.
The proprietary data utilized in this study to examine daily price changes has allowed the
examination of several key questions relevant to active market participants.
1. Of those parcels that did not undergo a price change, what quarter of the year and day of the week were
they sold in the shortest amount of time and closest to the list price?
2. Does the time of the year matter? Does listing in the summer direct more attention to a parcel? Or does
listing in the winter solicit attention from more serious buyers?
3. Of those parcels that changed prices, what was the average number of price changes, and what were the
sizes of these price changes?
4. Does the day of the week a price change occurs impact the speed of a sale or the sale price as it relates to
the final market price?
Page 255
The following literature review focuses on only the first two research questions since the
third and fourth relate to individual price changes, which have not been examined in literature.
While examining price changes is common in stock market studies, data on real estate pricing
must be hand-collected by a licensed broker from each MLS, of which there are hundreds. For
this reason, work in this area is limited.
LITERATURE REVIEW
Previous research has studied the relationships between list prices, sale prices, and time
on market (Horowitz, 1992; Yavas & Yang, 1995). For example, Yavas and Yang (1995)
examined the relationship between optimal list prices and time on market. The study focused on
the original list price, and the data examined was on single-family home sales. The authors found
that listing price has no impact on the listing duration of low- and high-priced houses. Kaplanski
and Levy (2012) examined real estate prices to determine if the time of the year is relevant. The
study focused on several countries and found a persistent seasonality. However, it analyzed only
broad indices (such as the Case–Shiller index), which examine only single-family home sales
based on a repeat sales approach, greatly limiting its value among practitioners. The current
paper determines whether these findings on seasonality and list prices hold in a sample of multi-
family parcels while extending the examination to individual price changes.
Price changes were not examined before 2002, and when they were, the focus was on the
reason for their occurrence instead of on the changes themselves. Knight (2002) studied single-
family houses and considered the causes and effects of changes in list price in relation to the
time-on-market measure. The author found that the homes most likely to undergo changes are
those with high initial markups, and vacant homes. The data in that study included only one
change in listing price, the final price change. If more than one change occurred, only the most
recent list price and date were used. Given that properties are not constrained to only one price
change, the dataset was possibly significantly reduced by this restriction. In fact, some parcels in
this dataset underwent nearly three dozen price changes. As stated in the introduction, this focus
on single-family homes also leads to more emotionally driven or at least more irregular
participants. The current paper extends the focus of price changes into the investor arena, where
participants are presumably less driven by emotion and more actively engaged in the market.
Given that listing prices begin as opinions supported by a general comparison analysis or
broker/seller sentiment, it is not appropriate to ignore the price changes that were necessary to
sell a parcel. Properties can undergo dozens of price changes to identify the market price, and
this study is the first to examine the daily price changes that lead to a sale.
DATA AND METHODOLOGY
This paper seeks to go beyond the public data points that are available (the original list
price, sale price, time on market, and last price), and instead include all available data. The
Chicago market is chosen for this initial examination because it is one of the largest metropolitan
areas in the United States and is included in the Case–Shiller index, a prime metric in the real
estate industry. Another reason is that every area in this market operates on an independent MLS.
For example, Illinois alone has dozens of MLSs, each requiring a separate membership. In fact,
Chicagoland utilizes at least six MLSs. The data from the choice MLS examined here singles out
Chicago proper. The data is hand-collected from all available “closed” properties with two to
four units in Chicago. “Closed” is an MLS term for a property that has been sold. The complete
Page 256
dataset is gathered from the MLS serving Chicago, Multiple Listing Service of Northern Illinois,
and spans approximately 7 years, starting from August 2004. The Financial Crisis officially
began in 2007 when Freddie Mac announced it would no longer buy the most risky subprime
(Center for Responsible Lending, 2007). This provides approximately 30 months’ worth of data
prior to the crisis. Some may argue that it actually began in September 2008, when the
government allowed Lehman to collapse, signifying they would not be an unlimited backstop. If
this is the case, the data precedes this by 49 months. The National Bureau of Economic Research
(2010) concluded that the recession ended in June 2009. The data collected for the current study
runs through December 2011, therefore providing approximately 30 months of data after that
publicized end. This creates three distinct windows to examine: pre-crisis, crisis, and post-crisis.
The focus here is buildings with two to four units (i.e., multi-unit buildings) and a closing (i.e.,
selling) price of $50,000 to $500,000. This expansive closing price range is chosen after the
average closing price of all of Chicago’s 77 neighborhoods over the time period has been
examined. Only five neighborhoods have average selling prices that approach the upper limit of
$500,000: Kenwood and Lake View, $365,500; Fuller Park, $379,000; Lincoln Park, $417,750;
and North Center, $476,000. A greater number of such neighborhoods would have created
upward bias.
The total number of properties in the sample, after the omission of several parcels
because of a lack of information, is 11,159. Of these, 4,748 have undergone price changes to
arrive at a sale. Complete descriptive statistics on the retrieved data can be found in Exhibit 1 in
Appendix A.
Using each of the final price changes available, we calculate the percentage changes as
follows:
(1)
Where is the percentage price change, Vt−1 is the previous asking price, and Vt is the final
price presented on the MLS before a sale occurs. Given that this final price change results in a
sale, this is one of the data points examined in this study.
To assess whether the day of the week increases the percentage change, the following
base regression equation is used:
(2)
Where PCF is the percentage change of the listing price and β1 through β7 represent the average
percentage price change on each day of the week beginning with Saturday. The dummy variables
indicate the day of the week on which return is observed: D1 = Saturday, D2 = Sunday, D3 =
Monday, etc. With the omission of the intercept, this equation identifies any daily effect that may
exist. Specifically, any effect on a given day is indicated by a statistically significant t value on
the dummy coefficient. Since Redfin (Ellis, 2013) and Fischler (2012) have shown that Friday
and Wednesday are the best days to list properties, we assume that this logic extends to price
changes too. Further, since a price change results in a top listing on the choice MLS for 5 days,
Wednesday would be ideal since this listing holds that top position during the entire weekend.
This expectation will change depending on the MLS since every system sets the rules for the
amount of time a listing is top-listed. For example, an MLS that allows a top listing position for 7
Page 257
days is unlikely to produce an optimal day to change a price since a broker can alter a price on
any given day and that listing remains top-listed every day of the following week.
To assess whether time on market is impacted by the day of the change, the following
base regression equation is used:
(3)
Where TOM represents the time on market between the final price change and the sale and β1
through β7 represent the average percentage price change on each day of the week beginning
with Saturday. The dummy variables indicate the day of the week on which return is observed:
D1 = Saturday, D2 = Sunday, D3 = Monday, etc. Similar to the previous equation, this equation,
with the intercept omitted, identifies any daily effect that may exist. Specifically, any effect on a
given day is indicated by a statistically significant t value on the dummy coefficient. As above,
both Wednesday and Friday are expected to produce significant variables.
Given that previous studies (Ellis, 2013; Fischler, 2012) have found that Friday is the
optimal day to list a parcel, the two regressions are also examined with Friday omitted.
Additionally, since research (e.g., French, 1980) examining the day-of-the-week effect in the
stock market has shown that Monday provides significantly low return, the two regressions are
also examined with Monday omitted. These six regressions, of which the results are reviewed in
Table 9, allow us to examine the differential intercepts. Since this study focuses on price changes
themselves and how they impact time on market and the ultimate discount, no effort is made to
control for factors such as the seller’s motivation, opportunities for owner financing, or the
distinction between partial and full vacancy. However, because the examined time frame
encompasses the real estate crisis, care is taken in identifying and controlling for properties that
are the result of a court sale, foreclosure, or short sale. Admittedly, this does not ensure that these
factors have been adequately controlled for since the listing parties may not have taken the time
to code these scenarios into the MLS. Additionally, the data is stratified based on three periods—
pre-crisis, crisis, and post-crisis—and on the price points identified in Panel C of the appendix to
ensure asking price is not a determining factor. Finally, in examining the price changes, we
include in the original regressions both the quantity of the change and the time to the last price
change as controls to gauge whether these factors contribute to time on market following the last
price change and final discount. The results obtained are not significant, so only the complete
sample is presented here.
RESULTS
Sample 1: Price change not undertaken
In this section, no empirical work is made. Instead, the intent is to provide insight into the
data being examined. When a parcel does not undergo a price change (because it does not need
to) to make a sale, we can assume that the initial price is appropriate, as determined by the
market. In practice, one may assume that the price is set too low, but if this were the case, we
would expect no discounts from the asking price and frequent overbidding. However, neither of
these events occurs with any regularity. Therefore, reducing the sample to those parcels that do
not undergo a price change leaves a total of 6,410 properties to examine. Table 1 shows that over
the period examined, the differences in inventory segmented by the quarter listed and sold are
negligible.
Page 258
Table 1
This table shows the 6,410 parcels sold without requiring a price change. The parcels are segmented by quarter to
gauge activity over the year. The list date is the date on which the parcel enters the market, and the sell date is the
date on which a purchase contract is accepted by both parties.
List Date Sell Date
Quarter 1 1,706 1,661
Quarter 2 1,716 1,715
Quarter 3 1,565 1,585
Quarter 4 1,423 1,449
We can identify some broad trends in the aggregated data in Table 1, such as the majority
of listings and sales during the first and second quarters. If a practitioner is trying to garner
maximum attention, the first half of the year is probably ideal. However, since the third and
fourth quarters have more sales than listings, sellers may hold a stronger bargaining position in
the second half of the year. From the information above, buyers may reason that the fourth
quarter is optimal since few properties have high demand (i.e., are sold) and they will be taken
most seriously then. Another finding regarding this subset of parcels that may act as further
evidence of an accurate price is the average day-to-sale of 34.06 days, in contrast to the average
for the complete sample, which is 80.6 days. This difference is significant and speaks of the
benefits of an initial price that requires no adjustments.
While the differences across quarters differ only slightly, it is worth identifying how
distressed sales may have impacted the numbers. The quantity of short sales and foreclosures is
shown in Table 2. In both cases, the majority of distressed sales occurred in 2009 and 2011. With
the data as large as it is, the omission of these years did not alter the quarterly dispersion
identified above.
Table 2
This table examines the 6,410 parcels sold without requiring a price change. The parcels are segmented by both
the quarter and the distress code.
Properties noted as short sales
Quarter 1 Quarter 2 Quarter 3 Quarter 4 SUM
140 153 151 145 589
Properties noted as foreclosures
Quarter 1 Quarter 2 Quarter 3 Quarter 4 SUM
252 338 309 400 1,299
Having looked at the timing of these listings and the speed at which they have moved, we
can now determine how a potentially accurate price translates to the sales price. It is worth
knowing not only when a parcel should be listed or purchased but also what discount from the
original list price can be expected. Table 3 shows that the average discount for those properties
that require no change in price is approximately 5%, regardless of the quarter. To ensure that this
simple average is not overly weighted by extremes, the upper and lower deciles are omitted, but
the result does not change substantially. Additionally, the median for the average discount based
on the list date is −4.97%, and that based on the sell date is −4.88%. This shows that the original
asking prices are not overly discounted, resulting in bidding wars, but that even when a parcel is
priced right, we can still expect an approximate discount of 5%.
Page 259
Table 3
This table shows the average difference between the list price and closing price for the parcels that do not
undergo a price change to make a sale. Both columns examine the same dataset with the time classification based
on either the original list date or the sell date.
List Date Sell Date
Quarter 1 −5.66% −5.26%
Quarter 2 −4.89% −5.69%
Quarter 3 −4.99% −5.56%
Quarter 4 −4.93% −5.26%
As shown above, a buyer seems to have a marginal benefit in negotiating a purchase in

the second quarter. Similarly, a seller has a marginal benefit in listing in this quarter. Examining
these price changes over time (Figure 1), we can see that, as expected, discounts are greatest
around the real estate collapse. However, even when the years of the collapse are omitted, the
average still hovers at 5%.
Figure 1
This figure shows the average discount (percentage discount from the asking price) based on the
close dates for the sample requiring no price change to make a sale.
Average Discount
0.00%
Q2 2006
Q3 2006
Q4 2006
Q1 2007
Q2 2007
Q3 2007
Q4 2007
Q1 2008
Q2 2008
Q3 2008
Q4 2008
Q1 2009
Q2 2009
Q3 2009
Q4 2009
Q1 2010
Q2 2010
Q3 2010
Q4 2010
Q1 2011
Q2 2011
Q3 2011
Q4 2011
-1.00%
-2.00%
-3.00%
-4.00%
Average Discount
-5.00%
-6.00%
-7.00%
-8.00%
-9.00%
Sample 2: Price change undertaken
Similar to the previous section, this discussion provides details on the underlying data.
The remainder of the sample, 4,748 parcels, underwent at least one price change before arriving
at a sale. These parcels had 12,806 price changes in total. This should imply a simple average of
2.7 price changes per listing, but that is misleading. In reality, the majority of the parcels
underwent only one price change, while others underwent nearly three dozen. Table 4 shows that
nearly 77% of the total price changes were undergone by parcels that required three price
changes or fewer. The table does not show all parcels but accounts for approximately 98% of the
total sample.
Page 260
Table 4
This table summarizes the information on the final price changes prior to the sale.
Number of Changes Average Discount to Sale % Change Number of Parcels
1 $(14,640.98) −5.69% 1983
2 $(16,896.62) −6.93% 1036
3 $(16,796.02) −7.68% 629
4 $(16,623.92) −7.20% 357
5 $(17,856.24) −8.21% 264
6 $(12,896.19) −6.48% 145
7 $(12,892.92) −6.82% 97
8 $(13,191.48) −8.00% 62
9 $(11,102.34) −6.09% 53
10 $(16,615.76) −7.32% 38
Table 5 shows the quarterly breakdown of this subset. The time on market is not
significantly different from one quarter to the next. However, the properties listed in the fourth
quarter sell at less of a discount, supporting the anecdotal evidence purported by practitioners
that only serious buyers engage in real estate in the winter and around holidays. Examining this
further, we see the greatest number of sales during the first half of the year and the greatest
discount given by those selling in the first quarter. While this is commonly the level studied by
researchers, it is clear that we lose a significant amount of information when we are concerned
only with the listing price, selling price, and time on market. In this study alone, such an
omission would have discarded nearly 8,000 data points.
Table 5
This table shows the averages of the parcels requiring a price change prior to being sold, segmented by both the
quarter the parcels are listed and the quarter they are sold. The total discount is the discount from the original list
to the sale price, and the time on market (TOM) reflects the total time from the original list to the sale date.
Quarter 1 Quarter 2 Quarter 3 Quarter 4
TOM based on quarter listed 167.37 160.07 173.82 161.52
Total discount based on quarter listed −20.95% −22.56% −24.04% −16.06%
TOM based on quarter sold 149.35 150.06 128.59 138.15
Total discount based on quarter sold −24.98% −21.44% −20.37% −22.34%
Table 6 also shows the properties partitioned by year. As expected, discounts escalated
during the real estate collapse. However, discounts remained relatively high even after the
collapse.
Table 6
This table shows the averages of the parcels requiring a price change prior to being sold, segmented by both the year the parcels are listed and
the year they are sold. The total discount is the discount from the original list to the sale price, and the time on market (TOM) reflects the total
time from the original list to the sale date.
2005 2006 2007 2008 2009 2010 2011
TOM based on year listed 414.5 214.96 128.75 160.70 151.28 137.79 77.28
Total discount based on year listed −18.01% −12.18% −21.13% −30.49% −25.30% −22.82% −14.51%
TOM based on year sold — 109.44 115.88 129.16 147.32 154.80 169.56
Total discount based on year sold — −9.00% −14.00% −27.61% −28.38% −24.12% −24.40%
Page 261
Day-of-the-week examination
While the observations presented thus far are of interest to market participants, the
primary contribution of this article extends beyond these. The remainder of the results section
examines the daily changes and the day-of-the-week effect. The first examination of daily price
changes is completed to determine any significant differences from the last price change, based
on the day. Table 7 shows the descriptive statistics of the last price change for each parcel
divided by the day of the week. In other words, the table examines the price change that leads to
the final list price. This is significant since it is this price that leads to the sale.
Table 7
Descriptive statistics for the final price change as measured by the percentage change from the previous price.
The day of the week is the day when the last price change is initiated.
Monday Tuesday Wednesday Thursday Friday Saturday Sunday
Mean −9.21% −8.66% −7.36% −8.03% −9.31% −7.35% −8.49%
Min −74.35% −67.69% −77.78% −68.33% −73.33% −73.37% −53.13%
Max 50.91% 111.76% 108.51% 50.04% 175.00% 43.53% 14.90%
St Dev 11.10% 12.63% 32.65% 31.93% 14.46% 11.79% 8.65%
Count 902 941 882 830 841 220 132
While this data omits a significant amount of the price changes, it is worth focusing on
since it is these price changes that ultimately result in the market price that makes the sale. The
frequency of changes on Monday and Tuesday is logical because most parcels are shown on the
weekend and if the parcels do not receive adequate attention in these showings, participants are
likely to request a price change to garner more attention. For the same reason, the most sizeable
price changes occur on Friday since these may result in more attention over the weekend. These
changes may be an attempt to get the parcel on top of the list and create interest in weekend
viewings. The initial test that extends this discussion is the testing of the null hypothesis: that the
price changes across days are not significantly different from one day to the next. Comparison of
the averages shows that only Monday, compared to Saturday, results in a noteworthy difference
at a 5% level of significance. The remaining days are not significantly different from one
another.
When comparing the discounted final sale price to the last list price, we arrive at the
descriptive statistics in Table 8, which essentially examines the difference between the last list
price and the final agreed-upon price at which the sale culminates.
Table 8
Descriptive statistics for the final sale price versus the most recent list price. The days of the week are the days
when the last price change is initiated.
Monday Tuesday Wednesday Thursday Friday Saturday Sunday
Mean −6.12% −5.92% −6.69% −6.81% −7.28% −8.20% −7.44%
Min −63.03% −70.00% −71.43% −65.60% −66.80% −76.78% −50.00%
Max 100.80% 103.95% 90.64% 125.56% 107.96% 53.26% 55.52%
St Dev 11.87% 11.73% 13.07% 12.53% 12.74% 11.58% 11.32%
Count 902 941 882 830 841 220 132
Page 262
Table 8 shows that while Monday and Tuesday seem to be the most popular days to
initiate a price change, the most significant average discounts (as related to the final sale) are
offered on the weekend, Friday through Sunday. Practitioners may consider that not only are
price changes on the weekend less common, but they also result in the most significant average
discounts on the sale price. The average discount across all properties is 6.65%, in contrast to the
5% among parcels requiring no price change. This implies that after undertaking price changes
and ultimately arriving at an appropriate reserve price, sellers become more willing to take a
lower offer. These findings may lead sellers to avoid price changes on the weekends and buyers
to offer less for properties that have undergone price changes.
Table 9 shows how the day of the week a price change occurs can impact both the time
on market and the final list price as compared to the market price. Sellers would like to sell at a
price close to the list price and would prefer shorter time on market. The table shows which day
a price change should occur to minimize the difference between the marketed price and the sale
price and which day will reduce the time on market. The results of the six regressions are
provided in this table.
Table 9
The results of the regression analyses, where a dummy variable is created for each day of the week, are
shown. To address the dummy variable trap, Regressions 1 and 4 omit the intercept, Regressions 2 and 5
omit Monday, and Regressions 3 and 6 omit Friday in order to examine the differential intercepts.
Model 1 2 3 4 5 6
Dependent Final Price Change (%) TOM from Change to Sale
Intercept NO INT −0.092* −0.093* NO INT 38.181* 35.905*
Saturday −0.073* 0.019 0.02 37.332* −0.849 1.428
Sunday −0.085* 0.007 0.008 36.038* −2.143 0.133
Monday −0.092* — 0.001 38.181* — 2.277
Tuesday −0.087* 0.006 0.006 37.917* −0.264 2.012
Wednesday −0.074* 0.019** 0.019** 32.99* −5.191* −2.915
Thursday −0.080* 0.012 0.013 35.688* −2.494 −0.217
Friday −0.093* −0.001 — 35.905* −2.277 —
R-Square 0.132 0.001 0.001 0.411 0.008 0.002

F-Value 102.57* 0.92 0.92 473.23* 1.41 1.41
* = significant at the 5% level; ** = significant at the 10% level
Table 9 shows that both Model 1 and Model 4 produce significant coefficients, which can
be interpreted as significant daily effects. Model 2 omits Monday and Model 3 omits Friday to
examine the final price change. These omissions are made to examine the differential
coefficients with respect to the omitted day. In both models, only Wednesday is significantly
different, at the 10% level. Given the conclusion of previous work that Friday is optimal, we
expected to see significant differences between other days. However, this does suggest that
changing prices on a Wednesday reduces the discount by approximately 1.9%. This finding is
expected since the MLS examined creates a top listing for 5 days. As previously stated, this
result is likely to change if the MLS adheres to different listing rules.
Models 5 and 6 use time on market as the dependent variable and omit Monday and
Friday, respectively. However, only Model 5 shows a significant difference, at the 10% level,
Page 263
between Monday and Wednesday. This can be interpreted as the time on market’s reduction by
approximately 5 days when the final price change occurs on a Wednesday, as compared to that
on a Monday. This supports the observation that Wednesday is the optimal day to change a list
price.
CONCLUSION
What conclusions can be drawn from this initial study on price changes? First and
foremost, when a parcel is priced “right,” an approximate discount of 5% can still be expected.
Additionally, when priced right, a parcel should sell in approximately 30 days regardless of the
quarter it is listed. When a parcel requires a price change, the discount from the last list price to
the sale price increases to 6.65%, implying that buyers may offer less for properties that have
undergone price changes.
When price changes become necessary, 77% of all parcels that undergo price changes
require three or fewer changes, and the average price change is a reduction in the list price of 6–
8%. Additionally, while the daily effects in Models 1 and 4 show that all days are significant, the
remaining regressions showing the differential coefficients do not support significant differences
between days. Even though previous research maintains that Friday is the optimal day to list
single-family homes, this cannot be defended by the statistical differences between the days. In
fact, the marginal benefits to a seller seem to come from price changes that occur on a
Wednesday since these changes reduce the discount by approximately 1.9%, selling marginally
closer to the asking price, and reduce the time on market to approximately 5 days—both
reductions significant at the 10% level. However, this result may be a function of the fact that the
MLS for Chicago provides a top listing of 5 days for price changes and conclusions are likely to
differ under different rules. For example, if the MLS permitted top listing for only 4 days as a
result of a price change, Thursday would be ideal since this would permit a listing to carry a top
position over the weekend. However, testing how the rules of the MLS may be dictating the
optimal day is an issue because the data is not readily available and must be hand-collected by a
licensed agent with access to each MLS.
This introduction to price changes delivers some actionable points for practitioners. For
example, Monday and Tuesday are the most popular days to initiate price changes. However, the
discounts are not significantly different from those on other days. From a practical point of view,
this may encourage practitioners not to change prices on these days since they are likely to face
more competition for attention then. From a buyer’s perspective, the story changes slightly.
When price changes occur on a Friday, the greatest resulting average discount is 9.31%.
Similarly, discounts from the market price are greatest when a price change is submitted between
Friday and Sunday. This means that buyers may be inclined to look for properties that change
prices on the weekend if they want to secure the largest discount. While this is by no means the
summation of findings, it does help to outline the significance of such research for participants.
Information such as this is imperative to improving not only the way in which practitioners
engage in the market but also their understanding of this asset class and the efficiency of the
market.
Page 264
REFERENCES
Ellis, T. (2011). Best day to list real estate for sale: Friday. Inman News. Retrieved June 2014, from
https://www.redfin.com/research/reports/special-reports/2011/what_day_of_the_week_should_i_list_
my_home.html#.ViG0q36rTIV
Center for Responsible Lending. (2007, February 27). Freddie Mac bans unaffordable subprime home loans [Press
release]. CSR News. Retrieved from http://www.csrwire.com/press_releases/16071-Freddie-Mac-Bans-
Unaffordable-Subprime-Home-Loans
Ellis, T. (2013). Top five home selling tips. Redfin (Special Report). Retrieved June 2014, from Redfin Research
Center website: https://www.redfin.com/research/reports/specialreports/2013/top_five_home_selling_
tips_from_redfin.html
Fischler, M. S. (2012, January 26). Best time to list a home? Midweek. The New York Times. Retrieved from
http://www.nytimes.com/
French, K. (1980). Stock returns and the weekend effect. Journal of Financial Economics, 8(1), 55–69.
Horowitz, J. L. (1992). The role of the list price in housing markets: Theory and an econometric model. Journal of
Applied Econometrics, 7(2), 115–129.
Kaplanski, G., & H. Levy (2012). Real estate prices: An international study of seasonality’s sentiment effect.
Journal of Empirical Finance, 19(1), 123–146.
Knight, J. R. (2002). Listing price, time on market, and ultimate selling price: Causes and effects of listing price
changes. Real Estate Economics, 30(2), 213–237.
National Bureau of Economic Research. (2010). Business Cycle Dating Committee, National Bureau of Economic
Research [Press release]. Retrieved from http://www.nber.org/cycles/sept2010.html
U.S. Census Bureau. (2013). Detailed tables on wealth and asset ownership. Retrieved from
https://www.census.gov/people/wealth/data/dtables.html
Yavas, A. & S. Yang (1995). The strategic role of listing price in marketing real estate: Theory and
evidence. Real Estate Economics, 23(3), 347–368.
APPENDIX A
Exhibit 1
Select Descriptive Statistics on Properties with Price Changes
These panels provide select descriptive statistics on properties that initiated price changes.
Panel A
This table shows the total number of unique properties included and separates them into those that initiated a
price change and those that did not. The table also shows how each property is classified when listed. The listing
classification of foreclosure (F), short sale (S), and court-approved sale (C) is the responsibility of the broker.
Closed with No Price Change Closed with Price Change
Foreclosed (F) 1,809 1,009
Short Sale (S) 791 1,093
Court-Approved (C) 31 23
No Specification Noted (NA) 3,779 2,623
Total Closed 6,410 4,748*
*Two properties are omitted due to an apparent abuse of the top-listing function. These two properties alone count for nearly 100 price changes
over a 2-year period.
Panel B
This table displays the total number of properties that initiated a price change (4,748) from the above panel. Each
daily directional change is shown as a unique observation. A small number of owners have changed prices twice
within a day with no net change, implying merely a goal to top the list. These events are ignored in this analysis
since the goal is only to receive a status change.
Unique Price Changes
Negative Movements 12,183
Positive Movements 624
Total Changes 12,807
Page 265
Panel C
This table shows additional descriptive details about the 4,748 properties with regard to their list and final prices.
The first table partitions the properties in $50,000 increments for both the final selling price and the original list
price. The second table displays the total discount from the original list price and the market time for each
property in the sample. The table continues to divide the properties by their classification type (NA, F, S, and C).
Final Price ($) N List Price ($) N
50,000–100,000 1,280 <50,000 10
100,000–150,000 854 50,000–100,000 554
150,000–200,000 716 100,000–150,000 709
200,000–250,000 552 150,000–200,000 837
250,000–300,000 464 200,000–250,000 585
300,000–350,000 318 250,000–300,000 584
350,000–400,000 245 300,000–350,000 419
400,000–450,000 177 350,000–400,000 349
450,000–500,000 142 400,000–450,000 229
450,000–500,000 206
500,000–550,000 157
550,000–700,000 102
700,000–900,000 7
Sum 4,748 Sum 4,748
Sum Average St Dev

Discount in Price (NA) $144,712,766 $30,479 $48,112
Discount in Price (F) $43,964,451 $9,260 $27,401
Discount in Price (S) $96,893,564 $20,413 $51,387
Discount in Price (C) $1,839,671 $387 $7,919
Market Time (NA) 524,634 days 110 days 162 days
Market Time (F) 204,319 days 43 days 120 days
Market Time (S) 274,466 days 58 days 141 days
Market Time (C) 4,448 days 1 days 17 days
Page 266
Page 267
THE EFFECTS OF VOLATILITY AND LEVERAGE ON

THE EARNINGS-GDP RELATION
Jagjit S. Saini, Western Michigan University
Todd P. White, University of New Mexico
ABSTRACT
We find the ability of aggregate accounting earnings to predict future GDP growth is
influenced significantly by changes in 1) market volatility and 2) average firm leverage. More
specifically, we find increasing levels of market volatility are associated with earnings being
more predictive of future GDP growth. We attribute this finding to the constraining influence of
volatility on the discount rate signaling effect of earnings. Next, we find aggregate earnings
predict GDP growth less when debt-to-equity levels have peaked and we attribute this result to
the inability of earnings to predict GDP growth when capital is constrained. Results demonstrate
the importance of controlling for macroeconomic factors such as the level of market volatility
and firm leverage when assessing the ability of aggregate earnings to forecast growth in GDP.
Keywords: aggregate earnings, GDP forecasting, market volatility, debt-to-equity
JEL Classification: D82, D84, G20
INTRODUCTION
Research has demonstrated the value of accounting earnings in predicting corporate stock
returns. This research depicts accounting earnings as a leading indicator of the economic
performance of a company. It seems intuitive then that aggregate accounting earnings for the
market as a whole would be indicative of future economic growth. However, this conclusion is
not easily deduced from prior research. Cready & Gurun (2010), for example, show that
aggregate accounting earnings is actually negatively related to returns in the stock market.
Despite this surprising relationship between aggregate earnings and returns, Konchitchki &
Patatoukas (2014), (hereafter KP (2014)), document that aggregate accounting earnings are
indeed predictive of future economic growth. Their study makes an important contribution by
providing robust evidence that aggregate corporate earnings are a leading indicator of economic
development. However, the study of KP (2014) does little to identify the types of
macroeconomic conditions that are more (or less) conducive to earnings predicting future GDP
growth. An analysis of the effects of macroeconomic conditions on the earnings-GDP
relationship would serve to further knowledge regarding exactly how aggregate earnings predict
future GDP growth. That is the aim of the current study. This study seeks to examine the effects
of macroeconomic conditions of (i) market volatility and (ii) firm leverage on the relationship
between aggregate earnings and GDP growth.
The theoretical framework utilized in this paper hypothesizes that both volatility and
leverage could impact the earnings-GDP relation through one of two mechanisms. These two
mechanisms are 1) discount rate signaling effects, and 2) forecasting future investment. Discount
rate signaling effects refers to the fact that aggregate accounting earnings contain not only news
about cash flows but also discount rate news. Cready & Gurun (2010) show aggregate
Page 268
accounting earnings are negatively related to returns and hypothesize the reason for this is
aggregate earnings forecast increases in the discount rate. Additionally, market volatility and
firm leverage may impact future investment. Research has shown investment in equipment,
R&D, and infrastructure forecasts future growth in GDP (Long & Summers, 1990; Lichtenberg,
1992; Herrerias & Orts, 2008). Aggregate earnings forecast GDP due to positive signals about
investment. If volatility and leverage constrain investment then earnings would be less predictive
of GDP.
The results of this study demonstrate that both (1) changing levels of market volatility
and, (2) changing levels of average leverage affect the predictability of earnings on GDP. This is
consistent with the hypothesis that macroeconomic conditions have a significant impact on the
degree to which aggregate earnings predict future GDP growth. Further analysis demonstrates
that increasing levels of volatility are actually associated with earnings being more predictive of
future GDP growth. This result is consistent with the hypothesis that increasing levels of
volatility actually constrain the discount rate signaling effect in earnings, leading to a higher
correlation between earnings and GDP growth.
With respect to average firm leverage, it is shown that earnings are less predictive of
GDP growth when the level of debt-to-equity is high but on the downward slope. This result
follows the theoretical foundation of Mendoza (2010) showing that over-borrowing triggers
constraints in lending which reduce asset prices and capital availability. Aggregate earnings are
less likely to translate into growth in GDP in this type of capital-constrained environment.
This paper represents an important contribution to the extant literature by documenting
that macroeconomic conditions play a significant role in the ability of earnings to forecast GDP
growth. The analysis provides an indication that the predictability of earnings on GDP is
constrained by the discount rate news in earnings. When market uncertainty is higher, the
discount rate news in earnings is constrained and earnings are more predictive of GDP.
Additionally, the study shows that restrictions on the availability of capital negatively affect the
degree to which earnings signal GDP growth. Overall, the study provides further insight into the
mechanism with which earnings predict GDP by identifying the effects of discount rate signaling
and investment forecasting components of earnings.
The next section of the paper presents the background and hypothesis development. This
is followed by the section describing the methodology utilized. Subsequently, the sample is
presented and this is succeeded by the results of the analysis. The final section concludes the
study.
BACKGROUND AND HYPOTHESIS DEVELOPMENT
It is well known that corporate earnings are an important component of national GDP.
However, there exists debate about the degree to which positive future economic output can be
predicted via positive corporate earnings. KP (2014) make an important contribution to this
debate. In their study, they show aggregate accounting earnings in the market are a leading
indicator of growth in GDP. The authors explain the theoretical foundation for this finding in that
1) corporate profits are correlated with other components of GDP and 2) accounting earnings
proxy for corporate profits. The empirical results of KP (2014) support the above assertions and
demonstrate the importance of aggregate accounting earnings in forecasting economic growth.
The current study extends the research of KP (2014) by analyzing macro-economic conditions
that cause earnings to be more (or less) predictive of future GDP. This will enable us to more
closely examine the factors involved in the predictability of earnings on GDP.
Page 269
To understand the effects of macroeconomic conditions on the predictability of earnings

on GDP growth, it is important to understand the relation between earnings and other
components of the economy. One important relationship to analyze is that of earnings and stock
returns. Extant research has demonstrated that aggregate earnings are negatively related to
market returns (Cready & Gurun 2010). The reason for this negative relation concerns the fact
that earnings contain information about discount rates. Stock market returns can be decomposed
into three components: 1) cash flow news, 2) discount rate news, and 3) previous returns.
Aggregate accounting earnings contain both cash flow news and discount rate news (Gkougkousi
2013; Kothari et al. 2013; Penman 2015). It is the discount rate signal in positive aggregate
earnings that causes market returns to react negatively (Cready & Gurun 2010). Forecasts of
higher borrowing news are impounded negatively by market participants. This finding is
corroborated by additional studies demonstrating a negative market reaction to the potential
increases in borrowing rates forecasted in economic news. Patatoukas (2014) demonstrate
empirically that increases in GDP are surprisingly not associated with positive market returns.
This lack of association is attributed to the higher borrowing rates signaled by GDP growth.
If accounting earnings signal increases in the discount rate which cause stock prices to
react negatively and affect other components of the economy, it is probable that the discount rate
signaling effect weakens the predictability of earnings on GDP growth. KP (2014) show
accounting earnings to be a leading indicator of economic news. It is possible, however, that the
earnings-GDP relationship is attenuated by the discount rate signaling effect of earnings which
negatively affects stock prices. Therefore, when analyzing how macro-economic conditions
affect the ability of aggregate earnings to predict GDP growth, it is important to identify how
those conditions will affect the degree to which earnings signal changes in borrowing rates. The
discount rate signaling effect of accounting earnings could counteract the cash flow news in
earnings. Therefore, macro-economic conditions which constrain (or enhance) the discount rate
signaling effect in earnings could alter the predictability of earnings on GDP.
The first macro-economic scenario considered is that of a volatile market. It is
hypothesized that volatile market conditions could actually decrease the propensity of positive
earnings to signal future discount rate increases. When market volatility is high, monetary policy
is more likely to pursue actions to lower borrowing rates (Mishkin, 2010; Perrera, 2010).
Therefore, when positive earnings are announced, the market and other components of GDP are
less likely to react negatively in expectation of higher borrowing rates. Since earnings contain
both cash flow news as well as discount rate news, market conditions which constrain the effects
of the discount rate news should increase the predictability of earnings on GDP. Patatoukas
(2014) shows GDP growth forecast errors are correlated with both cash flow news and discount
rate news but these two sources have offsetting impacts on stock prices over the sample period.
If volatile markets mitigate the discount rate signaling effect then the ability of earnings to
forecast GDP growth should be strengthened.
While volatile markets could constrain the discount rate signaling effect of earnings, it is
important to consider the negative effects that volatile market conditions could have on
investment. Prior studies have demonstrated that investment in equipment, R&D, and
infrastructure are important predictors of future growth in GDP (Long & Summers, 1990;
Lichtenberg, 1992; Herrerias & Orts, 2008). If volatile market conditions negatively impact
investment through a higher cost of capital, this could cause earnings to be less predictive of
GDP growth.
Page 270
In volatile markets, the cost of equity capital may increase as investors expect larger
rewards to compensate for risk. Guo (2002) shows that higher market volatility results in lower
economic output. He conjectures a reduction in economic output arises from a higher cost of
capital. This is supported by Ramey and Ramey (1995) who show countries with greater
volatility demonstrate less economic growth. The higher cost of capital in volatile markets could
constrain positive corporate profits and this would attenuate the earnings-GDP relation.
Therefore, the direction of the effect of market volatility on the earnings-GDP relation is
uncertain. The constraining effect of volatility on the discount rate and the potential increase in
the cost of capital lead to conflicting hypotheses. Therefore, the first hypothesis is non-
directional. It is stated in the null form as follows:
H1: Market volatility will have no effect on the ability of earnings to predict GDP growth.
The second macroeconomic scenario considered is that of the changing levels of average
debt-to-equity. Returning to the discount rate signaling effect of earnings, it is conjectured that
any scenario that alters this effect could change the ability of earnings to forecast future GDP
growth. If companies have greater debt, this should magnify the discount rate signaling effect of
earnings. If firms are taking on more debt, the amplification of the discount rate signaling effect
would cause an even more negative effect of aggregate earnings on market returns. If the market
reacts negatively to earnings due to greater firm leverage, this could weaken the relation between
earnings and future GDP growth. This would cause greater levels of firm debt to negatively
affect the predictability of earnings on GDP.
An additional reason why greater firm leverage could negatively affect the ability of
earnings to predict future GDP growth is that taking on too much debt may signal poor economic
growth. Bianchi & Mendoza (2011) argue that over-borrowing occurs in competitive
equilibriums because agents fail to internalize the amplification effects. This eventually triggers
a decline in asset prices and decreases lending available. Mendoza (2010) argues that debt
financing rises during expansions, and when it has reached the peak, it triggers the constraints
that tighten debt financing and reduce credit, which result in a decline of investment and GDP.
Additionally, Metiu et al. (2014) conjecture that credit constraints in the U.S. have a spillover
effect which negatively affects global economic conditions. Earnings will be less likely to
forecast higher GDP growth when debt financing is tighter and credit is less available. Therefore,
in addition to a magnification of the discount rate signaling effect of earnings, an additional
reason why greater leverage may negatively affect the ability of earnings to predict GDP is that
over-borrowing may signal poor economic growth.
Alternatively, increasing leverage may provide a positive signal that investment
opportunities are more attractive. In this case, future economic output would be expected to rise.
An increase in leverage may be indicative of an increasing set of investment opportunities. KP
(2014) maintain that one reason earnings is predictive of GDP is that earnings are correlated with
other components of GDP. If investment opportunities are increasing, positive earnings are more
likely to be correlated with other components of GDP. Since there are reasons to hypothesize
increasing leverage will have a negative effect on the earnings-GDP relation as well as reasons to
believe the effect will be positive, it is unclear which effect will dominate. Therefore, the second
hypothesis is stated in the null form as follows:
H2: The change in average firm leverage will have no effect on the ability of earnings to predict GDP
growth.
Page 271
METHODOLOGY
To investigate the effects of market volatility and leverage on the predictability of

earnings on GDP, we employ the methodology of KP (2014). This requires running time-series
regressions of GDP growth on the change in aggregate earnings. To examine the effects of
volatility and debt-to-equity, we include additional regressors as explained below.
The first hypothesis of this study states changes in the level of market volatility will have
no effect on the predictability of earnings on GDP. To test this hypothesis, we use a regression of
GDP on aggregate earnings similar to KP (2014). Since our hypothesis looks at the effect of
volatility on the earnings-GDP relation, we include the change in volatility as a regressor. We
use both GDP at time t+1 and GDP at time t+2 as dependent variables in the regression. Our
models are as follows:
GDPt+1 = α1 + β1∆Earningst + β2∆VIXt + β3(∆Earningst*ΔVIXt) + ε (1a)
GDPt+2 = α1 + β1∆Earningst + β2∆VIXt + β3(∆Earningst*ΔVIXt) + ε (1b)
Similar to KP (2014), GDP is equal to nominal GDP growth (at time t+1 for 1a and t+2
for 1b). ∆Earnings is equal to aggregate earnings growth for quarter q. Finally, ∆VIX is equal to
the one-year change in market volatility as measured by the CBOE index. Our first hypothesis
stated in the null form was that volatility would not affect the earnings-GDP relation. This
hypothesis is rejected if the coefficient on β3 is significant. This would indicate that a change in
the level of volatility affects the degree to which aggregate earnings predict future GDP growth.
The second hypothesis of this study states changes in the average level of firm debt-to-
equity will have no effect on the predictability of earnings on GDP. To test this hypothesis we
run the KP (2014) time-series regression of GDP growth on earnings and include debt-to-equity
as an additional regressor. The model is as follows:
GDPt+1 = α1 + β1 ∆Earningst + β2 ∆DTEt + β3 (∆Earningst*∆DTEt) + ε (2a)
GDPt+2 = α1 + β1 ∆Earningst + β2 ∆DTEt + β3 (∆Earningst*∆DTEt) + ε (2b)
GDP and ∆Earnings are as described previously. ∆DTE is equal to the change in the
aggregate debt-to-equity ratio and is included as a measure of leverage. Following KP (2014), all
regressions use Newey-West heteroskedasticity- and autocorrelation-consistent standard errors.
The second hypothesis is rejected if the coefficient on β3 is significant. This would indicate that a
change in the average level of debt-to-equity impacts the predictability of earning on returns.
While the above tests will allow us to infer whether volatility and debt-to-equity have an
effect on the earnings-GDP relation, it may be difficult to infer the direction of the effect on the
basis of those tests. For this reason, we include an additional test for each hypothesis. To perform
this test, we include four wave indicator variables. The four indicator variables take a value of 1
depending on whether the level of the variable is above the median and whether the variable is
increasing or decreasing. The variables are as follows: Wave_AI, Wave_AD, Wave_BI, and
Wave_BD. Wave_AI is equal to 1 if the amount is above the median and the change is positive.
Wave_AD is equal to 1 if the amount is above the median and the change is negative. Wave_BI is
Page 272
equal to 1 if the amount is below the median and the change is positive. Wave_BD is equal to 1 if
the amount is below the median and the change is negative. The bottom wave dummy
(Wave_BD) is excluded to avoid multicollinearity issues.
The first set of tests seeks to determine which period or wave of volatility causes earnings
to be more (or less) predictive of GDP growth. The specification with the three wave dummy
variables representing volatility and the interactions is shown below:
GDPt+1 = α1 + β1∆Earningst + β2VIX_AIt + β3 VIX_ADt + β4 VIX_BIt + β5(∆Earningst*VIX_AIt)

+ β6(∆Earningst*VIX_ADt) + β7(∆Earningst*VIX_BIt) + ε (3a)
GDPt+2 = α1 + β1∆Earningst + β2VIX_AIt + β3VIX_ADt + β4VIX_BIt + β5(∆Earningst*VIX_AIt)

+ β6(∆Earningst*VIX_ADt) + β7(∆Earningst*VIX_BIt) + ε (3b)
The second set of tests seeks to determine which period or wave of changes in average
debt-to-equity causes earnings to be more (or less) predictive of GDP growth. The specification
with three wave dummy variables represent debt-to-equity changes and the interactions is shown
below:
GDPt+1 = α1 + β1∆Earningst + β2DTE_AIt + β3DTE_ADt + β4DTE_BIt +

β5(∆Earningst*DTE_AIt) + β6(∆Earningst*DTE_ADt) + β7(∆Earningst*DTE_BIt) + ε (4a)
GDPt+2 = α1 + β1∆Earningst + β2DTE_AIt + β3DTE_ADt + β4DTE_BIt +

β5(∆Earningst*DTE_AIt) + β6(∆Earningst*DTE_ADt) + β7(∆Earningst*DTE_BIt) + ε (4b)
No prediction is made with respect to these indicator variables or the interaction

variables. The aim of these tests is to isolate the type of macro-economic environments that most
greatly affect the ability of earnings to predict GDP growth. Significance on the interaction
variables should yield greater insight toward this end.
SAMPLE
Our sample contains 94 quarterly observations spanning from Q1:1988 to Q2:2011. We

obtain data from the Compustat Quarterly US dataset. Earnings growth (∆Earnings) is measured
as the year-to-year change in scaled quarterly income. Following KP (2014), earnings are scaled
by sales. An aggregate quarterly time series of earnings is constructed using value-weighted
cross-sectional averages with weights based on market capitalization as of the beginning of the
quarter. Furthermore, following KP (2014), for a firm to be included in our sample, it must have
non-missing data for market value of equity, earnings, and the quarterly earnings announcement
date. The sample is restricted to firms with December fiscal year-ends. Finally, we delete firm-
quarter observations that fall in the top and bottom one percentile of each quarterly cross-section
of Earnings and ∆Earnings.
GDP is equal to the “final” estimate of realized GDP growth in nominal terms. This
number comes from the Real-Time Data Set for Macroeconomics of the Federal Reserve Bank of
Philadelphia. ∆VIX is equal to the yearly change in volatility as proxied by the Chicago Board of
Options Exchange Index, commonly known as the “fear index”. DTE is equal to the year-over-
year change in the debt-to-equity ratio measured as long-term debt over the market value of
equity. Data to create the DTE variable was obtained from both CRSP and Compustat.
Page 273
RESULTS
Descriptive statistics are presented in Table 1. These statistics match KP (2014) with
minor discrepancies. The median level of volatility (VIX) in our study is 19.23 and the median
level of average DTE is 0.379. The median for ∆VIX is 0.200 while the median for ∆DTE is -
0.026.
Table 1
Variable N Mean Median Std. Dev. 1rst Quartile 3rd Quartile
GDP 94 4.931 5.091 2.445 3.822 6.557
Earnings 94 0.090 0.092 0.036 0.077 0.116
∆Earnings 94 0.001 0.002 0.030 -0.009 0.013
VIX 93 20.929 19.230 7.986 14.410 24.590
∆VIX 93 -0.266 0.200 9.139 -3.320 4.700
DTE 94 0.380 0.379 0.090 0.306 0.441
∆DTE 94 -0.038 -0.026 0.198 -0.069 0.032
Table 2
PEARSON CORRELATION MATRIX
Variable GDP Earnings ∆Earnings VIX ∆VIX DTE ∆DTE
GDP 1 0.035 0.327 -0.384 -0.323 -0.316 -0.189
(0.737) (0.001) (0.001) (0.002) (0.002) (0.069)
Earnings 0.035 1 0.614 -0.232 -0.075 0.032 -0.171
(0.737) (<.0001) (0.025) (0.474) (0.759) (0.099)
∆Earnings 0.327 0.614 1 -0.270 -0.276 0.004 -0.178
(0.001) (<.0001) (0.009) (0.007) (0.968) (0.085)
VIX -0.384 -0.232 -0.270 1 0.563 0.245 0.270
(0.000) (0.025) (0.009) (<.0001) (0.018) (0.009)
∆VIX -0.323 -0.075 -0.276 0.563 1 -0.044 0.397
(0.002) (0.474) (0.007) (<.0001) (0.677) (<.0001)
DTE -0.316 0.032 0.004 0.245 -0.044 1 0.104
(0.002) (0.759) (0.968) (0.018) (0.677) (0.320)
∆DTE -0.189 -0.171 -0.178 0.270 0.397 0.104 1
(0.069) (0.099) (0.085) (0.009) (<.0001) (0.320)
GDP is nominal GDP growth for the quarter as reported by the Bureau of Economic Analysis.
Earnings is the amount of quarterly earnings.
∆Earnings is the year-to-year change in aggregate quarterly earnings.
VIX is the level of market volatility as measured using the VIX or "fear index".
∆VIX is the year-to-year change in market volatility measured using the VIX or "fear index".
DTE is quarterly debt-to-equity ratio.
∆DTE is the year-to-year change in the quarterly debt-to-equity ratio.
Page 274
Table 2 presents Pearson correlation statistics for the variables used in the study. As
expected, a positive change in earnings (∆Earnings) is positively related to GDP growth (coeff.=
0.327; p=0.001). Higher degrees of market volatility and debt-to-equity, however, appear
negatively related to GDP growth. Both the level of volatility (VIX) and increases in volatility
(∆VIX) are negatively related to GDP growth. The correlation coefficient on VIX is -0.384
(p<.001) while that on ∆VIX is -0.323 (p=0.002). The level of debt-to-equity (DTE) is negatively
and significantly related to GDP growth while changes in the level of debt-to-equity (∆DTE) are
only marginally negatively related to GDP growth. The correlation coefficient on DTE is -0.316
(p=0.002) while that on ∆DTE is -0.189 (p=0.069). The relation between both market volatility
and debt-to-equity with earnings appears to be negative. Both the level of volatility (coeff.= -
0.27; p=0.009) and the change in volatility (coeff.= -0.276; p=0.007) are negatively related to the
change in earnings (∆Earnings). With respect to debt-to-equity, however, only the change in
debt-to-equity (∆DTE) is marginally related to the change in earnings (coeff.=-0.178; p=0.0855).
With respect to volatility and leverage, there is a strong correlation between the level of debt-to-
equity and the level of volatility (coeff.=0.245; p=0.018). Finally, the change in debt-to-equity is
positively correlated with both the level of volatility (coeff.=0.270; p=0.009) and the change in
volatility (coeff.= 0.397; p<.0001).
Our first hypothesis stated in the null form is that the change in volatility has no effect on
the predictive ability of earnings on future GDP. This hypothesis is tested by regressing GDP
growth on earnings, volatility, and the interaction between these two variables as presented in
equation 1a. The results of this regression are presented in Table 3.
Panel A of Table 3 examines the effect of the change in volatility (∆VIX) on the earnings-
GDP relationship when the dependent variable is the next quarter’s GDP (GDPt+1). First, it is
interesting that when ∆VIX is included in the regression, ∆Earnings is now insignificant
(p=0.187). This contrasts with the results of KP (2014) where earnings was significant at the 1%
level. This indicates a significant portion of the effect of earnings in predicting GDP is subsumed
by market volatility. As expected, the coefficient on ∆VIX is negative (-0.11) and significant
(p=0.008) indicating that generally, higher volatility is associated with lower GDP growth. The
variable of interest, the interaction between ∆Earnings and ∆VIX, is highly significant (p=0.001).
This demonstrates the significant impact of changing volatility levels on the predictive effect of
earnings on GDP growth. While the coefficient on this interaction variable is positive (2.68), it is
difficult to draw conclusions regarding the direction of the relationship due to the fact the two
variables being interacted are both change variables. Further tests will attempt to more accurately
isolate the direction of the effect of volatility on the earnings-GDP relation. However, using this
initial test, it appears the null hypothesis that volatility does not affect the predictability of
earnings on GDP can be rejected.
Panel B of Table 3 examines the effect of the change in volatility (∆VIX) on the earnings-
GDP relationship when the dependent variable is GDP growth two quarters ahead (GDPt+2). The
first point of note when looking at this table is that ∆Earnings is now more significant than in
Panel A (p=0.002). This is probably due to the fact ∆VIX measures the change in volatility over
the prior year. Therefore, the farther out you forecast GDP, the less relevant the prior year’s
change in volatility becomes. This is supported by the fact that the coefficient on ∆VIX is now
only significant at the 5% level (p=0.03). The coefficient on ∆Earnings*∆VIX remains positive
and significant (coeff.= 1.89; p=0.001). Again, this means the null hypothesis which states
volatility has no effect on the ability of earnings to predict GDP can be rejected. Further tests
will seek to demonstrate the direction of this effect.
Page 275
Table 3
EFFECT OF MARKET VOLATILITY ON THE EARNINGS-GDP RELATION
GDP=α1 + β1∆Earningst + β2∆VIXt + β3(∆Earningst*∆VIXt) + ε (1a and 1b)
Panel A: Dependent Variable = GDPt+1
Est.
Variable +/- T-Stat p
Coeff.
Intercept 5.07 23.66 <.0001

∆Earnings + 14.34 1.33 0.187
∆VIX - -0.11 -2.71 0.008
∆Earnings*∆VIX ? 2.68 3.33 0.001
N 94
Adj. R-Squared 29.46%
Panel B: Dependent Variable = GDPt+2
Est.
Coeff.
Intercept 4.93 20.89 <.0001

∆Earnings + 22.57 3.12 0.002
∆VIX - -0.07 -2.22 0.029
∆Earnings*∆VIX ? 1.89 3.50 0.001
N 94
∆VIX is the year-to-year change in market volatility measured using the VIX or "fear index".
Our second hypothesis stated in the null form is that changes in the average level of
firms’ debt-to-equity do not affect the predictive ability of earnings on GDP growth. To test this
hypothesis, we regress GDP growth on the change in aggregate earnings (∆Earnings), the change
in average DTE (∆DTE), and the interaction of those two variables (∆Earnings*∆DTE) as
depicted in equations 2a and 2b. The results are exhibited in Table 4.
Panel A of Table 4 shows the results of the regression using GDP growth at time t+1
(GDPt+1). Unlike in Panel A of Table 3, the coefficient on ∆Earnings in this model is highly
significant (p=0.015). This means that unlike volatility, growth in leverage does not subsume the
predictive effect of earnings on GDP growth. The coefficient on ∆DTE is negative and
significant (coeff.= -3.06; p=0.018). This means, generally, greater aggregate leverage is
associated with a decrease in GDP growth. Finally, the coefficient on ∆Earnings*∆DTE is
positive and significant at the 10% level (coeff.= 175.07; p=0.052). Therefore, the null
hypothesis that the aggregate DTE level does not affect how earnings predict GDP growth can be
rejected at the 10% level of significance. Further tests, however, are needed to determine the
direction of this relationship.
Panel B of Table 4 shows the results of the regression using GDP growth at time t+2
(GDPt+2). The coefficient on ∆Earnings remains positive and significant. The coefficient on
∆DTE, however, is no longer significant (p=0.247). Finally, the interaction between ∆Earnings
Page 276
and ∆DTE is again positive and significant (coeff.= 116.42 and p=0.007). This means a change
in the prior year’s aggregate level of debt-to-equity has an impact on how earnings predict future
GDP.
Table 4
EFFECT OF LEVERAGE ON THE EARNINGS-GDP RELATION
GDP=α1 + β1∆Earningst + β2∆DTEt + β3(∆Earningst*∆DTEt) + ε (2a and 2b)
Est.
Coeff.
Intercept 4.98 16.45 <.0001

∆Earnings + 19.08 2.47 0.015
∆DTE - -3.06 -2.41 0.018
∆Earnings*∆DTE ? 175.07 1.97 0.052
N 94
Est.
Coeff.
Intercept 4.94 15.53 <.0001

∆Earnings + 21.68 2.66 0.009
∆DTE - -0.96 -1.17 0.247
∆Earnings*∆DTE ? 116.42 2.78 0.007
N 94
∆DTE is the year-to-year change in the quarterly debt-to-equity ratio.
The next set of tests attempts to more accurately identify the direction of the effect that
volatility and leverage have on the earnings-GDP relation. As discussed in the methodology
section, these tests implement four indicator variables to separate the directional changes or
“waves” of volatility and debt-to-equity. Two of the indicator variables represent periods when
the level of volatility (or DTE) is above the median with one of those indicators representing an
increasing level and the other representing a decreasing level. Additionally, two indicator
variables represent periods when the level of volatility (or DTE) is below the median with one of
those indicators representing an increasing level and the other representing a decreasing level. By
breaking out the changes or waves in the volatility, it is easier to identify the conditions under
which the earnings-GDP relationship is most affected.
Table 5 presents the results using the volatility wave indicator variables. For Panel A, the
dependent variable is GDP at time t+1 (GDPt+1). The variables of interest in this panel are the
interaction variables between earnings and the volatility indicators. The only significant
interaction among these variables occurs between earnings (∆Earnings) and the wave indicator
Page 277
variable representing a level of volatility that is below the median and increasing (VIX_BI). The
interaction is positive (coeff.= 29.70) so this indicates that when volatility is low and increasing,
earnings is more predictive of GDP. Panel B of Table 5 examines the effect of volatility on GDP
at time t+2 (GDPt+2). In this panel, it is evident that the only significant interaction with earnings
occurs when volatility is above the median and increasing (VIX_AI). The coefficient on this
variable is again positive and significant (coeff.= 27.59; p=0.02). This means a level of volatility
that is high and increasing is associated with a stronger earnings-GDP relation.
Table 5
EFFECT OF MARKET VOLATILITY ON THE EARNINGS-GDP RELATION
GDP=α1 + β1∆Earningst + β2VIX_AIt + β3 VIX_ADt + β4 VIX_BIt + β5(∆Earningst*VIX_AIt) +
β6(∆Earningst*VIX_ADt) + β7(∆Earningst*VIX_BIt) + ε (3a and 3b)
Variable +/- Est. Coeff. T-Stat p
Intercept 5.49 15.28 <.0001
∆Earnings + 9.91 0.89 0.377
VIX_AI - -1.13 -1.59 0.115
VIX_AD + -0.24 -0.32 0.748
VIX_BI - -0.69 -1.54 0.127
∆Earnings*VIX_AI ? 27.28 1.12 0.266
∆Earnings*VIX_AD ? -10.92 -0.45 0.652
∆Earnings*VIX_BI ? 29.70 1.84 0.069
N 93
Intercept 5.63 21.44 <.0001
∆Earnings + 11.59 1.98 0.051
VIX_AI - -1.33 -2.15 0.035
VIX_AD + -0.93 -1.54 0.128
VIX_BI - -1.05 -1.40 0.164
∆Earnings*VIX_AI ? 27.59 2.32 0.022
∆Earnings*VIX_AD ? 5.34 0.22 0.823
∆Earnings*VIX_BI ? 17.31 0.49 0.624
N 93
VIX_AI is a wave indicator variable that equals 1 if the level of volatility is above the median and the change
in volatility is positive (increasing).
VIX_AD is a wave indicator variable that equals 1 if the level of volatility is above the median and the change
in volatility is negative (decreasing).
VIX_BI is a wave indicator variable that equals 1 if the level of volatility is below the median and the change
in volatility is positive (increasing).
Therefore, for both GDP growth at time t+1 and GDP growth at time t+2, the implication
seems to be that an increasing level of volatility is associated with a stronger earnings-GDP
relation. This follows the theory that higher volatility mitigates the discount rate signaling effect.
In other words, higher volatility decreases the probability that positive earnings forecast
increases in borrowing rates. Therefore, in volatile markets, the “cash-flow news” in earnings is
not so strongly offset by the “discount-rate news”. This causes the correlation between earnings
and corporate profits to be stronger and earnings to be more predictive of GDP growth.
Page 278
Table 6 presents the results using DTE wave indicator variables. For Panel A, the
dependent variable is GDP at time t+1 (GDPt+1). Panel A shows the only significant interaction
between earnings and DTE occurs when DTE is above the median and decreasing (DTE_AD).
The sign on this interaction variable (∆Earnings*DTE_AD) is negative (coeff.= -37.85;
p=0.024). This means that when the DTE ratio has peaked and begins to decrease, the predictive
ability of earnings on GDP is smaller. Panel B shows a similar result when the dependent
variable is GDPt+2. The coefficient on the interaction variable (∆Earnings*DTE_AD) is again
negative and significant (coeff.= -45.79; p=0.007). Thus, in both model specifications, earnings
predict GDP less when DTE is above the median and decreasing. The most likely explanation for
this is that over-borrowing has occurred and is causing negative economic effects. This scenario
is explained in Mendoza (2010), Bianchi and Mendoza (2011), and Akinci and Queralto (2014).
Table 6
EFFECT OF LEVERAGE ON THE EARNINGS-GDP RELATION
GDP=α1 + β1∆Earningst + β2DTE_AIt + β3 DTE_ADt + β4 DTE_BIt + β5(∆Earningst*DTE_AIt) +
β6(∆Earningst*DTE_ADt) + β7(∆Earningst*DTE_BIt) + ε (4a and 4b)
Intercept 5.33 21.42 <.0001
∆Earnings + 25.39 3.07 0.003
DTE_AI - -0.60 -0.99 0.323
DTE_AD + -0.26 -0.52 0.601
DTE_BI - -1.33 -1.12 0.265
∆Earnings*DTE_AI ? 5.83 0.32 0.751
∆Earnings*DTE_AD ? -37.85 -2.31 0.024
∆Earnings*DTE_BI ? 44.45 0.73 0.468
N 94
Intercept 5.03 12.27 <.0001
∆Earnings + 33.855 3.43 0.001
DTE_AI - -0.12 -0.19 0.851
DTE_AD + 0.19 0.37 0.715
DTE_BI - -1.43 -1.95 0.054
∆Earnings*DTE_AI ? -7.73 -0.42 0.677
∆Earnings*DTE_AD ? -45.79 -2.75 0.007
∆Earnings*DTE_BI ? 39.12 0.79 0.430
N 94
DTE_AI is a wave indicator variable that equals 1 if the level of debt-to-equity is above the median and the
change in debt-to-equity is positive (increasing).
DTE_AD is a wave indicator variable that equals 1 if the level of debt-to-equity is above the median and the
change in debt-to-equity is negative (decreasing).
DTE_BI is a wave indicator variable that equals 1 if the level of debt-to-equity is below the median and the
change in debt-to-equity is positive (increasing).
Due to over-borrowing, a collateral constraint is triggered which causes agents to

liquidate capital thereby reducing the prices of assets and the availability of credit. These
conditions are more likely to lead to a financial crisis. Therefore, when over-borrowing has
Page 279
occurred and negative economic consequences ensue, corporate profitability is less likely to
forecast positive GDP growth.
ROBUSTNESS TESTS
Following KP (2014), we introduce familiar leading indicators of GDP to determine

whether volatility and leverage are incremental predictors of GDP. The three variables used are
as follows: Yield, which is the yield on one-year constant maturity Treasury bill (T-bill),
measured one month after quarter q ends; Spread, the yield on the ten-year constant maturity
Treasury bond minus the yield on the one-year constant maturity T-bill measured one month
after quarter q ends; and Return, the quarterly buy-and-hold stock market return over the three
months leading to one month after quarter q ends. We obtain the data on treasury yields from the
Federal Reserve Board’s H15 Report and stock market returns data from the CRSP Monthly
Index file of value weighted returns. The results of the robustness tests are presented in Table 7.
Panel A of Table 7 shows the results of robustness tests utilizing market volatility (VIX).
When the appropriate control variables are included, volatility continues to significantly affect
the predictability of earnings on GDP growth. Under all specifications, the coefficient on
volatility is negative and significant. When all three control variables are included, the
coefficient on VIX is -0.10 (p=.003). In addition, earnings continue to be more predictive of
GDP growth when volatility is increasing. This is demonstrated by the coefficient on
Earnings*VIX which is significant for all specifications and is equal to 1.85 (p=.003) when all
control variables are included in the regression. When the dependent variable is GDPt+2, results
on VIX and Earnings*VIX are similar. The coefficients on both variables are significant
under all specifications, however, the magnitudes are marginally smaller. In summary, the results
analyzing the effect of changing volatility on the earnings-GDP relation are consistent when
additional predictors of GDP are included in the model. Increasing levels of volatility actually
increase the predictability of earnings on GDP. The hypothesized reason for this concerns the
constraining effect of increasing volatility on discount rates.
Panel B of Table 7 presents the results of robustness tests analyzing debt-to-equity. With
GDPt+1 as the dependent variable, DTE continues to be negatively related to future GDP for all
specifications. Furthermore, the interaction variable of interest, DTE*Earnings, continues to
be significant at the 10% level in specifications including Spread and Return. However, the
coefficient on DTE*Earnings is insignificant when Yield is included in the regression. This is
likely the result of the manner with which the level of aggregate debt co-varies with the T-bill
yield. When GDPt+2 is assessed as the dependent variable, DTE continues to significantly
impact the earnings-GDP relation. In specifications using Spread and Returns, the coefficient on
DTE*Earnings remains significant at the 5% level. Overall, results including control variables
known to predict GDP are consistent with previous results. Changing levels of debt-to-equity
have a significant impact on the degree to which earnings predict GDP.
Page 280
Table 7
ROBUSTNESS TESTS
Panel A: GDP = α1 + β1∆Earningst + β2∆VIXt + β3∆Earningst*∆VIXt + β4Yieldt + β5Spreadt + β6Returnst + ε
Dependent Variable = GDPt+1 Dependent Variable = GDPt+2

Variable A B C D A B C D
Intercept 4.01*** 5.65*** 4.91*** 4.37*** 4.18*** 5.25*** 4.80*** 4.02***
∆Earnings 17.27* 15.39* 12.59 15.31* 24.64*** 23.15*** 21.14*** 23.27***
∆VIX -0.11*** -0.12*** -0.09*** -0.10*** -0.07** -0.08** -0.06** -0.06**
∆Earnings*∆VIX 2.00*** 2.32*** 2.42*** 1.85*** 1.41*** 1.69*** 1.68*** 1.25**
Yield 0.24*** - - 0.17 0.17 - - 0.17
Spread - -0.43** - -0.15 - -0.24 - 0.03
Return - - 5.60* 4.78 - - 4.56 4.1
N 93 93 93 93 93 93 93 93
Adj. R-Squared 33.40% 32.47% 31.10% 33.98% 21.69% 20.48% 20.95% 21.21%
Panel B: GDP = α1 + β1∆Earningst + β2∆DTEt + β3∆Earningst*∆DTEt + β4Yieldt + β5Spreadt + β6Returnst + ε

Dependent Variable = GDPt+1 Dependent Variable = GDPt+2
Variable A B C D A B C D
Intercept 3.73*** 5.42*** 4.69*** 3.40*** 4.20*** 5.08*** 4.73*** 3.46***
∆Earnings 21.70*** 19.73*** 13.25* 16.10*** 23.24** 21.88** 17.28** 19.20**
∆DTE -2.90** -3.11** -2.93** -2.78** -0.87 -0.97 -0.86 -0.72
∆Earnings*∆DTE 103.39 147.19* 152.8* 90.67 73.86 107.81** 99.63** 63.06
Yield 0.28** - - 0.27 0.17 - - 0.22
Spread - -0.33* - 0.07 - -0.1 - 0.22
Return - - 10.27** 9.49** - - 7.74 7.23
N 94 94 94 94 94 94 94 94
Adj. R-Squared 19.88% 15.84% 23.19% 26.84% 10.51% 8.45% 13.84% 14.08%
*, **, and *** indicate p-values less than 0.10, 0.05, and 0.01, respectively.
GDP, ∆Earnings, ∆VIX, and ∆DTE are as defined previously.
Yield is equal to the one-year constant maturity treasury bill measured one month after the quarter ends.
Spread is equal to the yield on the ten-year treasury bond minus the yield on the one-year treasury bill.
Returns is equal to the quarterly buy-and-hold stock market return.
In untabulated results, the “waves” of VIX and DTE are analyzed along with the
appropriate control variables. While the significance of the wave indicators is diminished by the
presence of the control variables, results are consistent with those reported previously. In the full
specification, with all three control variables included, the interaction between earnings and the
“wave” of volatility below the median and increasing (VIX_BI) is positive and significant
(coeff.=40.70; p=0.054). When GDPt+2 is used as the dependent variable, VIX_AI continues to be
negatively related to GDP but the interaction variable (Earnings*VIX_AI) is insignificant.
Similarly, when “waves” of DTE are analyzed after controls for Yield, Spread, and Return are
included, results are consistent with previous results but the significance is diminished. When
Page 281
GDPt+1 is used as the dependent variable Earnings*DTE_AD is no longer significant.

However, when GDPt+2 is used as the dependent variable, the coefficient on Earnings*DTE is
negative and significant (-33.77; p-value=0.034). This is consistent with previous tests.
Therefore, while the inclusion of additional predictors of GDP mitigates the strength of our
results when breaking volatility into “waves”, the inferences drawn are consistent with previous
tests.
CONCLUSION
The current analysis answers the call of studies such as Ball and Sadka (2015) to further
our understanding of aggregate earnings. The findings demonstrate the importance of
considering macroeconomic conditions when analyzing the ability of aggregate corporate
earnings to predict future GDP growth. The study demonstrates that both 1) market volatility and
2) average firm leverage as measured by the debt-to-equity ratio significantly impact the ability
of earnings to predict GDP. The study extends research seeking to understand exactly how
growth in earnings transfers into growth in GDP. More specifically, the analysis shows the
degree to which aggregate earnings forecast GDP growth is mitigated by the discount rate
signaling effects of earnings as well as the degree of average firm leverage. The results of the
study should serve as a catalyst for further research into the manner with which changing interest
rates and constraints in capital impact the economic information content of corporate earnings.
REFERENCES
Akinci, O., & A. Queralto (2014). Banks, Capital Flows and Financial Crises. International Finance Discussion
Papers No. 1121, Federal Reserve Board.
Ball, R., & G. Sadka (2015). Aggregate earnings and why they matter. Journal of Accounting Literature. 34, 39-57.
Bianchi, J., & E. Mendoza (2011). Over-borrowing, Financial Crises and ‘Macro-Prudential Policy’. IMF Working
Paper Series.
Cready, W. M., & U. G. Gurun (2010). Aggregate market reaction to earnings announcements. Journal of
Accounting Research 48(2), 289-334.
Gkougkousi, Z. (2013). Aggregate Earnings and Corporate Bond Markets. Journal of Accounting and Economics
52(1), 75-106.
Guo, H. (2002). Stock market returns, volatility, and future output. The Federal Reserve Bank of St. Louis,
September/October, 75-86.
Herrerias, M.J., & V. Orts (2008). Equipment Investment, Output, and Productivity in China. Empirical Economics
42(1), 181-207.
Konchitchki, Y. (2011). Inflation and nominal financial reporting: Implications for performance and stock prices.
The Accounting Review 86(3), 1045-1085.
Konchitchki, Y. (2013). Accounting and the macroeconomy: The case of aggregate price-level effects on individual
stocks. Financial Analysts Journal 69(6), 40-54.
Konchitchki, Y., & P N. Patatoukas (2014). Accounting earnings and gross domestic product. Journal of Accounting
and Economics 57, 76-88.
Kothari, S.P, L. Shivakumar, & O. Urcan (2013). Aggregate Earnings Surprises and Inflation Forecasts. Working
Paper, MIT and London Business School.
Lichtenberg, F.R. (1992). R&D Investment and International Productivity Differences. NBER Working Paper No.
4161.
Long, J.B. & L.H. Summers (1991). Equipment Investment and Growth. Quarterly Journal of Economics 106(2),
445-502.
Mendoza, E. G. (2010). Sudden stops, financial crises, and leverage. American Economic Review 100, 1941-1966.
Mishkin, F. S.(2011). Over the Cliff: From the Subprime to the Global Financial Crisis. Journal of Economic
Perspectives 25(1), 49-70.
Page 282
Metui, N., Bjorn, H., & M. Grill (2014). Financial Shocks, Credit Regimes, and Global Spillovers. American
Economic Association Working Paper Series.
Patatoukas, P.N. (2013). Detecting news in aggregate accounting earnings: implications for stock market valuation.
Review of Accounting Studies 19(1), 134-160.
Penman, S.H. & Yehuda, N. (2015). A Matter of Principle: Accounting Reports Convey Both Cash-Flow News and
Discount-Rate News. Working Paper, Columbia University and University of Texas at Dallas.
Perera, A. (2010). Monetary Policy in Turbulent Times: Impact of Unconventional Monetary Policies. Working
Paper, available at SSRN: http://ssrn.com/abstract=1669948.
Ramey, G. & V.A. Ramey (1995). Cross-country evidence on the link between volatility and growth. American
Economic Review 85, 1138-1151.
Page 283
CAPITAL BUDGETING PRINCIPLES: BRIDGING

THEORY AND PRACTICE
Atul K. Saxena, Georgia Gwinnett College
ABSTRACT
One of the important qualities that accrediting bodies, e.g., the AACSB are looking for in
business schools is learning outcomes that relate theory with practice. In fact, they encourage
business schools to include pedagogical tools in the curriculum that foster practical applications
of complex theoretical concepts, thereby making them intuitive and somewhat easier to grasp by
students. Additionally, prospective employers, college professors, and students themselves are
interested in learning valuable skills such as conducting research, team building, leadership,
and interdependence that they can take with them to their job. This paper describes a capital
budgeting project that is a real world simulation of a new business startup. It allows students to
acquire the valuable skills mentioned above. The proposed project is suitable for graduate
(MBA) and upper-level undergraduate courses. The project has been assigned in an MBA
program with great success in the core corporate finance. But it can also be amended and
utilized in the capstone strategic management course. For undergraduate finance students, this
project can be assigned in the second (intermediate) finance course. The project is particularly
appealing to non-traditional business students, who often desire to establish their own firms. The
project directs their focus on the achievement and profitability of their future dreams while
applying in practice what they learn in theory.
INTRODUCTION
Social scientists and academicians have stressed offering students multiple techniques of
pedagogy for better learning outcomes. These techniques include one-minute papers, more
detailed research papers, simulations, power point presentations, and real-world projects among
others. Research has shown that generation-X actually prefers experiential learning to the more
traditional lecture-based pedagogy (Bale and Dudney, 2000). Frequently professors spend a lot of
time and effort searching for projects to supplement their lectures to enrich their coursework. The
accrediting bodies encourage schools to include such pedagogical tools that bridge theory with
practice. While instructors have always desired such tools, lately there is an increased demand
from employers and student graduates to obtain these valuable hands-on experiences by
simulating the real world before entering it.
While useful to both graduate and undergraduate students, practical learning experience is
more important for the former group. This paper describes a project that can be used in upper
level undergraduate finance (and strategic management) courses, but is particularly geared
towards graduate students. The project requires students to apply financial analysis to the startup
of a small company. This project has already been assigned successfully in MBA (and
undergraduate) courses at a business school for a number of years with good results. Since most
of the students enrolled in the core MBA corporate finance course are classified as
nontraditional, they frequently have a dream of establishing their own company, of being
Page 284
entrepreneurs. This project provides them the opportunity to apply theoretical concepts and focus
on the costs and benefits of their future plans.
Viewing various business functions on a small scale provides insight in understanding the
interactions among functions in larger, established firms. While this paper involves the
application of financial analysis, the project can be modified for any other business discipline,
such as management or marketing. Over the course of study in an MBA program, it can also be
used as a continuing project, adding facets to the study in each discipline. It may be modifi67ed
to provide an examination of the practicalities in setting up businesses for other professionals,
such as medical offices, engineering firms, etc. Its main benefit for the student is to encourage
the disciplined thought and planning required in establishing a successful business.
The rest of the paper is organized as follows. Section II reviews the relevant literature.
Section III discusses some desirable attributes of a class project assignment. Section IV explains
the project in detail. Based upon procedural logic, Section IV is further divided into four
sequential sub-sections. Section V summarizes the paper and provides some concluding remarks.
LITERATURE REVIEW
In a classic study on how to frame classroom learning experiences that model necessary
attributes for the foundations of success, Bruner et al. (1999) found the following as important:
1. Select cases that employ, exercise or explore a tool or concept

2. Highlight the dilemmas of the decision maker
3. Set the numbers and critique them
4. Embrace uncertainty
5. Demand the action recommendations arising from analysis
6. Look for unintended consequences
7. Explore opportunities for further work
If a project possesses several of the above attributes, it is considered good. Remarkably,

the project explained in this paper meets most of the above criteria.
Bale and Dudney (2000) surveyed students to research their preferred mode of learning.
They find from their survey results that for Generation X students (born between 1961 and 1981)
“hybrid” teaching models incorporating both andragogy (self-directed, self-motivated) and
pedagogy methods are most effective. Making reference to another related study they go on to
conclude that Generation X wants to see value and relevance in education, otherwise they are not
motivated to learn new skills. They prefer experiential learning using as many of the five senses
as possible (Caudron 1997). The startup project explained here is an example of hybrid teaching
model that incorporates both andragogy and pedagogy.
Deeter-Schmelz, Kennedy, and Ramsey (2002) conclude that team projects play a vital
role in modern pedagogy. Moreover, as team projects become even more common in business
courses, an increased understanding of factors contributing to team effectiveness is necessary for
instructors to assist students in realizing the potential benefits of this pedagogical tool. Their
results support the positive and direct role of cohesion as an input variable on teamwork. Ashraf
(2004) finds that in business schools across the United States, one of the most common
pedagogical tools is the use of group projects. "Passive" instruction (i.e., lecture only) is
Page 285
considered to be an inferior mode of teaching. He highly recommends the use of group-based

projects as pedagogical tools. Since we suggest that our project be preferably given as a team
assignment, recommendations of both Deeter-Schmelz et al (2002) and Ashraf (2004) are met.
A simulation, like any pedagogical tool, must be evaluated in terms of its effectiveness in
achieving course objectives. In a study, Chapman and Sorge (1999) investigate how well a
particular simulation does in achieving course objectives and compare its performance to the
textbook and papers used in the course. They find that compared to the textbook and research
papers, students consistently gave simulation the highest ratings. In another study, Olson et al
(2006) discuss and encourage the use of simulation as a pedagogical tool. While their simulation
is developed for Eastern European transition economies, it is applicable to any pedagogical
learning situation, specifically to the general operations of the firm at the microeconomic level of
decision-making. Our project conforms to both studies, Chapman and Sorge and Olson et al.
While most of the above studies pertain to general education and business courses, there
is some literature that is specifically relevant to finance courses. For example, Gurnani (1984)
extensively reviews and compares capital budgeting concepts as advocated in theory with the
methods employed by industry. Capital budgeting is an interdisciplinary function, involving
diverse areas such as engineering, finance, and management. The ability of a firm to make sound
decisions in this area rests not only on the theoretical techniques employed but also on the
judgment, intuition, and creativity of the analysts and decision makers. He claims that the
academic literature has concentrated heavily on developing and refining quantitative evaluation
criteria, methods of measuring return, risk analysis techniques, and procedural aspects of capital
investment decision making. However, academic research has been criticized because it tends to
be essentially concerned with accuracy of analysis, sophistication of methodology, and
improving conditions in a laboratory setting without regard to the realities of corporate decision
making. One reason for the gap is a lack of bridging theory with practice at the school level. We
feel that this project is the perfect bridge.
Benton Gup (1994) surveys academics and practitioners and ranks those finance concepts
considered most important for students to acquire. The academics rank time value of money
capital budgeting, CAPM, capital structure, and valuation as the top five financial concepts for
this purpose. It is striking that all five are included to some degree in the project discussed in this
paper. The practitioners ranking excluded CAPM and valuation but included accounting and cost
of capital. This project requires a critical understanding of the cost of capital concept.
In what has to be one of the most comprehensive and impressive studies in corporate
finance, Harvey and Graham (2001) sampled 4440 firms receiving responses from 392 chief
financial officers (CFO’s) to examine the proverbial bridge between theory and practice. Their
findings are both reassuring and surprising. It is reassuring to them that NPV is dramatically
more important now as a project evaluation method than it was 10 or 20 years ago. The CAPM is
also widely used in the real corporate world. However, they find it surprising that more than half
of the respondents would use their firm's overall discount rate to evaluate an investment in an
overseas market, even though the investment likely has different risk attributes than the overall
firm. This indicates that practitioners might not apply the CAPM or NPV rule correctly, perhaps
indicating a need for a better bridge between theory and practice. A class assignment such as
proposed in this paper would be useful to reinforce this bridge.
Weaver and Michelson (2004) suggest a project that could accompany a corporate finance
course to enhance the learning of theoretical concepts. It is a simple Excel model that provides
Page 286
measures of the standard deviation of forecasted internal rate of return (IRR) given traditional
data inputs such as annual cash flows, terminal values and equity. The model first calculates IRR
using traditional discounted cash flow methods and then provides heuristic estimates of
variability measured in terms of "high," "low" and "most likely" values. It also provides an actual
measurement of risk in terms of mean and standard deviation and upper and lower quartiles,
along with a graphical presentation of various risk parameters. While the Excel model just
described is a good class project, our startup project is more comprehensive in nature covering a
wider variety of financial concepts.
DESIRABLE ATTRIBUTES OF A CLASS PROJECT
Project assignments vary widely in their complexity and the amount of time needed for
completion. For example, an economic ordering quantity (EOQ) model with imperfect quality
items can be rather challenging for a typical corporate finance course, it may be well suited for a
decision science course (Wang, Tang, and Zhao, 2007). Most finance class projects do not
necessarily have to be as complex as EOQ models. The project outlined in this paper is rigorous
yet relatively simple. It is a real world simulation of a firm and the decision making that goes on
within it by its financial managers. As discussed above, Chapman and Sorge (1999) recommend
the prudent use of such pedagogical tools. However, designing an appropriate project can be
tricky and time consuming. From our own experiences in the classroom, we have found that
certain key factors must be considered when designing a project assignment.
First, a well-designed class project must logically follow the concepts learned in class
and/or the text. There ought to be opportunities for students to clearly and easily relate to certain
key theoretical concepts and apply them in practice through the project. Second, it must be
doable within the term of the course, which is the case of the proposed assignment. Another issue
is whether a project can be done individually or in a group setting. Most instructors encourage
projects to be done in small groups of 3 or 4, depending on the class size. Despite the potential
for the classic free-rider problem (Ashraf 2004), group projects support the important goals of
team building, leadership, responsibility and mutual trust. Business program accrediting bodies,
e.g., AACSB, put enormous weight on these values. Moreover, there are alternative means of
mitigating free ridership, e.g., peer evaluation by team members. However, a situation may arise
that is not suitable for teams and group assignment. For instance, if the class size is very small or
students are extremely busy (executives, etc.) who do not have enough flexible time to meet in
teams. A desirable project can be done individually, as is ours.
THE PROJECT
There are several steps involved in this project assignment. The first step involves
selecting the type of business to be established. Step two entails setting the assumptions under
which the financial analysis will be performed. The third step involves calculating a financing
rate (the cost of capital), estimating the revenues and expenses over an extended period of time
(say a 5-year period). The fourth step consists of applying various capital budgeting techniques
to reach an accept/reject decision. The final step consists of evaluating and assessing the risk
involved in the cash flows and profitability. Each step is explained in detail in the following
subsections.
Page 287
A: Selection of the Business Type
It is helpful to select a business that does not depend on results of research and
development activities, exploration, etc. These unknown or future factors add considerable
complexity to the project and undermine the task of estimating probable cash flows from the
business by making the whole project seem unreal. Business types such as retail, most
manufacturing, consulting, construction, or service make the project more manageable for the
student. For those students who do not have a specific type of business they would like to
establish, a business run by a family member or friend can be a good choice since discussions
with these owners can provide a solid base for estimating the startup requirements, revenues,
costs and growth potential.
Occasionally, students run into problems with certain business selections. For instance,
franchises can be problematic if estimates of revenues, costs, franchise fees, and other data are
not provided by the franchiser. Buying an existing business for project analysis moves the
student outside the procedures provided in classroom discussion in the MBA’s core corporate
finance course and therefore makes the project more difficult for them. This activity is best
analyzed with acquisition procedures rather than capital budgeting used in this project. Indeed,
this variation of the project can be used for a finance course on Mergers and Acquisitions.
Not-for-profit businesses are frequently avoided by students because they assume that
they are not suitable for a profit analysis. However, since these businesses must take in at least as
much money as they spend to stay in existence, they are as appropriate for this project as a for-
profit business. Businesses that require very large capital outlays at startup for assets with lives
longer than the project horizon (say 5 years) will generally not be profitable within the analysis
period. This problem can be overcome and is discussed in Section IV-C.
B: Statement of Assumptions
A statement of assumptions used to estimate cash flows is an important habit for students
to build. While in the project its function is strictly to build the initial cash flow estimates and
provide a base for risk analysis, in an actual establishment of a firm it allows periodic
reassessment of the progress expected. Should what initially appeared to be a profitable venture
fail to meet projections or economic conditions worsen beyond expectations, the owner may
need to either take alternative measures or shut down before losses become excessive. For a
project manager in an established firm, changing assumptions may invalidate prior capital
budgeting cash flow estimates. It is the responsibility of the project manager to keep upper
management informed of these changing circumstances and to re-estimate the probable profit of
the project. Failure to do so can significantly impact the profitability of the firm and in turn have
a devastating effect on the career of the project manager. Finally, assumptions are also required
for the instructor to evaluate the student’s ability to apply the concepts. Assumptions generally
include such things as the economic conditions, growth in revenues/costs, hiring of employees,
increases in fixed assets, cost of capital, termination revenues and expenses, initial inventories
and fixed assets, etc. Table 1 contains an example of the set of assumptions to be used for this
case.
Page 288
As suggested in the simplified example in Table 1, the best estimate for sales growth is
projected to be 10% annually. Students might more reasonably predict sales growth of 25% in
year 2, 15% in year 3 followed by 5% growth in the last two years. As examples, assumptions
might also include a significant increase in payroll in year 3 as a planned administrative staff
addition occurs. At the same time one might see increased office expenses and depreciation.
Students need to be encouraged to be creative, imaginative, yet realistic when making these
assumptions.
Table 1
AN EXAMPLE OF A SET OF ASSUMPTIONS
Business Type: Retail Outlet
Expected Case Assumptions*:
1. Sales Growth 10% of Sales per year

2. Cost of Goods Sold 60% of Sales per year
3. Utilities $5,000 per year
4. Advertising $10,000 per year
5. Miscellaneous Expenses $9,000 per year
*Note that this is only a partial set of assumptions for illustration purpose.
C: Cost of Capital and Cash Flow Estimates
Since the project involves a startup company, a basic assumption is that at least initially,
it is a sole proprietorship and the cost of capital is composed of the student’s own required rate of
return plus the cost of borrowing money. Students are asked to call a financial institution to
determine what lending rate would be required for a business of the type chosen. The weighted
average of these two rates is used as the discount rate for capital budgeting purposes. Students
may wish to assume additional investors and incorporate their required rates as well when
computing the overall cost of capital.
Students are also asked to estimate cash flows for the initial startup costs and
revenue/expenses for five years at which time the business is shut down or sold. The five year
life span may appear somewhat arbitrary at first. However from experience, this is a long enough
horizon to include most of the changes a new company may encounter so students have the
opportunity to manage the growth. At the same time, a 5-year life span of the project is not so
long as to make long-term estimates of cash flows too unrealistic and far-fetched. The process
and organization of cash flows in this paper follow that presented by Titman, Martin, and Keown
(2014).
To demonstrate knowledge of technology (a desirable tool by AACSB), spreadsheets are
required for the organization and estimation of cash flows. The initial outlay includes all cash
flows that occur at the beginning. Table 2 provides a complete output of the capital budgeting
analysis. It shows that our sample project requires modifications to the proposed property as well
as furniture and fixtures to open. It also has deposits and opening expenses. These could be
Page 289
utility and phone deposits, operating licenses, and the initial advertising campaign. Working
capital requirements might include cash.
The next cash flow category includes revenues and expenses occurring throughout the
five-year life of the project on an annual basis. Generally called after-tax cash flows, these
include annual revenues, annual expenses, depreciation, and taxes. The format of these cash
flows follows the general format of an income statement except that interest expense is not
included. All after-tax financing expenses are recovered by the level of the interest rate used to
discount the cash flows. The final cash flow category is the terminating cash flows. These
include all one-time cash flows occurring at shut down and could include after-tax salvage value,
disposal/restoration expenses, sale of business revenue, etc. Since these cash flows occur in year
5, they should be netted with the year five after-tax cash flows. At this point students should have
six cash flows: total initial outlay and cash flows for years 1-5 (year 5 includes the terminal cash
flow). Additional instructions given to students in this phase can include:
- After-tax cash flows in years 1-5 must vary. Texts frequently repeat the use of year 1 cash flows
in all succeeding years of the project life for ease of classroom instruction. Requiring variability
forces a more realistic picture of a firm.
- Record cash flows as they occur. While the after-tax cash flows format resembles an accounting
income statement, it does not follow accounting practices. Cash flows should coincide with cash
going into and out of a bank account.
- At termination students can assume a complete shutdown with or without salvage value or the
sale of the company. For firms that had costly and long-lived fixed assets, realistic profitability
will require the sale of the assets or the company in year 5.
- Categories estimated in the after-tax cash flows should be moderate in breadth. For instance,
estimates for total revenue and total cost are too broad. For a retail outlet, estimating revenue and
costs for every item sold is too detailed.
- Straight line depreciation or MACRS can be used.
Students who are seriously considering starting the business analyzed in the project are permitted
and encouraged to be as detailed as they feel necessary.
Page 290
Table 2
AN EXAMPLE OF A COMPLETE CAPITAL BUDGETING SPREADSHEET
Initial Expenses ($) Year 0 Year 1 Year 2 Year 3 Year 4 Year 5
Fixed Assets $100,000
Renovations $20,000
Working Capital $10,000
Bank Balance $25,000
Inventory $30,000
Miscellaneous $9,000
Initial Cash Outlay $194,000
Revenues $500,000 $550,000 $605,000 $665,500 $732,050

Cost of Goods Sold $300,000 $330,000 $363,000 $399,300 $439,230
Payroll $50,000 $52,500 $55,125 $57,881 $60,775
Utilities $5,000 $5,000 $5,000 $5,000 $5,000
Lease/Rent $10,000 $10,000 $10,000 $10,000 $10,000
Advertising $10,000 $10,000 $10,000 $10,000 $10,000
Maintenance $1,000 $1,000 $1,000 $1,000 $1,000
Insurance $500 $500 $500 $500 $500
Other Overheads $500 $500 $500 $500 $500
Depreciation $2,000 $2,200 $2,420 $2,662 $2,928
Earnings before Tax $121,000 $138,300 $157,455 $178,657 $202,116
Less Taxes @ 40% $48,400 $55,320 $62,982 $71,463 $80,847
Earnings after Tax $72,600 $82,980 $94,473 $107,194 $121,270
Add Depreciation $900 $900 $900 $900 $900
Salvage Value $51,000
Net Cash Flow ($194,000) $73,500 $83,880 $95,373 $108,094 $173,170
Cumulative NPV* ($194,000) ($127,182) ($57,860) $13,796 $87,625 $195,150
*Assuming a 10% discount rate
D: Capital Budgeting Techniques and Acceptability Analysis
Once the net cash flows are obtained, the acceptability of the business is evaluated. Students
are required to use several decision criteria methods: payback period, discounted payback period,
net present value (NPV), profitability index, internal rate of return (IRR), and modified internal
rate of return.
- Payback period provides the number of years required for the initial outlay to be recovered from
the after-tax cash flows. Since this is strictly an accumulation of the cash flows in years 1-5, it
fails to account for the time value of money and is considered to be a less than accurate method
and, financially speaking, a naïve way of evaluating the acceptability of the project. Acceptability
of the business depends on owner-set criteria. For example, the initial outlay must be recovered
within 3 years. If the pay back is equal to or less than this hurdle, the business is acceptable.
Despite its limitations, the payback period method remains a popular capital budgeting technique
(Harvey and Graham, 2001). It is frequently used as a preliminary screening measure in large
firms and as the sole requirement in cash poor firms.
Page 291
- Discounted pay back corrects for the lack of use of the time value of money in the pay back
method by discounting each year’s cash flow to year zero using the cost of capital as the discount
rate. Therefore, this technique is regarded as an improvement on its predecessor and not as naïve.
It is interpreted in the same manner as pay back but will obviously take longer to recover the
initial outlay since the cash flows are in present value terms. Once again, the owner must set the
acceptability criterion.
- Net present value (NPV) is the present value of the cash inflows minus the present value of the
cash outflows and provides the dollar estimate of the change in the value of the firm. The business
is acceptable if the NPV is positive.
- Profitability index is the present value of the cash inflows divided by the present value of the cash
outflows and provides the dollar return for each dollar invested. The business is acceptable if the
profitability index is greater than one.
- The internal rate of return (IRR) is the discount rate that equates the present value of the future
cash flows to the initial outlay. It provides the percent return on funds invested assuming that the
cash flows are reinvested at the internal rate of return as they flow into the firm. This is known as
the reinvestment rate assumption. If these funds cannot be reinvested at that rate, the return will
not be achieved. For this reason, sometimes the IRR rule is regarded as too optimistic, and the
modified IRR is computed as discussed in the next paragraph. The internal rate of return must be
greater than the firm’s cost of capital for the business to be profitable.
- When the reinvestment rate assumption cannot be met, or when a relatively more conservative
technique is desired, the modified internal rate of return is calculated. All the cash flows are
compounded to the final year (year 5 for the project) using a reasonable rate for reinvestment,
generally the cost of capital, and totaled to arrive at the future value of all cash flows. The
modified-IRR is the implied rate that equates the initial outlay with the future value just
calculated. This modified-IRR must be greater than the cost of funds.
If the business is unprofitable, students are asked to discuss some methods that might
make it profitable. For example, operating from a home office or obtaining lower cost facilities
might delay costs, or slowing/increasing the growth rate might provide a greater spread between
revenues and costs. Students are not required to apply these suggestions.
E Risk Assessment
Students are also asked to analyze business risk using one of four risk analysis techniques
and to discuss their findings. The methods suggested are sensitivity analysis, scenario analysis,
decision tree analysis, and simulation. In all cases, the student can also determine the probability
of the net present value falling below zero since this requires the average of several estimates of
the net present value and its standard deviation. Although these techniques carry different
nomenclature depending on the source, their definitions below should be familiar to faculty.
- In sensitivity analysis, the assumptions used in the analysis are changed one at a time to
determine those with high impact on the net present value. These are called driver variables and
generally require a high degree of confidence in the estimate or the ability to be well managed for
an overall assessment of low business risk.
- Scenario analysis involves modifying the expected scenario already presented with the worst case
and best case estimates of the assumptions used to create the model. This has the advantage of
incorporating the interactions of all the variables into the analysis.
Page 292
- Decision tree analysis provides re-evaluation points as the establishment of the business
progresses. Owners can incorporate their experience at these points to re-estimate profitability.
They may decide to expand/contract the business, modify facilities, shut down, etc. The decision
tree provides “legs” to determine the net present values for each of the possible paths that the firm
might take. The expected net present value and its standard deviation can assist in the risk
assessment.
- Simulation provides estimates of the net present value by randomly selecting a value from each
variable’s probability distribution and combining them for the trial NPV calculation. Computer
simulation software is generally instructed to make 1,000 to 10,000 trail runs, creating a net
present value probability distribution. The area under the curve below a net present value of zero
provides an assessment of the risk of the business.
Summarization of the acceptability of the business including both the decision criteria and
the risk analysis concludes the project. Since risk analysis provides no definitive answer for how
much risk is acceptable, students must apply their own risk preferences to this decision.
Depending upon the preparedness of students, this section can be excluded from undergraduate
finance courses if it becomes too overwhelming for them.
SUMMARY
This paper describes a capital budgeting project for the startup of a new business (e.g., a
sole proprietorship). It is a real-world project that is do-able in a semester. It is preferably
assigned as a group project, but can be adapted for individual student assignment. The
company/business type is chosen by the student(s). Based on the types of assets and services
required, students estimate the initial startup cost, the recurring revenues and expenses over the
life of the business and any terminating cash flows. Once the cash flows are estimated, the
business is evaluated for profitability and risk using the capital budgeting techniques of the net
present value (NPV) and the internal rate of return (IRR). Students then must decide if they
would proceed with that “dream” business.
The project can be assigned to MBA students in their core corporate finance course or
with slight modifications it can also be included in courses such as management, marketing or
entrepreneurship. A remarkable characteristic is that the project can be used as a thread
connecting much of the MBA curriculum, creating a management business plan, a marketing
plan, a cash budget, etc. in different classes. The described project has also been used in
undergraduate finance classes by eliminating the risk analysis. Certain non-business professional
programs, such as health care or engineering, where students frequently plan to open their own
business, may also find it beneficial to include it in their curriculum.
Page 293
REFERENCES
Almer, E. D, K. Jones, and C. Moeckel, (1998). Impact of one-minute papers on learning in an introductory
accounting course. Issues in Accounting Education, 13( 3), 485-497.
Ashraf, M., (2004). A Critical Look at the Use of Group Projects as a Pedagogical Tool. Journal of Education for
Business, 79(4), 213-216.
Bale, J. M. and D. Dudney, (2000). Teaching Generation X: Do Andragogical Learning Principles Apply to
Undergraduate Education?” Financial Practice and Education, 10(1), 216.
Bruner, R, B. Gup, B. H Nunnally, and L. C. Pettit, (1999). Teaching with cases to graduate and undergraduate
students. Financial Practice and Education, 9(2), 138.
Caudron, S., (1997). Can generation X’ers be trained? Training and Development, 51(3), 20-25.
Chapman, K. J. and C. L. Sorge, (1999). Can a simulation help achieve course objectives? An exploratory study
investigating differences among instructional tools, Journal of Education for Business, 74(4), 225-230.
Deeter-Schmelz, D.R, K.N. Kennedy and R. P. Ramsey, (2002). Enriching our understanding of student team
effectiveness. Journal of Marketing Education, 24(2), 114-124.
Gup, B., (1994). The five most important finance concepts: A summary, Financial Practice and Education, 4(2),
106-109.
Gurnani, C., (1984). Capital Budgeting: Theory and Practice. The Engineering Economist, 30(1), 1-19.
Harvey, C. and J. Graham, (2001). The theory and practice of corporate finance: Evidence from the field. Journal
of Financial Economics, 60, 187-243.
Olson, D.L., M.F. Shipley, M. Johnson, P. Dimitrova, (2006). Simulation as a pedagogical tool for managerial
decision-making in a transition economy. The Journal of the Operational Research Society, 57(9), 109.
Titman, S., J.D. Martin, and A.J. Keown, (2014). Financial Management Principles and Applications. Pearson
Prentice Hall, 12th edition.
Wang, X., W. Tang, and R. Zhao, (2007). Random fuzzy EOQ model with imperfect quality items. Fuzzy
Optimization and Decision Making, 6(2), 139-154.
Weaver, W. and S. Michelson, (2004). A Pedagogical Tool to Assist in Teaching Real Estate Investment Risk
Analysis. Journal of Real Estate Practice and Education, 7(1), 43-52.
Page 294
Page 295
AN EXPERIMENT TO ASSESS THE EFFECT ON

INVESTING OF SWITCHING AUDITORS FROM BIG 4
FIRMS TO REGIONAL FIRMS
Arnold Schneider, Scheller College of Business
ABSTRACT
This study examines whether knowledge about companies switching auditors from Big 4
firms to regional firms affects decisions to invest in those companies. Participants were given a
scenario involving an investing decision and were first asked to assess the level of risk
associated with investing in the company. Next, they were asked to allocate $10,000 between
investing in the company versus a money market account. Four different questionnaires were
created by varying information about auditor switches. Results indicated that neither risk
assessments nor investment amounts differed for companies that switch auditors from Big 4 firms
to regional firms as compared to companies that did not switch auditors. For companies that did
switch auditors, providing a reason for the switch did not impact investing decisions. Finally,
when a reason was provided for the switch, participants perceived a lower risk associated with
switching auditors to acquire more expertise as opposed to switching to obtain lower audit fees,
but this perception did not lead to increased amounts of investment.
INTRODUCTION
In recent years, many companies have switched their audit firms from Big 4 to smaller
ones (Cullinan, Du, and Zheng, 2012). A study conducted by the United States Government
Accountability Office (2008) reports that, from 2002 to 2006, there was a material increase in the
market share of publicly-traded client firms with $500 million or less in revenue by non-Big 4
audit firms, and a corresponding increase in the market share during that period by Big 4 audit
firms. Ettredge, Li, and Scholz (2007) report that from their sample, the majority of Big 4 clients
who dismissed their auditors subsequently hired non-Big 4 audit firms. Among the many reasons
cited for this phenomenon are client-auditor disagreements, lower audit fees, and a desire to
obtain better services (Fontaine, Ben Letaifa, and Herda, 2013; Stefaniak, Robertson, and
Houston, 2009).
The main purpose of this research is to ascertain whether knowledge about companies
switching auditors from Big 4 firms to regional firms affects decisions to invest in those
companies. Often, Big 4 firms are perceived as performing higher quality audits than regional
firms (Nichols and Smith, 1983). Thus, a switch may be perceived as a negative signal coming
from the company that switches auditors in this manner, as suggested by prior studies of stock
market reactions to these types of auditor switches (e.g., Knechel, Naiker, and Pacheco, 2007). In
addition, “the selection of a non-Big 4 auditor could signal that the company is not an attractive
client for a Big 4 firm, which may be more selective in their choice of clients. In this case,
moving from a Big 4 to non-Big 4 may be an unfavorable signal to the marketplace about the
client’s future prospects” (Cullinan, Du, and Zheng, 2012, 7). In recent years, however, these
perceptions may be changing due to events such as Arthur Andersen’s failure, the passage of the
Sarbanes-Oxley Act of 2002, and the creation of the Public Company Accounting Oversight
2
Page 296
Board (PCAOB). The PCAOB has encouraged the use of second-tier audit firms as alternates to
Big 4 firms, suggesting that there is no difference in audit quality between Big 4 and second-tier
firms (Jenkins and Velury, 2011). Cassell et al. (2013) find that, post-Andersen, perceived
financial reporting credibility of second-tier audit firm clients is indistinguishable from that of
Big 4 clients. Moreover, some studies (e.g., Ettredge, Li, and Scholz, 2007) indicate that
companies sometimes change audit firms to obtain audit fee savings or better services. “Under
such circumstances, a nonnegative or even a positive market reaction could be expected for a
switch from a Big 4 to a non-Big 4 auditor” (Chang, Cheng, and Reichelt, 2010, 84).
When a company switches auditors, it need not disclose the reason for the auditor change,
so companies often do not make these disclosures. The current study also examines whether
disclosing a reason for the switch impacts investors’ judgments and whether investors perceive a
reason relating to lower audit fees differently from a reason pertaining to a desire to obtain better
audit services.
The issues examined in this study are important to companies considering switching
auditors from Big 4 to smaller audit firms, and if they do decide to switch in this manner, the
results of this study can provide guidance on whether it is worthwhile to disclose the reason for
the switch. Findings from this study indicate that neither risk assessments nor investment
amounts differed for companies that switched auditors from Big 4 firms to regional firms as
compared to companies that did not switch auditors. For companies that switched auditors, the
results show no differential impact on investing decisions between the disclosure of a reason for
an auditor switch versus no disclosure of the reason for the switch. Furthemore, for companies
that provided a reason for an auditor switch, respondents perceived a lower risk associated with
switching auditors to acquire more expertise as opposed to switching auditors to obtain lower
audit fees. This perception, however, did not lead respondents to increase the amount they would
invest in the company.
The remainder of this paper is organized as follows. The next section reviews previous
literature dealing with the effects of auditor switches in general and then the effects of auditor
switches from from Big 4 firms to smaller firms. The following section develops three
hypotheses relating to the effects of auditor switches from from Big 4 firms to smaller firms.
Afterwards, the selection and demographic information about participants are provided. Then,
the experimental task undertaken by the participants is discussed. Next, the paper presents the
results of manipulation checks, hypothesis tests, and resposes to post-experimental questions.
Finally, the paper concludes with a summary of results, limitations, and directions for future
research.
PRIOR STUDIES
A long line of literature has investigated auditor switching in general (i.e., not necessarily
from Big 4 to other firms or visa versa). A number of studies have shown negative reactions to
auditor switching. Fried and Schiff (1981) find a negative market reaction to auditor switches,
but do not find a difference in market reactions to auditor switches accompanied by disagreement
disclosures versus those without such disclosures. Smith and Nichols (1982) report that the
market reacts more negatively to firms that change auditors and disclose disagreements with its
auditor than for firms that change auditors but do not disclose such disagreements. DeFond,
Ettredge, and Smith (1997) report negative returns for the period between auditors’ resignations
and the SEC’s receipt of the related Form 8-K filing and for the five day period beginning on the
SEC receipt date. Wells and Louder (1997) obtain negative stock price reactions for companies
3
Page 297
associated with auditor resignations and find that reasons provided for resignations or other
information related to resignations provide no additional information to the market. The authors
conclude that it is the resignation disclosure itself, not any other related information, that causes
the negative reaction. Shu (2000) reports that investors react negatively to auditor resignations
and that the magnitude of these negative reactions increase with higher client litigation risk.
Whisenant, Sankaraguruswamy, and Raghunandan (2003) show that disclosure of auditor
resignations together with reportable events dealing with financial statement reliability issues are
associated with negative returns above and beyond the effect of the resignation disclosure itself.
Disclosure of auditor dismissals together with those reportable events are associated with
negative returns above and beyond the effect of the dismissal disclosure itself for a seven-day
trading period, but not for a three-day trading period surrounding the disclosure date. Beneish et
al. (2005) demonstrate that when resignations are accompanied by disclosures about auditor-
client disagreements over accounting treatment or over the adequacy of internal controls, the
resignations are associated with negative abnormal returns for the former client. Griffin and Lont
(2010) also find that investors react negatively to auditor resignation announcements. The
negative reaction increases for companies with prior securities litigation and higher bankruptcy
risk. Also, the negative reaction is explained mostly by two reportable event disclosures – auditor
client-disagreements and nonreliance on management. Results in Schneider (2015) reveal that
auditor switches produce higer investment risk assessments and marginally lower amounts
invested than no auditor switches, but the effects of auditor resignations were not significantly
different from the effects of dismissals.
On the other hand, several studies have found no reactions or positive reactions to auditor
switching. Johnson and Lys (1990) find no contemporaneous abnormal stock price reaction to
auditor switching announcements. Klock (1994) also finds no significant stock price effect
associated with the switching of auditors. Asthana, Balsam, and Krishnan (2010) show that
clients switching from Arthur Andersen just prior to its demise experienced positive abnormal
returns during the three day window surrounding the announcement of the switch. The authors
attribute this positive response to the reduction in uncertainty associated with the cost of finding
a new auditor. Mande and Son (2013) provide evidence that stock markets have a positive view
of auditor switches for companies that have experienced financial statement restatements. Taken
as a whole, the evidence is very mixed as to whether auditor switches in general adversely impact
investing decisions.
Some researchers have addressed the effects of switches from Big 4 firms to smaller
firms. Eichenseher, Hagigi, and Shields (1989) show that market reactions surrounding auditor
switches to Big N firms are slightly positive, while auditor switches away from Big N firms yield
negative market reactions. Dunn, Hillier, and Marshall (1999) find a negative market reaction to
auditor resignations on the date of the resignation letter and that the loss of a Big N auditor is
associated with a greater negative reaction. Knechel, Naiker, and Pacheco (2007) provide
evidence that companies changing from specialist Big 4 auditors to non-big 4 auditors have the
largest adverse capital market reactions, while the market reacts most favorably when companies
change from non-Big 4 auditors to Big 4 audit firms that are not specialists. Boone, Khurana, and
Raman (2010) find that client-specific ex ante risk premium is lower for Big 4 clients than for
second-tier audit firm clients. This result is consistent with the notion that investors perceive Big
4 audit quality to be higher. Blau et al. (2013) show that short sellers increase activity following
changes from Big 4 to non-Big 4 auditors and that short selling during the post-announcement
period leads to higher short-selling revenues. This indicates that short sellers view these auditor
changes as negative events.
Page 298 4
While these studies demonstrate that switching away from Big 4 audit firms have
negative consequences for investors, other studies provide different results. Chang, Cheng, and
Reichelt (2010) find that between 2004 and 2006 the market responded non-negatively to auditor
switches from Big 4 to third-tier accounting firms, as well as to switches from Big 4 to Medium
2 firms and from Big 4 to other Big 4 firms. However, the market reacted negatively to other
switches. Cullinan, Du, and Zheng (2012) find negative market reactions for auditor switches,
but note that the market does not react more negatively when companies switch from a Big 4
firm to a second tier firm than when companies switch from a Big 4 firm to another Big 4 firm.
These prior studies on auditor switches all use archival data and focus on stock market
reactions. The current study involves investing decisions by individuals rather than a market-
based approach. Individual investors merit study because they represent a significant portion of
the investment community and potentially impact stock prices (DeLong et al., 1991), their
welfare is of concern to the SEC, and knowledge about individual investor decision-making is
useful for developing theoretical models relating to stock markets (Daniel, Hirshleifer, and
Subrahmanyam, 1998).
The current study uses an experimental approach to examine the effects of Big 4 to
regional firm auditor switches on individuals’ investing decisons. An experimental approach
enables one to completely control the information available to investor participants. One can
control for factors that might create confounding effects in archival studies, such as concurrent
information disclosure, firm-specific characteristics, and self selection.
HYPOTHESES
Lu ( 2006) analytically shows that auditor switching sends a negative signal to capital
markets that can result in stock price declines for switching companies. Reasons for the negative
signal may include that the new auditor selected would be perceived as less critical of the
company management, managers might be looking after their own interests rather than
shareholder wealth maximization, and corporate governance could be weak. In line with this
notion that auditor switches convey negative signals, an auditor switch is a factor considered by
the Public Company Accounting Oversight Board when selecting specific audits for inspection
(Stefaniak, Robertson, and Houston, 2009). As discussed earlier, results from some prior studies
on auditor switches are consistent with this negative signal view (i.e., adverse stock price
reactions occur for companies switching auditors), but other studies have shown no adverse
impact or sometimes a positive impact when switching audit firms.
Switching from a Big 4 firm to a regional firm could signal that the company is not an
attractive client for a Big 4 firm, which may be more selective in its choice of clients (Cullinan,
Du, and Zheng, 2012). On the other hand, the perception may be that the switch is a positive
development. For instance, Louis (2005) provides evidence that non-Big 4 firms have a
comparative advantage over Big 4 firms in assisting clients with merger transactions. As noted
earlier, most prior studies have found more negative market reactions for companies switching
away from Big 4 firms than for other auditor switches. Some studies, however, have shown
neutral or even positive reactions for companies switching away from Big 4 firms. Therefore,
non-directional hypotheses are offered. Two components of investing judgments will be
examined, the risk associated with the investment and the amount to invest in the company, as
follows:
Page 299 5
H1A: Risk assessments will be differentially impacted for companies that switch auditors from Big 4
firms to regional firms as compared to companies that do not switch auditors.
H1B: Amounts invested will be differentially impacted for companies that switch auditors from Big 4
firms to regional firms as compared to companies that do not switch auditors.
The Securities and Exchange Commission (SEC) requires registrants to report in its Form
8-K filing any change in auditor (SEC, 2014), as well as whether the change was client-initiated
or auditor-initiated, but the registrant need not disclose the reason for an auditor change. Indeed,
reasons for auditor switches are often not provided. Sankaraguruswamy and Whisenant (2004)
show that reasons for auditor switches were provided in only 26 percent of Form 8-K filings
containing auditor switches. Grothe and Weirich (2007) find that of 1,011 auditor switches
reported to the SEC in 2006, 72.5 percent disclosed no information about the reason for the
switches.
Since providing a reason for changing an auditor promotes transparency, one might
expect that investors should view these disclosures favorably. However, prior research has
produced mixed results. Sankaraguruswamy and Whisenant (2004) examine auditor switches
where companies voluntarily reported their reasons for switching. They classified reasons as
verifiable and non-verifiable and found that the latter were positively associated with abnormal
returns while the former were unrelated. Hackenbrack and Hogan (2002) report that earnings
response coefficients do not differ pre- to post-switch for companies that provide uninformative
or no reasons for the switch, while earnings response coefficients are lower after switches for
some types of reasons and higher for other types of reasons. Consistent with these findings,
Hossain, Mitra, and Rezaee (2014) provide evidence that when auditor switches are accompanied
by preexisting red-flag situations, capital market effects are negative, implying that disclosure of
reasons for switching auditors would be informative to investors. Based on the mixed findings of
prior research, non-directional hypotheses are offered. As before, investment risk and amounts
invested are examined separately, as follows:
H2A: For companies switching auditors from Big 4 firms to regional firms, risk assessments will be
differentially impacted for companies that provide a reason for the switch than for companies that
do not provide a reason.
H2B: For companies switching auditors from Big 4 firms to regional firms, amounts invested will be
differentially impacted for companies that provide a reason for the switch than for companies that
do not provide a reason.
The next hypothesis examines two commonly cited reasons for why companies switch
from Big 4 to smaller audit firms -- to obtain better service in the form of more industry expertise
and to reduce their audit fees (Chang, Cheng, and Reichelt, 2010; Stefaniak, Robertson, and
Houston, 2009; Ettredge, Li, and Scholz, 2007). These two reasons may evoke very different
perceptions by investors, although each may be viewed both positively and negatively. More
industry expertise may be viewed as enhancing audit quality, but investors may question whether
it involves spending extra money to do so. On the other hand, investors may percieve lower audit
fees as desirable from the standpoint of maximizing the company’s profits, but they could also be
concerned about an associated decline in audit quality since lower fees could result from auditors
spending less time conducting the audit.
Few studies have investigated the impact of auditor switches on investing by examining
reasons such as more expertise or lower audit fees. Hackenbrack and Hogan (2002) find that
earnings response coefficients are lower subsequent to the auditor change for companies that
Page 300 6
switched for fee-related reasons, but higher for those that switched for service-related reasons.
Knechel, Naiker, and Pacheco (2007) show that companies switching auditors experience
positive abnormal returns when the successor auditor is an industry specialist, but when the
successor audit firm is not a specialist, negative abnormal returns are experienced. Chang,
Cheng, and Reichelt (2010) find that a relatively more positive stock market reaction to clients
switching from Big 4 to smaller firms reflects companies seeking better services rather than
lower audit fees. Because no other study has produced evidence on the comparative effects of
these two reasons for switching auditors, and since each reason can be viewed as both positive
and negative by investors, non-directional hypotheses are offered. Again, investment risk and
amounts invested are examined separately, as follows:
H3A: For companies that provide a reason for switching auditors from Big 4 firms to regional firms,
risk assessments will be differentially impacted for reasons relating to obtaining lower audit fees
than for reasons relating to the desire to obtain better service in the form of more industry
expertise.
H3B: For companies that provide a reason for switching auditors from Big 4 firms to regional firms,
amounts invested will be differentially impacted for reasons relating to obtaining lower audit fees
than for reasons relating to the desire to obtain better service in the form of more industry
expertise.
PARTICIPANTS
Particpants were obtained from websites of Institute of Management Accountants (IMA)

chapters located throughout the United States. Members of the IMA represent individuals who
are generally knowledgeable about audited financial statements and who would also have some
awareness about issues involving auditor changes. IMA members were contacted by email and
asked whether they were willing to participate in a study dealing with the effects of financial
information on investing decisions, and if so, whether they would also be willing to distribute
some questionnaires to accounting/finance colleagues at their place of work. Based on their
responses (i.e., willingness to participate, willingness to distribute questionnaires), 338
questionnaires were sent to 106 IMA members. Completed questionnaires were later returned by
101 people from at least 38 different organizations (some respondents did not reveal their
organizational affiliations). If all of the questionnaires were distributed, the response rate would
be 30 percent (101/338).
The average age of the participants is 46 years, 68 percent are male, 41 percent have
masters degrees, they average 23 years of full-time work experience, and 88 percent indicate
accounting/finance as the job category for their most recent full-time position. Most are
experienced investors -- 82 percent state that they have previously purchased or sold stock
relating to a personal investment. These participants would seem to have characteristics and
backgrounds that are appropriate for the current study. Statistical tests were conducted to
determine whether the demographic data across all four groups are similar. Because some groups
did differ with respect to age and experience, supplemental analyses were conducted with these
two variables as covariates (the results of these supplemental analyses are reported later in this
paper).
Page 301
7
TASK
The participants were given a case scenario involving an investing decision pertaining to
a hypothetical company that designs, develops, manufactures, and markets power protection for
computer communications and electronic applications worldwide. The case provided background
information about the company, financial information (revenues, assets, cash flow from
operations, and earnings per share), a consensus analyst forecast of earnings per share for the
coming year together with the consensus analyst recommendation, and stock price data for the
company covering the past twelve months. The case also stated that the company’s financial
statements were audited and that a clean (i.e., unqualified) opinion audit report had been issued
on its most recent year financial statements.
Participants first assessed the level of risk associated with investing in the common stock
of the company. This assessment was done on a 10-point scale ranging from no risk at all to very
high risk. Next, assuming an investment horizon of one to two years, the participants allocated
$10,000 between investing in the common stock of the company versus investing in a relatively
risk-free money market account.
Four different questionnaire versions were created by varying the information about an
auditor switch and the reason for the switch. The questionnaires were pre-tested with nine
business school faculty members, resulting in some minor changes. For one version of the
questionnaire (NO SWITCH), there was no auditor switch at all. A second version
(SWITCH/NO-REASON) had an auditor switch with no reason given for the switch. A third
version (SWITCH/LOWER-FEES) involved an auditor switch due to the company obtaining a
lower audit fee. A fourth version (SWITCH/MORE-EXPERTISE) contained an auditor switch
because of a desire to obtain auditors with more industry expertise. This results in a 1 x 3
research design with a control condition (NO SWITCH) and three treatment conditions involving
auditor switches. Each participant received only one of these four questionnaire versions. As
shown in Table 1, the numbers of participants responding for each questionnaire version are as
follows: 30 respondents in the NO SWITCH group, 20 respondents in the SWITCH/NO-
REASON group, 23 respondents in the SWITCH/LOWER-FEES group, and 28 respondents in
the SWITCH/MORE-EXPERTISE group.
RESULTS
As a manipulation check, participants were asked at the end of the experiment to recall
the case scenario information about any auditor change. Of the 101 participants, eight did not
correctly recall the auditor change information. When these participants are excluded from the
data anlaysis, the results remain essentially unchanged. Data analysis was also performed on a
data set that excluded participants who had no previous investing experience. Results were
similar to those of the full data set. Therefore, the following data analysis includes all
participants. Since each of the three hypotheses are non-directional, all statistical tests reported in
this section are two-tailed.
Page 302
8
Table 1
DEPENDENT VARIABLE MEANS
Mean Mean
Risk Investment
Group (and sample size) Assessment* Amount*
NO-SWITCH (n=30) 5.77 $4,233

(1.98) ($3,048)
SWITCH/NO-REASON (n=20) 6.19 $3,975

(1.95) ($3,354)
SWITCH/LOWER-FEES (n=23) 6.23 $4,978

(1.51) ($2,140)
SWITCH/MORE-EXPERTISE (n=28) 5.32 $4,759

(1.72) ($2,887)
*Standard deviations are reported in parentheses below the means.
Note: For Risk Assessment, the rating scale ranged from 1 = “not risky at all” to 10 = “very risky.”
The average responses to the two dependent variables for each questionnaire group
appears in Table 1. Overall, the average risk assessment (1=not risky at all; 10=very risky) was
5.80, with a range of 5.32 for the SWITCH/MORE-EXPERTISE group to 6.23 for the
SWITCH/LOWER-FEES group. The average amount allocated to investing in the common stock
of the company was $4,495, ranging from $3,975 for the SWITCH/NO-REASON group to
$4,978 for the SWITCH/LOWER-FEES group. A MANOVA revealed no significant differences
across the four groups (Wilks’ Lambda = .443). Likewise, ANOVAs indicated no significant
differences across the four groups for the risk assessment variable (p = .262) and for the amount
allocated to investing (p = .624). These analyses were also performed with age and work
experience as covariates and the results from ANCOVAs showed no significance for either of the
two dependent variables.
To test the first set of hypotheses, H1A and H1B, which address the impact on investing
resulting from knowledge of auditor switches from Big 4 to regional firms, the responses to the
NO-SWITCH (i.e., conrol) group were compared to those of the combined three treatment
groups, all of which involved auditor switches. The risk assessment for the control group
averaged 5.77, while that of the combined treatment groups averaged 5.85. The amount invested
for the control group averaged $4,233, while the average for the combined treatment groups was
$4,607. The direction of these differences would suggest a negative signal regarding the auditor
switch for the risk assessment variable, but a positive signal regarding the auditor switch for the
amount invested. However, neither of these two differences is statistically significant (p = .844
for the risk assessment; p = .553 for the amount invested). Therefore, neither H1A nor H1B is
supported. Hence, there is no evidence that risk assessments or investment amounts differ for
companies that switch auditors from Big 4 firms to regional firms as compared to companies that
do not switch audit firms.
Page 303
9
The second set of hypotheses, H2A and H2B, examine whether investing judgments will
be differentially impacted for companies that provide a reason for an auditor switch versus
companies that do not provide a reason for the switch. To test H2A and H2B, the responses to the
SWITCH/NO-REASON group were compared to those of the combined SWITCH/MORE-
EXPERTISE and SWITCH/LOWER-FEES groups. The risk assessment for the SWITCH/NO-
REASON group averaged 6.19, while that of the combined two groups that provided reasons for
the switches averaged 5.70. The amount invested for the SWITCH/NO-REASON group
averaged $3,975, while the average for the combined two groups that provided reasons for the
switches was $4,860. For both of these response variables, the direction of the differences would
suggest a negative signal sent by not providing a reason for a switch from a Big 4 audit firm to a
regional audit firm. However, neither of these two differences is statistically significant (p = .328
for the risk assessment; p = .236 for the amount invested). Therefore, for both risk assessments
and amounts invested, there was no differential impact on investing between the disclosure of a
reason for an auditor switch versus no disclosure of the reason for the switch. Hence, neither
H2A nor H2B is supported.
To examine the possibility that the two reasons (i.e., lower audit fees and more industry
expertise) could have opposite effects and thereby cancel each other out, two supplementary tests
were conducted. First, the responses to the SWITCH/NO-REASON group were compared to
those of the SWITCH/MORE-EXPERTISE group. The risk assessment for the SWITCH/NO-
REASON group averaged 6.19, while that of SWITCH/MORE-EXPERTISE group averaged
5.32. The amount invested for the SWITCH/NO-REASON group averaged $3,975, while the
average for the SWITCH/MORE-EXPERTISE group was $4,759. Neither of these differences is
statistically significant (p = .118 for the risk assessment; p = .395 for the amount invested).
Second, the responses to the SWITCH/NO-REASON group were compared to those of the
SWITCH/LOWER-FEES group. The risk assessment for the SWITCH/NO-REASON group
averaged 6.19, while that of SWITCH/LOWER-FEES group averaged 6.23. The amount invested
for the SWITCH/NO-REASON group averaged $3,975, while the average for the
SWITCH/LOWER-FEES group was $4,978. Neither of these differences is statistically
significant (p = .952 for the risk assessment; p = .243 for the amount invested). Therefore, the
finding that disclosing a reason for an auditor switch has no impact on investing judgments does
not appear to be attributable to offsetting effects of the two different types of reasons for
switching.
The third set of hypotheses, H3A and H3B, focus on companies that provide a reason for
an auditor switch by examining whether investing judgments are differentially impacted for
reasons relating to obtaining lower audit fees versus reasons relating to the desire to obtain better
service in the form of more industry expertise. To test H3A and H3B, the responses to the
SWITCH/MORE-EXPERTISE group were compared to those of the SWITCH/LOWER-FEES
group. The risk assessment for the SWITCH/MORE-EXPERTISE group averaged 5.32, while
that of the SWITCH/LOWER-FEES group averaged 6.23. This difference is marginally
significant (p = .057). Hence, H3A is weakly supported. This suggests that disclosing a reason
for an auditor switch dealing with seeking more expertise may send a more positive signal to
investors than disclosing a reason for an auditor switch dealing with seeking lower audit fees.
The amount invested for the SWITCH/MORE-EXPERTISE group averaged $4,759, while the
average for the SWITCH/LOWER-FEES group was $4,978. This difference is not statistically
significant (p = .765). Hence, H3B is not supported. So, while it appears that participants
perceive a lower risk associated with switching auditors to acquire more expertise as opposed to
Page 304 10
switching auditors to obtain lower audit fees, this perception did not lead them to increase the
amount they would invest in the company.
After the experiment was concluded, respondents rated the importance of eight different
factors in making their investing decisions (1 = no importance; 10 = very important), as shown in
Table 2. All of the factors except “Stability/instability of company’s relationship with its auditor”
are significantly different from the scale midpoint of 5.5 at the .05 level of significance. This
suggests that a company changing auditors is considered neither important nor unimportant in
participants’ judgments about investing. The most important one of these factors was “financial
data”. “Having a Big 4 firm vs. a regional firm conduct the audit” was ranked last (tied with
“standard industrial classification”). This ranking appears to be consistent with the current
study’s main findings that investors do not strongly consider factors relating to switches from Big
4 audit firms to regional firms when making judgments about investing.
Table 2
FACTOR RATINGS
Factor Avg. Rating Standard

Deviation
Firm description 6.7 2.1
Standard industrial classification 4.0 2.5
Stability/instability of company’s relationship with its auditor 5.3 2.4
Auditor’s opinion on financial statements 7.8 2.1
Financial data 8.9 1.2
Analyst recommendation 6.2 2.1
Historical stock prices 7.4 1.8
Having a Big 4 firm vs. a regional firm conduct the audit 4.0 2.1
Rating scale: 1 = no importance; 10 = very important
CONCLUSION
The current study examines whether knowledge about auditor switches from Big 4 to
regional firms affects judgments about investing in those companies that switch auditors. Results
indicate that neither risk assessments nor investment amounts differed for companies that
switched auditors from Big 4 firms to regional firms as compared to companies that did not
switch auditors at all. This finding is consistent with Chang, Cheng, and Reichelt (2010), whose
archival study found that the stock market responded non-negatively to auditor switches from Big
4 to third-tier accounting firms. For companies that switched auditors, the current study shows
that there was no differential impact on investing judgments between the disclosure of a reason
Page 305 11
for an auditor switch versus no disclosure of the reason for the auditor switch. Finally, for
companies that provided a reason for an auditor switch, participants perceive a lower risk
associated with switching auditors to acquire more expertise as opposed to switching auditors to
obtain lower audit fees, but this perception did not lead particpants to increase the amount they
would invest in the company.
One implication of these findings is that companies need not be wary of switching
auditors from Big 4 to regional firms. An implication for companies that switch from Big 4 firms
to regional firms is that they need not be concerned about whether or not to provide a reason for
the audit firm switch. Different reasons that are disclosed, though, may convey different
messages about the riskiness of the company. Specifically, an auditor switch to obtain more
expertise may be viewed more favorably (in terms of risk) by investors than an auditor switch to
obtain lower audit fees. Despite that, amounts that would be invested in a company switching
from a Big 4 firm to a regional firm are not likely to be affected.
The usual limitations of experimental approaches apply to the current study as well. One
limitation is that this research was conducted in the context of one company scenario only and
therefore is not necessarily generalizable to other types of company scenarios. Therefore, future
studies should investigate the impact of auditor switches with company settings having different
characteristics regarding industry, competitive environment, risk, financial performance, analyst
forecasts and recommendations, and historical stock prices. Another limitation is that investors
can usually obtain more information about a company than was provided in the current study’s
questionnaire. A third limitation is that economic incentives such as incurring a financial loss
from poor investing decisions were not present in this experiment. Participants did not have any
actual funds at risk in the current study. Future studies can go beyond investigating participants’
intentions by examining investing decisions using an experimental economics approach where
the amount of compensation to participants would depend on the outcomes of their investing
decisions. A fourth limitation is that IMA members may not be representative of the typical
individual investor, so care must be taken in generalizing this study’s results to other types of
respondents. Finally, the total sample size of 101 participants, divided into four cells ranging
from 20 to 30, may not have been adequate.
Future research should also investigate effects of disclosing reasons for auditor switches
from Big 4 firms to regional firms other than to oabtain lower audit fees or to acquire auditors
with more industry expertise. Other reasons include disagreements between the client and its
auditor, the client’s business strategy, going concern issues, management’s reputation,
reputation/experience of the auditor, scope limitations, and independence impairment (Turner,
Williams and Weirich, 2005; Calderon and Ofobike, 2008; Chang, Cheng, and Reichelt, 2010).
The current study’s setting involved switching from Big 4 audit firms to regional audit firms, so
future research could also examine scenarios involving switches from regional audit firms to Big
4 firms. Another avenue for future research would be to investigate the effects of switching from
Big 4 audit firms to regional audit firms on other types of judgments such as commercial lending
decisions.
Page 306
12
REFERENCES
Asthana, S.C., S. Balsam, and J. Krishnan (2010). Corporate governance, audit firm reputation, auditor switches, and
client stock price reactions: The Andersen experience. International Journal of Accounting 14, 274-293.
Beneish, M.D., P.E. Hopkins, I.P. Jansen, and R.M. Martin (2005). Do auditor resignations reduce uncertainty about
the quality of firms’ financial reporting? Journal of Accounting and Public Policy 24, 357-390.
Blau, B.M., T.J. Brough, J.L. Smith, and N.M. Stephens (2013). An examination of short-selling activity
surrounding auditor changes. Journal of Accounting, Auditing & Finance 28(4), 348-368.
Boone, J.P., I.K. Khurana, and K.K. Raman (2010). Do the big 4 and second-tier firms provide audits of similar
quality? Journal of Accounting and Public Policy 29(4), 330-352.
Calderon, T.G. and E. Ofobike (2008). Determinants of client-initiated and auditor-initiated auditor changes.
Managerial Auditing Journal 23(1), 4-25.
Cassell, C.A., G. Giroux, L.A. Myers, and T.C. Omer (2013). The emergence of second-tier auditors in the US:
Evidence from investor perceptions of financial reporting credibility. Journal of Business Finance &
Accounting 40(3) & (4), 350-372.
Chang, H., C.S.A. Cheng, and K.J. Reichelt (2010). Market reaction to auditor switching from big 4 to third-tier
small accounting firms. Auditing, A Journal of Practice & Theory 29(2), 83-114.
Cullinan, C.P, H. Du, and X. Zheng (2012). Barriers to entry to the big firm audit market: Evidence from market
reaction to switches to second tier audit firms in the post-SOX period. Research in Accounting Regulation
24, 6-14.
Daniel, K., D. Hirshleifer, and A. Subrahmanyam (1998). Investor psychology and security market under- and
overreactions. Journal of Finance 53, 1839-1886.
DeFond, M.L., M.L. Ettredge, and D.B. Smith (1997). An investigation of auditor resignations. Research in
Accounting Regulation 11, 25-45.
Dunn, J., D. Hillier, and A.P. Marshall (1999). The market reaction to auditor resignations. Accounting and Business
Research 29(2), 95-108.
DeLong, J., A. Shleifer, L. Summers, and R. Waldman (1991). The survival of noise traders in financial markets.
Journal of Business 64, 1-19.
Eichenseher, J.W., M. Hagigi, and D. Shields (1989). Market reaction to auditor changes by OTC comopanies.
Auditing, A Journal of Practice & Theory 9(1), 29-40.
Ettredge, M.L., C. Li, and S. Scholz (2007). Audit fees and auditor dismissals in the Sarbanes-Oxley era. Accounting
Horizons 21(4), 371-386.
Fontaine, R. S. Ben Letaifa, and D. Herda (2013). An interview study to understand the reasons clients change audit
firms and the client’s perceived value of the audit service. Current Issues in Auditing 7(1), A1-A14.
Fried, D. and A. Schiff (1981). CPA switches and associated market reactions. The Accounting Review 56(2), 326-
341.
Griffin, P.A. and D.H. Lont (2010). Do investors care about auditor dismissals and resignations: What drives the
response? Auditing, A Journal of Practice & Theory 29(2), 189-214.
Grothe, M. and T.R. Weirich (2007). Analyzing auditor changes: Lack of disclosure hinders accountability to
investors. The CPA Journal 77(12), 14-23.
Hackenbrack, K. and C. Hogan (2002). Market response to earnings surprises conditional on reasons for an auditor
change. Contemporary Accounting Research 19(2), 195-223.
Hossain, M., S. Mitra, and Z. Rezaee (2014). Voluntary disclosure of reasons for auditor changes and the capital
market reaction to information disclosure. Research in Accounting Regulation 26, 40-53.
Jenkins, D.S. and U. Velury (2011). The emergence of second-tier auditors in the post-SOX era, An analysis of
accounting conservatism. Research in Accounting Regulation 23, 172-176.
Johnson, W. and T. Lys (1990). The market for audit services: Evidence from voluntary auditor changes. Journal of
Accounting & Economics 12(1-3), 281-308.
Klock, M. (1994). The stock market reaction to a change in certifying accountant. Journal of Accounting, Auditing &
Finance 9(2), 339-347.
Knechel, W.R., V. Naiker, and G. Pacheco (2007). Does auditor industry specialization matter: Evidence from
market reaction to auditor switches. Auditing, A Journal of Practice & Theory 26(1), 19-45.
Louis, H. (2005). Acquirers’ abnormal returns and the non-Big 4 auditor clientele effect. Journal of Accounting &
Economics 40, 75-99.
Lu, T. (2006). Does opinion shopping impair auditor independence and audit quality? Journal of Accounting
Research 44(3), 561-583.
Page 307 13
Mande, V. and M. Son (2013). Do financial restatements lead to auditor changes? Auditing, A Journal of Practice &
Theory 32(2), 119-145.
Nichols, D.R. and D.B. Smith (1983). Auditor credibility and auditor changes. Journal of Accounting Research
21(2), 534-544.
Sankaraguruswamy, S. and J.S. Whisenant (2004). An empirical analysis of voluntarily supplied client-auditor
realignment reasons. Auditing, A Journal of Practice & Theory 23(1), 107-121.
Schneider, A. (2015). Does information about auditor switches affect investing decisions? Research in Accounting
Regulation 27, 39-44.
Securities and Exchange Commission (SEC). (2014). Regulation S-K, Item 304(a), SEC, Washington, DC., avaliable
at, http,//www.ecfr.gov/cgi-bin/text-
idx?SID=b86657e71943763a7eb9f8333c6536f0&node=17,3.0.1.1.11.4.35.4&rgn=div8.
Shu, S. (2000). Auditor resignations: Clientele effects and legal liability. Journal of Accounting & Economics 29(2),
173-205.
Smith, D. and D. Nichols (1982). A market test of investor reaction to disagreements . Journal of Accounting &
Economics 4(2), 109-120.
Stefaniak, C.M., J.C. Robertson, and R.W. Houston (2009). The causes and consequences of auditor switching: A
review of the literature. Journal of Accounting Literature 28, 47-121.
Turner, L.E., J.P. Williams, and T.R. Weirich (2005 Suplenment). An inside look at auditor changes. The CPA
Journal 75(11), 12-21.
United States Government Accountability Office. (2008). Audits of public companies, Continued concentration in
audit market for large public companies does not call for immediate action. Washington, US Government
Printing Office. Available at, http,//www.gao.gov/new.items/d08163.pdf.
Wells, D.W. and M.L. Louder (1997). The market effect of auditor resignations. Auditing, A Journal of Practice &
Theory 16(1), 138-144.
Whisenant, J.S., S. Sankaraguruswamy, and K. Raghunandan (2003). Market reactions to disclosure of reportable
events. Auditing, A Journal of Practice & Theory 22(1), 181-194.
Page 308
Page 309
BANK SIZE EFFECT IN THE DEPOSIT MARKET:

EVIDENCE FROM BANK BRANCH OWNERSHIP
CHANGES
Huabing (Barbara) Wang, West Texas A&M University
ABSTRACT
This paper examines the effect of bank size in the deposit market by utilizing a sample of
branch offices that have changed ownership. We find that branches with larger new owners are
associated with deposit decrease in the branch, while branches with smaller new owners are
associated with deposit increase in the branch. This is consistent with the competitiveness of
smaller banks in the deposit market. However, we also find that if the branches have changed
ownership to the four largest banks, deposits increase too at the branch level, suggesting unique
niches of these largest banks.
INTRODUCTION
The most recent financial crisis has put large banks under the spotlight for the systematic
risks they bring to the economy. The massive bailout of these banks has been troubling for
regulators. During a senate hearing on establishing a framework for systemic risk regulation in
2009, Sheila Bair, then Chairman of the U.S. Federal Deposit Insurance Corporation (FDIC),
states that “the notion of too big to fail creates a vicious circle that needs to be broken.” She
further states in her testimony the need to “develop a resolution regime that provides for the
orderly wind-down of large, systemically important financial firms, without imposing large costs
to the taxpayer.” While debatable, many even have advocated the breaking-up of large banks to
avoid the “too-big-to-fail” problem and future financial catastrophe. The Progressive Change
Institute, who recently released a poll result showing bipartisan backing for breaking up big
banks, is an example of such advocate groups (Schroeder, 2015). The Economist magazine held
a debate on whether big banks should be broken up on May 14, 2013, and the post-debate poll
indicated favorable public sentiment with 83% yes vote.
However, technology advances and heightened regulation have favored large banks as
never before. For example, Wheelock and Wilson (2010) suggest that information technology
has tended to favor larger institutions for two reasons. First, larger institutions can more easily
bear the relatively high fixed cost of information processing equipment and software. Secondly,
the new technologies have eroded the value of close proximity and personal relationships in
gaining soft information, which has been an advantage of small institutions. Regarding
regulation, Peirce et al (2014) conduct a small bank survey covering 200 banks across 41 states,
and report substantially increased compliance costs in the wake of new regulations such as
Dodd-Frank for participating banks. Indeed, according to FDIC data, U.S. banking assets and
deposits have continued to consolidate since the financial crisis of 2008. Tracy (2013) reports in
a Wall Street Journal article that the number of federally insured institutions has shrank to 6,891
in the third quarter of 2013, and community banks might be “too small to survive”.
While technology and regulation both put small banks at a disadvantage, community
banks have many advantages compared with large banks in the deposit market. The ability to
source low-cost stable core deposits can significantly lower bank funding cost, contributing to
Page 310
the bank’s financial well-being. Small banks are known for their more personalized services,
lower checking fees, and more lending flexibility (Kadlec, 2014). According to the 2013
Banking satisfaction study by J. D. Powers, smaller banks have higher scores regarding fees and
problem resolution. At the same time, the ability of banks to attract transaction and savings
deposits from businesses and consumers is an important measure of a bank’s acceptance by the
public (Rose and Hudgins, 2012, P. 397). Community banks are viewed as part of the local
economy, and indeed, they tend to keep their sourced deposits in the local community. In the
wake of the recent global financial crisis, the resentment towards the largest banks of the country
has grown, and as mentioned earlier, the idea of breaking up large banks has gained support and
advocated by many advocate groups. Therefore, the depositor sentiment may also play in the
smaller banks’ favor. Of course, large banks do offer more convenience thanks to their larger
branch and ATM network. The scale of economy allows them to provide more cutting-edge
technology and security features appreciated by customers (Kadlec, 2014). On balance, it is an
empirical question whether larger or smaller banks have an upper hand in the competition for
deposits.
This paper attempts to shed light on this question. Specifically, we examine the effect of
bank size on deposit market competition, by utilizing a natural laboratory setting where branch
offices experience ownership change among banks of different sizes. We find that branches with
larger new owners are associated with deposit decrease in the branch, while branches with
smaller new owners are associated with deposit increase in the branch. This is consistent with the
competitiveness of smaller banks in the deposit market. However, we also find that if the
branches have changed ownership to the Top 4 banks, deposits increase too at the branch level,
suggesting unique niches of these Top 4 banks. This is consistent with Hirtle (2007)’s finding
that mid-sized branch networks may be at a competitive disadvantage in branching activities.
LITERATURE REVIEW
Our paper is related to several streams of literature in the banking field. First of all, it is a
critical question whether smaller community banks can survive the competition with larger
banks. As shown in prior literature, community banks play a vital role in local economic growth,
especially in the rural areas. For example, Hakenes, Hasan, Molyneux, and Xie (2014)
theoretically show that small regional banks are more effective in promoting local economic
growth. Their empirical evidence based on a sample of German banks lends support to their
model. Burgess, Robin and Pande (2005) also document that branch expansion into rural
unbanked locations in India significantly reduces rural poverty. This is consistent with the
observation of Brooks (2014) in the U.S. banks, which reports that small businesses in the rural
areas of the U.S. hurt by the decline of local community banks.
However, despite their contribution to local economic growth, smaller community banks
may be at a competitive disadvantage due to the economy of scale. For example, in terms of the
adoption of new information technology, Wheelock and Wilson (2010) suggest that larger
institutions have advantages in two regards. First, they can more easily bear the high fixed cost
of information infrastructure and software. Secondly, the new technologies have eroded the value
of close proximity and personal relationships, which has been an advantage of small institutions.
In the wake of the most recent financial crisis and the passage of Dodd Frank, community banks
may face even harsher competition. Marsh and Norman (2013) suggest that Dodd-Frank will
lead to greater asset concentration in a small number of financial institutions, furthering the trend
of “too big to fail.” Peirce et al (2014) conduct a small bank survey covering 200 banks across 41
Page 311
states, and report substantially increased compliance costs in the wake of new regulation such as
Dodd-Frank for participating banks.
Prior research on the interaction of small versus large banks offers mixed evidence on the
competition dynamics between large and small banks. Pilloff (1999) documents reduced
competition in the rural markets where big banks operate, allowing all banks in those markets to
earn greater returns. Filbeck et al (2010) find that community banks have gained market share at
the expense of larger, regional banks in small metropolitan Statatistical areas (MSAs). Pilloff
(1999), however, does not find clear and consistent patterns of variations in the relationship
between the profitability of small banks and the presence of big banks. Filbeck, Preece, and Zhao
(2012) also document fewer advantages to large banks making market share gains in terms of
subsequent performance. Hirtle (2007) reports that mid-sized branch networks may be at a
competitive disadvantage in branching activities relative to larger or smaller branch networks.
On the other hand, Hannan and Prager (2009) document that when large primarily-out-of-market
banks increase their presence in the rural market, the profits of small single-market banks are
significantly affected. Overall, while some studies suggest that community banks may still be
viable in the competition with larger banks, prior literature does show concern especially in the
rural areas.
While the prior studies approach the issue at the bank level, focusing on bank
profitability and market share, we take a micro view in our study by focusing on branch deposit
of banks of various sizes. Specifically, we examine branch deposit changes after bank ownership
change between large and small banks. While larger banks have many advantages, smaller banks
also have their strong suit, especially in customer service. For example, Eastman et al (2010) find
that consumers are significantly more comfortable with community banks than the other forms of
banking, including national banks, regional banks, investment houses, and brokerage firms. In
our study, we examine whether the new ownership change will affect branch deposit level and
share. For example, when a branch changes hand from a small community bank to a large
national bank, will the branch deposit level/share be negatively or positively affected? The
answer to such questions will shed light on the dynamics of competition between community and
large banking institutions in the deposit market.
Finally, our research is also embedded in the literature on the determinants of the market
share of deposit, a vital source of funding for banking institutions. Berger and Dick (2007)
document the early-mover advantage in the banking market. Specifically, they find that the
earlier a bank enters a market, the larger its market share relative to other banks. It is worth
noting that Berger and Dick (2007) also examine bank entry by merger, and find that entering a
market through a merger gives the bank a higher market share because the bank is buying up
another bank’s existing branch network in the market. Furthermore, our study is related to the
literature on bank switching behavior. Kiser (2002) documents based on survey data that the
average household uses the same bank for 10 years, and household relocation is the main cited
reason of switching banks. Lees, Garland, and Wright (2007) examine bank customer loyalty and
switching behavior, and report utility maximization (‘a better deal’) and expectation
disconfirmation (usually service failure) as the main switching reasons.
Page 312
DATA AND METHODOLOGY
Data, Sample, and Sample Characteristics
We start by obtaining the branch office deposit data1 over the period 2000-2013 from the
Federal Deposit Insurance Corp (FDIC). We then focus on branch acquisitions through either
asset acquisition or bank merger and acquisition (M&A) over the eleven-year period of 2002-
2012 period so that we have one year of branch deposit data available post-acquisition and two
years pre-acquisition2. We delete observations representing branch offices that have changed
hands more than once during the sample period. We also require the branch to have non-zero
reported branch deposit over the sample period. Our final sample includes 11,118 branch
ownership changes.
Table 1
SAMPLE CHARACTERISTICS
(n=11,118; All numbers are in thousands except the Metro Area Dummy and branch market share)
Mean Median Min Max

Branch deposit at the
acquisition year 67,013 36,439 4 17,890,014
Total assets of selling

bank 150,871,106 18,545,750 4,858 1,132,840,000
Total assets of acquiring

bank 432,800,384 104,413,336 26,914 1,450,829,889
Metro area
1 .86 0 1
Dummy
Branch deposit:
Branch deposit after the
acquisition year 65,135 36,312 3 16,730,800
Branch deposit before

the acquisition year 69,199 38,210 1 16,889,232
Change in branch deposit -4,063 -1,211 -9,015,802 7,711,817
Market share:
Market share after the
acquisition year
16.18% 5.25% 0.00% 100%
Market share before the

acquisition year 17.56% 5.96% 0.00% 100%
Change in market share -1.38% -0.15% -99.99% 99.98%
Table 1 contains the major characteristics of the sample. The average bank branch in the
sample has approximately $67 million in total deposits the year of the branch acquisition. Half of
Page 313
the branches have deposits less than $36 million. On average, the selling bank has the average
total asset of $151 billion, while the acquiring bank has the average total asset of $433 billion,
almost three times the size of the selling bank. In terms of the median asset size, the difference is
even more prominent with $18.5 billion for the seller and $103 billion for the acquirer.
Regarding geographic location, the Metro area3 has been more active in branch acquisitions or
bank M&A, accounting for 86% of the sample.
Table 1 also illustrates the change in branch deposit and market share around the
acquisition year. We first aggregate the reported deposits from the FDIC Summary of Deposit
database by county for each year to obtain the total market volume. Then we divide the branch
deposit by the total market to obtain the branch’s market share. On average, both branch deposit
volume and branch market share have dropped the year after the change of ownership in the
branch.
Table 2 demonstrates the transition matrix of bank branches in the sample. Following
FDIC convention, we define institutions with more than $10 billion in assets as large banks, and
institutions with less than $1 billion as community banks. The remainder of the institutions are
mid-size banks. Large banks as a buyer accounts for 73.60% of the observations. Large banks are
also the most likely seller, accounting for 55.97% of the sample. It is rare that the branches go
from a large bank to a community bank, and there are only 20 such cases in the sample.
Table 2
TRANSITION MATRIX OF BRANCHES (2002-2012)
Seller Community Mid-size Large Total
Buyer
580
Community
734 61 20 (7.33%)
2,120
Mid-size 1,264 767 89
(19.07%)
8,183
Large 386 1,683 6,114
(73.60%)
2,384 2,511 6,223
Total 11,118
(21.44%) (22.59%) (55.97%)
(100%)
Methodology and Hypothesis Development
We estimate the following model to test the relationship between bank size and post-
acquisition branch deposit change:
Chg_Depositi=α+β1*Dif_Sizei+ β2*Top4_buyeri+ β3*Metroi+ β4*Chg_area_depositi+
β5*Prior_growthi+εi (1)
Chg_MKTSharei=α+β1*Dif_Sizei+ β2*Top4_buyeri+ β3*Metroi+ β4*Chg_area_depositi+
β5*Prior_growthi+εi (2)
In Model (1), Chg_Deposit is our dependent variable measured as the volume of branch
deposit the year after the ownership change minus the volume of branch deposit the year before
the ownership change. Therefore, higher deposits after the branch ownership transfer will result
in a positive Chg_Deposit, suggesting that the branch attracts more deposits under the new
ownership.
Page 314
In Model (2), our dependent variable is Chg_MKTShare , which is measured as the

market share of the branch at the county level after the ownership change minus that before the
ownership change. Similar to the case of Chg_Deposit, higher branch market share after the
ownership change will result in a positive Chg_MKTShare, indicating better performance of the
branch in the deposit collecting market under new ownership.
We introduce five explanatory variables in the models as follows that explain the deposit
level (share) changes.
First, Dif_Size equals the asset size of the acquiring bank minus the asset size of the
selling bank. This variable will be positive when the ownership of a branch changes from a
larger bank to a smaller bank, and negative is the ownership changes from a smaller bank to a
larger bank. The sign of the coefficient indicates the effect of bank size in the deposit market.
Community banks have many advantages over larger banks in their more personalized services,
lower checking fees, and more lending flexibility (Kadlec, 2014). However, large banks do have
larger branch and ATM network and often offer more cutting-edge technology and security
features. Ultimately, it is an empirical question whether larger or smaller banks are more
competitive in the deposit market.
Secondly, we define the following banks as the Top 4 banks: Wells Fargo (CERT=3511),
JPMorgan Chase (CERT=628), Citigroup (CERT=7213), Bank of America (CERT=3510).
These banks constantly top of the list of large banks. They are among the first groups of banks
receiving bailout funds and have received the most4. Combined, these banks have received the
bailout funding of more than $100 billion (including the funds initially allocated to Merrill
Lynch and later absorbed by Bank of America) out of the $200 billion total bailout fund
allocated. Top4_buyer is a dummy variable that takes the value one if the buying bank is among
the Top 4 banks, and zero otherwise. Approximately 41.73% of the branches in the sample were
acquired by the Top 4 banks. As industry leaders, the Top 4 banks offer comprehensive services
and products, and feature the most cutting-edge technology. However, they are also more likely
to draw resentment in the wake of the most recent financial crisis.
Our third independent variable is Metro, which is a dummy variable that takes the value
one if the branch is in a metro area and zero otherwise. Branches in metro areas might face more
intense competition, and we expect a deeper loss of deposits after ownership changes in these
areas. For example, Berger and Dick (2007) note the difference between metropolitan markets
and rural markets and only focus on the metropolitan markets in their study of bank entry and
early-mover advantage. Superficially, they argue that metropolitan markets are generally
considered more competitive with eaiser entry/exit and acquisition of local market share.
Finally, we control for the area deposit growth and the branch’s own deposit trend prior
to the change in ownership. Chg_area_deposit equals the total volume of county deposits the
year after the ownership change minus that before. We expect that the coefficient of
Chg_area_deposit is positive as branch deposit positively correlates with the area deposit trend.
Prior_growth is measured as the growth rate of branch deposit from year -2 to year -1, which
captures the branch’s deposit trend prior to the ownership change. We expect the coefficient of
Prior_growth to be positive since the historical trend might continue into the future, positively
affecting the dependent variables.
It is worth noting that because we examine a specific branch with location unchanged, we
can ignore the location effect, although it can be a significant driver for deposit level (share). In
addition, deposit rates might have direct bearing on branch deposit volume. However, we view
deposit rates as endogenous in our analysis because banks have the discretion to raise or lower
Page 315
the deposit rate after branch acquisition. Prior literature also indicates that deposit rates and bank
funding costs correlates with bank size. For example, Jacewitz and Pogach (2013) document
significant and persistent pricing advantages at the largest banks for comparable deposit products
and deposit risk premiums.
EMPIRICAL RESULTS
We run the regression following Model (1) as described in Section III, and report our
result in Table 3 Panel A. For the combined sample, the difference in size is negatively
associated with an increase in deposit volume at the branch, after accounting for the area deposit
change and the branch’s previous deposit trend. We rerun the regression on the two sub-samples
of “large banks as buyers” and “small banks as buyers” to further test the issue. We find that the
coefficient of Dif_size is significantly negative for branches bought by large banks. However, the
coefficient of Dif_size is significantly positive for branches bought by small banks. Combined,
the negative association between bank size and branch deposit indicates that smaller banks might
have an advantage in attracting deposits. In Panel B, we report our results from Model (2) with
the change in branch market share as the dependent variable. Overall, our results are similar with
the coefficient of size differences negative and significant.
It is interesting to note that the coefficient of Top4_buyer is significantly positive for both
models. This suggests that a top 4 bank as the buyer actually enhances the deposit flow into the
branch following the ownership change. While smaller banks generally have an advantage in the
deposit market, the Top 4 Bank status also has a positive impact on the deposit flow.
Finally, a note of caution is in order. There are numerous factors affecting the deposit
flow at the branch level, such as deposit rate, service quality at the branch, and the bank’s ability
to provide credit to deposit customers at the branch. However, due to limitation of data, we do
not have measurement or proxy for these variables. Therefore, we can only include the limited
explanatory variables available to us, and this may attribute to the low explanatory power of our
models, as suggested by the low R-square values.
Table 3
MULTIVARIATE REGRESSIONS: RESULTS
Combined sample Large banks as buyers Small banks as buyers
Coefficient T Coefficient t Coefficient t

Panel A: y= Chg_Deposit
Constant -13559 -2.98*** -13777 -1.97** -21052 -1.32
Dif_Size 0.00 -2.2** 0.00 -2.30** 0.00 1.69*
Top4_buyer 17259 2.49** 17481 2.46**
Chg_area_
deposit 0.00 4.91*** 0.00 4.91*** 0.00 1.84*
Prior_growth 0.84 0.08 0.68 0.07 20.24 0.06
*
Metro 9112 1.85 10259 1.45 18648 0.84
N 11,118 8,183 815
Adjusted R2 0.027 0.034 0.038
Page 316
Panel B: y= Chg_Mktshare
Constant -0.06 -6.79*** -0.02 -4.32*** -0.01 -2.15**
Dif_Size -0.00 -5.21*** -0.00 -5.22*** -0.00 -0.94
Top4_buyer 0.02 4.75*** 0.02 4.47***
Chg_area_
-0.00 -3.81*** -0.00 -3.74*** -0.00 -2.61***
deposit
Prior_growth 0.00 0.32 0.00 0.27 0.00 0.50
Metro 0.00 1.19 0.00 1.23 -0.01 -1.04
N 11,118 8,183 815
Adjusted R2 0.033 0.043 0.064
Note: Top_4 buyer is not included for the subsample of “Small banks as buyers” for lack of variation. ***
indicates significance at the 99 percent level; ** indicates significance at the 95 percent level; * indicates
significance at the 90 percent level
CONCLUSION
This paper examines how bank size affects deposit flow at the branch level by utilizing a
natural experimental setting of branch ownership change. We find that bank size difference of
the acquiring and selling banks is significantly negatively associated with the branch deposit
change after the ownership, after accounting for the area deposit change and the branch’s
previous deposit trend. In other words, branches with larger new owners are associated with
deposit decrease in the branch, while branches with smaller new owners are associated with
deposit increase in the branch. Despite the concern that small banks might be too small to
survive, their smaller size appear to be advantageous in the deposit market. However, we also
find that if the branches have changed ownership to the four largest banks, the deposits increase
too at the branch level.
Our paper sheds light on the competitiveness of small banks in the deposit market. While
small community banks play positive roles in local economic growth, they may be at a
disadvantage due to the economy of scale, and therefore their survival is of concern to regulators
and the public. The findings from our paper may partly ease this concern as small community
banks seem to have an edge in the game of sourcing core funding sources, a key factor of bank
success.
ENDNOTES
1 All FDIC-insured institutions have been required to fill out an annual summary of deposit (SOD) survey
since 1934 to report branch office deposits as of June 30 every year.
2 In later analysis, we control for the deposit trend prior to the ownership change and therefore need at least
two years of deposit data prior.
3 A metro area is an area that is centered on a single large city (or sometimes two large cities) with
substantial economic influence over the surrounding areas.
Page 317
4 For more details, please see the Treasury’s bank bailout list from http://money.cnn.com/news/specials/
storysupplement/bankbailout/
REFERENCES
Berger, A. N. and A. A. Dick (2007). Entry into banking markets and the early-mover advantage, Journal of Money,
Credit and Banking 39(4), 775-807
Brooks, C. (2014). Rural small businesses hurt by decline of local banks. Fox Business, published Feb. 27, 2014,
retrieved from http://smallbusiness.foxbusiness.com/finance-accounting/2014/02/27/rural-small-
businesses-hurt-by-decline-local-banks/
Burgess, Robin and Rohini Pande (2005). Can rural banks reduce poverty? Evidence from the Indian social banking
experiment, American Economic Review 95 (3), 780-795.
Eastman, J., W. Denton, M. Thomas, and L. Denton (2010). Consumer perceptions of community banks: an
exploratory study, Marketing Management Journal 20 (1), 204-216.
Filbeck, G., D. Preece and X. Zhao (2012). Market share growth and performance measures: The case of large
versus community banks, Banking and Finance Review 4(2), 29-45.
Filbeck, G., D. Preece, S. Woessner, and S. Burgess (2010). Community banks and deposit market share growth,
International Journal of Bank Marketing 28 (4), 252-266.
Hakenes, H., I. Hasan, P. Molyneux & R. Xie (2014). Small banks and local economic development, Review of
Finance, forthcoming.
Hannan, T. H., and R. A. Prager (2009). The profitability of small single-market banks in an era of multi-market
banking, Journal of Banking & Finance 33, 263-271.
Hirtle, B. (2007). The impact of network size on bank branch performance. Journal of Banking and Finance 31(12),
3782-3805.
Kadlec, D. (2014). It’s time to forgive the big banks: Consumers still blame big banks for many of their economic
woes. But it’s time to move on. Time, Feb. 24, 2014, retrieved from http://business.time.com/2014/02/24/
its-time-to-forgive-the-big-banks/
Kiser, E. (2002). Household switching behavior at depository institutions: Evidence from survey data. Antitrust
Bulletin, 47, 619-622
Lees, G., R. Garland, and M. Wright (2007). Switching banks: Old bank gone but not forgotten. Journal of Financial
Services Marketing. 12 (2), 146-156
Marsh, T. and J. Norman (2013). The Impact of Dodd-Frank on Community Banks, American Enterprise Institute
white paper
Peirce H., I. Robinson, and T. Stratmann (2014). How are small banks faring under Dodd-Frank? Working paper,
George Mason University, retrieved from
http://mercatus.org/sites/default/files/Peirce_SmallBankSurvey_v1.pdf
Pilloff, S. J. (1999). Does the presence of big banks influence competition in local market? Journal of Financial
Services Research, 15, 159-177.
Rose, P. S. and S. C. Hudgins (2012). Bank Management & Financial Services, ninth Edition, McGraw-Hill
Schroeder, P. (2015). Poll: Bipartisan backing for breaking up big banks. The Hill, published on Jan. 20, 2015,
retrieved http://thehill.com/policy/finance/230058-poll-bipartisan-backing-for-breaking-up-big-banks
Tracy, R. (2013). Tally of U.S. Banks sinks to record low: small lenders are having the hardest time with new rules,
weak economy and low interest rates, Wall Street Journal, published on Dec. 3, 2013, retrieved
http://online.wsj.com /news/articles/SB10001424052702304579404579232343313671258
Wheelock, D and P. Wilson (2011). Are Credit Union too Small? Review of Economics and Statistics 93 (4), 1343-
1359.
Page 318
Page 319
EFFICIENCY AND CORPORATE BANKRUPTCY

PREDICTION
Domenico Piatti, University of Bergamo
Peter Cincinelli, University of Bergamo
Davide Castellani, University of Bergamo
ABSTRACT
This study investigates the role of technical efficiency in predicting the probability of
default of a sample of Italian SMEs in the period 2007-2009. This specific period is of particular
interest because it is centered on the beginning of the Global Financial Crisis. We argue that
technical efficiency allows for a forward-looking perspective and can contribute to shed more
light on the reasons behind the default of many Italian SMEs in the particular period considered.
The technical efficiency is estimated with a stochastic frontier approach and the efficiency ratio
is used as independent variable, along with several financial ratios. Consistently with the
literature, the results suggest that efficiency is a good predictor when the financial ratios are also
considered. Several robustness checks support the preliminary findings.
INTRODUCTION
Although there are numerous empirical studies on business failure prediction in the
literature, only few of them purposely employ technical efficiency ratios to predict financial
distress (Becchetti & Sierra, 2003, Pusnik & Tajnikar, 2008; Iazzolino, Bruni & Beraldi, 2013;
Li & Wang, 2014). Most models, in fact, rely on hard information embedded in the balance sheet
in terms of financial ratios. Conversely, information provided by the level of technical efficiency
can improve understanding of the role played by qualitative or soft variables in corporate default
prediction (Becchetti & Sierra, 2003).
Based on a sample of Italian SMEs in the period 2007-2009, this study therefore aims to
assess whether a firm’s technical efficiency can lead to a significant improvement in the
performance of business failure prediction models.
Efficiency is measured as the distance from the industry’s best efficient frontier. The least
efficient firms are expected to have a greater probability of default because inefficiency can
compromise their ability to generate profits and cash flows to meet their financial exposure.
This study contributes to the extant literature in the following respects. Firstly, the
purpose of adopting soft information, as a measure to detect early business failure, is to
overcome a merely backward-looking perspective based only on financial ratios, in favour of a
forward-looking perspective. A further reason concerns the Italian SMEs sample used in our
empirical analyses. Particularly, besides a “polycentric model” (Censis, 2008) expressed by
Italian SMEs, a remarkable aspect concerns the time period considered, i.e. 2007-2009, which
includes the onset of the financial crisis. As in many other developed countries, the crisis
increased the strain on the already stressed Italian economic environment.
Page 320
The study is organized as follows. Section 2 provides a literature review. Section 3

presents the sample and methodology. Section 4 shows the efficiency ratio. Sections 5 and 6
give, respectively, the empirical results and robustness checks, and section 7 sets out the main
conclusions.
LITERATURE REVIEW
Over the 1930-1965 time period, a few early studies attempted to use financial ratios for
bankruptcy prediction (Smith, 1930; FitzPatrick, 1932; Smith & Winakor, 1935; Merwin, 1942;
Chudson, 1945; Hickman, 1957; Jackendoff, 1962). In these studies, bankruptcy prediction was
conducted by means of statistical univariate approaches focused on comparison between the
financial ratios of failed companies with those of sound firms. However, although the univariate
approaches were unable to control for the interrelationships among several indicators (Beaver,
1966), they laid the bases for multivariate bankruptcy prediction models. The first multivariate
study was published by Altman (1968). In Altman’s seminal paper, the financial ratios of healthy
companies are different from those of insolvent ones. Moreover, this difference becomes
progressively stronger as the date of bankruptcy approaches. Several bankruptcy prediction
models are present in the literature. More than 165 bankruptcy prediction studies were published
from 1966 (Beaver, 1966) to 2007 (Bellovary, Giacomino & Akers, 2007). However, all these
models differ in terms of input variables, prediction timeframe, statistical methodology and
definition of default. Specifically, these studies, as regard the input variables, sought to test the
effectiveness of financial ratios in forecasting insolvencies. Various financial ratios were used to
forecast insolvencies (Ezzamel, Brodie & Mar-Molinero, 1987). The lack of an economic theory
of firm crisis has induced several authors to test the effectiveness of financial ratios from balance
sheet data in predicting insolvencies through empirical analysis (Skogsvik, 1990) and on the
basis of specific research objectives (Edmister, 1972; Keasey & Watson, 1987). However,
despite the numerous and different financial ratios used in the literature, there is an unexpressed
but evident agreement on the best, and therefore most widely adopted, ratios. Out of the overall
752 different financial ratios used in the literature (Bellovary, Giacomino & Akers, 2007), the
return on assets (ROA), that is, the ratio of net income to total assets, and the current ratio, that
is, the ratio of current assets to current liabilities, are commonly adopted. As suggested by
Tamari (1966), companies that show current ratios less than one are closer to bankruptcy.
Moreover, the ratio between cash flow and total debt is the best ratio with the greatest failure
predictive capacity (Beaver, 1966). As Tamari suggests, this ratio is characterised by a margin of
error, in terms of bankruptcy prediction, ranging from 13% in the first year to 24% in the fifth
year. Regardless of the specific financial ratios, a model that predicts insolvencies as early as
possible is more appropriate and valuable. For instance, Deakin (1972) and Dwyear (1992)
provide a model in which bankruptcy can be predicted two years prior the failure with an
accuracy of 96% and 97% respectively. El Hennawy & Morris (1983) show a prediction ability
of five years earlier with 100% accuracy. Several contributions provide diverse prediction
timeframes: 6 years (Martin, 1977; Skogsvick, 1990; Gilson & Vetsuypens, 1993), 5 years
(Deakin, 1972; Wilcox, 1973; Altman, Haldeman & Narayanan, 1977; Frydman, Altman & Kao,
1985), 4 years (Sinkey, 1975; Kahya & Theodossiou, 1999) and 3 years (Appetiti, 1984; Izan,
Page 321
1984; Levitan & Knoblett, 1985; Lo, 1986). However, the predictive ability of the model is
strongly affected by both the financial ratios and the extent of the time period considered. The
predictive ability of the model can decline as the number of years prior to the failure increases
(Deakin, 1972), or it can be affected by biases due to positive serial correlation (Kahya &
Theodossiou, 1999).
With reference to methodology, most of the techniques adopted to date in the literature
converge into two specific categories (Kumar & Ravi, 2007). The first category, i.e. statistical
techniques, includes linear discriminant analysis, multivariate discriminant analysis, quadratic
discriminant analysis, logistic and probit regression and factor analysis. The second category
encompasses intelligent techniques, such as neural network architectures (e.g. multi-layer
perception, probabilistic neural networks, auto-associative neural networks, self-organizing
maps, learning vector quantization), decision trees, case-based reasoning, evolutionary
approaches, soft computing (hybrid intelligent systems), operational research techniques (e.g.
linear programming, data envelopment analysis, quadratic programming) and other intelligent
techniques (e.g. support vector machine, fuzzy logic techniques). Other techniques, rough set
theory (Slowinski & Zopounidis, 1995), hazard models (Shumway, 2001), Bayesian network
models (Sarkar & Sriram, 2001) and genetic programming (McKee & Lensberg, 2002) have
been also used in the bankruptcy field.
Regarding the definition of default, it is widely known that a generally-accepted
definition does not exist in the literature (Sharma & Mahajan, 1980; Koenig, 1985; Guilhot,
2000; Crutzen, 2010). Some studies consider default to be either a simply bankruptcy
(Fernández-Castro, 1988; Laitinen, 1991; Everett & Watson, 1998), or a Chapter XI bankruptcy
petition (Altman, 1981) linked with the going concern perspective (Taffler & Tishaw, 1977), or
the occurrence of certain events, such as bankruptcy, bond default, overdrawn bank account and
non payment of a preferred stock dividend (Beaver, 1966). From a regulatory perspective,
instead, situations such as the obligor’s unlikelihood of paying its credit obligations and when
the obligor is past due more than 90 days are warnings of potential default (Basel Committee on
Banking Supervision, 2006). Moreover, lower profitability levels, in comparison with the risk
free rate or a substantial amount of loans past due, represent a different default perspective
(Unal, 1988).
Within this framework, the Italian environment has been characterised by a considerable
number of studies on failure prediction. On comparing statistical methodologies for prediction
with neural networks, Altman, Marco & Varetto (1994) find a balanced degree and beneficial
characteristics. Amendola, Bisogno, Restaiano & Sensini (2011), investigate the main aspects
related to the bankruptcy model of 63 failed industrial firms in the Campania region of Italy.
Bisogno (2012) analyses Italian bankruptcy procedures from an ex post and an ex ante
perspective. The latter, as suggested by the author, benefits from prediction models for default
risks. Gentry, Newbold & Whitford (1985), Raja, Nosworthy & Goureia (1980) Gombola,
Haskins, Ketz & Williams (1987), Gilbert (1990), Charitou (2004) and Piatti (2014b) investigate
the role of cash flow ratios in order to improve firm assessments in business failure prediction
models.
Page 322
A key aspect, which emerges from the literature outlined above, concerns the kind of
information adopted. The several dimensions on which information is recorded, collected,
compared, coded and catalogued allow a twofold classification to be made between hard and soft
information (Petersen, 2004). Financial ratios, contained in balance sheets and income
statements, are characterised by hardness, precision, verifiability and manipulability (Godbillon-
Camus & Godlewski, 2005). On the other hand, soft information is considered to be a stock of
information that embraces economic, social, financial, political and personal aspects (Cosma,
2002). Soft information provides more advantages in decentralised organizational structures than
in larger hierarchies, where higher levels of bureaucracy, such as division managers, have been
characterised by several and well documented reports, which, however, are “not terribly
informative” (Stein, 2002). The harmonization between hard and soft information allows to have
a forward-looking perspective (qualitative components), as opposed to the backward-looking one
(quantitative elements), in order to obtain richer and more sustainable predictions of business
failures and, indeed, to avoid an “excess of automatisms” (Draghi, 2010). Over the years, only a
few bankruptcy prediction studies have considered both quantitative and qualitative factors.
Some authors have analysed qualitative credit risk regarding German banks (Weber, Krahnen &
Vossmann, 1999; Brunner, Krahnen & Weber, 2000; Günther & Grüning’s, 2000) and German
companies (Hesselmann, 1995; Blochwitz & Eigermann, 2000; Lehmann, 2003; Grunert, Norden
& Weber, 2005). Altman, Sabato & Wilson (2008), on a sample of 5.8 million sets of accounts of
unlisted UK SMEs along the 2000-2007 time period, investigate the role played by qualitative
information as predictive variables of company distress. The authors argue that qualitative
information, such as legal action by creditors to recover unpaid debts, company filing histories,
comprehensive audit report data and firm-specific characteristics, can significantly improve the
accuracy of the distress assessment by up to 13%. Combining hard and soft information on a
sample of Italian North-Eastern SMEs, Gibilaro & Piatti (2012) argue that both the company’s
confidence and relationship banking (as qualitative variables) improve its position in a prediction
failure perspective. On a sample of 1,446 private firms obtained from a Taiwanese finance
company in September 2000 and October 2005, Chen, Huang & Tsai (2013) show that soft
information, such as employee loyalty, the degree of the borrowing firm’s sales to long-term
customers, and the sum of the adjustments made by loan officers on the scores for financial
ratios regarding both leverage and profitability, have a crucial capacity to predict defaults of
small business borrowers. Using a sample of 389 small loans granted by a French credit
cooperative bank, Cornée (2013) emphasises that management quality and project quality,
together with hard information variables, improve credit default prediction.
The benefit from the early detection of corporate failure signals has highlighted, over the
years, the crucial concept of efficiency relative to its origin and operational variations. In the past
twenty years, efficiency analysis has become an “appealing theme” for academic research (Resti,
1997). In the rich and heterogeneous literature, Data Envelopment Analysis (DEA) and the
Stochastic Frontier Approach (SFA) are the most widely adopted approaches. The advantages of
these non-parametric and parametric tools derive from the possibility to improve on
methodologies such as discriminant analysis (Retzlaff-Roberts & Puelz, 1996) in order to
investigate a firm’s inefficiency (through distance from the “best practice”) as an ex-ante
Page 323
indicator of business failure and, moreover, to strengthen the role of qualitative variables in
predicting business failure (Becchetti & Sierra, 2003).
In parallel with DEA methodology, several authors have adopted a parametric technique,
i.e. the stochastic frontier approach (SFA), developed by Aigner, Lovell & Schmidt (1977), to
assess the efficiency of companies in the financial environment (Aly, Grabowski, Pasurka &
Rangan, 1990; Kaparakis, Miller & Noulas, 1994; Allen & Rai, 1996; Berger & Mester, 1997)
and as a tool with which to detect inefficiency measures (Jondrow, Knox Lovell, Materov &
Schmidt, 1982; Ferrier & Lovell, 1990; Hunt-McCool, Koh & Francis, 1996; Kwan & Eisenbeis,
1996). Becchetti & Sierra (2003) adopted the SFA to test whether productive efficiency could
predict bankruptcy. Using data from Mediocredito Centrale surveys (from 1989 to 1997), they
concluded that firm efficiency, measured as the distance from the efficient frontier, can have
explanatory power in predicting bankruptcy, and that the predictive capacity can be further
improved with both balance sheet and qualitative variables. Hwang, Siao, Chung & Chu (2011)
have obtained similar results. Comparing the probability of bankruptcy on the Merton model and
on a discrete-time hazard model, both of them used in the SFA framework, they found that as the
time of bankruptcy draws near, the technical efficiency derived from SF decreases. Styrin (2005)
applied the SFA to obtain X-efficiency for all Russian banks during the period between
1999(Q1) and 2002(Q4). Their X-efficiency scores were positively correlated with the share of
big risks in a bank’s loan portfolio and negatively correlated with the share of related borrowers.
Recently, some authors have tried to associate the Mixture Hazard Model (Almanidis & Sickles,
2012) and the distance to default (Merton’s model) (Saeed & Izzeldin, 2014) with the SFA.
In the above studies, the combination of hard information, soft information, and estimates
of technical efficiency proxies has further enriched and broadened out the discussion on
bankruptcy prediction. Nonetheless, the adoption of soft information, rather than financial ratios,
both to detect bankruptcy and to conduct efficiency estimation (through non-parametric and
parametric techniques) is still rare. Ceteris paribus, methods that take account only of hard
information to predict corporate bankruptcy are strongly oriented to a backward-looking
perspective. Recently, although within limits, the academic literature has paid attention to soft
information and its ability to predict early bankruptcy. This study aims to fill this gap by gaining
better understanding of a forward-looking perspective based on the role of soft information
captured by a stochastic frontier approach.
SAMPLE AND METODOLOGY
The data used in the empirical analysis were entirely drawn from AIDA, a financial
database maintained by Bureau Van Dijk which consists of the balance sheet data of all the
Italian firms. To be pointed out is that, while banks usually generate their statistical models on
the basis of private data, this study considers only available public data. Two issues were
considered when building the sample. The first was the size of firms, and the second was the
default definition. As regards the former issue, given the objective of this study, we focused only
on SMEs. To this end, according to Basel 2 and European Union definition rules, we consider
firms with a turnover of less than 50 million euros and/or with fewer than 250 employees. The
latter issue is more challenging because there are several definitions of default in the literature
Page 324
(Everett & Watson, 1998; Daubie & Meskens, 2001), including bankruptcy (Altman, 1968),
crisis (Gilbert, Menon & Schwartz, 1990; Chowdhury & Lang, 1993) organisational exit
(Swaminathan, 1996), collapse (Argenti, 1986), temporary and slight financial distress that does
not necessarily result in failure (Gentry, 1985; Johnsen & Melicher, 1994; Piatti, 2012; Piatti,
2014a). In this study, however, following Pederzoli & Torricelli (2010), default occurs when a
firm is insolvent and itself, one or more creditors, the Public Prosecutor or the Law Court can
request the beginning of bankruptcy proceedings.
According to the definitions of SMEs and of default, a sample for the period 2007-2009
was extracted by focusing on two groups: firms that did not go bankrupt in year 2009 and firms
that went bankrupt in year 2009. In this way it was possible to analyse the characteristics of firms
that influenced the probability of failure two years and one year prior to the default.
From the perspective of economic activity, the sample was built so that financial and
construction firms were excluded. Moreover, the sample did not include firms less than five
years of age. After dealing with missing values, a sample of 8,145 firms was selected from
AIDA. 253 firms in the sample, i.e. 3% of the firms selected, went bankrupt in 2009. Table 1
shows the distribution of firms by class of turnover, economic sector and location.
Table 1
FIRM DISTRIBUTION BY TURNOVER CLASS, ECONOMIC SECTORS AND MACRO-AREA
LOCATION
Turnover Firm Economic Firm Macro- Firm
class (EUR distribution sector distribution area distribution
thousands) based on based on location based on
turnover productive macro-area
class (%) sectors (%) location
<5,000 0.73 Service 16.17 NW 41.5%
5,000-10,000 42.60 Commerce 37.24 NE 18.55%
10,000-15,000 23.13 Industry 46.59 M 28.48%
15,000-20,000 12.3 S 11.47%
20,000-30,000 13.40
30,000-40,000 5.47
40,000-50,000 2.36
Total 100 100 100%
Notes: NW = North-West; NE = North-East; M = Middle; S = South including Islands.
The dependent variable is dichotomous and takes the value of zero if the firm is sound
and the value of one if the firm is failed. With regard to the independent variables, given the
absence of a commonly accepted financial ratio framework, 25 financial ratios were selected
from among those most frequently used in the literature (Bellovary, Giacomino & Akers 2007;
Lin & Chen, 2011; Serrano-Cinca & Gutierrez-Nieto, 2013). They represent the following
profiles (Beaver, 1966; Ohlson, 1980; Levitan & Knoblett, 1985; Gentry, Newbold & Whitford,
1987; Charitou, Neophytou & Charalambous, 2004; Altman & Sabato, 2007;): 1) liquidity,
which provides the defensive cash resources for firms to meet claims for payment (Keasey &
Page 325
Watson, 1987); 2) activity, which proxies for the sale generating ability of the company’s assets;
3) profitability, which is an important indicator of a firm’s health (Zavgren, 1985); 4)
capitalization, which provides a cushion to absorb fluctuations in the borrower’s earnings and
asset values; 5) interest coverage, which determines how easily a company can pay interest on
outstanding debt. Besides the financial ratios, and given the aim of the analysis, an efficiency
ratio was introduced. It is discussed in detail in the next section. Financial and efficiency ratios
were calculated based on accounting data related to the period 2007-2008.
Owing to potential extreme value problems in the variables of financial ratios (excluding
efficiency ratio) that could alter the results of the analysis, we followed Tinoco & Wilson (2013)
by using the hyperbolic tangent transformation (tanh transformation) to provide a satisfactory
solution to the outliers issue instead of the traditional “windsorising” technique. The hyperbolic
function tanh(x) was used and tested in robust signalling processing as well as in statistical
estimation, and it was shown to be very useful in decreasing the effect of extreme values of a
specific variable (Godfrey, 2009).
A principal component analysis (Lattin, Carroll & Green, 2003) was applied to the set of
financial ratios with the purpose of identifying a latent structure characterized by components
able to reduce the complexity, to synthesise the information, and to tackle multicollinearity of the
independent variables. Components and efficiency ratios were merged in order to obtain a final
complete model characterised by the presence of both efficiency and financial ratios. In addition,
as an alternative to a static analysis we computed the averages of the efficiency and financial
ratios, as in Edminster (1972) and Appetiti (1984), over two years (2007 and 2008), and we used
the averages as independent variables in the model. For year 2008, the trend of the independent
variables computed over two years (from 2007 to 2008) was added as a regressor to the
independent variables of the final model. The trend included upward and downward financial
and efficiency ratio movements obtained by comparing values at the end of 2008 with those at
the end of 2007. Depending on this movement, a dichotomous variable was introduced for each
covariate. The variable takes the value of 1 for upward movement and 0 for downward
movement. Unlike the methodology of Edmister (1972), in our analysis the dichotomous variable
synthesising the trend did not replace the other independent variables but instead added to them.
In this way, the model simultaneously presents the static information of the year immediately
preceding the classification of firms (the year 2008) and the dynamic information synthesised
from the trend, which represent an implicit correction of any accounting manipulations. Finally,
several control variables were added to the regressors: 1) a dummy variable for each economic
sector (commerce and industry; service as baseline category); 2) a dummy variable for each
location in Italy (North-East, Middle, and South including Islands; North-West as baseline
category); 3) a dummy variable for each category of the sales size (between 5 and 10 million,
between 10 and 15 million, between 15-20 million, between 20-30 million, between 30-40
million and between 40-50 million; less than 5 million as a baseline category); the age of the
firm represented by its logarithm.
A logistic regression model including the financial components, the efficiency ratio and
the controls was estimated to determine the probability of a firm entering the zero class (sound)
or class one (failed). In particular, the model consists of a set of independent variables that
Page 326
explain a dependent dichotomous variable. The logit binary model (Wooldridge, 2009; Cameron
& Trivedi, 2010) assumes that the stochastic error follows a logistic distribution.
A critical value of 3% was used to discriminate firms between the two classes. Following
Tinoco & Wilson (2013) this value was the ratio between the number of failed firms and the total
number of sample firms. This threshold took into account that the number of sound and failed
firms was not balanced.
EFFICIENCY RATIO
In the economic literature, the idea of efficiency has taken deep root, giving rise to a
unanimous and a well-established accommodation (Resti, 1997). Within this framework, Varian
(1990) provides a production efficient plan definition: “a production plan in is called
efficient, if there is no in such that ; i.e., a production plan is efficient if there is no way
to produce more output with the same inputs or to produce the same output with less inputs”. In
addition, production efficiency leads to at least three efficiency definitions: technical, allocative,
and revenue efficiency (Fanti, 1997).
Focusing on technical efficiency, i.e. maximum output from a given combination of
factors, the efficiency ratio was determined by a cross-section analysis using a stochastic frontier
production function, which took into account a sample of 8,145 firms over the 2007-2008 time
period. The model that fits the production function is of the form
( )
The technical inefficiency effects, in accordance with Battese & Coelli (1995), are:
The function, estimated through the stochastic frontier approach, is a Cobb-Douglas production
function with two input factors, specified as follows:
( ⁄ ) ( ⁄ ) ( ⁄ ) ( )
where
( ⁄ mpl) is the natural log of the production value per worker for the th firm at time of
i,t
observation;
(Crmcm⁄ mpl) is a vector of the first factor input Crmcm, i.e. the natural log of cost of raw
i,t
material, consumables and merchandise per worker for the th firm at time ;
( ixAss⁄ mpl) : is a vector of the second factor input FixAss, i.e. the natural log of fixed assets
i,t
per worker for the th firm at time ;
i,t : is assumed to be independent and identically distributed (i.i.d.) (0 ) random errors;
Page 327
i,t : is assumed to be independent and identically distributed (i.i.d.) non-negative truncations of

the ( ) distribution that accounts for technical inefficiency in production. Note that i,t is
obtained by truncation at zero of the normal distribution with mean i,t and variance ;
i,t : is a vector of explanatory variables associated with the technical inefficiency of production
of each firm over the time period considered, and is a vector of unknown coefficients. The
technical inefficiency effect i,t , in the stochastic frontier model outlined above, is regressed on
the following variables assumed to influence efficiency:
( ) ( ) ( ) ( )
where:
Sizei,t is the natural log of the total employees for the th firm at time ;
Agei,t is the natural log of the total years in activity for the th firm at time ;
Geogi are dummy variables that refer to the geographic location (North-East, Middle, and South
including Islands; North-West as baseline category);
Sectori are dummy variables that refer to the economic sector (commerce and industry; service
as baseline category).
As the technical inefficiency component is estimated, the measure of technical efficiency is

defined as follows:
( ) ( )
As suggested by Battese & Coelli (1988), the technical efficiency value is determined as
( )
( )
where:
1
i,t is the production for the th firm at time .
The technical efficiency measure takes a value between 0 and 1. or example, if a firm’s
technical efficiency is 0.78, this means that, on average, it uses 78% of its potential production
capacity, compared with a fully efficient firm with the same mix and size of inputs.
The adoption of a parametric approach requires selection of a specific functional form of the
error distribution and of the inefficiency component. If we assume orthogonality between
inefficiency components, random errors, input and output factors, three kinds of inefficiency
components distribution are offered by the literature: half-normal, normal-truncated and normal-
exponential.
We adopt the parameterization provided by Frontier2 where and are replaced with
and ( ⁄ ) respectively, and where takes a value between 0 and 1. In
Page 328
particular, when takes values close to 1, we assume that the deviations from the frontier are
only due to technical inefficiency. On the contrary, when takes values close to 0, we assume
that the deviations are related only to random error.
The Cobb-Douglas production function consists of two inputs: cost of raw material,
consumables and merchandise per worker; and fixed assets per worker. All the coefficients of the
input factor variables are statistically significant at 1% level. These coefficients, interpretable as
partial elasticity, show a greater reaction of the production value per worker to the cost of raw
materials (0.32), for both 2007 and 2008, than to the fixed assets (0.11 and 0.08) in 2007 and
2008 respectively. A higher value for the cost of raw materials highlights the important role
played by the input variable factors used in the production process (Trestini, 2006). The decrease
in the marginal productivity of fixed assets is supported by the engineering industry downturn
that occurred in the last quarter of 2008. Overall, the deterioration of the marginal productivity of
the Italian industrial sector is due in particular to the shrinkage of productive capacity and the
credit crunch (Banca d’Italia, 2008). A more exhaustive analysis concerns the assessment of
some production sets’ properties, such as constant, and non-increasing/decreasing returns to
scale. The production technology exhibits constant returns to scale if, on using the double
amount of each input the quantity produced is also double. However, the Wald test rejects the
null hypothesis at 1% statistical level and hence suggests that the production function does not
exhibit constant returns to scale.3 Moreover, in order to verify the existence of non-increasing or
non-decreasing returns to scale, a point estimate for linear combination test was conducted. The
sum of the coefficients estimated is significantly less than one (0.43 in 2007 and 0.41 in 2008).4
Therefore, the production function exhibits non-increasing returns to scale, i.e. if the cost
of raw materials, consumables and merchandise together with fixed assets are doubled, the
output is less than double and any feasible input-output vector can be scaled down (Mas-Colell,
Whinston & Green, 1995).5
Table 2 shows the empirical results.
Table 2
EMPIRICAL RESULTS STOCHASTIC FRONTIER APPROACH
Frontier Production 2007 2008
Crmcm/Empl 0.3256*** 0.3229***
(0.005) (0.005)
FixAss/Empl 0.1106*** 0.0875***
(0.006) (0.005)
Constant 8.1675*** 8.4170***
(0.081) (0.080)
Inefficiency Components
Size 1.5845*** 1.5962***
(0.056) (0.057)
Age 0.0471 0.0647
(0.041) (0.041)
North-East 0.2316* 0.2119*
(0.101) (0.101)
Middle 0.0221 0.0032
(0.096) (0.096)
South 0.5425*** 0.4484***
(0.125) (0.127)
Commerce Sector 1.1023*** 1.1885***
(0.152) (0.155)
Industry Sector 1.4056*** 1.4764***
(0.138) (0.143)
Page 329
N. Observations 8,129 8,136

̅ 0.403 0.403
0.585 0.588
0.504 0.5095
0.321 0.319
Log-Likelihood -7,492.41 -7,554.22
Stochastic Frontier Estimations (SFA). This table presents the production function efficiency regression equations using the stochastic frontier
method for cross-section. The entire sample is composed of 8,145 Italian firms (firm-year data from balance sheets and income statements) over
the period 2007-2008, split up by geographical area for all sectors. The industry sectors, where each firm belongs to, are: Commerce, Industry
and Service. The dependent variables is: PRODUCTION VALUE= natural log of production value per worker for the th firm at time of
observation; independent variables-input factors (CRMCM/EMPL) COST RAW MATERIAL CONSUMBABLE & MERCHANDISE= natural
log of cost raw material consumables and merchandise per worker for the th firm at time of observation; (FIXASS/EMPL) FIXED ASSETS=
natural log of fixed assets per worker for the th firm at time of observation; inefficiency components are: SIZE= natural log of the total
employees for the th firm at time of observation; AGE= the natural log of the total years of life for the th firm at time of observation;
NORTH-EAST= dummy variable equals to one if the firm is located in western-northern Italy, zero otherwise; CENTRE= dummy variables
equals to one if the firm is located in centre Italy; SOUTH & ISLANDS= dummy variable equals to one if the firm is located in southern Italy
and in its islands, zero otherwise; COMMERCE SECTOR= dummy variable equals one if the firm belongs to the commerce sector, zero
otherwise; INDUSTRY SECTOR= dummy variable equals to one if the firm belongs to industry sector, zero otherwise. The standard errors are
on brackets. = represents the identically and distributed random error; = is the standard deviation of technical inefficiency; = is the
standard deviation of random error; = is the sum between and ; = ( ⁄ ). The levels of significance are the following: *,**,***
which denote respectively, 10%,5% and 1% significance level.
The coefficients of the variables affecting the inefficiency measure were estimated
simultaneously with the frontier. If the coefficient is positive, it increases the inefficient term ,
and therefore decreases the efficiency technical value. Vice versa, if the coefficient is negative, it
reduces the inefficiency and positively affects the efficiency value on the whole. According to
the results, larger firm size has a negative impact on production efficiency. Furthermore, firms
located in North-Eastern and Southern Italy are significantly less efficient than North-Western
Italy (Becchetti & Sierra, 2003). These findings point to other explanatory factors such as
weaknesses in the civic community, civic engagement, solidarity, honesty, trust, law-
abidingness, stronger crime control, and social capital (Putnam, 1993). With regard to the
economic sector, the results hint that both the commerce and industry sectors show a lower
production efficiency than the service sector.
EMPIRICAL RESULTS
An exploratory analysis was carried out, and the results are shown in Table 3. Mean,
standard deviation and median of the financial ratios are presented on the basis of two groups:
sound firms and failed firms. The t-test on the mean and the Wilcoxon-Mann-Whitney test on the
median were performed to evaluate the univariate discriminatory power of each ratio. The
statistics of the financial ratios highlight, with a few exceptions, that the means and medians are
significantly different between the two subsets.
Page 330
Table 3
DESCRIPTIVE STATISTICS FOR RATIOS SPLIT IN SOUND FIRMS AND FIRM WITH FINANCIAL
DISTRESS
2008 2007 2008
Ratio Sound firms Failed firms Sound firms Failed firms T- Median
Mean Median Std Mean Median Std Mean Median Std Mean Median Std test Test
CA/CL 0.826 0.834 0.122 0.769 0.785 0.114 0.828 0.835 0.127 0.748 0.768 0.131 0.00 0.00
QUICK 0.692 0.716 0.193 0.568 0.575 0.175 0.690 0.715 0.202 0.548 0.542 0.185 0.00 0.00
CASH/TA 0.070 0.030 0.097 0.027 0.008 0.048 0.064 0.024 0.094 0.020 0.005 0.045 0.00 0.00
Working capital/total asset 0.273 0.291 0.207 0.305 0.317 0.168 0.275 0.288 0.202 0.295 0.291 0.178 0.09 0.33
Operating cash flow/financial
0.095 0.058 0.123 0.043 0.032 0.079 0.094 0.053 0.137 -0.003 0.014 0.119 0.00 0.00
debt
Inventories/average daily
0.902 1.000 0.284 0.942 1.000 0.219 0.900 1.000 0.287 0.937 1.000 0.234 0.04 0.00
sales
Net working capital flow/debt 0.107 0.078 0.115 0.071 0.066 0.081 0.106 0.076 0.126 0.031 0.044 0.112 0.00 0.00
Net working capital/sales 0.213 0.202 0.183 0.327 0.318 0.219 0.224 0.207 0.188 0.359 0.342 0.247 0.00 0.00
Sales/total asset 0.853 0.896 0.152 0.725 0.740 0.193 0.830 0.880 0.169 0.665 0.657 0.212 0.00 0.00
Ebitda/totale asset 0.099 0.083 0.079 0.067 0.064 0.059 0.086 0.073 0.077 0.028 0.045 0.093 0.00 0.00
Depreciation/fixed asset 0.198 0.152 0.162 0.126 0.088 0.116 0.181 0.127 0.168 0.139 0.087 0.155 0.00 0.00
Return on asset (ROA) 0.065 0.050 0.075 0.037 0.038 0.056 0.053 0.043 0.074 -0.008 0.021 0.108 0.00 0.00
Return on debt 0.023 0.021 0.021 0.038 0.038 0.019 0.025 0.024 0.019 0.035 0.035 0.015 0.00 0.00
Return on equity 0.085 0.074 0.223 -0.013 0.011 0.218 0.044 0.043 0.231 -0.224 -0.035 0.384 0.00 0.00
Return on sales 0.045 0.035 0.068 0.037 0.038 0.080 0.038 0.031 0.071 -0.021 0.022 0.159 0.00 0.00
Net income/total asset 0.025 0.012 0.055 0.000 0.001 0.041 0.020 0.008 0.060 -0.046 -0.005 0.106 0.00 0.00
Net income/sales 0.018 0.008 0.058 -0.005 0.001 0.068 0.013 0.005 0.068 -0.072 -0.006 0.160 0.00 0.00
Earning before taxes/total
0.053 0.032 0.078 0.012 0.010 0.051 0.040 0.022 0.078 -0.038 0.000 0.111 0.00 0.00
asset
Equity/total asset 0.228 0.191 0.159 0.180 0.156 0.119 0.265 0.239 0.168 0.176 0.155 0.124 0.00 0.00
Equity/debt 0.803 0.992 0.470 0.621 0.858 0.544 0.704 0.968 0.572 0.241 0.696 0.793 0.00 0.00
Sales/debt 0.325 0.234 0.270 0.240 0.184 0.196 0.387 0.312 0.288 0.236 0.183 0.203 0.00 0.00
Ebit/interest expenses 0.018 0.016 0.014 0.031 0.031 0.015 0.019 0.017 0.014 0.029 0.029 0.014 0.00 0.00
Interest expenses/ebitda 0.016 0.010 0.031 0.040 0.032 0.047 0.018 0.012 0.033 0.042 0.035 0.039 0.00 0.00
Interest expenses /total assets 0.918 0.964 0.121 0.799 0.826 0.171 0.914 0.963 0.129 0.738 0.746 0.198 0.00 0.00
Interest expenses /sales 0.016 0.010 0.031 0.040 0.032 0.047 0.018 0.012 0.033 0.042 0.035 0.039 0.00 0.00
All financial ratios were transformed through hyperbolic tangent transformation whose range was from -1 to 1. The F-test was used in the context of
variance analysis (ANOVA). The T-Test was replaced by Welch statistics if the hypothesis of variance homogeneity between groups (Leven’s test) was
rejected. The Welch test provided robust results even in the presence of the heterogeneity of variance between the two groups. A non parametric test, the
Wilcoxon-Mann-Whitney test, was conducted for the median.
Through principal component analysis, the 25 financial ratios described above were
reduced to their main components. Table 4 shows the components, and the names allocated to
them.
Page 331
Table 4
PATTERN MATRIX WITH THE COMPONENTS
Profitability
Capitalization Asset Working
and debt Debt cost
Financial ratios service
and liquidity turnover capital
capacity
Operating cash flow/financial debt 0.73
Net working capital flow/debt 0.77
Ebitda/totale assets 0.89
Return on assets (ROA) 0.92
Return on equity 0.75
Return on sales 0.83
Net income/total assets 0.89
Net income/sales 0.73
Earning before taxes/total assets 0.89
Ebit/interest expenses 0.68
Interest expenses/ebitda -0.64
Current asset/current liabilities 0.62
(Current asset-inventories)/current
0.63
liabilities
Liquidity/total assets
Equity/total assets 0.89
Equity/debt 0.90
Sales/total assets 0.89
Depreciation/fixed assets 0.49
Sales/debt 0.86
Working capital/total assets 0.87
Net working capital/sales 0.90
Inventories/average daily sales 0.58
Return on debt 0.90
Interest expenses /total assets 0.86
Interest expenses /sales 0.66
xtraction method: principal component analysis. Bartlett’s test of sphericity was conducted with rejection of the null hypothesis. KMO test
results were 0.79, higher than 0.7. The factors loading matrix was orthogonally rotated with the Varimax method. The component number was
not defined on the basis of eigenvalues with a value greater than 1 but by the parallel analysis (Zwick & Velicer, 1986). The components
explained 73.88% of the variance for the financial ratios. Low commonality, less than 0.4 (Haslam et al. 1992) is not reported.
Note that the 25 financial ratios initially considered are reduced to 5 components, with
great interpretative advantage. The components are named by considering the values assumed by
factor loadings in the rotated matrices. In particular, the first component, which explains 29.05%
of the variance, is named “Profitability and debt service capacity” because it includes
profitability and cash flow ratios that reflect the ability of the firm to generate profits and funds
from its operations to meet its financial obligations. The second component, named
“Capitalization and liquidity” includes ratios related to the financial structure and short-term
liquidity. Higher loadings of equity and liquidity positively affect the component. The third
component, named “Asset turnover”, consists of ratios that measure how effectively the firm
Page 332
deploys its resources. The fourth component, named “Working capital”, contains three
indicators, which proxy for the relationship between sales or assets and the need to finance
working capital. The last component, named “Debt cost”, approximates the economic impact of
debt and the cost of borrowings. Means, standard deviations and medians of the financial
components in year 2008 are presented in Table 5 on the basis of two groups: sound firms and
failed firms.
Table 5
PATTERN MATRIX WITH THE COMPONENTS: YEARS 2008
Sound firms Failed firms
Median
Components T-test
Test
Mean Median Std Mean Median Std
Profitability and debt

-0.072 -0.075 1.009 -1.003 -0.289 1.7842 0.00 0.00
service capacity
Capitalization and
0.131 -0.069 1.016 -0.196 -0.212 0.792 0.00 0.00
liquidity
Asset turnover -0.026 0.173 1.008 -1.054 -1.013 1.483 0.00 0.00
Working capital -0.019 -0.009 0.989 0.411 0.387 1.453 0.09 0.33
Debt cost 0.057 -0.014 0.982 0.367 0.382 1.004 0.00 0.00
T-test on the mean and Wilcoxon-Mann-Whitney test on the median are performed to evaluate the univariate discriminatory power of
each component.
The statistics of the components differ markedly between the two subsets.
The present study reports ex-ante models developed for the estimation of default
likelihood. Accordingly, estimation was made of the probability of failure in the year prior to the
observation of default (t-1; 2008) as well as two years prior to the financial distress event (t-2;
2007). Consequently, first the models provided evidence about the predictors that best
discriminated between failed and healthy enterprises; second, they tested their predictive power.
Table 6 reports the results of the logit model with regard to three different specifications.
In the first specification, we used only variables related to the efficiency ratios. In the second
specification, only the components of the financial ratios were used, and in the third
specification, we estimated the model with both the efficiency ratios and the components of the
financial ratios. To interpret the results, negative coefficients indicate that the variable has a
positive influence on the probability of a firm being assigned to the class of sound firms (the
class with the value of zero).
Page 333
Table 6
LOGISTIC REGRESSION RESULTS. THE DEPENDENT VARIABLE HAS A VALUE OF 0 FOR SOUND
FIRMS AND A VALUE OF 1 FOR FIRMS IN FINANCIAL DISTRESS
2008 2007
Default
1° spec 2° spec 3° spec 1° spec 2° spec 3° spec
Efficiency -3.571*** -3.721*** -4.009*** -3.912***
(0.341) (0.381) (0.365) (0.392)
Commercial sector -0.582** -0.199 -0.185 -0.596** -0.284 -0.256
(0.263) (0.327) (0.325) (0.261) (0.325) (0.328)
Industrial sector 1.001*** 1.266*** 1.283*** 0.987*** 1.256*** 1.289***
(0.214) (0.288) (0.287) (0.212) (0.277) (0.28)
North-East -0.16 -0.2 -0.166 -0.135 -0.177 -0.163
(0.192) (0.199) (0.202) (0.19) (0.195) (0.199)
Middle 0.121 -0.0481 -0.00535 0.157 0.0217 0.0415
(0.156) (0.167) (0.169) (0.156) (0.164) (0.167)
South 0.268 -0.0194 0.0384 0.311 0.0209 0.0274
(0.218) (0.243) (0.246) (0.221) (0.232) (0.233)
Sales 5-10 -2.218*** -1.189** -1.420*** -0.751** -0.886** -0.969***
(0.39) (0.511) (0.519) (0.335) (0.361) (0.361)
Sale 10-15 -2.525*** -1.383*** -1.666*** -0.580* -0.642* -0.660*
(0.406) (0.524) (0.531) (0.344) (0.369) (0.368)
Sales 15-20 -2.172*** -0.967* -1.242** -0.483 -0.51 -0.567
(0.418) (0.538) (0.547) (0.364) (0.388) (0.389)
Sales 20-30 -2.195*** -0.884* -1.155** -0.158 -0.303 -0.358
(0.413) (0.531) (0.54) (0.348) (0.38) (0.381)
Sales 30-40 -2.040*** -0.841 -1.228** -0.118 -0.0525 -0.141
(0.451) (0.562) (0.578) (0.387) (0.404) (0.412)
Sales 40-50 -2.732*** -1.422** -1.664**
(0.592) (0.695) (0.717)
Age -0.108 -0.143* -0.147** -0.136* -0.139* -0.182**
(0.071) (0.074) (0.075) (0.07) (0.073) (0.074)
Profitability & debt serv capacity -0.778*** -0.787*** -0.530*** -0.548***
(0.068) (0.069) (0.077) (0.079)
Capitalization and liquidity -0.821*** -0.828*** -0.642*** -0.609***
(0.115) (0.118) (0.106) (0.105)
Asset turnover -0.484*** -0.488*** -0.534*** -0.535***
(0.05) (0.051) (0.052) (0.055)
Working capital 0.308*** 0.305*** 0.332*** 0.305***
(0.060) (0.06) (0.063) (0.06)
Debt cost 0.369*** 0.363*** 0.486*** 0.477***
(0.057) (0.061) (0.069) (0.065)
Constant 1.278** -3.278*** -0.324 -0.0616 -3.557*** -0.524
(0.559) (0.658) (0.725) (0.513) (0.524) (0.589)
N. Observations 8,145 8,145 8,145 8,145 8,145 8,145
Pseudo R2 0.09 0.228 0.257 0.096 0.182 0.218
Log-Likelihood ratio test 204.06*** 513.9*** 580.32*** 215.37 410.16*** 491.44
1° type error cut-off 0.03 0.257 0.205 0.182 0.257 0.205 0.202
Accuracy ratio cut-off 0.03 0.648 0.757 0.773 0.659 0.732 0.758
Sensitivity 0.743 0.794 0.818 0.743 0.794 0.798
Specificity 0.645 0.755 0.771 0.657 0.729 0.754
Area under ROC curve 0.751 0.856 0.871 0.758 0.838 0.857
BIC -86.99 -360.81 -418.22 -107.32 -266.09 -338.37
2
Hosmer-Lemeshow 17.67* 11.7 8.64 12.6 14.49* 7.13
The results in the table were obtained with logit regression. All z statistics have been corrected to take into account
the heteroscedasticity and the autocorrelation of the errors. The standard errors are reported in parentheses. *, ** and
*** indicate levels of significance of 10 %, 5% and 1 %, respectively. The first specification of the model included
only the variables determined by the presence of efficiency ratio. The second specification represented the base
model obtained using only the components of financial ratios, and the third specification considered the basic model
in addition to the variables determined by the efficiency ratio. The log-likelihood ratio test was obtained from the
Page 334
following difference: (-2 log-likelihood of the base model) - (-2 log-likelihood of the full model). The first type error
represents the percentage of failed firms evaluated as sound firms from the model. The accuracy ratio was
determined by the ratio between the firms properly classified and the total of firms in the sample. The sensitivity
measures the ability of the model to predict failure events correctly, while the specificity defines the ability of the
model to predict non-financial distress events correctly. The area under the ROC curve is a measure of the predictive
accuracy of the model, with a value of 1 representing a perfect model. The cut-off of 0.03 was determined by the
ratio between the number of bankrupted enterprises in the sample and the total number of sampled enterprises.
If only the efficiency ratio is considered (first specification of the model), the results are
not encouraging. The coefficients of the efficiency ratio are negative and statistically significant
for both one year (2008) and two years (2007) prior to the default, as expected, but all the
statistical diagnostics are indicative of poor performance. In particular, the significance of the
Hosmer-Lemeshow χ2 test does not reveal a goodness of fit of the model, and the high
percentage of failed firms evaluated as sound firms by the model (first type error: 22.92%)
confirms that efficiency, although very important, in isolation is not a good predictor of default.
In the second specification of the model, only the predictive ability of financial ratios was
considered, without the contribution of efficiency. All the components that synthesize the
financial ratios show high statistical significance levels for both years, 2008 and 2007. In
particular, the negative signs of the coefficients associated with the components “Profitability
and debt service capacity”, “Capitalization and liquidity” and “Asset turnover” indicate that the
greater these ratios, the more likely a firm’s classification as sound. By contrast, the positive
signs of the coefficients of the “Working capital” and “Debt cost” components increase the
likelihood of a firm being classified as failed.
The behaviour of the above variables appears economically justifiable and correct. To be
noted is that the level of working capital and its management play a crucial role in generating
financial needs. Moreover, in SMEs like those analysed in this study fixed and large investment
in plants and equipment are less important than the day-by-day management of working capital.
The cost of debt is also very important, and it depends both on the leverage and on the cost of
debt itself. In fact, all financial ratios represented by their components maintain, in line with the
literature (Unal, 1988), high predictive power for 1 or 2 years prior to firm classification.
Regarding the control variables, such as the economic sector, geographic area and size,
the coefficients exhibit an interesting behaviour. The coefficient of the industrial sector variable,
in particular, has a high statistical significance level in all the specifications and for both years,
while the commercial sector variable is statically significant only in the first specification. Where
the firm is located in Italy seems not at all to affect the probability of default. Firms with a
turnover less than or equal to 15 million euros have a higher probability of being assigned to the
sound class, and applies for all specifications and years. Nevertheless, firms with a turnover
higher than 15 million euros also show a negative statistically significance coefficient, but only
for years 2008, that is, one year prior to default. This behaviour may be attributed to the lack of
financial planning often linked to growth that is typical of medium-sized firms. This may explain
the counterintuitive signs assumed by the coefficients linked to the size variable (Williamson,
1967). The age of the firm is also important, and the negative coefficient indicates that the longer
the firm’s life, the lower its probability of default.
Page 335
After separately analysing the roles of efficiency and financial ratios, the third
specification of the model shows the combination of both of them. As Table 6 evidences, adding
efficiency ratios to financial ratios increases the predictive capacity of the final model compared
to the second specification, for years, 2007 and 2008. All the coefficients related to the efficiency
ratios and the financial components are statistically significant. The log-likelihood ratio test
(Cameron & Trivedi, 2010) is significant for both years, 2008 and 2007, and the BIC level
provides strong support for the third specification with respect to the first and second
specifications.
In addition, going from the first to the third specification, there is an improvement in the
accuracy ratio, the first type error, the sensitivity, and the specificity for both years. However, it
seems that the model of the year before failure performs better in predicting whether a firm is
classified as defaulted. Therefore, the use of efficiency ratios can furnish more information than
the use of only financial ratios, and this implies a better prediction ability in terms of firm
creditworthiness, which is in line with the findings of Becchetti & Sierra (2003) and Psikillaki,
Tsolas & Margaritis (2010).
The results are further verified when a more dynamic model is considered. Tables 7 and 8
report the estimates obtained by using respectively the two-year simple arithmetic mean of
components and the trend ratios over the two years added to the basic model for the year. Each
trend variable is proxied by a dummy that takes the value of 1 in the case of an upward trend
over the two years and of 0 in the case of a downward trend.
Table 7
LOGISTIC REGRESSION RESULTS. THE DEPENDENT VARIABLE TAKES THE VALUE OF 0
FOR SOUND FIRMS AND 1 FOR FIRMS WITH FINANCIAL DISTRESS MEAN
Mean 1° spec 2° spec 3° spec
Efficiency -7.421*** -7.559***
(0.508) (0.556)
Commercial sector -0.551** -0.296 -0.263
(0.263) (0.318) (0.32)
Industrial sector 1.033*** 1.225*** 1.265***
(0.215) (0.276) (0.276)
North-East -0.127 -0.202 -0.172
(0.193) (0.198) (0.201)
Middle 0.146 -0.026 0.021
(0.158) (0.165) (0.171)
South 0.292 -0.027 0.034
(0.223) (0.238) (0.245)
Sales 5-10 -2.326*** -1.423*** -1.713***
-0.407 -0.453 -0.421
Sales 10-15 -2.603*** -1.648*** -1.931***
(0.422) (0.467) (0.436)
Sales 15-20 -2.287*** -1.239** -1.544***
(0.436) (0.482) (0.455)
Sales 20-30 -2.268*** -1.171** -1.419***
(0.431) (0.477) (0.445)
Sales 30-40 -2.097*** -1.158** -1.502***
(0.469) (0.514) (0.495)
Sales 40-50 -2.847*** -1.780** -2.131**
(0.608) (0.675) (0.697)
Age -0.124* -0.151** -0.201***
(0.072) (0.074) (0.076)
Page 336
Profitability & debt service capacity -0.808*** -0.832***

(0.079) (0.082)
Capitalization and liquidity -0.778*** -0.753***
(0.122) (0.128)
Asset turnover -0.524*** -0.536***
(0.051) (0.055)
Working capital 0.341*** 0.299***
(0.061) (0.062)
Debt cost 0.520*** 0.497***
(0.059) (0.062)
Constant 4.191*** -2.901*** 3.018***
(0.62) (0.61) (0.705)
N. Observations 8,145 8,145 8,145
Pseudo R2 0.13 0.216 0.279
Log-Likelihood ratio test 292.79 487.64*** 629.87***
1° type error cut-off 0.03 0.205 0.213 0.186
Accuracy ratio cut-off 0.03 0.701 0.75 0.789
Sensitivity 0.794 0.786 0.814
Specificity 0.697 0.749 0.788
Area under ROC curve 0.792 0.835 0.882
BIC -175.72 -334.55 -467.77
Hosmer-Lemeshow 2 18* 21.67* 8.43
The table shows the results obtained using as input of the regression the simple arithmetic mean of the
covariates computed on 2 years. The results of the table were obtained with logit regression. All z statistics are
corrected to take into account the heteroscedasticity and the autocorrelation of the errors. The standard errors
are reported in parentheses. *, ** and *** indicate levels of significance of 10 %, 5% and 1 %, respectively.
The first specification of the model includes only the variables determined by the presence of efficient ratio.
The second specification represents the base model obtained using only the components of financial ratios,
and the third specification considers the basic model in addition to the variables determined by the efficiency
ratio. The log-likelihood ratio test is obtained from the following difference: (-2 log-likelihood of the base
model) - (-2 log-likelihood of the full model). The first type error represents the percentage of failed firms
evaluated as sound firms from the model. The accuracy ratio is determined by the ratio between the firms
properly classified and the total firms in the sample. The sensitivity measures the ability of the model to
predict failure event correctly while the specificity defines the ability of the model to predict non-financial
distress event correctly. The area under ROC curve is a measure of the predictive accuracy of the model, with
a value of 1 representing a perfect model. The cut-off of 0.03 is determined by the ratio between the number of
companies bankrupted of the sample and the total number of companies observed.
On considering the mean values over the two years, the efficiency ratios and financial
components show high statistically significance levels in all the specifications, which further
supports the use of a model that includes both efficiency and financial ratios.
Page 337
Table 8
LOGISTIC REGRESSION RESULTS. THE DEPENDENT VARIABLE TAKES THE VALUE OF 0 FOR SOUND FIRMS AND
1 FOR FIRMS WITH FINANCIAL DISTRESS TREND
Trend 1° spec 2° spec 3° spec
Efficiency -5.368*** -5.389***
(0.454) (0.497)
Trend efficiency 1.039*** 0.950***
(0.174) (0.187)
Commercial sector -0.564** -0.228 -0.224
(0.263) (0.327) (0.327)
Industrial sector 1.025*** 1.237*** 1.236***
(0.215) (0.288) (0.288)
North-East -0.155 -0.228 -0.201
(0.194) (0.203) (0.207)
Middle 0.128 -0.07 -0.042
(0.157) (0.168) (0.171)
South 0.249 -0.002 0.025
(0.218) (0.239) (0.244)
Sales 5-10 -2.249*** -1.051* -1.422***
(0.416) (0.543) (0.511)
Sales 10-15 -2.531*** -1.200** -1.576***
(0.431) (0.555) (0.522)
Sales 15-20 -2.199*** -0.807 -1.202**
(0.443) (0.567) (0.538)
Sales 20-30 -2.199*** -0.693 -1.062**
(0.439) (0.561) (0.531)
Sales 30-40 -2.039*** -0.649 -1.129**
(0.476) (0.589) (0.569)
Sales 40-50 -2.773*** -1.217** -1.588**
(0.607) (0.716) (0.7)
Age -0.114 -0.155** -0.184**
(0.071) (0.075) (0.076)
Profitability and debt service capacity -0.795*** -0.798***
(0.067) (0.069)
Capitalization and liquidity -0.860*** -0.827***
(0.118) (0.121)
Asset turnover -0.469*** -0.480***
(0.052) (0.055)
Working capital 0.277*** 0.262***
(0.059) (0.06)
Debt cost 0.451*** 0.435***
(0.053) (0.055)
Trend profitability & debt service capacity -0.103 -0.073
(0.174) (0.179)
Trend capitalization and liquidity 0.144 0.137
(0.177) (0.184)
Trend asset turnover -0.059 -0.045
(0.165) (0.17)
Trend working capital 0.337** 0.329**
(0.152) (0.154)
Trend debt cost -0.770*** -0.724***
(0.147) (0.15)
Constant 2.092*** -3.179*** 0.727
(0.574) (0.705) (0.758)
N. Observations 8,145 8,145 8,145
Pseudo R2 0.109 0.244 0.286
Log-Likelihood ratio test 246.51*** 549.65*** 644.57***
1° type error cut-off 0.03 0.237 0.202 0.182
Accuracy ratio cut-off 0.03 0.676 0.771 0.790
Sensitivity 0.763 0.798 0.818
Specificity 0.673 0.770 0.798
Area under ROC curve 0.771 0.858 0.879
Page 338
BIC -120.44 -351.54 -428.45

2
Hosmer-Lemeshow 15.84* 20.76* 10.3
The table shows the results obtained using as input to the regression the covariates specified in Table 5 and in
addition for each covariate, its trend. The trend is represented by dummy variables that take the value of 1 in
the case of a growing trend and 0 in the case of a decreasing trend in the ratios. The results of the table were
obtained with logit regression. All z statistics were corrected to take into account the heteroscedasticity and the
autocorrelation of the errors. The standard errors are reported in parentheses. *, ** and *** indicate levels of
significance of 10 %, 5% and 1 %, respectively. The first specification of the model included only the
variables determined by the presence of efficient ratio. The second specification represented the base model
obtained using only the components of financial ratios, and the third specification considered the basic model
in addition to the variables determined by the efficiency ratio. The log-likelihood ratio test was obtained from
the following difference: (-2 log-likelihood of the base model) - (-2 log-likelihood of the full model). The first
type error represents the percentage of failed firms evaluated as sound firms from the model. The accuracy
ratio is determined by the ratio between the firms properly classified and the total of firms in the sample. The
sensitivity measures the ability of the model to predict failure events correctly while the specificity defines the
ability of the model to predict non-financial distress events correctly. The area under the ROC curve is a
measure of the predictive accuracy of the model, with a value of 1 representing a perfect model. The cut-off of
0.03 was determined by the ratio between the number of companies bankrupted of the sample and the total
number of companies observed.
On adding the variables that proxy for the trend of the financial components and
efficiency ratios to the model (Table 8), all the estimates remain statistically significant, but the
coefficients of the trends are statistically significant only for the efficiency ratio, the working
capital component and the debt cost component, which suggests that not only the levels but also
the dynamics of these variables affect the likelihood of a firm being classified as sound.
To sum up, all the estimated coefficients of the efficiency ratios and the financial
components are statistically significant at 1% in both 2008 and 2007. This suggests that both
types of variables have a high discriminating and predicting power.
Finally, for the Roc Curve point of view, the model that considers the mean of ratios
seems to perform better: indeed, the area under the Roc Curve is 88.19% for the mean model in
comparison to 87.05% and 87.88% respectively for the model of Table 6 and the model with
trends (Table 8).
ROBUSTNESS CHECK: BOOTSTRAPPING PROCEDURE
In order to support the estimates obtained with logistic regressions, we conducted a

comparative analysis of robustness by applying a bootstrapping procedure (Efron, 1979). The
bootstrap method allows the application of a general resampling technique that helps to assess
whether sample statistics or estimated regression parameters are biased. The underlying
motivation is the concern that previous logistic estimates may have be biased because the
distributional assumptions were violated or the regressors and the error terms were correlated. In
particular, the empirical estimations of the sample distribution concern the first (only efficiency
ratios considered) and third (both efficiency and financial ratios) specification of each logistic
regression conducted (technical efficiency component for 2007 and 2008 respectively, average
technical efficiency and both average and trend technical efficiency component from 2007 to
2008).6
The bootstrapping procedure is carried out in three steps: 1) identify the proper number of
replications for the estimation of the standard errors by comparing the results with the robust
Page 339
standard errors and with a few number of replications, i.e. 1,000 replications; 2) compare the
bootstrapped confidence intervals, the normal confidence intervals (N), the percentile confidence
interval (P) and the bias-corrected confidence intervals (BC); 3) investigate the relationship
between the bootstrapped standard errors and the observed coefficients of each standard error of
each independent variable, in order to detect possible noise in standard error estimates of each
variable considered.
In order to show the effects of the number of replications on standard error estimations,
Tables 9 and 10 report the comparisons between the results obtained with the robust standard
error procedure and the bootstrapping procedure (with 1,000 replications). The delta percentage,
on the right side, indicates the change in the standard errors.
Table 9
BOOTSTRAPPED STANDARD ERRORS:
1° MODEL 2007-2008 TECHNICAL EFFICIENCY RATIOS
R.S.E. B=1,000 R.S.E. B=1,000 Δ% Std Error
Variable
2007 2007 2008 2008 2007 2008
Efficiency -4.009 -4.009 -3.571 -3.571
(0.365) (0.384) (0.341) (0.349) 5.26% 2.35%
Constant -0.062 -0.062 1.278 1.278
(0.513) (0.568) (0.559) (0.592) 10.80% 5.76%
Notes: R.S.E.:Robust Standard Error; B= 1,000: bootstrapping procedure with 1,000 replications.
Table 10
3° MODEL 2007-2008 TECHNICAL EFFICIENCY COMPONENT
Variable
2007 2007 2008 2008 2007 2008
Efficiency -3.9123 -3.9123 -3.7209 -3.7209
(0.392) (0.39) (0.381) (0.385) -0.41% 1.02%
Constant -0.5245 -0.5245 -0.3239 -0.3239
(0.589) (0.609) (0.725) (0.763) 3.41% 5.17%
Table 11 reports the post-estimation results obtained by running the logistic regressions
with the bootstrapping procedure. The results, with regard to only the first and third
specifications, are equivalent to those provided by Table 10.
Table 11
POST ESTIMATION RESULTS - BOOTSTRAP PROCEDURE - EFFICIENCY COMPONENT
1° Spec. 3° Spec. 1° Spec. 3° Spec.
Description
2007 2007 2008 2008
1° type error cut-off 0.03 25.69% 20.16% 25.69% 18.18%
Accuracy Ratio cut-off 0.03 66.03% 75.61% 64.85% 77.29%
Sensitivity 74.31% 79.84% 74.31% 81.82%
Specificity 65.77% 75.47% 64.55% 77.14%
Area under ROC Curve 0.758 0.857 0.751 0.87
Like Table 10 and Table 11, Table 12 presents the results of the bootstrapped standard
errors with reference to the technical efficiency ratios and the financial components (both mean
and trend). Moreover, Table 13 provides the results of the post-estimation coefficients that,
Page 340
compared with those obtained from logistic regression without bootstrapping, have similar
values.
Table 12
1°&3° MODEL 2007-2008 TECHNICAL EFFICIENCY COMPONENT MEAN&TREND
Variable
1° Spec. 1° Spec. 3° Spec. 3° Spec. 1° Spec. 3° Spec.
Efficiency (Mean) -7.4213 -7.4213
(0.508) (0.504) -0.88%
Efficiency -5.3894 -5.3894
(0.492) (0.497) 0.95%
Efficiency (Trend) 0.9502 0.9502
(0.189) (0.187) -0.85%
Constant 4.1913 4.1913 0.7268 0.7268
(0.619) (0.656) (0.788) (0.757) 5.82% -3.93%
Table 13
POST ESTIMATION RESULTS - BOOTSTRAP PROCEDURE - EFFICIENCY COMPONENT
(MEAN&TREND)
Efficiency Component &
Efficiency (Mean)
Efficiency (Trend)
Description
1° Spec. 3° Spec. 1° Spec. 3° Spec.
2007-2008 2007-2008 2007-2008 2007-2008
1° type error cut-off 0.03 20.55% 18.58% 23.72% 18.18%
Accuracy Ratio cut-off 0.03 70.06% 78.87% 67.59% 79.03%
Sensitivity 79.45% 81.42% 76.28% 81.82%
Specificity 69.75% 78.79% 67.31% 78.94%
Area under ROC Curve 0.792 0.882 0.77 0.878
Table 14 and Table 15 show the results of the estimation of the different confidence
intervals. The confidence intervals estimated are: normal (N), percentile (P) and bias corrected
(BC). The confidence intervals for the technical efficient variable are similar in all the
specifications analysed. In particular, percentile and bias corrected methods provide similar
results. This confirms that the statistics are unbiased. Moreover, with unbiased statistics, the
bootstrap distribution becomes approximately normal. Table 14 (bis) and Table 15 (bis) report
the ratios of the estimated bias to standard error. In accordance with the finding of Efron and
Tibshirani (1993), when this ratio is small – that is, less than 0.25 – the bias is suitable.
Table 14
BOOTSTRAPPED CONFIDENCE INTERVALS EFFICIENCY RATIOS 3rd SPECIFICATION 2007-2008
[95% Conf.
Model Variable Observed Coefficient Bias Bootstrap Std Error Interval Type
Interval]
-4.697 -3.126 (N)
3rd Spec. 2007 Efficiency -3.912 -0.033 (0.400) -4.715 -3.154 (P)
-4.657 -3.124 (BC)
-4.484 -2.957 (N)
3rd Spec. 2008 Efficiency -3.721 -0.025 (0.389) -4.514 -2.978 (P)
-4.444 -2.907 (BC)
Notes: Confidence Intervals: N= Normal; P= Percentile; BC= Bias Corrected.
Page 341
Table 14 (bis)
BOOTSTRAPPED CONFIDENCE INTERVALS EFFICIENCY COMPONENTS 3rd SPECIFICATION
2007-2008
Observed Bootstrap Std. Bias/Std [95% Conf.
Model Variable Bias Interval Type
Coefficient Error Err. Int.]
-4.697 -3.126 (N)
3rd Spec. 2007 Efficiency -3.912 -0.033 0.400 -0.08 -4.715 -3.154 (P)
-4.657 -3.124 (BC)
-4.484 -2.957 (N)
3rd Spec. 2008 Efficiency -3.721 -0.025 0.389 -0.06 -4.514 -2.978 (P)
-4.444 -2.907 (BC)
Table 15
BOOTSTRAP CONFIDENCE INTERVALS EFFICIENCY COMPONENTS (MEAN & TREND)
3rd SPECIFICATION 2007-2008
Observed Bootstrap Stand. [95% Conf.
Model Variable Bias Interval Type
Coefficient Error Interval]
-8.671 -6.445 (N)
3rd Spec.
Efficiency (Mean) -7.559 -0.057 0.567 -8.740 -6.539 (P)
2007-2008
-8.626 -6.452 (BC)
-6.354 -4.423 (N)
3rd Spec.
Efficiency -5.389 -0.062 0.492 -6.438 -4.561 (P)
2007-2008
-6.334 -4.377 (BC)
0.577 1.323 (N)
3rd Spec.
Efficiency (Trend) 0.950 0.010 0.190 0.587 1.348 (P)
2007-2008
0.578 1.328 (BC)
Table 15 (bis)
BOOTSTRAP CONFIDENCE INTERVALS EFFICIENCY COMPONENTS (MEAN & TREND) 3rd
SPECIFICATION 2007-2008
Observed Bootstrap Stand. [95% Conf.
Model Variable Bias Bias/Std.Err Interval Type
Coefficient Error Int.]
-8.671 -6.445 (N)
3rd Spec. Efficiency
-7.559 -0.057 0.567 -0.10 -8.740 -6.539 (P)
2007-2008 (Mean)
-8.626 -6.452 (BC)
-6.354 -4.423 (N)
3rd Spec.
Efficiency -5.389 -0.062 0.492 -0.13 -6.438 -4.561 (P)
2007-2008
-6.334 -4.377 (BC)
0.577 1.323 (N)
3rd Spec. Efficiency
0.95 0.010 0.190 0.06 0.587 1.348 (P)
2007-2008 (Trend)
0.578 1.328 (BC)
Furthermore, Table 16 and Table 17 show the results of the estimations with reference
not only to the beta efficiency components, but also to the beta standard error of efficiency
components. In so doing, the bootstrapped results make it possible to detect any noise in the
estimated standard errors of the efficiency ratios (Cameron & Trivedi, 2010). The results show
that, because the bootstrapped standard errors of beta technical efficiency are close to the mean
of the bootstrapped standard errors, the noise in all beta technical efficiency standard errors is
sufficiently small. Moreover, this slight difference corroborates the estimates of the default
logistic regressions with robust standard errors.
Page 342
Table 16
BOOTSTRAP EFFICIENCY COMPONENTS Β EFFICIENCY COMPONENTS Β STD. ERR.
1st AND 3rd SPECIFICATION 2007-2008
Observed Bootstrap Std.
Model Variable Δ β-β Std. Error
Coefficient Error
Efficiency -4.0094*** (0.382)
1st Spec. 2007 0.02
Efficiency Std. Error 0.4054*** (0.012)
Efficiency -3.9123*** (0.405)
3rd Spec. 2007 0.03
Efficiency -3.5712*** (0.353)
1st Spec. 2008 0.06
Efficiency -3.7209*** (0.396)
3rd Spec. 2008 0.06
Table 17
BOOTSTRAP EFFICIENCY COMPONENTS Β EFFICIENCY COMPONENTS Β STD. ERR.
3rd SPECIFICATION 2007-2008 (MEAN & TREND)
Observed Bootstrap Stand.
Model Variable Δ β-β Std. Error
Coefficient Error
Efficiency (Mean) -7.5587*** (0.549)
3rd Spec. 2007-2008 0.10
Efficiency -5.3894*** (0.515)
0.04
3rd Spec. 2007-2008
Efficiency (Trend) 0.9502*** (0.196)
0.02
SUMMARY AND CONCLUSIONS
An increase in a firm’s technical efficiency in the use of inputs can result in either a
decrease in the overall costs or an increase in production. The level of technical efficiency is
therefore an important determinant of the firm’s financial sustainability. In particular, we argue
that the qualitative or soft information conveyed by the level of technical efficiency can improve
the capacity to predict the firm’s probability of default. Measuring the level of technical
efficiency is however tricky since it can be affected by several unobserved factors and the real
production function is unknown. The most advanced methodologies are data envelopment
analysis, which is a non parametric approach, and the stochastic frontier analysis, which is a
parametric approach. In this study, we have estimated technical efficiency with the latter
methodology as implemented in Battese & Coelli (1995). We have then used the estimates of the
technical efficiency along with the more traditional financial ratios to predict a firm’s default in
the frame of a logit model. Hence this study furnishes a comparison of the classification accuracy
and predictive power of two types of variables, efficiency and financial ratios. The sample
employed consisted of 8,145 Italian SMEs in the period 2007-2009. The results suggest that
efficiency is a good predictor of the probability of default only when the financial ratios are also
included in the model. This is consistent with the findings of Becchetti & Sierra (2003) and
Psikillaki, Tsolas & Margaritis (2010). In particular, the model that consists of the average
values over the period considered seems partially to outperform the other models in its ability to
classify a firm as unsound.
Page 343
The policy implications are several and can be drawn from the interpretation of the
results. We can assume that when a firm employs the factors of production efficiently, that is,
maximizing the potential output, it also means that the owned and, especially, borrowed funds
are allocated efficiently and used to properly generate value. The greater the production value the
greater the ability to repay the borrowed funds. In this study we inferred our technical efficiency
ratios from only two subsequent fiscal years but in terms of default prediction, the performance
of the efficiency in the medium-long run can be more important. We believe that the information
of whether a firm has been efficient over time is mostly embedded in the income statement and
balance sheet. This somehow explains why including the financial ratios dramatically improve
the prediction capacity. Besides, the financial ratios proxy for the firm’s ability to generate cash
flow and meet its obligations. We indeed found further that the financial ratios that fall within
the “Profitability and debt service” component, the “Capitalization and liquidity” component,
and the “Asset turnover” component appear to decrease the probability of default. On the other
hand, the ratios in the “Working capital” component and “Debt cost” component increase the
probability of default. We point out that good management of working capital is extremely
important given the small size of the Italian firms analysed. Moreover, commercial enterprises
and enterprises with a turnover of between 5 and € 15 Mln also show a lower probability of
default compared to firms with a turnover of less than € 5 Mln. Where the enterprise is located is
instead irrelevant. So, to sum up, current and potential financiers of Italian SMEs have to be
concerned by the firm’s technical efficiency level. The technical efficiency is though
unobserved. This also limits its inclusion in credit scoring models unless some proxies are
identified. However, traditional financial ratios seem to still provide the most comprehensive
information.
Finally, this study has contributed to the default prediction literature in several ways.
First, by allowing for the estimated efficiency ratios in the empirical model, it has further
confirmed that a forward-looking perspective is indeed an improvement on the only backward-
looking perspective. Second, our results show that the hard information provided by the financial
ratios is essential, and that it cannot be left out of any model of default prediction. Third, this
study adds to the empirical studies that analyse the Italian SMEs in a very particular period, i.e.
the Global Financial Crisis. In particular, it suggests some of the factors, including technical
efficiency, that may have caused several SMEs to default in the year after the beginning of the
financial crisis – that is, 2009. Further research should focus on longer and updated panel data on
Italian SMEs that enable investigation of whether the level of technical efficiency before the
onset of the financial crisis is a good predictor of the probability of default during the crisis that
is still negatively affecting the Italian economic system.
Page 344
ENDNOTES
1. The value of production is if the dependent variable is expressed in original units, ( ) if the
dependent variable is expressed in log.
2. Frontier is the command implemented in Stata sofware, which fits stochastic production or cost frontier
models both for cross-section (frontier) and panel data (xtfrontier). In addition, Stata provides other two
commands, sfcross and sfpanel (Belotti, Daidone, Ilardi & Atella, 2012), which enable estimation of
Battese & Coelli (1995) model.
3. The Wald test shows a equal to 6719,63 with a p-value= 0,000 for 2007 and a equal to 7672,47
with a p-value= 0,000 for 2008. Moreover, since the production function is expressed as a Cobb-Douglas
function, constant returns to scale imply that the sum of the coefficients on the cost of raw materials,
consumables & merchandise per worker together with fixed assets per worker is one.
4. The test conducted is called lincom. It computes point estimates, standard errors, or statistics, p-values,
and confidence intervals for linear combinations of coefficients.
5. In accordance with Mas-Colell, Whinston & Green (1995), a production technology exhibits non-
increasing returns to scale if for any , y for all scalars [0,1].
6. The robustness check considers only the technical efficiency components. The remaining explanatory
variables are available under request.
REFERENCES
Aigner, D., C.A.A. Lovell & P. Schmidt (1977). Formulation and estimation of stochastic frontier production
function models. Journal of Econometrics, 6(1), 21-37.
Almanidis, P. & R. Sickles (2012). Banking Crises, Early Warning Models, and Efficiency. Working Paper.
Allen, L. & A. Rai (1996). Operational efficiency in banking: An international comparison. Journal of Banking &
Finance, 20(4), 655-672.
Altman, E. (1968). Financial ratios, discriminant analysis and prediction of corporate bankruptcy. Journal of
Finance, 22, 589-610.
Altman, E., R.G. Haldeman & P. Narayanan (1977). Zeta analysis. Journal of Banking & Finance, 1, 29-54.
Altman, E., R.B. Avery, R.A. Eisenbeis & J.F. Jr. Sinkey (1981). Applications of Classification Procedures in
Banking and Finance. Greenwich, JAI Press.
Altman, E., G. Marco & F. Varetto (1994). Corporate distress diagnosis: Comparisons using linear discriminant
analysis and neural networks (the Italian experience). Journal of Banking & Finance, 18, 505-529.
Altman, E.I. & G. Sabato (2007). Modelling credit risk for SMEs: Evidence from the U.S. market. Abacus, 43, 332-
357.
Altman, E., G. Sabato & N. Wilson (2008). The Value of Qualitative Information in SME Risk Management.
Journal of Financial Services Research, 40(2), 15-55.
Aly, H.Y., R. Grabowski, R.C. Pasurka & N. Rangan (1990). Technical, scale, and allocative efficiencies in US
banking: an empirical investigation. The Review of Economics and Statistics, 211-218.
Amendola, A., M. Bisogno, M. Restaino & L. Sensini (2011). Forecasting corporate bankruptcy: empirical evidence
on Italian data. EuroMed Journal of Business, 6(3), 294-312.
Appetiti, S. (1984). Identifying unsound firms in Italy. An attempt to use trend variables. Journal of Banking &
Finance, 8, 269-279.
Argenti, J. (1986). Predicting Corporate Failure. Accountancy, 97, 157-158.
Banca d’Italia, (October, 2008). Bollettino Economico, N. 54.
Battese, G.E. & T.J. Coelli (1988). Prediction of firm-level technical efficiencies with a generalized frontier
production function and panel data. Journal of econometrics, 38(3), 387-399.
Page 345
Battese, G. E. & T.J. Coelli (1995). A model for technical inefficiency effects in a stochastic frontier production
function for panel data. Empirical economics, 20(2), 325-332.
Basel Committee on Banking Supervision, (2006). International Convergence of Capital Measurement and Capital
Standards. A revised Framework. Comprehensive Version Basel.
Beaver, W. (1966). Financial ratios as predictors of failure. Supplement to Journal of Accounting Research, 5, 71-
127.
Becchetti, L. & J. Sierra (2003). Bankruptcy risk and productive efficiency in manufacturing firms. Journal of
Banking & Finance, 27, 2099-2120.
Bellovary, J., D.E. Giacomino & M. Akers (2007). A review of Bankruptcy Prediction Studies: 1930 to Present.
Journal of Financial Education, 33, 56-95.
Belotti, F., S. Daidone, G. Ilardi & V. Atella (2012). Stochastic frontier analysis using Stata. CEIS Tor Vergata
Research Paper Series, 251.
Berger, A.N. & L.J. Mester (1997). Inside the black box: What explains differences in the efficiencies of financial
institutions?. Journal of Banking & Finance, 21(7), 895-947.
Bisogno, M. (2012). The Accessibility of the Italian Bankruptcy Procedures: An Empirical Analysis. Eurasian
Business Review, 2(2), 1-24.
Blochwitz, S. & J. Eigermann (2000). Unternehmensbeurteilung durch Diskriminanzanalyse mit qualiativen
Merkmalen. Zeitschrift für betriebswirtschaftliche Forschung, 52, 58-73.
Bureau of Business Research, (1930). A Test Analysis of Unsuccessful Industrial Companies. Bulletin N. 31.
University of Illinois Press. Urbana.
Brunner, A., J.P. Krahnen & M. Weber (2000). Information production in credit relationships: on the role of internal
ratings in commercial banking. Working Paper N. 2000/10,Center for Financial Studies, Frankfurt/Main.
Cameron, C.A. & P.K. Trivedi (2010). Microeconometrics Using Stata. College Station: Stata Press Publication.
Censis, (2008). Rapporto sulla situazione sociale del Paese 2008. Roma: Franco Angeli.
Charitou, A., E. Neophytou & C. Charalambous (2004). Predicting Corporate Failure: Empirical Evidence for the
UK. European Accounting Review, 13(3), 465-497.
Chen, Y., R.J. Huang & J. Tsai (2013). Soft Information and Small Business Lending. Journal of Financial Services
Research, 1-19.
Chowdhury, S. & J. Lang (1993). Crisis, decline and turnaround: a test of competing hypotheses for short-term
performance improvement in small firms. Journal of Small Business Management, 31(4), 8-17.
Chudson, W. (1945). The Pattern of Corporate Financial Structure. National Bureau of Economic Research. New
York.
Cornée, S. (2013). The Relevance of Soft Information for Predicting Small Business Credit Default: Evidence from
a Social Bank. Working Paper. University of Rennes.
Cosma, S. (2002). Il rapport banca-impresa: le variabili relazionali e comportamentali nella valutazione del rischio
di credito. Torino: Giappichelli Editore.
Crutzen, N. (2010). Essay on the prevention of small business failure: taxonomy and validation of five explanatory
business failure patterns (EBFPs). Académie Universitaire Wallonie. HEC Ecole de Gestion de l’Université
de Liège.
Daubie, M. & N. Meskens (2001). Bankruptcy Prediction: Literature Survey of the Last Ten Years. Belgian Journal
of Operations Research, Statistics and Computer Science, 41, 43-58.
Deakin, E. (1972). A discriminant analysis of predictors of business failure. Journal of Accounting Research, 10(1),
167-179.
Draghi, M. (July, 2010). I v d v a d a Ba a d’I a a, ABI, Italy.
Dwyear, M. (1992). A comparison of statistical techniques and artificial neural network models in corporate
bankruptcy prediction. Ph.D. Dissertation. University of Wisconsin-Madison.
Edmister, R.O. (1972). An Empirical Test of Financial Ratio Analysis for Small Business Failure Prediction.
Journal of Financial and Quantitative Analysis, 7, 1477-1493.
Efron, B. (1979). Bootstrap methods: another look at the jackknife. The annals of Statistics, 1-26.
El Hennawy, R. & R. Morris. (1983). The significance of base year in developing failure prediction models. Journal
of Business Finance and Accounting, 10(2), 209-223.
Page 346
Everett, J. & J. Watson (1998). Small business failure and external risk factors. Small Business Economics, 11, 371-
390.
Ezzamel, M., J. Brodie & C. Mar-Molinero (1987). Financial Patterns of Manufacturing Companies. Journal of
Business Finance & Accounting, 14(4), 519-536.
anti, L. (1997). Inefficienza tecnica e di scala: Un’applicazione all’industria italiana di una frontier flessibile di
produzione. L’ du a XVIII(2), 271-282.
Fernández-Castro, A. (1988). A Spanish model for credit risk classification. Studies in Banking and Finance, 7, 115-
125.
Ferrier, G.D. & C.K. Lovell (1990). Measuring cost efficiency in banking: econometric and linear programming
evidence. Journal of Econometrics, 46(1), 229-245.
FitzPatrick, P. (1932). A Comparison of ratios of successful industrial enterprise with those of failed companies. The
Certified Public Accountant: 598-605, 656-662, 727-731.
Frydman, H., E. Altman & D. Kao (1985). Introducing recursive partitioning for financial classification: The case of
financial distress. Journal of Finance, 40(1), 269-291.
Gentry, J.A., P. Newbold & D.T. Whitford (1985). Classifying bankruptcy firms with funds flow components.
Journal of Accounting Research, 23(1), 146-160.
Gentry, J.A., P. Newbold & D.T. Whitford (1987). Funds flow components, financial ratios and bankruptcy. Journal
of Business Finance and Accounting, 14(4), 595-606.
Gibilaro, L. & D. Piatti (2012). Il ruolo delle variabili hard finanziarie e non finanziarie nella Business Failure
Prediction. Sinergie, Rapporti di Ricerca, 36, 135-184.
Gilbert, L.R., K. Menon & K.B. Schwartz (1990). Predicting bankruptcy for firms in financial distress. Journal of
Business Finance, 17, 161-171.
Gilson, A.C. & M.R. Vetsuypens (1993). CEO compensations in financially distressed firm: An empirical analysis.
Journal of Finance, 48, 425-458.
Godbillon-Camus, B. & C. Godlewski (2005, December). Credit risk management in banks: Hard information, soft
Information and manipulation. Soft Information and Manipulation.
Godfrey, M.D. (2009). The TANH transformation. Standford University, Information Systems Laboratory.
Gombola, M., M. Haskins, J. Ketz & D. Williams (1987). Cash flow in bankruptcy predictions. Financial
Management, 16(4), 55-65.
Grunert, J., L. Norden & M. Weber (2005). The role of non-financial factors in internal credit ratings. Journal of
Banking and Finance, 29, 509-531.
Guilhot, B. (2000). Défaillance d’entreprise: Soixante-dix ans d’analyses théoriques et empiriques. Revue Française
de Gestion, 52-67.
Günther, T. & M. Grüning (2000). Einsatz von Insolvenzprognoseverfahren bei der Kreditwürdigkeitsprü im
Firmenkundenbereich. Die Betriebswirtschaft, 60, 39-59.
Haslam, S.A., J.C. Turner, P.J. Oakes, C. McGarty & B.K. Hayes (1992). Context‐ dependent variation in social
stereotyping 1: The effects of intergroup relations as mediated by social change and frame of reference.
European Journal of Social Psychology, 22(1), 3-20.
Hesselmann, S. (1995). Insolvenzprognose mit Hilfe qualitative Faktoren. Aachen.
Hickman, W.C. (1957). Corporate Bond Quality and Investment Performance. National Bureau of Economic: New
York.
Hunt-McCool, J., S.C. Koh & B.B. Francis (1996). Testing for deliberate underpricing in the IPO premarket: A
stochastic frontier approach. Review of Financial Studies, 9(4), 1251-1269.
Hwang, R. C., J.S. Siao, H. Chung & C.K. Chu (2011). Assessing bankruptcy prediction models via information
content of technical inefficiency. Journal of Productivity Analysis, 36(3), 263-273.
Iazzolino, G., M. Bruni & P. Beraldi (2013). Using DEA and financial ratings for credit risk evaluation: an empirical
analysis. Applied Economics Letters, 20(14), 1310-1317.
Izan, H. (1984). Corporate distress in Australia. Journal of Banking & Finance, 8, 303-320.
Jackendoff, N. (1962). A Study of Published Industry Financial and Operating Ratios. Bureau of Economic and
Business Research. Philadelphia: Temple University.
Page 347
Johnsen, T. & R.W. Melicher (1994). Predicting corporate bankruptcy and financial distress: information value
added by multinomial logit models. Journal of Economics and Business, 46, 269-286.
Jondrow, J., C.A. Knox Lovell, I.S. Materov & P. Schmidt (1982). On the estimation of technical inefficiency in the
stochastic frontier production function model. Journal of Econometrics, 19(2), 233-238.
Kahya, E. & P. Theodossiou (1999). Predicting corporate financial distress: A time series CUSUM methodology.
Review of Quantitative Finance & Accounting, 13(4), 323-345.
Kaparakis, E.I., S.M. Miller & A.G. Noulas (1994). Short-run cost inefficiency of commercial banks: A flexible
stochastic frontier approach. Journal of Money, Credit & Banking, 875-893.
Keasey, K. & R. Watson (1987). Non-financial symptoms and the prediction of small company failure: A test or
Argenti’s hypotheses. Journal of Business Finance & Accounting, 14, 335-354.
Koenig, G. (1985). Entreprises en difficultés: des symptoms aux remèdes. Revue Française de Gestion, 84-92.
Kumar, P.R. & V. Ravi (2007). Bankruptcy prediction in banks and firms via statistical and intelligent techniques –
A Review. European Journal of Operational Research, 1-28.
Kwan, S. H. & R.A. Eisenbeis (1995). An analysis of inefficiency in banking: A stochastic cost frontier approach.
Working paper, Working Papers in Applied Economics, Federal Reserve Bank of San Francisco.
Laitinen, E.K. (1991). Financial ratios and different failure processes. Journal of Business, Finance & Accounting,
649-673.
Lattin, J.M., J.D. Carroll & P.E. Green (2003). Analyzing multivariate data. Pacific Grove, CA: Thomson
Brooks/Cole.
Lehamann, B. (2003). Is It Worth the While? The Relevance of Qualitative Information in Credit Rating. Working
Paper, from http://ssrn.com/abstract=410186
Levitan, A. & J. Knoblett (1985). Indicators of exceptions to the going concern assumption. Auditing: A Journal of
Li, S. & S. Wang (2014). A financial early warning logit model and its efficiency verification approach. Knowledge-
Based Systems, from http:77dx.doi.org710.10167j,knosys.2014.03.017.
Lin, F.D. & E. Chen (2011). Financial ratio selection for business crisis prediction. Expert Systems with Application,
38(12), 15094-15102.
Lo, A.W. (1986). Logit versus discriminant analysis. A specification test and applications to corporate bankruptcies.
Journal of Econometrics, 31, 151-178.
Martin, D. (1977). Early Warning of bank failure: A logit regression approach. Journal of Banking & Finance, 1,
249-276.
Mas-Colell, A., M.D. Whinston & J.R. Green (1995). Microeconomic Theory. Oxford: University Press.
McKee, T.E. & T. Lensberg (2002). Genetic Programming and Rough Sets: A Hybrid Approach to Bankruptcy
Classification. European Journal of Operational Research, 436-451.
Merwin, C. (1942). Financing small corporations in five manufacturing industries 1926-1936. New York: National
Bureau of Economic Research.
Ohlson, J. (1980). Financial ratios and the probabilistic prediction of bankruptcy. Journal of Accounting Research,
18(1), 109-131.
Pederzoli, C. & C. Torricelli (2010). A parsimonius default predictions model for Italian SMEs. Banks and Banks
Systems, 5(4), 5-9.
Petersen, M.A. (2004). Information: Hard and Soft, Mimeo.
Piatti, D. (2012). Il ruolo dei flussi di cassa nella valutazione del merito di credito, Bergamo: University Press,
Sestante Ed..
Piatti, D. (2014a). Corporate Social Performance and Social Disclosure: Evidence from Italian Mutual Banks.
Accademy of Accounting and Financial Studies Journal, 18(1).
Piatti, D. (2014b). Operating Cash Flow and Creditworthiness Assessment. Accademy of Accounting and Financial
Studies Journal, 18(2).
Psillaki, M., Tsolas I.E. & Margaritis D. (2010). Evaluation of credit risk based on firm performance. European
Journal of Operational Research, 201, 873-881.
Pusnik, K. & M. Tajnikar (2008). Technical and Cost Efficiency as Determinants of Business Failures of Small
Firms. Eastern European Economics, 46(1), 43-62.
Page 348
Putnam, R.D. (1993). Making Democracy Work. Princeton: University Press.

Raja, A., M. Nosworthy & D. Goureia (1980). Diagnosis of Financial Health by Cash Flow Analysis. Working
Papers, London Business School.
Resti, A. (1997). “ valuating the Cost-Efficiency of the Italian Banking System: What Can Be Learned from the
Joint Application of Parametric and Non-Parametric Techniques. Journal of Banking & Finance, 21(2), 221-
250.
Retzlaff-Roberts, D. & R. Puelz (1996). Classification in automobile insurance using a DEA and discriminant
analysis hybrid. Journal of Productivity Analysis, 7(4), 417-427.
Saeed, M. & M. Izzeldin (2014). Examining the relationship between default risk and efficiency in Islamic and
conventional banks. Journal of Economic Behavior & Organization.
Sarkar, S. & R. Sriram (2001). Bayesian Models for Early Warning of Bank Failure. Management Science, 1457-
1475.
Serrano-Cinca, C. & B. Gutierrez-Nieto (2013). Partial Least Square Discriminant Analysis for bankruptcy
prediction. Decision Support Systems, 54, 1245-1255.
Sharma, S. & V. Mahajan (1980). Early warning indicators of business failure. Journal of Marketing, 44(4), 80-89.
Shumway, T. (2001). Forecasting Bankruptcy More Accurately: A Simple Hazard Model. The Journal of Business,
74(1), 101-124.
Sinkey, J.jr. (1975). A multivariate statistical analysis of the characteristics of problem banks. Journal of Finance,
30(1), 21-36.
Skogsvik, K. (1990). Current cost accounting ratios as predictors of business failure: The Swedish case. Journal of
Business Finance & Accounting, 17, 137-160.
Slowinski, R. & C. Zopounidis (1995). Application of the Rough Set Approach to Evaluation of Bankruptcy Risk.
Intelligent Systems in Accounting, Finance and Management, 4, 27-41.
Smith, R. & A. Winakor (1935). Changes in Financial Structure of Unsuccessful Industrial Corporations. Bureau of
Business Research, 51. Urbana: University of Illinois Press.
Stein, J.C. (2002). Information Production and Capital Allocation: Decentralized versus Hierarchical Firms. The
Journal of Finance, 5, 1981-1921.
Styrin, K. (2005). What explains differences in efficiency across Russian banks? Economics Education and
Research Consortium.
Swaminathan, A. (1996). Environmental conditions at founding and organizational mortality: a trial by-fire model.
The Academy of Management Journal, 39(5), 1350-1377.
Taffler, R.J. & H. Tisshaw (1977). Going, Going Going – Four Factors Which Predict. Accountancy, 50.
Tamari, M.S. (1966). Financial Ratios as a Means of Forecasting Bankruptcy. Management International Review, 4,
15-21.
Tinoco, M.H. & N. Wilson (2013). Financial distress and bankruptcy prediction among listed companies using
accounting, market and macroeconomic variables. International Review of Financial, 30, 394-419.
Trestini, S. (2006). Un modello di frontiera stocastica per la valutazione dell’efficienza tecnica. Una applicazione
agli allevamenti bovini da carne del Veneto. Rivista di Economia Agraria, 3, 315-338.
Unal, T. (1988). An early warning model for predicting firm failure and bankruptcy. Studies in Banking and
Finance, 7, 141.
Varian, H.L. (1990). Microeconomic analysis (Second edition), New York: W.W. Norton.
Weber, M., J.P. Krahnen & F. Vossmann (1999). Risikomessung im Kreditgeschäft: Eine empirische Analyse
bankinterner Ratingverfahren. Zeitschrift für betriebswirtschaftliche Forschung, Sonderheft 41, 117-142.
Wilcox, J. (1973). A prediction of business failure using accounting data. Journal of Accounting Research:
Supplement on Empirical Research in Accounting, 163-190.
Williamson, O.E. (1967). Hierarchical control and optimum firm size. Journal of Political Economy, 75, 123-138.
Wooldridge, J.M. (2009). Introductory Econometrics. A modern Approach. Mason: International Student Edition.
Zavgren, C.V. (1985). Assessing the Vuknerability to failure of American Industrial Firms: A logistic Analysis.
Journal of Business and Finance Accounting, 12(1), 19-45.
Zwick, W., Velicer, R. & F. Wayne (1986). Comparison of five rules for determining the number of components to
retain. Psychological Bulletin, 99(3), 432-442.

Academyof Acctg Financial Ratio Paper

Uploaded by

Copyright:

Available Formats

Academyof Acctg Financial Ratio Paper

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Academyof Acctg Financial Ratio Paper

Uploaded by

Copyright:

Available Formats

Volume 19, Number 3 Print ISSN: 1096-3685

Online ISSN: 1528-2635

Authors execute a publication permission agreement and assume all liabilities.

The Academy of Accounting and Financial Studies Journal is owned and

Copyright 2015 by Jordan Whitney Enterprises, Inc., Weaverville, NC, USA

EDITORIAL REVIEW BOARD

Robert Marley Liang Song

Robert Graber Steve Moss

Michael Grayson Chris Harris

Sudip Ghosh Atul K. Saxena

Rufo R. Mendoza Anthony Yanxiang Gu

Alkali Yusuf Hema Rao

Junaid M. Shaikh Marek Gruszczynski

Linda Bressler Natalie Tatiana Churyk

Suzanne Pinac Ward Dan Ward

Eugene Calvasina Ron Stunda

P.N. Saksena Martha Sale

Askar Choudhury Rafik Z. Elias

Harold Little Dawn Mead Hukai

James A. DiGabriele Desti Kannaiah

Philip H. Siegel Treba Marsh

Frank Plewa Mukunthan Santhanakrishnan

THE PRIVATE SECURITIES LITIGATION REFORM ACT AND LITIGATION BACKED

POTENTIAL FINANCIAL EFFECTS OF THE 2015 CORPORATE TAX REFORM OF

DISENTANGLING THE IFRS 1 DISCLOSURE: OBSERVATIONS FROM THE EARLY

LIQUIDITY IN LEBANESE COMMERCIAL BANKS AND ITS DETERMINANTS……….57

UNDERSTANDING THE LIMITATIONS OF FINANCIAL RATIOS………………….…….75

FINANCIAL REPORTING QUALITY OF FAMILY-CONTROLLED SMALL AND

BIG 4 AUDITOR OFFICE SIZE, ANALYSTS’ ANNUAL EARNINGS FORECASTS AND

AUDIT FEES, EARNINGS MANAGEMENT, AND LITIGATION RISK: EVIDENCE FROM

ABNORMAL AUDIT FEES AND AUDITOR SIZE IN THE JAPANESE AUDIT

TIER 1 CAPITAL RATIOS: THE REGULATION OF RISK BEHAVIOR

ASSOCIATION BETWEEN PRE IPO EARNINGS MANAGEMENTS AND POST IPO

THE ASSOCIATION BETWEEN FINANCIAL LITERACY AND TRUST IN FINANCIAL

THE EFFECTS OF COMPENSATION SCHEME, SOURCE CREDIBILITY, AND

OWNERSHIP STRUCTURE AND INFORMATION DISCLOSURE: AN APPROACH AT

DAILY PRICE CHANGES IN THE REAL ESTATE MARKET……………………………..253

THE EFFECTS OF VOLATILITY AND LEVERAGE ON THE EARNINGS-GDP

CAPITAL BUDGETING PRINCIPLES: BRIDGING THEORY AND PRACTICE…………283

AN EXPERIMENT TO ASSESS THE EFFECT ON INVESTING OF SWITCHING

EFFICIENCY AND CORPORATE BANKRUPTCY PREDICTION………………………...319

PREDICTORS OF STOCK RETURNS: SOME EVIDENCE

Keywords: Stock returns, dividends, emerging markets, Ghana Stock Exchange.

VARIABLES AND THEIR DETERMINATION

Dividend Price Ratio (DPRICER) – Dividend Yield

Dividends per Share (DPS)

Price Earnings Ratio (PER)

Book to Market Ratio (BMR)

Size is the market capitalization of the company and is defined as:

Stock Returns (STOCKR)

HYPOTHESES AND METHODOLOGY

a. Dependent variable: STOCKR

THE PRIVATE SECURITIES LITIGATION REFORM

HYPOTHESIS ON THE IMPACT OF PSLRA ON FIRM VALUE

DATA AND RESEARCH METHODOLOGY

Scaled Market-to-Book Low 895 621.05 285.16 2.611 0.563

Four Factor - HML Low 0.0096*** 0.0000 -0.0177*** 0.0000

Market Model Low 0.0053*** 0.0059 -0.0199*** 0.0000

Unadjusted Low 0.0068*** 0.0004 -0.0209*** 0.0000

Four Factor - HML Low 0.0096* 0.0000 -0.0177* 0.0000

Market Model Low 0.0053* 0.0059 -0.0199* 0.0000

Unadjusted Low 0.0068* 0.0004 -0.0209* 0.0000

Four Factor - HML Low 0.0186* 0.0000 -0.0259* 0.0000

Market Model Low 0.0118* 0.0000 -0.0245* 0.0000

Unadjusted Low 0.0124* 0.0000 -0.0243* 0.0000