Stubben 2006
Stubben 2006
Stubben 2006
Revenue Targets?
Stephen R. Stubben*
February 2006
Abstract:
This paper addresses two questions related to the use of discretion over revenues. First, do firms
use discretion in revenues to manage earnings to meet earnings targets? Second, do firms
manage revenues to meet revenue targets? To answer these questions, I model a common form
of discretionary revenues and its effect on the relation between revenues and accounts receivable.
Using this model, I find that firms with earnings just above analysts’ consensus forecasts report
positive discretionary revenues. Firms with greater incentives to use discretion in revenues as
opposed to expenses (i.e., growth firms and firms with high gross margins) do so to a greater
extent than other firms. I find limited evidence that growth firms use discretionary revenues to
meet revenue forecasts.
* I would like to thank my dissertation committee of Mary Barth, Bill Beaver, and Maureen McNichols
for invaluable comments and suggestions, and Chris Armstrong, Yonca Ertimur, Fabrizio Ferri, Alan
Jagolinzer, Wayne Landsman, Dave Larcker, Nate Sharp, Mark Soliman, and workshop participants at
Stanford University and the 2005 Accounting Research Symposium at Brigham Young University. I
thank Huron Consulting Group for providing data on restatements and SEC enforcement actions.
1. Introduction
This paper addresses two questions related to the use of discretion in revenues. First, do
firms use discretion in revenues to manage earnings to meet earnings targets? Second, do firms
Firms have incentives to meet earnings targets, and evidence suggests that they manage
earnings to do so (Burgstahler and Dichev, 1997). Firms can manage earnings using revenues,
expenses, or both. However, earnings management using revenues is likely to be more costly
than other forms of earnings management. Earnings management using revenues is more likely
to be detected and has a greater cost given detection (Marquardt and Wiedman, 2004), which
suggests that firms might prefer to manage earnings using expenses. However, certain firms,
such as growth firms and firms with high gross margins, may manage earnings using revenues
because the potential benefits are greater. Growth firms reap greater benefits from managing
earnings using revenues because investors value revenues of growth firms significantly more
highly than expenses (Ertimur, Livnat, and Martikainen, 2003). Firms with high gross margins
reap greater benefits than other firms because each dollar of discretionary revenue has a greater
impact on earnings.
Studying revenues rather than net earnings has two advantages. First, studying earnings
components can provide insights into how earnings are managed. Revenues is an ideal
component to examine; it is the largest earnings component for most firms, and it is subject to
relative to other forms of earnings management. For example, Dechow and Schrand (2004, page
42) documents that over 70% of SEC Accounting and Auditing Enforcement Releases involve
1
Throughout the paper I use “revenue management” to describe the use of discretion in revenues to meet revenue
targets. “Revenue manipulation” or “discretionary revenues” alone could indicate either revenue management or
earnings management using revenues.
1
misstated revenues, and revenues are the most common type of financial statement restatement
earnings has the potential to provide more precise estimates of discretion. The aggregate accrual
models that are commonly used to study discretion in earnings have been criticized for their low
power and inaccurate estimates of discretion (e.g., Dechow, Sloan, and Sweeney, 1995; Guay,
Kothari, and Watts, 1996; McNichols, 2000; and Thomas and Zhang, 2000). Although prior
studies have examined whether firms use discretionary accruals to meet earnings benchmarks
(e.g., Burgstahler and Eames, 2002; Dechow, Richardson, and Tuna, 2003), it is possible that
biased and low-powered estimates of discretion from misspecified accrual models affect the
conclusions of these studies. For example, Dechow, Richardson, and Tuna (2003) concludes that
if firms use discretionary accruals to avoid losses, their model is not powerful enough to detect it.
incentives to meet earnings targets. However, for growth firms it may not be sufficient to meet
only earnings targets; revenue targets are also important. Because revenue increases are more
sustainable than cost reductions (Ghosh, Ju, and Jain, 2005), investors rely on revenues more
than expenses to evaluate growth firms’ future growth potential. For example, among growth
firms that just meet earnings forecasts, those that miss revenue forecasts have significantly
negative stock returns during the earnings announcement period (Ertimur, Livnat, and
Martikainen, 2003). Therefore, it is likely that growth firms have incentives to manage revenues
to meet revenue forecasts, in addition to meeting earnings forecasts. Consistent with this idea,
Plummer and Mest (2001) finds a discontinuity in the revenue forecast error distribution, with
more than expected small positive revenue forecast errors. However, I am not aware of any
2
study that has provided direct evidence of the use of discretionary revenues to meet revenue
forecasts.
I hypothesize that firms, particularly growth firms and firms with high gross margins, use
revenues to manage earnings to meet earnings forecasts. Specifically, I test whether firms that
report earnings equal to or slightly above the consensus forecast have positive discretionary
revenues. Among firms with small positive earnings forecast errors, I also test whether growth
firms have higher discretionary revenues than non-growth firms, and whether firms with high
gross margins have higher discretionary revenues than other firms. I also hypothesize that
growth firms manage revenues to meet revenue targets. I test whether growth firms that report
revenues equal to or slightly above the consensus forecast have positive discretionary revenues.
revenues and accounts receivable. In this paper, prematurely recognized revenues are sales
recognized before GAAP criteria are met and before any cash is collected. My revenue model is
similar to existing discretionary accrual models (Jones, 1991; Dechow, Sloan, and Sweeney,
1995), but with three key differences. First, I model the receivables accrual, rather than
aggregate accruals, as a function of the change in revenues. As I argue and show, receivables
have a stronger and more direct relation with revenues than the other accrual components. Thus,
the inclusion of other accrual components leads to noisy and biased estimates of discretion.
Because I model receivables instead of aggregate accruals, the model is one of revenues rather
than earnings. Second, I model the receivables accrual as a function of the change in reported
revenues, rather than the change in cash revenues (Dechow, Sloan, and Sweeney, 1995).
3
likely to overstate estimates of discretion for growth firms. Third, I model the change in annual
receivables as a linear function of two components of the change in annual revenues: change in
revenues of the first three quarters, and change in fourth-quarter revenues. Because sales in the
early part of the year are more likely to be collected in cash by the end of the year, these have
different implications for receivables than does a change in fourth-quarter sales. Discretion in
revenues is captured by discretionary receivables, which is the difference between the actual
change in receivables and the predicted change in receivables based on the model.
To the extent that discretionary revenues are not offset by corresponding expenses,
accrual models should detect them. Therefore, for comparison I present results using the term-
adjusted modified Jones model of aggregate accruals (Teoh, Wong, and Rao, 1998). Prior
research finds that the Jones (1991) model and its variants are misspecified for firms with
extreme performance (e.g., Dechow, Sloan, and Sweeney, 1995; McNichols, 2000). Because I
study growth firms, it is possible that the revenue and accrual models produce biased estimates
of discretion. Using simulations of manipulation (Kothari, Leone, and Wasley, 2005), I find that
the revenue model produces estimates of discretion that are well specified for growth firms. The
benchmark accrual model does not. For this reason, I also use performance-matched
discretionary accrual estimates, which are purported to be well specified in the presence of
extreme performance (Kothari, Leone, and Wasley, 2005). I also find that the revenue model is
more likely to detect revenue manipulation than the accrual model. Using a sample of firms
subject to enforcement actions by the Securities and Exchange Commission and subsequent
restatements, I find that only the revenue model detects discretionary revenues for firms that
4
Regarding using revenues to meet earnings targets, I find the following. Discretionary
revenues are significantly positive for firms that just meet analysts’ consensus forecasts of
earnings, indicating firms use revenues to manage earnings to meet analysts’ forecasts.
Furthermore, discretionary revenues are significantly higher for growth firms and firms with
high gross margins than for other firms with small positive earnings forecast errors. Thus, firms
with greater benefits from managing earnings using revenues rather than expenses are willing to
bear the greater costs associated with this type of earnings management. The performance-
matched discretionary accrual estimates do not detect earnings management to meet earnings
forecasts.
Regarding using revenues to meet revenue targets, I find that discretionary revenues are
significantly positive for growth firms that just meet analysts’ consensus forecasts of revenue,
but not for other firms that just meet revenue forecasts. This finding suggests that growth firms
meet revenue targets by prematurely recognizing revenue. However, discretionary revenues for
growth firms with small positive revenue forecast errors are not significantly higher than those
for growth firms with small negative revenue forecast errors, which casts doubt on a revenue
estimate of discretion in revenues that can be used to detect revenue management. This revenue
estimate can also be used as a measure of earnings management that is more powerful and less
biased than estimates from accrual models. Even though it does not detect discretionary
expenses, it detects earnings management (via revenues) to meet earnings forecasts where the
5
The paper continues as follows. The motivation and hypotheses are discussed in section
2. The research design is presented in section 3. Section 4 details the data and descriptive
statistics, section 5 evaluates the revenue and accrual models, and section 6 discusses the primary
Discretion in revenues can be used to achieve two financial reporting goals. First, firms
can use revenues to manage earnings to meet earnings targets (i.e., earnings management).
Second, firms can manage revenues to meet revenue targets (i.e., revenue management).
Burgstahler and Dichev (1997) argues that firms have incentives to meet earnings
expected frequency of firms with zero or small earnings and increases in earnings in cross-
sectional distributions. Dechow, Richardson, and Tuna (2003) extends Burgstahler and Dichev
(1997) by testing whether firms with small profits use discretionary accruals to avoid reporting a
loss. They find similar magnitudes of discretionary accruals for small loss and small profit
firms, and they conclude that if firms overstate earnings to report profits, their accrual model is
Dechow, Richardson, and Tuna (2003) also examines the relative importance of
reporting profits, earnings increases, and positive earnings forecast errors. They show that
although the discontinuities in the annual earnings and earnings change distributions have
decreased over time, the discontinuity in the analyst forecast error distribution has increased.
These results suggest that meeting analysts’ consensus forecasts is becoming the more important
6
benchmark.2 Several studies have attempted to find evidence that firms manage earnings to meet
consensus forecasts, with mixed results (Burgstahler and Eames, 2002; Matsumoto, 2002;
Phillips, Pincus, and Rego, 2003; and Dhaliwal, Gleason, and Mills, 2004). For example,
Burgstahler and Eames (2002) finds that firms with small positive forecast errors have higher
abnormal accruals using the Jones (1991) model, whereas Phillips, Pincus, and Rego (2003),
Firms can manage earnings to meet benchmarks using revenues, expenses, or both.
Studying components of earnings can provide insights into how firms manage earnings.
However, most studies rely on measures of discretion in aggregate earnings. Three exceptions
are Plummer and Mest (2001), Marquardt and Wiedman (2004), and Roychowdhury (2004).4
Plummer and Mest (2001) studies the discretion over earnings components using distributional
tests similar to those of Burgstahler and Dichev (1997). They find evidence that suggests firms
manage earnings upward to meet earnings forecasts by overstating revenues and understating
they do not test whether discretionary revenues explain the discontinuity they find in the revenue
forecast error distribution. Marquardt and Wiedman (2004) estimates the unexpected portions of
firms manipulate. They find evidence that firms with small earnings increases understate special
2
Graham, Harvey, and Rajgopal (2005) finds that capital market incentives dominate CFOs’ reasons for managing
earnings; CFOs think meeting benchmarks leads to credibility in the market and higher stock prices. Although it is
possible that investors are able to completely “unravel” the discretionary portion of reported financial results at least
in some cases, the survey results suggest that managers perceive a benefit for using discretion to meet benchmarks,
and this perception leads them to do so.
3
Managers can also meet revenue forecasts by guiding analysts to lower their forecasts prior to the revenue
announcement. Because of the difficulty of measuring managerial guidance, I do not control for its impact on firms’
ability to meet forecasts. To the extent firms guide analysts, finding earnings and revenue management is more
difficult. However, Matsumoto (2002) finds growth firms manage earnings upward but not forecasts downward.
4
Other studies that examine discretion over particular earnings components to meet earnings benchmarks have been
conducted in the banking industry (Beatty, Ke, and Petroni, 2002) and the property-casualty insurance industry
(Beaver, McNichols, and Nelson, 2003). Phillips, Pincus, and Rego (2003) and Dhaliwal, Gleason, and Mills (2004)
examine the manipulation of income tax expense to meet earnings benchmarks.
7
items but do not overstate revenues. They also find evidence that firms use discretion in
Roychowdhury (2004) finds evidence that firms offer sales discounts to avoid reporting losses,
Of the earnings management choices available, revenues is one of the most costly.
Marquardt and Wiedman (2004) discusses the costs associated with earnings management and
conclude that the manipulation of recurring items, especially revenues, results in the most severe
costs through an increased probability of detection and more negative pricing consequences if
detected. Beneish (1999) finds a positive association between overstated revenues and the
probability a firm will be targeted by an SEC enforcement action. Furthermore, although the fact
that revenues are the most common type of financial statement restatement (Turner, Dietrich,
Anderson, and Bailey, 2001) could mean that revenue manipulation is commonly attempted, it
also could mean that revenue manipulation is more likely to be detected. In addition to
increasing the probability of detection, revenue manipulation increases the stock price
consequences if the manipulation is detected. Wu (2002) finds that restatements of revenues are
associated with significantly more negative stock returns than other types of restatements, and
Palmrose, Richardson, and Scholz (2004) finds that the likelihood of litigation after a restatement
It is possible that the greater costs associated with earnings management using revenues
would discourage firms from engaging in this form of reporting manipulation. However, growth
firms and firms with high gross margins potentially realize greater benefits than other firms from
overstating revenues, which makes these firms more likely to bear the higher costs associated
8
The competitive strategy literature predicts that firms pursuing growth through revenue
increases are different from those pursuing growth through expense reductions from cost cutting
or productivity gains (Porter, 1985). Revenue growth indicates growth in product demand rather
than merely cost control and is more sustainable. Thus, for growing firms, it is especially
important to report earnings growth and earnings surprises that are driven by revenues.
Consistent with this idea, Ertimur, Livnat, and Martikainen (2003) finds that investors value
revenue surprises more highly than expense surprises, especially for growth firms.5 Firms with
high gross margins also have incentives to manage earnings using revenues. For these firms,
H1: Discretionary revenues are positive for firms with small positive earnings forecast
errors
H1a: Among firms with small positive earnings forecast errors, growth firms have higher
discretionary revenues
H1b: Among firms with small positive earnings forecast errors, firms with high gross
Many studies address firms’ incentives to meet earnings targets. However, because
earnings components differ in persistence, they can be differentially informative about firm
performance (Lipe, 1986). Consequently, the source of the earnings surprise can be important.
Because revenue increases are more sustainable than expense decreases, revenue surprises are a
5
Evidence from the financial press corroborates this emphasis on revenues. One Wall Street Journal article notes,
“Only earnings generated by revenue improvements are getting investors excited.” (Zuckerman, 2000).
9
better indicator of future growth than expense surprises (Ghosh, Gu, and Jain, 2005). Ertimur,
Livnat, and Martikainen (2003) finds that among growth firms that just meet earnings forecasts,
those that miss revenue forecasts have significantly negative announcement-period abnormal
returns, whereas those that meet revenue forecasts have significantly positive abnormal returns.
For value firms, those that meet earnings forecasts have significantly positive abnormal returns
regardless of whether the revenue forecast is met. Thus, growth firms have incentives to meet
Because growth firms have incentives to meet revenue forecasts, it is likely that they
manage revenue to meet revenue forecasts. Magrath and Weld (2002) argues that the pressure to
meet revenue forecasts is particularly intense (as compared to pressure to meet earnings
forecasts) and may be the primary catalyst leading to “questionable, improper, or fraudulent
revenue-recognition practices.”
In response to this pressure on firms with respect to revenues, regulators have focused on
revenue accounting for several years. In 2005, the Financial Accounting Standards Advisory
Council’s annual survey listed revenue recognition as the number one concern for the Financial
Accounting Standards Board for the fourth consecutive year. Arthur Levitt, the former chairman
of the Securities and Exchange Commission, argues that revenue recognition is one of the
Consistent with firms facing pressure to meet revenue targets, several sources suggest
that revenue manipulation is relatively common. Nelson, Elliott, and Tarpley (2002) surveys
auditors and find that revenue manipulation is one of the most commonly attempted forms of
2000, approximately 66% of all accounting litigation cases allege some sort of revenue
10
recognition violations. Dechow and Schrand (2004, page 42) documents that over 70% of SEC
Accounting and Auditing Enforcement Releases involve overstated revenue, and Turner,
Dietrich, Anderson, and Bailey (2001) finds that revenue recognition restatements are the most
frequent. Despite this evidence, and the many studies on earnings management, there is little
Plummer and Mest (2001), which finds more firms than expected reporting small positive
revenue forecast errors.6 I test whether growth firms overstate revenues to meet this target.
H2: Discretionary revenues are positive for growth firms with small positive revenue
forecast errors
3. Research Design
3.1 Identification of Firms Suspected to Have Used Discretionary Revenues to Meet Forecasts
My hypotheses require measures for high growth, high gross margin, small positive
earnings and revenue forecast errors, and discretionary revenues. I define growth firms as firms
in the highest quartile of revenue growth each industry and year, measured as revenues in year t-
1 divided by revenues in year t-2. I measure gross margin as the difference between annual sales
and cost of sales, divided by sales (all in year t-1). High gross margin firms are those in the
highest quartile of the gross margin distribution each industry and year.
I define small positive revenue forecast errors as revenue realizations that are greater than
the consensus forecast by less than 1% of beginning-of-year market value of equity. Similarly,
small positive earnings forecast errors are earnings realizations that are greater than the
6
Plummer and Mest (2001) examines the revenue forecast error distribution to determine whether firms overstate
revenues to meet earnings forecasts. However, their finding more than expected firms with small positive revenue
forecast errors could also indicate revenue management to meet revenue forecasts.
11
consensus forecast by less than 0.3% of beginning-of-year market value of equity. The choice of
0.3% reflects approximately the median gross margin for sample firms (0.3). That is, on average
the scaled earnings forecast error distribution is less disperse than the scaled revenue forecast
error distribution, this choice also serves to produce similar proportions of small positive
earnings and revenue forecast errors.7 To provide additional confidence in the results, I compare
estimates of discretion by firms with small positive forecast errors to estimates of discretion by
firms with small negative forecast errors. I define small negative revenue (earnings) forecast
errors are revenue (earnings) realizations that miss the consensus forecast by less than 1% (0.3%)
revenues can take a number of forms. Some involve the manipulation of real activities (e.g.,
sales discounts, relaxed credit requirements, channel stuffing, and bill and hold sales), and others
do not (e.g., sales recognized before recognition criteria are met, fictitious revenues, and revenue
deferrals). In this paper, I model premature revenue recognition and its effect on the relation
between revenues and accounts receivable. Premature revenue recognition includes channel
stuffing and bill and hold sales, if customers do not pay cash for the inventory, and sales
recognized before recognition criteria are met. It also includes fictitious sales recorded on
account.
7
I find similar results when scaling by average total assets or defining small earnings forecast errors as those greater
than the consensus forecast by less than either 0.5% or 1% of market value of equity.
12
I focus on premature revenue recognition because evidence suggests it is a common form
of revenue management.8 For example, Levitt (1998) argues that premature revenue recognition
is one of five fundamental problems with financial reporting, and Feroz, Park, and Pastena
(1991) finds that more than half of SEC enforcement actions issued between 1982 and 1989
addition, other forms of revenue manipulation, such as sales discounts, could be profit-
maximizing business decisions and not merely attempts to meet a performance benchmark.
S it = S itUM + δ itRM
By assumption, there are no cash collections of discretionary sales during the current
period; these managed sales increase reported ending accounts receivable (AR) and reported sales
by the same amount. Thus, discretionary receivables equals the discretionary portion of sales (δ
RM 10
). Ending accounts receivable equals the portion of current nondiscretionary sales that were
not collected in cash (c × SUM) plus all discretionary sales. I assume that current accruals are
resolved within one year and receivables relating to sales in prior years are no longer
collectible.11
8
In a future version of the paper, I plan to include measures of additional types of discretionary revenues.
9
If revenues are continually managed, δ RM could be interpreted as net revenue management (i.e., revenue
management net of reversals from the prior period).
10
Because I estimate this model using net accounts receivable, the estimated discretion also includes any discretion
in the allowance for doubtful accounts.
11
I address the impact of this assumption in untabulated tests by adding additional lags of sales to the model.
Results are similar.
13
Because nondiscretionary sales are not observable, I express ending receivables in terms
of reported sales.
ARit = c × S it + (1 − c ) × δ itRM
I estimate the following equation, which I refer to as the annual equation because it is
based on the change in annual revenues. The estimate of a firm’s discretionary revenues is the
∆ARit = α + β × ∆S it + ε it ( Ann. AR )
The advantage of modeling receivables as opposed to accruals is that they are directly
related to sales. However, this may not be true for other current accruals (Kang and
Sivaramakrishnan, 1995). Accounts payable relates to purchases, and following the model of
Dechow, Kothari, and Watts (1998), inventory relates to forecasted sales for the next period, not
current actual sales. Forecasted sales equals current sales if sales follows a random walk, but this
is not true for growth firms. The relation between other accruals and change in sales is not clear.
Because sales alone does not explain payables, inventory, and other accruals, accrual models
based on sales alone produce noisy estimates of discretion. The estimates are also biased for
The coefficient β in Eq. (Ann AR) is an estimate of the portion of sales that are not
collected in cash by the end of the year, and the error represents scaled revenue management
12
My measure of premature revenue recognition is not independent of other forms of revenue management. For
example, relaxed credit requirements increases receivables relative to sales and will be detected by the model.
Channel stuffing where cash is received increases sales relative to receivables and will bias against finding
discretionary revenues with my revenue model.
14
(scaled by 1 – c).13 Because discretionary revenues are in reported revenues, the amount of
revenue management estimated by the annual revenue model will be understated (footnote 31 of
Jones, 1991).14 The modified Jones model (Dechow, Sloan, and Sweeney, 1995) conditions on
the change in cash sales rather than total sales, which avoids systematically understating the
amount of earnings management. However, this approach introduces another problem: credit
sales are treated as discretionary. Firms with a higher (lower) than average portion of
nondiscretionary sales that are credit will have discretionary accruals that are greater (less) than
zero. I condition on total sales because this understates estimated earnings management, which
biases against finding in favor of my alternative hypotheses. I report for comparison results
using the change in cash sales, which I refer to as the modified equation.
One limitation of accrual models is that they, by conditioning on annual sales, treat sales
made early in the year the same as sales made late in the year. Current accruals are generally
resolved within one year. Thus, sales made late in the year are more likely to be receivable at
year end. Therefore, I also estimate a version of the annual model allowing the estimated portion
of sales that are uncollected at year end to vary in the fourth quarter. I refer to this as the interim
In Eq. (Int. AR), S1_3 is sales in the first three quarters, and S4 is sales in the fourth
quarter. Even though Eq. (Int. AR) incorporates quarterly sales, I estimate discretion on an
13
The coefficient on sales is also affected by economic events such as a change in credit policy or factoring accounts
receivable (McNichols, 2000). Regarding factoring, Sopranzetti (1998) searches a database of over 4,000 publicly
traded firms from 1972-1993 and finds only 269 reports of factored accounts receivable by 98 firms. Nearly half of
the firms (47) were from the Textile and Apparel industries, and no other industry had more than 7 firms. To control
for the potential effects of factored receivables, I repeat my tests after excluding firms from the Textile and Apparel
industries and find similar results.
14
To mitigate the bias that arises because discretionary revenues are in the explanatory variable and the error, I
exclude firms suspected of manipulation (i.e., firms with small positive forecast errors) when estimating the models.
15
annual level. Any revenue management in early quarters that reverses by year end will not be
captured.
To the extent discretionary revenues are not offset by corresponding expenses, accrual
models should detect them. As a benchmark for the revenue models, I estimate discretionary
accruals using the term-adjusted modified Jones model (Teoh, Wong, and Rao, 1998), which is
similar to the modified Jones model (Dechow, Sloan, and Sweeney, 1995) but excludes
depreciation expense and property, plant, and equipment. The following equations for accruals
AC it = α + β × ∆S it + ε it ( Ann. AC )
AC it = α + β1 × ∆S 1_3it + β 2 × ∆S 4it + ε it ( Int. AC )
where AC represents current accruals. Estimates from the modified accrual model are calculated
using the estimated coefficients from the annual accrual model (Dechow, Sloan, and Sweeney,
1995). The Appendix summarizes the revenue and accrual models I use in this paper. Following
Kothari, Leone, and Wasley (2005), I estimate nondiscretionary accruals with scaled and
unscaled intercepts (by assets), to control for scale differences among firms (Barth and Kallapur,
1996).
management. McNichols (2000) finds that discretionary accrual estimates are biased for high
growth and highly profitable firms. Kothari, Leone, and Wasley (2005), following Teoh, Welch,
and Wong (1998) and Kasznik (1999), suggests performance-matched discretionary accrual
estimates to remedy this concern. That is, rather than relying on a firm’s raw discretionary
accrual estimate, they subtract the discretionary accrual estimate of a firm in the same industry
with smallest absolute difference in return on assets in the current year. Their results suggest
16
from earnings management research. I report results of performance-matched estimates for
and discretionary accruals of firms just above the forecast with those of firms just below the
forecast.
I perform the analysis using annual performance targets and discretion in annual
revenues.15 The sample includes firms on the Compustat annual file with available data between
1988 and 2003. My sample period begins in 1988 because prior to that date cash flow from
operations disclosed under Statement of Financial Accounting Standards No. 95 (FASB, 1987) is
unavailable. I exclude firms in regulated industries (financial, insurance, and utilities) because
their incentives to manage earnings and revenues likely differ from those of other firms.
I measure the change in receivables directly from the cash flow statement, and I calculate
accruals as earnings before extraordinary items plus depreciation and amortization less cash flow
from operations. I collect annual (quarterly) sales from the Compustat annual (quarterly) file.
Sales of the first three quarters is the difference between annual sales and fourth-quarter sales.
All sales and accrual variables are deflated by average total assets. Earnings growth is the
change in income before extraordinary items, deflated by average total assets. Industries are as
defined in Barth, Beaver, Hand, and Landsman (2005). I obtain analysts’ consensus earnings
and revenue forecasts and their realizations from I/B/E/S unadjusted summary file. Earnings and
15
In a future version of the paper, I plan to conduct similar tests based on quarterly benchmarks.
17
revenue forecasts are the last consensus (median) forecast prior to the earnings announcement.16
I winsorize at 2% gross margin, earnings growth, earnings and revenue forecast errors, and each
Table 1 presents distributional statistics. Panel A indicates that mean (median) accruals
are –1% (0%) of average assets. Many prior studies document slightly lower mean accruals.
However, unlike those studies, I do not include depreciation the accrual measure. The mean and
median change in receivables is 1% of average assets. Panel A also indicates that the mean
(median) change in sales is 10% (8%) of average assets. On average, the sales change is
approximately evenly distributed across quarters. The median change in sales of the first three
quarters is 5% of average assets (approximately 2% per quarter), and the median change in
Panel B of Table 1 presents correlations. Because the Pearson and Spearman correlations
are similar, I focus on the Pearson correlations. All correlations are significantly different from
zero, except the Spearman correlation between change in annual sales and accruals other than
However, change in receivables is more highly correlated with change in sales than are total
accruals. The correlation between annual sales change and change in receivables is 0.47
compared to the 0.26 correlation between annual sales change and accruals. Additionally,
change in receivables is more highly correlated with change in fourth-quarter sales than with the
16
I/B/E/S began tracking revenue forecasts in 1996, and the proportion of firms with revenue forecasts has increased
each year since then. By 2003, 94% of I/B/E/S firms had a revenue forecast (Ertimur and Stubben, 2005).
17
I use the term significance to denote statistical significance at less than the 0.05 level, based on a one-sided test
when I have signed predictions and a two-sided test otherwise.
18
change in sales of the first three quarters (0.51 versus 0.38). Also, the change in annual
receivables is more highly correlated with the change in fourth-quarter sales than it is with the
change in annual sales (0.51 versus 0.47). Taken together, these correlations suggest estimates
from models of receivables are less noisy than estimates from accrual models, and that using
quarterly data to disaggregate annual change in sales might lead to better specified discretionary
accrual models. However, I base my inferences on multivariate tests presented in the next
section.
Table 2 presents results from the estimation of the annual and interim equations. Panel A
presents results of pooled estimates of the annual equations with year and industry fixed effects.
The accrual model and the revenue model produce similar coefficients (0.10 and 0.09 in the
pooled estimation), but the t-statistic and adjusted r-squared are higher in the revenue model
(120.42 and 0.25 versus 56.83 and 0.11), consistent with the higher correlation between change
in receivables and change in sales shown in panel B of Table 1. Untabulated results reveal that
the coefficient in the revenue model is positive (significantly positive) in 285 (274) out of 285
Panel A also presents results of a variation of the annual equation with accruals other
than receivables as the dependent variable. The coefficient on change in sales is zero. This
finding indicates that the change in receivables drives much of the correlation between accruals
and change in sales. As expected, the relation between other accruals and sales change is weaker
19
than that of the receivables accrual and sales change, which leads to more noisy estimates of
Panel B presents results from estimations of the interim equations. In the revenue model,
the coefficient on change in fourth-quarter sales (0.26) is significantly higher—over six times
higher—than that of the change in sales of the first three quarters (0.04), although both are
significantly higher than that of the first three quarters (0.19 versus 0.07). Also, when allowing
for a separate coefficient on fourth-quarter sales, the adjusted r-squared of the revenue model
increases from 0.25 to 0.30, and the adjusted r-squared of the accrual model remains at 0.11.
Panel B also presents an estimation of the interim equation with accruals other than
receivables as the dependent variable. The coefficient on change in sales of the first three
significantly negative. Untabulated results reveal that this negative coefficient is largely
attributable to the payables accrual, which is positively correlated with the change in sales, but
subtracted in the calculation of accruals. Similar to panel A, the explanatory power of the model
Before testing the hypotheses, I use two approaches to evaluate estimates of discretion
from the various models. In the first approach, I simulate manipulation of revenues and
expenses and then assess the ability of the models to detect it. In the second approach, I rely on
actual earnings and revenue manipulation in a sample of firms that are known to have misstated
18
Finding little or no relation between aggregate accruals other than receivables and change in sales does not imply
that there is no relation between change in sales and individual accrual components. However, it does support
modeling specific accruals, such as receivables, rather than aggregate accruals.
20
their financial results. This approach assesses the ability of the models to detect revenue and
expense manipulation in a sample of firms that were investigated by the Securities and Exchange
I evaluate the specification and power of the revenue and accrual models using simulated
revenue and expense manipulation. Such simulations have been used by Dechow, Sloan, and
Sweeney (1995) and Kothari, Leone, and Wasley (KLW, 2005), among others, to test the power
of manipulation, I am able to obtain evidence of the bias, specification, and power of competing
models. I measure the bias of each model as the difference between the mean estimate of
discretion and the amount of manipulation I induce. If the model is unbiased, then the difference
will equal zero. I evaluate the specification of the models by computing how often tests reject
the null hypothesis of no manipulation for samples in which I induce no manipulation. Finally, I
evaluate the power of the models by computing how often tests detect manipulation when I
induce it.
discretion—i.e., subsamples with high growth (McNichols, 2000). I follow the approach
employed by KLW, with three exceptions I describe below. The procedure is as follows. In
each industry and year, I sort observations into quartiles of earnings growth and then repeat the
21
(2) Simulate revenue manipulation by adding 2% (of average total assets) to the change
in sales, the change in fourth-quarter sales, and the receivables accrual, and 2% times
the gross margin to current accruals of these 100 firm-years; or simulate expense
(3) Estimate the models using observations from all earnings growth quartiles, excluding
(4) Use each model’s coefficient estimates to calculate estimates of discretion for the 100
sample firm-years.
(5) Calculate the mean estimate of discretion from each model, and test whether the mean
The statistics from the 250 samples form the basis of the tests. I report the mean and
standard error of the 250 estimates of discretion, as well as the percent of the 250 times that the
model rejects the null hypothesis of no manipulation. A rejection rate of 5% is expected when
manipulation is not introduced, and the 95% confidence interval for the rejection rate of 5%
ranges from 2% to 8% (KLW). If the actual rejection rate is below 2% or above 8%, the test is
misspecified. When manipulation is introduced, however, the rejection rate should be 100%.
My procedure differs from that of KLW in three ways. First, I simulate combinations of
revenue and expense manipulation to evaluate the models under different forms of earnings
management. Second, I calculate accruals using items from the statement of cash flows. Hribar
and Collins (2002) finds that the error in the balance sheet approach of estimating accruals is
correlated with firms’ economic characteristics. As KLW note, this error not only reduces the
models’ power to detect earnings management, but also has the potential to generate incorrect
22
inferences about earnings management. Finally, I winsorize model variables before, rather than
after, estimating the models. This ensures that each models’ mean estimate of discretion is zero.
Table 3, panel A, presents descriptive statistics from the simulation. The table presents
estimates of discretionary accruals and discretionary revenues from the annual, modified, and
manipulation of 2% of assets, expense manipulation of 2%, and both revenue and expense
manipulation of 2%.
Table 3, panel A, reveals that each of the six equations produces a positive estimate of
discretion for growth firms with zero induced manipulation, which indicates a positive bias for
growth firms. However, the bias is smaller for the revenue models than for the accrual models.
The annual, modified, and interim accrual model estimates are 1.70, 1.92, and 1.63 percent of
assets; revenue model estimates are 0.41, 0.66, and 0.25 percent of assets. The larger estimates
for the accrual models are consistent with accruals other than receivables not being explained by
the change in sales alone, and the factors omitted from the models being correlated with growth.
For example, it is likely that growth firms invest in inventory beyond what would be predicted
The results in Table 3, panel A, indicate that the modified equation produces the most
biased estimate of discretionary revenues for both the accrual and the revenue models. For the
revenue models, the bias from the modified equation (0.66) is larger than that of the annual
(0.41) or interim (0.25) equation. This finding is consistent with growth firms having large
23
The results in Table 3, panel A, indicate that the interim equations produce the least
biased estimate of discretionary revenues for both the accrual and the revenue models. For the
revenue models, the bias from the interim equation (0.25) is less than that of the annual equation
(0.41). This finding is consistent with growth firms having a greater portion of annual revenues
Table 3, panel A, also presents standard errors across models. A model that produces
estimates with lower standard errors is more likely to detect revenue manipulation when it
occurs. The standard errors from the revenue models are less than half those of the accrual
models for each of the annual, modified, and interim equations. The standard errors of accrual
models are 1.06, 1.08, and 1.07 percent of assets, and those of the revenue models are 0.49, 0.54,
and 0.47 percent of assets. Also, for both the accrual models and the revenue models, the
modified equation produces estimates with the largest standard error, which confirms the lower
explanatory power of the change in cash from sales that is used in these equations.
Table 3, panel A, presents evidence on the bias of the competing models when I induce
revenue and expense manipulation. When revenue manipulation is induced, the bias of the
annual and interim equations decreases whereas that of the modified equations remains the same.
When revenue manipulation of 2% of assets is induced, the bias of the annual revenue equation
decreases from 0.41 to 0.23 percent of assets; revenue manipulation is estimated at 2.23% when
only 2% is induced. This decrease in the bias is a result of the annual equation treating a portion
of the manipulated revenue as nondiscretionary. The modified equation, however, is biased for a
different reason. It treats non-manipulated credit sales as discretionary, leading to a bias that is
larger than that of the annual equation (0.66 versus 0.41 percent of assets without revenue
manipulation and 0.66 versus 0.23 percent of assets with revenue manipulation of 2%).
24
By construction, all the expense manipulation is incorporated in the discretionary accrual
estimates, and none is incorporated in the discretionary revenue estimates. Thus, the success of
the revenue model in detecting earnings management depends on how much of the discretion
involves revenues.
Table 3, panel B, reports results on the specification and power of the models under the
null hypothesis of no discretion.19 Evidence on the specification of the models for growth firms
is presented in the first column of panel B. Each of the three accrual models over-rejects the null
hypothesis of no manipulation. Rejection rates for the annual, modified, and interim equations
are 40.0%, 44.8%, and 38.0%. In general, the revenue models are better specified than the
accrual models. Rejection rates for the annual, modified, and interim equations are 11.2%,
20.8%, and 8.0%. These findings indicate that only the interim revenue model produces well-
specified tests of revenue manipulation. All other models significantly over-reject the null
hypothesis of no manipulation.
With revenue manipulation of 2% of assets, the rejection rates for the revenue models
exceed their accrual model counterparts, indicating that the revenue models are more powerful
than the accrual models at detecting revenue manipulation. Rejection rates for the annual,
modified, and interim accrual models are 57.6%, 69.2%, and 48.8%, and rejection rates for the
annual, modified, and interim revenue models are 100.00%, 100.00%, and 94.8%. Thus, despite
the general tendency of accrual models to over-reject the null hypothesis, the revenue models
19
This analysis assumes zero discretionary revenues/accruals on average for growth firms. This does not, however,
assume no manipulation. Because models of discretion are estimated in cross section, estimated manipulation is
relative to the industry-year average. Therefore, the assumption of this analysis is that growth firms, on average, do
not manipulate more than other firms in the same industry and year. To the extent this is not true, I overstate the
bias and misspecification of the models.
25
When expense manipulation is added, the accrual models detect it most of the time.
Rejection rates for the annual, modified, and interim accrual models are 92.8%, 94.4%, and
91.2%. Finally, when both revenue and expense manipulation are added, each of the models
detects manipulation more than 90% of the time. However, the high rejection rates of the accrual
models are in part attributable to their general tendency to over-reject for growth firms.
Table 3, panel B, also tabulates rejection rates for performance-matched estimates from
the six models. For growth firms with no revenue manipulation, the performance-matched
estimates produce lower rejection rates for each of the models. However, the rejection rate for
each of the accrual models is significantly higher than 5% (18.0%, 20.4%, and 17.2% for the
annual, modified, and interim equations), indicating that each is misspecified, even after
performance matching. The rejection rates of the revenue models decrease from 11.2%, 20.8%,
and 8.0% to 8.8%, 10.0%, and 4.8%. This indicates that after performance matching, the interim
Table 3, panel B, reveals that performance-matched estimates from each model are less
rejection rates decrease from 57.6%, 69.2%, and 48.8% to 35.2%, 43.6%, and 26.0% for the
annual, modified, and interim accrual models. Rejection rates for the revenue models decrease
from 100.0%, 100.0%, and 94.8% to 90.8%, 92.4%, and 76.0%. For the best specified model,
the interim revenue model, performance matching has a small effect on specification and reduces
power. Although in theory both the revenue and accrual models should detect revenue
management, the interim revenue model is the only model that is well specified for growth firms,
26
The accrual models detect expense manipulation much of the time. Rejection rates for
the annual, modified, and interim accrual models are 72.8%, 73.6%, and 71.6%. Finally, when
both revenue and expense manipulation are added, the power of the accrual models increases
slightly (over that of expense manipulation alone), whereas that of the revenue models remains
The findings indicate that performance matching improves the specification of the
accrual models more so than the revenue models. Performance matching improves the
specification of the accrual models by controlling for factors omitted from the models—for
example, changes in purchases to explain the payables accrual and changes in expected sales to
explain the inventory accrual. The interim revenue model is more complete. Thus, it is less
important to control for omitted variables, thereby reducing the power of the model. The results
from Table 3, panel B, indicate that for growth firms, performance matching generally improves
the specification of the models, but only the interim revenue model is well specified with or
without performance matching. Furthermore, performance matching reduces the power each of
The second procedure I use to evaluate revenue and accrual models assesses their ability
to detect revenue and expense manipulation in a sample of firms that are known to have
misstated their financial results. The known manipulators are a sample of 68 firms that were
investigated by the SEC for accounting irregularities between 1997 and 2003 and then
subsequently restated their annual financial results. I study the intersection of SEC enforcement
actions and restatements in order to identify a sample of firms that are more likely to have
27
manipulated their financial results. When studying enforcement actions, it is possible that the
allegations are not true. With restatements, it is possible the restatement reflects a minor
correction that does not represent intentional manipulation by managers. A combined sample
I divide sample firms into two groups: those that manipulated revenues, and those that
manipulated expenses but not revenues.20 For each sample firm, I group observations into four
time periods: the manipulation period, the year before the manipulation, the year after the
manipulation, and all other years. I assume that, on average, sample firms overstate revenues
and earnings during the manipulation period.21 I also assume that no manipulation took place the
I make the following predictions. For the sample firms that manipulated revenues, if the
models are correctly specified, mean discretionary revenue and accrual estimates should not
differ from zero during the year before the manipulation. If the models are powerful, mean
discretionary accrual and revenue estimates should be significantly positive during the revenue
manipulation period.
For the sample firms that manipulated only expenses, I predict that the accrual models, if
correctly specified, will not detect discretionary accruals before the manipulation. If the accrual
models are powerful, they will detect positive discretionary accruals during the manipulation. If
different from zero before, during, and after the manipulation period.
Based on the findings in the previous section, I use the interim revenue model to measure
discretionary revenues because it is the most correctly specified of the three revenue models. I
20
Expenses include “Reserves/Accruals” and “Inventory”, as categorized by Huron Consulting Group.
21
Consistent with an overstatement on average, Dechow, Sloan, and Sweeney (1996) finds positive discretionary
accruals during the manipulation period by firms subject to SEC enforcement actions.
28
use performance-matched estimates from the modified annual accrual model to measure
discretionary accruals. This model serves as a benchmark because it is the most commonly used
of the accrual models. Although it is not as well specified as the annual and interim accrual
models, it is generally the most powerful. Based on the simulation analysis, the interim revenue
model is the only model correctly specified for high growth firms, and prior research documents
that firms targeted by SEC enforcement actions tend to be growth firms (Beneish, 1999).
Table 4, panel A, displays the distribution of sample firms through event time. Revenues
were manipulated over 42 firm-years, and expenses were manipulated over 24 firm-years,
consistent with revenue manipulation being one of the most common forms of earnings
management.
Panel B provides evidence consistent with both models being well specified for the entire
sample. Neither model detects discretion for the year before the manipulation (t = –0.28 for the
accrual model and t = –0.62 for the revenue model). Assuming the sample firms overstated
earnings and revenues during the manipulation period, only the revenue model is powerful
enough to detect this discretion. The mean discretionary accrual estimate is not significant
(0.45%, t = 0.36) and the discretionary revenue estimate is significantly positive (1.34%, t =
2.26).
Panel C presents results for the sample of firms that manipulated revenues. Both models
appear to be well specified; mean estimates are not significantly different from zero for the year
before the manipulation (t = 0.63 for the accrual model and t = –1.06 for the revenue model).
However, only the revenue model is powerful enough to detect revenue manipulation during
29
event year 0. The accrual model detects revenue manipulation of 2.26% (t = 1.35), and the
Panel D presents results on the detection of expense manipulation. Again, both models
appear to be well specified. Mean estimates are not significantly different from zero for the year
before the manipulation (t = 0.28 for the accrual model and t = 0.05 for the revenue model). As
expected, the revenue model does not detect discretion during the manipulation period (t = 0.87).
However, the accrual model does not either (t = 0.67). In sum, the revenue model detects
discretion by firms that manipulated revenues, but the accrual model is unable to detect the
manipulation of revenues or expenses. These findings suggest that the revenue model is superior
for detecting revenue manipulation, and also earnings manipulation in general for this sample of
firms.
6. Primary Results
Tables 5 and 6 present mean estimates of discretionary revenues and accruals for firms
just meeting earnings and revenue forecasts. For the same reasons as with Table 4, I use the
from the modified accrual model to measure discretionary accruals. I also report performance-
matched revenue model estimates to provide further assurance that the detected discretion is not
attributable to variables omitted from the revenue model. Because firms with analyst forecasts
have different characteristics than those without forecasts, I require that the matched firm also be
22
Inferences are not sensitive to this choice.
30
6.1 Earnings Management Using Revenues
benchmarks. Panel A presents descriptive statistics. Thirty-seven percent of the sample of firms
with earnings forecasts report earnings above the consensus forecast by less than 0.3% of equity
market value. The mean (median) revenue growth is 20% (10%), and the mean (median) gross
Panel B presents results for the pooled sample. Consistent with H1, the interim revenue
model detects discretionary revenues (0.32%, t = 9.91), and this estimate is significantly higher
than that of firms that just missed the earnings forecast. The modified accrual model detects
discretionary accruals (1.14%, t = 17.08), but the estimate is not significantly higher than that of
firms that just missed the target. After performance matching, it is no longer significantly
Panel C presents results for growth firms. The interim revenue model detects
discretionary revenues for growth firms (0.59%, t = 8.55), and this estimate is significantly
higher than that of growth firms that just missed the earnings forecast. Consistent with H1a, it is
also significantly higher than that of non-growth firms that just met the earnings forecast. The
modified accrual model detects discretionary accruals (0.96%, t = 6.55), but the estimate is not
significantly higher than that of firms that just missed the forecast. After performance matching
the estimate, it is no longer significantly positive. The performance-matched estimate from the
Panel D presents results for firms with high gross margins. The interim revenue model
detects discretionary revenues for firms with high gross margins (0.65%, t = 11.82; 0.22%, t =
2.81 after performance matching), and this estimate is significantly higher than that of high gross
31
margin firms that just missed the earnings forecast. Consistent with H1b, it is also significantly
higher than that of other firms that just met the earnings forecast. The modified accrual model
detects discretionary accruals (1.11%, t = 8.92). Although this estimate is significantly higher
than that of firms that just missed the forecast, it is no longer significantly positive after
performance matching.
benchmarks. Panel A presents descriptive statistics. Twenty-nine percent of the sample of firms
with revenue forecasts report revenues above the consensus forecast by less than 1.0% of equity
Panel B presents results for the pooled sample. The interim revenue model detects
discretionary revenues (0.20%, t = 3.59), but this estimate is not significantly higher than that of
firms that just missed the revenue forecast. The modified accrual model does not detect
discretionary accruals (0.08%, t = 0.47), and the estimate is not significantly higher than that of
firms that just missed the earnings forecast. Furthermore, neither the discretionary revenue
estimate nor the discretionary accrual estimate is significantly positive after performance
matching.
Panel C presents results for growth firms. Consistent with H2, the interim revenue model
detects discretionary revenues for growth firms (0.42%, t = 3.81), but not after performance
matching (0.21%, t = 1.35). Although this estimate is significantly higher than that of non-
growth firms that just met the revenue forecast, it is not significantly higher than that of growth
firms that just missed the revenue forecast. The modified accrual model does not detect
32
discretionary accruals (–0.72%, t = –2.14), and the estimate is not significantly higher than that
7. Conclusion
This paper studies whether firms use discretion in revenues to meet earnings and revenue
forecasts. One advantage to studying revenues rather than aggregate earnings is that it can
provide insights into how firms manage earnings. I find that firms prematurely recognize
revenues to meet earnings forecasts. I also find that firms with greater incentives to use
discretion in revenues as opposed to expenses (i.e., growth firms and firms with high gross
Another advantage to studying revenues is that a revenue model can provide more
precise estimates of discretion than existing discretionary accrual models. I use simulated
revenue and expense manipulation to show that the revenue model is better specified than
accrual models for growth firms. It is also more powerful than accrual models at detecting
revenue manipulation or an even combination of revenue and expense manipulation. This latter
finding is supported further using a sample of firms that were targeted by SEC enforcement
actions and subsequently restated their financial results. The accrual model, performance
matched to correct for misspecification, does not detect revenue or expense manipulation. The
revenue model detects discretion in both the full sample of SEC firms and the sub-sample of
The revenue model can also be used to test for revenue management. I find limited
evidence that firms in general and especially growth firms manage revenues to meet revenue
33
forecasts. This is the first study I am aware of that examines whether firms use discretionary
The revenue model in this paper is relevant for the estimation of discretionary accruals.
In developing the revenue model, I suggest solutions for two limitations of commonly used
accrual models. First, I suggest modeling nondiscretionary accruals as a function of the change
in reported revenues rather than the change in cash revenues, in order to avoid biased estimates
of discretion for growth firms. Second, I suggest separating the change in fourth-quarter
revenues from annual revenues when modeling discretionary accruals. This helps to avoid
biased estimates of discretion for growth firms with relatively high fourth-quarter revenues.
One limitation of this study is that it measures only one (albeit common) form of
discretionary revenues. Firms have a number of alternatives available to manage revenues, and I
plan to include additional measures of discretionary revenues in a future version of the paper.
34
Appendix: Summary of Revenue and Accrual Models
where
AR = End of fiscal year accounts receivable
AC = Annual current accruals (excluding depreciation), = earnings before extraordinary items
+ depreciation – cash from operations
S = Annual revenues
S1_3 = Revenues of the first three quarters
S4 = Revenues of the fourth quarter
∆ = Denotes annual change
35
References
Barth, M.; W. Beaver; J. Hand; and W. Landsman. 2005. Accruals, accounting-based valuation
models, and the prediction of equity values. Journal of Accounting, Auditing & Finance
20: 311-345.
Barth, M; and S. Kallapur. 1996. The effects of cross-sectional scale differences on regression
results in empirical accounting research. Contemporary Accounting Research 13: 527-
567.
Beatty, W.; B. Ke; and K. Petroni. 2002. Differential earnings management to avoid earnings
declines and losses across publicly and privately-held banks. The Accounting Review 77:
547-570.
Beaver, W.; M. McNichols; and K. Nelson. 2003. Management of the loss reserve accrual and
the distribution of earnings in the property-casualty insurance industry. Journal of
Accounting and Economics 35: 347-376.
Burgstahler, D.; and I. Dichev. 1997. Earnings management to avoid earnings decreases and
losses. Journal of Accounting and Economics 24: 99-126.
Burgstahler, D.; and M. Eames. 2002. Management of earnings and analyst forecasts to achieve
zero and small positive earnings surprises. Working paper, University of Washington.
Dechow, P.; S. P. Kothari; and R. Watts. 1998. The relation between earnings and cash flow.
Journal of Accounting and Economics 25 (May): 131-168.
Dechow, P.; S. Richardson; and I. Tuna. 2003. Why are earnings kinky? An examination of the
earnings management explanation. Review of Accounting Studies 8: 355-384.
Dechow, P., and C. Schrand. 2004. Earnings quality. The Research Foundation of CFA Institute.
Charlottesville, Virginia.
Dechow, P.; R. Sloan; and A. Sweeney. 1995. Detecting earnings management. The Accounting
Review 70: 193-225.
Dechow, P.; R. Sloan; and A. Sweeney. 1996. Causes and consequences of earnings
manipulation: An analysis of firms subject to enforcement actions by the SEC.
Contemporary Accounting Research 13:1-36.
Dhaliwal, D.; C. Gleason; and L. Mills. 2004. Last change earnings management: Using the tax
expense to achieve earnings targets. Contemporary Accounting Research 21: 431-459.
36
Ertimur, Y.; J. Livnat; and M. Martikainen. 2003. Differential market reactions to revenue and
expense surprises. Review of Accounting Studies 8, 185-211.
Ertimur, Y., and S. Stubben. 2005. Analysts’ incentives to issue revenue and cash flow forecasts.
Working paper, Stanford University.
Feroz, E.; K. Park; and V. Pastena. 1991. The financial and market effects of the SEC’s
accounting and auditing enforcement releases. Journal of Accounting Research 29
(Supplement):107-142.
Financial Accounting Standards Board. 1987. Statement of Financial Accounting Standards No.
95: Statement of Cash Flows (FASB, Stamford, CT).
Ghosh, A.; Z. Gu; and P. Jain. 2005. Sustained revenue and earnings growth, earnings quality,
and earnings response coefficients. Review of Accounting Studies 10:33-57.
Graham, J.; C. Harvey; and S. Rajgopal. 2005. The economic implications of corporate financial
reporting. Journal of Accounting and Economics 40: 3-73.
Guay, W.; S. P. Kothari; and R. Watts. 1996. A market-based evaluation of discretionary accrual
models. Journal of Accounting Research 34 (Supplement): 83-105.
Hribar, P., and D. Collins. 2002. Errors in estimating accruals: Implications for empirical
research. Journal of Accounting Research 40: 105-135.
Jones, J. 1991. Earnings management during import relief investigations. Journal of Accounting
Research 29: 193-228.
Kang, S., and K. Sivaramakrishnan. 1995. Issues in testing earnings management and an
instrumental variable approach. Journal of Accounting Research 33: 353-367.
Kasznik, R. 1999. On the association between voluntary disclosure and earnings management.
Journal of Accounting Research 37: 57-81.
Kothari, S. P.; A. Leone; and C. Wasley. 2005. Performance matched discretionary accrual
measures. Journal of Accounting and Economics 39: 163-197.
Levitt, A. 1998. The numbers game. Speech delivered at the NYU Center for Law and Business,
New York, NY, September 28.
Lipe, R. 1986. The information contained in the components of earnings. Journal of Accounting
Research (Supplement): 37-64.
Magrath, L; and L. Weld. 2002. Abusive earnings management and early warning signs. The
CPA Journal (August).
37
Marquardt, C.; and C. Wiedman. 2004. How are earnings managed? An examination of specific
accruals. Contemporary Accounting Research 21: 461-491.
Nelson, M.; J. Elliott; and R. Tarpley. 2002. Evidence from auditors about managers’ and
auditors’ earnings-management decisions. The Accounting Review 77 (Supplement): 175-
202.
Phillips, J.; M. Pincus; and S. Rego. 2003. Earnings Management: New Evidence Based on
Deferred Tax Expense. The Accounting Review 78: 491-521.
Plummer, E., and D. Mest. 2001. Evidence on the management of earnings components. Journal
of Accounting, Auditing & Finance 16: 301-323.
Porter, M. 1985. Competitive Advantage: Creating and Sustaining Superior Performance, New
York: The Free Press.
PricewaterhouseCoopers. 2001. Revenue is revenue, right? Under SAB No. 101, the answer is
not always apparent. Deal Flash 11.
Roychowdhury, S. 2004. Management of earnings through the manipulation of real activities that
affect cash flow from operations. Working paper, Massachusetts Institute of Technology.
Sopranzetti, B. 1998. The economics of factoring accounts receivable. Journal of Economics and
Business 50: 339-359.
Teoh, S.; I. Welch; and T. Wong. 1998. Earnings management and the long-run
underperformance of seasoned equity offerings. Journal of Financial Economics 50: 63-
100.
Teoh., S.; T. Wong; and G. Rao. 1998. Are earnings during initial public offerings opportunistic?
Review of accounting Studies 3: 175-208.
Thomas, J., and X. J. Zhang. 2000. Identifying unexpected accruals: a comparison of current
approaches. Journal of Accounting and Public Policy 19: 347-376.
Turner, L.; J. R. Dietrich; K. Anderson; and A. Bailey. 2001. Accounting restatements. Working
paper, Colorado State University.
38
Wu, M. 2002. Earnings restatements: A capital market perspective. Working paper, New York
University.
Zuckerman, G. 2000. Abreast of the market. Wall Street Journal. New York, N.Y.: Sep 25, 2000.
39
Table 1: Sample descriptive statistics (N = 56,403)
Variables are deflated by average total assets. All correlations in panel B are significantly
different from zero (p < 0.05), except the Spearman correlation between ∆S and AC – ∆AR (p =
–0.14).
40
Table 2: Estimation of Discretionary Accrual and Revenue Models (N = 56,403)
Dep. Var.it = α + β × ∆S it + ε it
β
Dep. Var. Estimate t-statistic Adj. R2
AC 0.10 56.83 0.11
∆AR 0.09 120.42 0.25
AC – ∆AR 0.00 2.25 0.04
β1 β2
Dep. Var. Estimate t-statistic Estimate t-statistic Adj. R2
AC 0.07 28.79 0.19 33.39 0.11
∆AR 0.04 41.82 0.26 102.97 0.30
AC – ∆AR 0.02 11.05 –0.07 –12.26 0.04
Variables are defined in Table 1. Each panel presents the results of pooled estimations of the
models, with an intercept scaled by average total assets and separate industry and year unscaled
intercepts. The scaled and separate intercepts are not tabulated.
41
Table 3: Specification and Power of Discretionary Accrual and Revenue Models for
Earnings Growth Firms
%Manip:
(Rev,Exp) (0,0) (2,0) (0,2) (2,2)
Model Mean Mean Mean Mean S.E.
Ann. AC 1.70 2.29 3.70 4.29 1.06
Mod. AC 1.92 2.67 3.92 4.67 1.08
Int. AC 1.63 2.03 3.63 4.03 1.07
Ann. AR 0.41 2.23 0.41 2.23 0.49
Mod. AR 0.66 2.66 0.66 2.66 0.54
Int. AR 0.25 1.71 0.25 1.71 0.47
Performance-matched Estimates
%Manip:
(Rev,Exp) (0,0) (2,0) (0,2) (2,2) (0,0) (2,0) (0,2) (2,2)
Model Rate Rate Rate Rate Rate Rate Rate Rate
Ann. AC 40.0 57.6 92.8 96.0 18.0 35.2 72.8 89.2
Mod. AC 44.8 69.2 94.4 99.2 20.4 43.6 73.6 90.4
Int. AC 38.0 48.8 91.2 93.6 17.2 26.0 71.6 82.8
Ann. AR 11.2 100.0 11.2 100.0 8.8 90.8 8.8 90.8
Mod. AR 20.8 100.0 20.8 100.0 10.0 92.4 10.0 92.4
Int. AR 8.0 94.8 8.0 94.8 4.8 76.0 4.8 76.0
Earnings growth firms are those in the highest quartile of change in earnings before
extraordinary items deflated by average total assets, where quartiles are determined by industry
and year. “%Manip” is the percent of revenue or expense manipulation induced in each of 250
random samples of 100 firm-years before estimation of the models—either 0% or 2% of average
assets. Sample firm-years are excluded from estimation of the models. “Mean” is the mean of
the mean discretionary revenue/accrual estimates from 250 samples of 100 firms, “S.E.” is the
standard deviation of the 250 sample means, and “Rate” is the percent of the 250 sample means
that were significantly greater than zero (α = 0.05). The accrual and revenue models are
described in the appendix. Performance-matched estimates are calculated for each sample firm
by subtracting the discretionary revenue/accrual estimate of the firm from the same industry and
year with the closest return on assets.
42
Table 4: Detection of Revenue and Expense Manipulation by Firms Subject to SEC
Enforcement Actions and Financial Statement Restatements
Sample firms are drawn from the intersection of the set of firms subject to SEC enforcement
actions between 1997 and 2003 and a database of financial restatements maintained by Huron,
Inc. REV indicates whether the sample firm restated revenues, and EXP indicates whether the
43
sample firm restated expenses (reserves, accruals, or inventory), but not revenues. Event years
include –1, the year preceding the first year of the manipulation, 0, one or more years during the
manipulation, and 1, the first year after the manipulation. Panels B through D present mean
estimates of discretion using the modified accrual model and the interim revenue model, which
are described in the appendix. Estimation of the models is carried out by industry-year groups,
excluding the sample firms from the estimation. “Mean” represents the mean regression residual
across all sample firm observations in the event year, and “t-statistic” is calculated as the mean
estimate divided by the standard error of the mean estimate. Performance-matched (PM)
estimates are calculated for each sample firm by subtracting the discretionary accrual estimate of
the firm from the same industry and year with the closest return on assets.
44
Table 5: Mean Estimates of Discretion for Firms Just Meeting Earnings Targets
Panel B: Firms with Small Positive Earnings Forecast Errors – Entire Sample
Panel C: Firms with Small Positive Earnings Forecast Errors – Growth Firms
45
Table 5 (continued): Mean Estimates of Discretion for Firms Just Meeting Earnings
Targets
Panel D: Firms with Small Positive Earnings Forecast Errors –Firms with High Gross Margins
FE is the difference between realized annual earnings and the last consensus (median) earnings
forecast before the earnings announcement, scaled by beginning-of-year market value of equity,
FE_SMP is an indicator variable that equals one if FE is between 0% and 0.3%, SG equals sales
in year t-1 divided by sales in year t-2, and GM is gross margin in year t-1, which equals sales
minus cost of goods sold, divided by sales. Growth (high gross margin) firms are those with SG
(GM) in the highest quartile of firms in the same industry and year. Discretionary revenue and
accrual estimates are expressed as a percent of average total assets. The discretionary accrual
and revenue models are described in the appendix. Performance-matched (PM) estimates are
calculated for each sample firm by subtracting the discretionary accrual estimate of the firm from
the same industry and year with the smallest absolute difference in return on assets. The final
column in panels C and D presents results from a two-sample t-test for differences between
growth (high gross margin) firms and other firms. * indicates that the estimate of discretion is
significantly higher than that of firms that just missed the forecast (FE greater than −0.03% but
less than 0%).
46
Table 6: Mean Estimates of Discretion for Firms Just Meeting Revenue Forecasts
Panel B: Firms with Small Positive Revenue Forecast Errors – Entire Sample
Panel C: Firms with Small Positive Revenue Forecast Errors – Growth Firms
RFE is the difference between realized annual revenues and the last consensus (median) revenue
forecast before the earnings announcement, scaled by beginning-of-year market value of equity,
RFE_SMP is an indicator variable that equals one if RFE is between 0% and 1%, SG equals sales
in year t-1 divided by sales in year t-2. Growth firms are those with SG in the highest quartile of
firms in the same industry and year. Discretionary revenue and accrual estimates are expressed
as a percent of average total assets. The discretionary accrual and revenue models are described
in the appendix. Performance-matched (PM) estimates are calculated for each sample firm by
subtracting the discretionary accrual estimate of the firm from the same industry and year with
the smallest absolute difference in return on assets. The final column in panel C presents results
from a two-sample t-test for differences between growth firms and other firms. * indicates that
the estimate of discretion is significantly higher than that of firms that just missed the forecast
(RFE greater than −1% but less than 0%).
47
Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.
Alternative Proxies: