Efficiency and the Determinants of Bank

Profitability in MENA Countries

Dennis Olson, Ph.D.
Professor of Finance
School of Business and Management
American University of Sharjah
Sharjah, United Arab Emirates

and

Taisier A. Zoubi, Ph.D, CMA, CFM
Professor of Accounting
School of Business and Management
American University of Sharjah
Sharjah, United Arab Emirates


PRELIMINARY DRAFT








Authors names are listed alphabetically. Dennis Olson and Taisier A. Zoubi


1

Efficiency and the Determinants of Bank
Profitability in MENA Countries



Abstract

This study compares accounting-based and economic-based measures of efficiency and
profitability of banks in Middle East and North Africa (MENA) countries. It considers
differences between banks in GCC countries (oil-based economies) and other MENA
countries, as well differences between Islamic and conventional banks. The data consist of
cost, profit, and accounting ratios for all national banks in MENA countries for the period
2000-2007. Accounting-based measures of performance are based on panel regressions
using the ratiosreturn on assets, return on equity, net interest margin, and the net
noninterest margin. Economic-based measures of bank performance are derived from
stochastic frontier analysis using cost and alternative profit functions. These functions are
first approximated using the the same nonhomothetic translog expansion considered in
many previous. Then the results are compared with those from three other functional
forms: Generalized Leontief, minflex Laurent-Generalized Leontief, and minflex Laurent-
translog. The minflex Laurent forms permit greater flexibility in the approximation
function and a larger regular region than the translog approximation, while the
Generalized Leontief form may be preferable to translog forms when there is low elasticity
of substitution between inputs. These functional forms are used to derive estimates of (1)
elasticities of substitution between the inputs deposits, labor, and fixed assets, (2) scale
economies, and (3) cost-based and profit-based measures of the degree of efficiency or
inefficiency for each bank. A comparison of the economic and accounting-based measures
of profitability is accomplished by second stage panel regressions using measures of
efficiency as the dependent variable and the various accounting ratios as independent
variables.
Preliminary results show that GCC banks and Islamic banks are more profitable than
other MENA banks using accounting based measures of performance. They are also more
cost and profit efficient based on stochastic frontier analysis. The accounting and
economic-based performance indicators provide somewhat consistent results among banks
and countries. However, there are some differences between the estimates from the four
different functional forms. There are also differences in results when cost and profit
functions are estimated separately for GCC and non-GCC countries for Islamic and
conventional banks, suggesting that there are considerable differences between the nature
of the banking industry across countries.



2
1. Introduction
During the last two decades the banking sector in the MNEA (Middle East and North
Africa) region has experienced major transformations in its operating environment. Both external
and domestic factors have affected its structure and performance. The role of banks remains
central in financing economic activity and providing for sustained economic growth and
development. Also, a sound and profitable banking is better able to withstand word-wide or
regional financial shocks. Given the importance of the banking sector it is not surprising that a
substantial body of academic now exists that examines cost efficiency and profitability in the
banking sector. Since most of the literature has focused upon the banking sectors in North
America and Europe, there is need for a more comprehensive examination of the determinants of
profitability and efficiency in developing countries and for the MENA region, in particular
The literature on profitability and efficiency in the banking industry can be classified into
two groups of studieseconomics-based and accounting-based. The economics-based research
uses using various mathematical models to measure the efficiency by which banks convert inputs
into one or more bank outputs. Cost and/or profit efficiency is then calculated as the distance
away from some ideal frontier measured relative to the lowest cost or highest profit firms in the
sample. The accounting-based research uses more comprehensive information from financial
statements to identify the determinants of bank profitability. Some studies focus upon the
external factors determining bank profit, such as market share, inflation, business cycles, and
industry return; while other work emphasizes the impact of bank-specific or internal factors of
profitability such as size, revenue growth, business risk, credit risk, and expenses. With some
notable exceptions, such as Grinyer, McKiernan, and Yasai-Ardekani, 1988; White, 1986; and


3
Lenz, 1981, most studies do not assess the joint impact of internal and external factors on bank
profitability. Similarly, few studies [with the exception of Berger and Mester (1997)] have
examined the relationship between cost or profit efficiency and the accounting-based
determinants of bank profitability.
The aim of this study is threefold. First, we examine the effect of bank-specific internal
factors and external factors on banks profitability, using an empirical framework that
incorporates the nonlinear relationship among the variables examined in this study. We apply
nonlinear models to a panel of MENA banks for the period 2000-2007. Second, we estimate cost
and profit efficiency for MENA banks based upon estimates from a parametric stochastic
frontiers calculated from the translog and other functional forms. Third, we check for
consistency between the results of the economics and accounting-based measures of bank
efficiency and profitability.
The focus on banking in the MENA region is important a number of reasons. It is a fast
growing region in terms of both population and wealth. Recently, several countries from the
Gulf Cooperation Council (GCC) have provided considerably liquidity for the world banking
system. The banking sector in the MENA region is relatively young, with most banks only
dating back to the 1950s and many being established in the 1970s and even more recently. The
region is particularly interesting because it provides a mix of conventional and Islamic banks and
because some economies are based on oil exports, while others are not.
The remainder of the paper is organized as follows. Section 2 provides a review of the
literature on bank efficiency, determinants of profitability, and studies of the MENA banking
industry. Section 3 describes the data and Section 4 defines the variables used to examine bank


4
profitability. Section 5 presents the accounting-based determinants of profitability derived from
panel estimates by fixed and random effects models. Section 6 presents the translog functional
form which is used to estimate cost and profit efficiency for Islamic versus conventional banks
and to identify efficiency differences for oil exporting nations versus other MENA economies.
Section 7 compares the results for the translog approximation with those of the minflex Laurent-
translog, Generalized Leontief, and the minflex Laurent-Generalized Leontief approximations for
the cost and profit functions of MENA banks. Section 8 compares accounting and economic-
based measures of efficiency by checking whether accounting ratios used in Section 5 explain
cost and profit efficiency measured in section 6. Section 9 makes some concluding remarks.
2. Literature Review
Several studies have examined the effect of the internal and external factors on bank
profitability [e.g., Berger et. al. (1987), Neely and Wheelock (1997), and Barajas et. al. (1999)].
Internal variables utilized include various measures of size, capital, risk management and
expense management. Molyneux and Thornton (1992), Bikker and Hu (2002), and Goddard et.
al. (2004) contend that size is closely related to profitability of a bank since relatively large banks
tend to raise less expensive capital and, hence, appear more profitable. Other studies, such as
Amato and Wilder (1985) Grant et. al. (1988), and Ramanujam and Varadarjan (1989) have
raised concerns that variables other than bank size have a major impact on profitability. For
example, leverage was found to be important and usually negatively correlated with profitability
as in Grant and Jammine (1988) and McDougall and Round (1984).
External determinants of bank profitability examined in prior studies include inflation,
interest rates, cyclical output, market concentration, industry size, and ownership status. For


5
example, Saunders and Schumacher (2000) showed that the degree of bank capitalization, bank
market structure, and interest rate volatility affect profitability, as measured by interest rate
margins, for banks in six European countries and the US over the period 1988-95. Brock and
Suarez (2000) examined the performance of banks for seven Latin American countries during the
1990s. They reported that bank spreads are influenced by liquidity and capital risk at the bank
level, and by interest rate volatility, inflation, and GDP growth at the macroeconomic level.
Previous research relevant to the MENA include Hakim and Neaime (2000) who
examined the effect of liquidity, credit risk, and capital adequacy on profitability of banks in
Egypt and Lebanon for the period 1993-1999. Return on equity was used as a measure of
profitability. Their results show that lending activities have strong impact on profitability of
banks in both countries, while the effect of capital adequacy on profitability was stronger for
Lebanese banks than Egyptian banks. Liquidity was not an important factor in explaining bank
profitability in either country.
Ahmed and Khababa (1999) studied the effects of size, business risk, and market
concentration on the profitability of eleven commercial banks in Saudi Arabia for the period
1992-1997. They employed a regression model using three measures of profitability--return on
assets, return on equity, and earnings per share. Their result showed that business risk and bank
size explained bank profitability in Saudi Arabia.
Essayyad and Madani (2003) examined the concentration, efficiency, and profitability of
10 commercial banks in Saudi Arabia for the years 1989-2001. They showed that the Saudi
Arabian banking industry is highly concentrated and has a four-firm concentration ratio ranging
between 69% and 87%. Additionally, profitability rises with increases in bank efficiency and


6
that Saudi Arabian bank profits are positively related to oil revenues.
Islam (2003) examined the development and performance of local and foreign banks in
three GCC countries (Bahrain, Oman, and UAE) for the period 1996-2000. He calculated the
return on average total assets of banks on those countries to measure the performance of the
banking sectors on those countries. No attempt was made to examine empirically the
determinants of bank profitability. He also examined the funding of commercial banks, deposit
insurance schemes, assets quality, audits and reporting, and the offshore banking activities.
Islams study can be characterized as descriptive of the economy and banking sectors in three
GCC countries.
Ben Nacer (2003) examined the effect of internal and external factors on the return on
assets of ten Tunisian banks for the period 1980 through 2000 using regression analysis.
Significant internal factors positively related to were the ratios of overhead expenses to total
assets, loans to assets, and equity to assets, and bank size. Important external determinants of
profitability included market concentration and the relative size of the banking sector in GDP,
while traditional variables such as inflation and the percentage annual growth in GDP did not
significantly affect profitability.
Tarawneh (2006) studied the impact of bank size (total assets), assets utilization ratio
(operational income divided by total assets) and operational efficiency (total operating expenses
divided by net interest income) on profitability measured by return on assets and interest income
for five Omani banks for the period 1999-2003. These three explanatory variables were
statistically significant in explaining profitability of Omani banks.
Ramanathan (2007) examined the performance of 55 GCC banks using Data


7
Envelopment Analysis (DEA) and Malmquist Productivity Index (MPI) for the period 2000-
2004. He used two output factors: loans and other earning assets and four input factors: fixed
assets, deposits, equity, and fixed assets. Only 15 banks in the GCC region were efficient,
although the MPI results indicated that there was an improvement in productivity during the
period 2000-2004 in four countries: Bahrain, Kuwait, Saudi Arabia, and UAE.
Angbazo (1997) examined the effect of bank reserves, the leverage ratio (capital to
assets), default risks (loan charge offs to total loans), efficiency (earning assets to total assets),
and liquidity risk (liquid assets to total liabilities) on banks performance as measured by the net
interest margin for US banks during the years 1988-1993. He found that leverage, efficiency, and
liquidity risks were positively related to bank performance.
Berger and Mester (1997) undertook a rather comprehensive analysis of cost and profit
efficiency in U.S. banking and also examined accounting correlates of economic efficiency.
Demirg-Kunt and Huizinga (1999) examined the internal and external determinants of
profitability of banks in 80 countries for the period 1988-1995. Internal bank characteristics
(such as size, percentage of foreign ownership, loan to assets ratio, equity to assets, overhead
costs to total assets ratio, and non-interest earning assets), macroeconomic conditions (such as
inflation and short term interest rates), and country-specific institutional features (such as explicit
and implicit bank taxation,

deposit insurance regulation, overall financial structure, and

the
underlying legal structure) all affect bank profitability.
Brock and Rojas-Suazez (2000) utilized a two step procedure to examine bank interest
rate spreads in five Latin American countries during the mid 1990s (Argentina, Bolivia,
Colombia, Chile, and Peru). In the first stage, they regressed the bank interest spread against


8
variables such as non-performing loans, capital ratio, operating costs, liquidity and time
dummies. They found positive coefficients for capital ratio, cost ratio, and the liquidity ratio;
while the effect of nonperforming loans was mixed (positive for some countries and negative for
others). In the second stage pure bank spreads were regressed against macroeconomic
variables. This procedure showed that profitability, as measured by bank spreads, was positively
related to interest rate volatility and inflation, but unaffected by the growth rate of GDP.
Kosmidou, Pasiouras, and Tsaklanganos (2007) examined the determinants of
profitability of Greek banks operating in 11 nations over the period 19952001 using regression
analysis. Bank size, asset quality (loan loss provision divided by net interest income), liquidity
(net loans divided by total assets), capital strength (equity divided by total assets), a cost variable
(noninterest expense divided by average assets), a concentration variable (total assets of 5 largest
banks divided by total assets of all banks), and market share (deposits of the bank divided by
total deposits of all the banks) were important factors in explaining bank profitability..
Valverde and Fernandez (2007) examined the determinants of bank margins for 19,322
bank-years in seven European countries (Germany, Spain, France, the Netherlands, Italy, the
United Kingdom and Sweden) for the period 19942001. The primary measure of bank
performance was the loan to deposits rate spread, while accounting gross income and the Lerner
Index provided a consistency check. Lagged values of the dependent bank margin variable,
credit risk, liquidity risk, interest rate risk, ratio of deposits to total liabilities, totals for other
earning assets and loan commitments, and operating inefficiency were positively correlated with
bank margins. Loans to total assets, other earning assets to total assets, and GDP were
negatively related to bank margins. (doesnt make sensereview paper)


9
(Review this on e too) Barros, Ferreira, and Williams (2007) estimated the probability of
bank performance, measured by cost efficiency and profit efficiency using a stochastic frontier
and Fourier flexible functional form for a sample of 1384 banks operating in 15 European
countries for the period 1993-2001. Several factors were used to explain the performance of a
bank using mixed logit. Explanatory variables are: national banking markets; the legal system of
the banks country, location of the bank; the country of origin of foreign banks; the amount of
loans, deposits, and assets of each bank; total assets of the bank to total assets of all the banks for
over the period, total loans of a bank to total loans of all the banks, deposits of a bank to total
deposits of all the banks. The results of this study indicate that better performer banks tend to be
larger than their worst performing banks. Worst performing banks has less loans and less
efficient in terms of cost and profit than best performing banks. However, the results indicated
that the location of the bank and the country of origin of the bank were insignificant variables for
best performing banks but significant for worst performing banks. Legal tradition is a significant
variable in explaining bank performance.
Kosmidou (2008) examined the effect of banks characteristics (internal factors) and the
macroeconomic and financial structure (external factors) on profitability, measured by return on
average total assets, of Greek banks for the period 1990-2002 using ordinary least squares. The
study indicate that the equity to assets ratio, loan loss reserves to loans, bank size, annual change
in GDP, inflation rates, ratio of banking system assets to GDP, ratio of market capitalization to
banks assets and concentration ratios were statistical significant and had the expected signs in
explaining profitability of banks. However, the money supply growth variable was not
significant in explaining profitability of banks.


10
Other studies of relevance to the MENA area include Bennaceur and Goaied (2008) who
investigated the profitability of Tunisian banks and Muharrami, Matthews, and Khabari (2006)
who examined market structure in the Arab GCC banking system. Our study goes beyond
previous work by considering the whole MENA region and we examine internal and external
factors in an accounting-based study of profitability based on a panel regression model. Such
results are compared with an economic-based study of cost and profit efficiency. The
economics-based results are derived using both the traditional translog function form for the
stochastic frontier as well as more flexible functional forms represented by various minflex
Laurent models.
3. Data
In order to examine the factors that explain the profitability of banks in the MENA region, we
collected data from the annual reports of banks in the MNEA region for the years 2000-2007.
The income statement, statement of change in stockholders equity, balance sheet, statement of
cash flows, and the notes to the financial statements were obtained from the annual report of each
bank as reported on their individual websites. External variables affecting bank performance
(e.g., inflation, GDP) were collected from International Monetary Fund (IMF).
The period of analysis represents the years for which electronic data are available for the majority
of banks in the MENA region. Foreign banks (such as Citibank, HSBC, ABN-Amro, etc.) have a
significant presence in the region, but they are excluded from the sample because country-
specific financial results are difficult to identify. Noting these caveats, we obtained data for 85
different banks for the years 2000 to 2007, for a total 470 bank-year observations. The
distribution of sample banks for the eleven Mena countries in our sample for the years 2000-2007


11
is shown in Table 1. Panel A indicates that the number of banks in the sample begins with 35 in
2000 and peaks at 82 in 2005. Annual sample size varies primarily because electronic data for
several banks in our sample were not available prior to 2002 and for some banks we could not
obtain data before 2004. There are only 23 banks in the sample for 2007 because some banks
only post annual reports with a considerable delay. Finally, the decline in the number of banks
after 2005 is due to bank mergers.
Looking at the data set by country, our sample ranges from 12 bank-years of data for Tunisian
banks up to 76 bank-years of data for banks in the United Arab Emirates. Panel B shows that
there are 308 bank-years of data for the oil producing countries of the GCC and 162 bank-years
of data for the non-oil producing countries of the MENA region. The sample contains 109
Islamic bank-years of data and 361 bank-years for conventional banks. Further subdivision
reveals that there are 104 bank-years of data for GCC Islamic banks and 204 bank-years of data
for conventional GCC banks. For the non-GCC region, there are only 5 bankyears of data for
Islamic banks and 157 bank-years of data for conventional banks.
3. Variables used in the study
The ratios and variables used in this study are summarized in Table 2. Accounting-based
studies of performance generally search for variables that explain differences in bank profitability
and based upon the review of the literature four profitability ratios are considered in this study.
The two most common measures of bank performance are return on assets (ROA), which is
defined as net income divided by total assets and return on equity (ROE), which is net income
divided by average shareholder equity. Two banking industry specific measures of performance
are the net interest margin (NIM), which is interest income minus interest expense divided by


12
total assets, and the net non-interest margin (NNIM), which is noninterest income minus
noninterest expense divided by total assets. In contrast, economics-based studies of cost and
profit efficiency generally take input quantities (e.g., number of workers, total deposits, fixed
assets) and aggregate costs for these categories of expenses (e.g. personnel expenses, interest
expenses, other non-interest expenses) to estimate some ideal minimum cost function.
Calculation of profit efficiency requires additional information about operating and interest
income and adjustments to net income. Following Maudos et al (2002) and other studies, we
measure profit efficiency using operating profit, which is defined as net income minus provisions
for loan losses. The various internal and external variables used as determinants of bank
profitability are described below.
4.1 Internal bank characteristics
Bank size (SIZE) is represented by the logarithm of total bank assets. In the banking
literature, bank size is the single most analyzed aspect of banking industry structure and bank
performance. Based on studies such as Hall and Weiss (1976), Grant, Jammine, and Thomas
(1988), and Kosmidou, Pasiouras, and Tsaklanganos (2007) the log of total assets is the most
frequently used measure of size. Generally, a positive relationship between profitability and
SIZE is expected, but this single measure may be too simplistic if profitability depends upon
relative size within a country, or if banks are subject to economies of scale up to a point and
diseconomies of scale thereafter. However, such nonlinearities can be checked using squared
and cubic terms in a separate test.
The loan specialization ratio (LOANS) is defined as net loans divided by total assets. It is
referred to as a liquidity ratio or an asset utilization ratio in some studies of the banking industry.


13
Loans generally provide the highest risk and highest return of any bank asset. Provided a bank is
not taking on unacceptable risk, the relationship between LOANS and bank profitability should
be positive.
The security specialization ratio (SECUR) is the ratio of other earning assets to total
assets. Other earning assets include all return-bearing assets other than loansmeaning various
types of securities. As noted by Staikouras, Mamatzakis, and Koutsomanoli-Filippaki (2008),
this ratio measures bank diversification and should be positively related to profitability.
The deposit specialization ratio (DEPLIAB) shows the importance of customer deposits
as a source of bank funds. It defined as total deposits divided by total liabilities and clearly
similar to the ratio of deposits to total assets as used in some studies. Since deposits are the
lowest cost source of funds, a positive relationship might be expected between DEPLIAB and
profitability. However, deposits are the least stable source of funds, so that the profitability
relationship may not be entirely clear. Valverde and Fernandez (2007) argue that the variable
could be either positively or negatively related to profitability depending upon whether deposits
are used as loss leaders to attract customers to other bank services and whether customers
actually purchase other services.
4.2 Internal measures of bank efficiency
Operating Expenses to Income, or the Inefficiency Ratio (INEFF), is defined as operating
expenses divided by gross income. Total bank costs can be separated into operating expenses
plus income expenses, while gross income is the sum of interest income and operating income.
Operating expenses are deemed to be under the control of management, so this expense ratio (or
variants such as operating expenses to operating income, or operating expenses to total assets,


14
etc.) measures bank inefficiency. Valverde and Fernandez (2007) contend that higher noninterest
expense relative to income implies lower efficiency (greater inefficiency)meaning that the
inefficiency ratio is negatively related to bank profitability.
The ratio of overhead, which is depreciation plus other expenses, to total assets (OVER)
would seem to be a variable that bank managers would want to control. However, Bennaceur
and Goaied (2008) discovered that overhead expenses were significantly positively related to the
net interest margin, as well as to return on assets. They argued that overhead costs are apparently
are passed on to depositors and lenders. Reasons for this relationship are not entirely clear, but
these expenditures may reflect amenities and other services that bank customers want, even if
costs are higher.
Higher values of noninterest bearing assets as a fraction of total assets (NIBA), where
non-earning assets include cash, fixed assets, and amounts due, should be negatively correlated
with profitability. As expected, Bennaceur and Goaied (2008) found NIBA to be negatively
related to profitability. Staikouras, Mamatzakis, and Koutsomanoli-Filippaki (2008) examined a
similar ratio, the cash to assets ratio and determined that it also was negatively correlated with
profitability.
Our final efficiency variable is the labor cost to gross income ratio (LCI), which shows
the efficiency of management in controlling labor cost relative to total revenue. Normally, higher
ratios suggest inefficiencies and there should be a negative relationship between this variable and
profitability. However, in the MENA region, wages and salaries are a lower part of total bank
expenses than in other countries, so that higher numbers may suggest better utilization of a
relatively less expensive input in production.


15
4.3 Internal measures of risk
Default risk (RISK) is calculated by dividing total debt by total assets. As suggested by
Vander Vennet (2002) and Altunbus et.al. (2001), this variable also indicates differences in risk
preferences between banks. Generally, a negative relationship between profitability and risk of
the bank is expected, although the two variables might be positively correlated a very low levels
of bank debt.
Credit risk (CRISK) is measured by the loan-loss provisions to loans ratio. Prior research
suggests that increased exposure to credit risk is normally associated with decreased bank
profitability. For example, Valverde and Fernandez (2007) found that the ratio of loan defaults
to total loans is a forward looking measure that is a significant negative determinant of bank
profitability. Kosmidou, Pasiouras, and Tsaklanganos (2007) use a similar ratio, the ratio of loss
provisions to net interest income as a measure of asset quality and obtain similar results.
Capital strength (CSTR) is defined as equity divided by total assets. It measures capital
adequacy and higher ratios show better capitalized bank having lower leverage and lower risk.
Many prior studies going back to Berger and Mester (1997) have found a positive relationship
between profitability and capital strength. Nevertheless, since equity is such a high cost source
of bank funds, and if a bank is already adequately capitalized or even over-capitalized, the impact
on return of greater equity may be negative. Such results have been shown by Kosmidou,
Pasiouras, and Tsaklanganos (2007) and Staikouras, Mamatzakis, and Koutsomanoli-Filippaki
(2008).

4.4 External explanatory variables


16
Change in gross domestic product (CGDP) is used a proxy for cyclical output. Prior
studies contend that GDP growth has a major impact on banks profitability over the business
cycle. For example, a cyclical downswing could reduce lending activities of banks, which in turn
reduces reduce interest income. High economic growth, in contrast, leads to more bank deposits
and more lending activity, which should increase bank profits.
Inflation (IFL) has generally been found to negatively impact bank profits. For example,
Boyd et al. (2001) find evidence of a strong negative correlation between inflation and lending
activity, perhaps because banks to no want loans repaid in inflated dollars. Bank profitability
ratios should be higher in countries with lower inflation rates.
Concentration (CONC) is the ratio of total assets of one bank to the total assets of all
banks in a country. Alternative measures of concentration, such as the deposits of a bank divided
by country total deposits of all banks in the country, or the ratio of a banks loans to total bank
loans may be equally good measures of concentration. However, all three measures are highly
correlatedmeaning that only one measure of concentration is needed to explain bank
profitability. Previous studies have found a positive relationship between profitability and
various measures concentration.
Finally, two dummy variables are introduced: GCC and TYPE. The GCC dummy is
equal to one if the bank is located in a GCC, or oil producing country. It is set equal to zero if a
bank is in a non-GCC country in the MENA region. TYPE is set equal to one for Islamic banks
and it is equal to zero for conventional banks.
4. Accounting-Based Examination of Profitability
Keeton and Matsunaga (1985) and Smirlock (1985), Islam (2003), Kosmidou (2008) have


17
argued that return on assets (ROA) is the most useful measure of profitability and bank
performance over time since assets have a direct affect on both components of profit--income
and expenses. Our data sample consists of 470 bank-years of data from 87 different banks for
eight different years from 2000 to 2007. Since the number of years of data varies by bank, our
sample is considered an unbalanced panel of cross sectional and time series observations.
Econometric estimation of the model to discover the determinants of ROA (as well as the
determinants of ROE, NIM, and NNIM) is performed using a generalized least squares panel
estimator. Although results are often similar, the panel estimator provides an efficiency gain
over least squares estimation of panel data because it uses more information. Following standard
practice in the literature, we then examine two types of panel models. The more complex fixed
effects model requires the estimation of a bank specific effectmeaning an individual intercept
for each of the 87 banks. The computationally simpler random effects model assumes
homogeneity between banks and only requires a single intercept term across all banks.
The basic framework for the panel models is:
Y
it
=
i
+ X
it
+
it
, (for i = 1,470 and t = 2000,2007) (1)
where Y
it
is the dependent variable (either ROA, ROE, NIM, or NNIM),
i
is the firm specific
intercept, X
it
is a vector of the (up to) 16 internal and external variables described in Table 2 that
affect each bank-year observation, is simplified notation for the (up to) 16 regression
coefficients for the variable X
it
, and
it
is the disturbance term that is assumed to be normally
distributed with a mean of zero. Since many of the internal and external variables are highly
correlated, only some subset of independent variables (label the number as k) will be significant


18
in determining the best models for each profitability ratios. The independent variables for each
profitability ratio are selected from an exhaustive search based on maximizing the F statistic of a
panel model using version 7.0 of the RATS statistical package.
The models that best explain the four profitability ratios are presented in Table 4, Panels
A D. The choice between a fixed and random effects model for each of the four profitability
ratios is made using a Hausman test. The null hypothesis for the Hausman test is that all of the
i

intercept terms are equal. Insignificant values of the Hausman test statistic, which is distributed
as chi-square with k degrees of freedom, imply failure to reject the simple random effects model.
In contrast, significant values for the Hausman statistic mean the more complex fixed effect
model should be used to examine bank profitability. In the economics and finance literature, in
general, the simpler random effects model is preferred in the majority of cases using panel data,
but certainly not for all data sets.
The first profitability ratio, ROA, is best described by a fixed effects panel model. The
simple random effects model is rejected at the 1% significance level. As expected, the loan
specialization ratio (LOANS) is positive and statistically significant. The overhead ratio (OVER)
also is significant and positive. This is somewhat unexpected, but Bennaceur and Goaied (2008)
reported a similar result for Tunisian banks. It probably shows that overhead expenses are being
used to generate profit. The inefficiency ratio (INEFF), as expected based on many previous
studies, is significantly negatively related to ROA. Capital strength (CAPSTR), as found in most
studies, is significantly positively correlated with ROA. Although somewhat surprising, inflation
(INFL) is positively correlated with ROA. However, strong significance for this variable
requires that the GCC dummy variable also be included. The GCC and TYPE dummies indicate


19
that banks in oil producing countries enjoy higher profitability and that Islamic banks have higher
return on assets than conventional banks. Remaining variables not selected in the final model
generally have expected signs, but are subsumed by included variablesoften because of
multicolinearity problems. For example, NIBA would be negatively statistically significant if
LOANS is removed.
ROE can be explained by a random effects panel model. The significant explanatory
variables INEFF, OVER, INFL, GCC, and TYPE have the same signs as in the ROA equation.
There is a positive relationship between ROE and SIZE, while CONC is negatively related to
ROE. Both the SIZE and CONC variables have the same sign as in the Bennaceur and Goaied
(2008) study. This result suggests that larger banks have higher ROE across all countries, but
that within a country that concentration or relative size may negatively impact profitability as
measured by ROE.
The remaining two ratios (NIM and NNIM) capture two generally complementary, but
occasionally conflicting, components of the first profitability ratio (ROA). NIM is best
represented by a fixed effects model. As expected, the variables LOANS, INEFF, OVER, and
CAPSTR have the same signs as in the previous ROA and ROE regressions. The TYPE dummy
variable is negativeshowing that Islamic banks rely less upon interest related revenues than
conventional banks. The variables LCI and CRISK are positive showing that a larger net interest
margin is needed to maintain profitability as labor expenses proportionately increase (LCI) and
when credit risk is deemed higher (CRISK). DEPLIAB is positively related to NIMmeaning
that banks more dependent upon deposits can obtain a higher net interest margin because
deposits are a low cost source of funds.


20
NNIM can be represented by a random effects model. INEFF and CAPSTR have the
same signs as in previous models. However LOANS and SECUR have a negative sign. LOANS
are positively related to the NIM, so loan specialization is negatively related to the noninterest
margin. A negative relationship between SECUR and profitability, in general, occurs because
banks that specialize more in securities have less loans as a percent of total assets. However, the
specific negative relationship between SECUR and NNIM may not be as easy to explain (but Im
Taiser can do it). Finally, TYPE is positive and statistically significant showing that Islamic
banks rely more upon noninterest sources of income.
5. Economic-Based Examination of Cost and Profit Efficiency
Following a long tradition in the banking literature, we initially adopt a translog flexible
functional form to estimate cost and profit functions for MENA banks. As in Berger and
Mester (1997), the intermediation approach is adopted so that assets on the bank balance sheet
are treated as outputs and liabilities and physical factors of production are treated as inputs. In
our study, banks are assumed to use the inputs x
1
= deposits, x
2
= labor, and x
3
= physical capital
to produce the outputs y
1
= net loans and y
2
= dollar value of securities and other earning assets.
In some formulations of cost and profit functions, the total dollar value of these two earning
assets is simply added together and called output (Y), where Y= y
1
+y
2
. Deposits are the sum of
all checking, savings, and time deposits at an institution as measured in U.S. dollars. Its unit
price (p
1
) is defined as interest expense divided by total deposits, while the share of the deposit
input in a banks total cost (s
1
) is interest expensed divided by total cost (where total cost is the
sum of interest expense plus other expenses). The labor input is measured by total personnel
expenditures. Labors share of total cost is s
2
= personnel expenditures/ total cost, and its price


21
(p
2
) is proxied by personnel expenditures/ total assets. This definition of price, also adopted by
Maudos et al (2002), can be used when data on the number of employees are not readily
available. Finally, physical capital is defined as expenditures on plant and equipment, as
measured by depreciation plus other expenses on the income statement. Capitals share of total
cost is s
3
= (depreciation + other expenses)/ total cost and its price is estimated by p
3
= other
noninterest (and non-personnel) expenses/ book value of fixed assets.
In addition to the rather standard inputs and outputs defined above, we follow Berger and
Mester (1997) and include the dollar value of financial equity capital for each bank as a quasi-
fixed netput quantity (E) that enters into the translog unit cost function.
The translog cost function estimates the natural log of total cost for each bank in each
year (with the bank and time subscripts deleted for notational convenience) as a function of the
natural logs of input prices, outputs, and netputs as follows:
3 3 3 2 2 2
0
1 1 1 1 1 1
1 1
ln ln ln ln ln ln ln
2 2
i i ij i j n n nm n m
i i j n n m
C p p p y y y o o | o o
= = = = = =
= + + + +



3 2 3
1 1 1
ln ln ln ln ln ln ln
in i n E EE Ei i
i n i
p y E E E E p m
= = =
+ + + +


2
1
ln ln
Ei n
n
E y u c
=
+ + +

. (2)
The Greek letters refer to parameters that will be estimated using nonlinear least squares
regression on the system of equations that include the cost function plus the share equations for
s
1
and s
2
as follows:


22

3 2
1 1 1 1 1 1
1 1
ln ln ln
j j n n E
j n
s p y E o | m c
= =
= + + + +

(3)

3 2
2 2 2 2 2 2
1 1
ln ln ln
j j n n E
j n
s p y E o | m c
= =
= + + + +

. (4)
Since the share equations sum to one, the third share equation (s
3
) for physical capital is omitted.
The share equations are included in the system of equations to improve efficiency of estimation.
The cost function can be estimated by itself, but since the share equations add no new
parameters not already included in the cost function, the additional information provided
improves the precision of parameter estimates. The terms ,
1
and
2
represent stochastic error
terms for each firm in each time period in the respective regression equations (2), (3) and (4),
while u is a nonnegative term measuring potential inefficiency. The term u in equation (2)
cannot be directly estimated from the system of three equations. Instead, the error term in
equation (2) for any firm k is actually u
k
+
k
. By estimating a stochastic efficient frontier instead
of some average cost function, one assumes that the best practice or lowest cost firm has zero
inefficiency, or that u
k
= 0. Stochastic errors average out to zero, so assume that
k
= 0 for a
typical bank and that u
k
> 0 for all but the best practice bank. A common measure of inefficiency
(IN) is the percentage difference in total cost to produce any level of output for a given bank (C
k
)
versus the minimum possible cost for the best practice bank (C
min
). Mathematically, percentage
inefficiency for any firm is expressed by
IN = 100 (C
k
/C
min
1). (5)
To obtain a determinate solution to the system of equations, some further restrictions are


23
commonly imposed on the estimation of the translog cost function. First,
3
1
1
i
i
o
=
=

ensures that
factor shares sum to one. Then, symmetry requires that
12 21
= and
ij ji
| | = for all ij. Finally,
linear homogeneity in input prices imposes the following restrictions:
3 3 3 3
1 1 1 1
0
ij ij in Ei
i j i i
| | m
= = = =
= = = =

. (6)
Note that the cost function in equation (2) is nonhomothetic, meaning that no restrictions are
imposed on the relationship between cost and outputs. This means that returns to scale can vary
with output level and that different factor proportions might be efficiently employed at different
output levels.
To measure overall returns to scale, the translog cost function can be differentiated with
respect to the outputs y
1
and y
2
. Assuming that the dollar value of loans and securities can
simply be added together, a measure of scale economies (SE) is

2 2 2 3 2 2
1 1 1 1 1 1
1
ln ln ln ln
2
n nm n m in i Ei
n n m i n n
SE y y p E o
= = = = = =
= + + +

. (7)
If SE<1, there are increasing returns to scale, which is alternatively referred to as economies of
scale. If SE>0, there exist decreasing returns to scale or diseconomies of scale. Finally, SE=1
means constant returns to scale and no economies nor diseconomies of scale. This, in a sense,
implies optimal bank size. Because the cost function is non-homothetic and the approximation is
around a point, the measured scaled economies may be different for banks much smaller or larger
than the average bank in the sample.
The translog function is said to be a flexible functional form and it is the most popular cost


24
function used to model cost and efficiency in banking. Hence, it is adopted for a first look at the
banking sector in the MENA countries. The major underlying assumption of the translog model
is that the underlying cost and production functions can be represented by a specific type of
logarithmic relationship between input prices, input quantities, output quantities and total cost of
production. The translog approximation has proven reasonable for a variety of cost, production,
and consumption data in many, but not in all studies. There is some controversy regarding
whether other functional forms significantly improve upon the translog approximation and that
issue is explored in Section 7.
The translog profit and alternative profit functions have been discussed and modeled in
Berger and Mester (1997), Maudos et al (2002), and in many other studies. Since data on output
prices are not available, the specific form adopted in most studies is to estimate the alternative
profit function. Now define = operating profit, which is net income minus provisions for loan
losses as defined in Maudos et al (2002). For the cost function of equation (2), we replace ln C
on the right-hand side of the equation with + , where is a positive number added to the profit
of the least profitable (most unprofitable) bank so that its operating profit equals zero. This
avoids the problem of trying to take a natural log of a negative number. Profit efficiency is
calculated in the similar manner to cost efficiency except that banks are compared against the
most profitable and not the minimum profit bank. Hence, equation (5) is modified as follows:
IN = 100 (1 -
k
/
max
). (8)
Maximum likelihood estimates of the parameters of the translog cost and alternative profit
functions are presented in Table 5. The log of the likelihood function for the estimated translog


25
cost function is 485.73 and 16 of the 21 estimated coefficients are significant at the 5% level.
Scale economies estimated using equation (7) are SE = 1.018 > 1, suggesting very minor
diseconomies of scale. The log of the likelihood function for the alternative profit function is
LLF = 206.29, which confirms the results of previous studies that the translog cost functions are
estimated with greater precision than the profit functions. However, the fit still appears to be
reasonable since 15 of 21 estimated coefficients are significant at the 5% level. Scale economies
as measured by the alternative profit function are SE = 0.9987 < 1, suggesting almost constant
returns to scale.
Another use of cost and profit functions, and perhaps the main objective of many economic
studies, is to determine substitution possibilities between the various factors of production. The
most common measure of input substitution possibilities between any two factors i and j is the
Allen elasticity of substitution (AES
ij
). The Allen cross-price is defined as AES
ij
=
ij
/S
i
S
j
for i
j and the Allen own-price elasticity is AES
ii
=
ii
/S
i
Si
j
-1/S
i
for all factors. The notation is the
same as in equations (2), (3), and (4) where as
ij
and
ii
are regression parameters and S
i
and S
j

are the factor shares in either the cost or profit function, depending upon which equation is being
estimated.
Based upon the cost function parameters presented in Table 5, the average Allen elasticities
of substitution across all years are presented in Table 6. The own-price elasticities are on average
all negative as required by economic theory, showing that an increase in factor price reduces
factor usage. For the first factor (deposits), AES
11
= -0.51, while AES
22
= -1.85 for labor and
AES
33
= -2.30 for fixed assets. These results show that deposits are least elastic factor in
production, which seems logical since deposits are needed to fund the purchase of securities or


26
the granting of loans. The cross-price elasticities are AES
12
=.36, AES
13
= .71 and AES
23
= .79
showing that factors are somewhat substitutable, but nevertheless inelastic.
The own-price Allen elasticities for the profit function are AES
11
= -0.67, AES
22
= -3.40, and
AES
33
= -2.49. The cross-price elasticities are AES
12
= 0.84, AES
13
= 0.68, and AES
23
= 0.93.
These again show inelastic factor demand, but slightly greater substitution possibilities than
suggested by the cost function.
6. A comparison of the translog and minflex Laurent functional forms
The translog approximation to the true cost function is only locally flexible and Berger and
Mester (1997) note that the potentially globally flexible Fourier cost function may be preferable
when there is a wide range of bank sizes across a data sample. Another globally regular
functional form is Barnett, Geweke and Wolfes (1991) asymptotically ideal model (AIM). It has
recently been implemented by Fisher, Fleissing, and Serletis (2001) and Feng and Serletis (2008)
to model a variety of consumption and cost data. While these globally regular approximations
may hold promise for improved modeling of efficiency in the banking industry, they are beyond
the scope of this paper.
Globally regular forms may impose too rigid a structure on the cost and profit functions
particularly if some banks truly do not try to minimize costs or maximize profits. An alternative
is to examine other locally flexible functional forms such as the Generalized Leontief form
developed by Diewert (1971). As argued by Dumont (2006), the Generalized Leontief has good
regularity properties when the true elasticities of input substitution are near zero, while translog
is more appropriate when all elasticities of substitution are near one. Both the translog and


27
Generalized Leontief models are based upon Taylor series expansions. The various minflex
Laurent models developed by Barnett (1985) and Barnett, Lee, and Wolfe (1985) are based upon
the Laurent series expansion. Although a Taylor series expansion of some order can produce a
smaller remainder term than any other expansion series, this property need not hold for finite
orders of expansion, such as third order or less as would be required in empirical work in
economics and finance. Similarly, large orders of expansion, or many expansion terms may be
needed to accurately model data using a Fourier or a Muntz-Szatz expansion. For a second order
expansion series, which is most common in the literature, a Laurent expansion will generally
provide a smaller remainder term than a Taylor series expansion, at the cost of only adding a few
regression parameters. The regular region also will generally be larger than for functional forms
approximated by the taylor series expansion. The minflex Laurent functional form is a simplified
version of the complete Laurent expansion that can be used to encompass both the translog and
Generalized Leontief functional forms. The minflex Laurent-translog is estimated in logarithms
just like translog. The minflex larent-Generalized Leontief is an expansion in the square roots of
prices and other factors, just like the original Generalized Leontief functional form.
For our data set the minflex Laurent-translog generalization of the translog cost function of
equation (2) is expressed by






28
3 2 2 2 3 2
0
1 1 1 1 1 1
1
ln ln ln ln ln ln ln
2
i i n n nm n m in i n
i n n m i n
C p y y y p y o o o o
= = = = = =
= + + + +



3 2
1 1
ln ln ln ln ln ln ln
E EE Ei i Ei n
i n
E E E E p E y m
= =
+ + + +


3 3 3 2 3
1 1 1 1 1
1 1 1 1
2 ln ln ln ln ln ln
ij in Ei
i j i n i
i j i n i
b c d u
p p p y E p
c
= = = = =
+ +

. (9)
The share equations are represented by
3 2 3 2
1 1 1 1 1 1 1 1 1
1 1 1 1
ln ln ln ln ln ln
j j n n E j j n n E
j n j n
s p y E b p c y d E o | m c
= = = =
= + + + + + + +

(10)
3 2 3 2
2 2 2 2 2 2 2 2 2
1 1 1 1
ln ln ln ln ln ln
j j n n E j j n n E
j n j n
s p y E b p c y d E o | m c
= = = =
= + + + + + + +

. (11)

The minflex Laurent-Generalized Leontief cost function is represented by
C =
Since share equation form for the Generalized Leontief model is highly nonlinear, both it and the
minflex Laurent-Generalized Leontief model are estimated using the three factor input quantity
equations without including the cost equation. The values for the log of the likelihood function
and the Allen elasticities for all four functional forms are presented in Table 6.
7. Comparison of Accounting and Economic-Based Profitability Measures
In order to compare the economic and accounting based results, we take the efficiency measures
from equation (5) as measured by cost and profit efficiency using the translog functional form


29
and use these values as the dependent variable in a second stage panel regression of all of the
accounting variables used in Section 4. Results are shown in Table 7 and can be compared to the
panel regression results for ROA and ROE in Table 4.


8. Summary and Conclusion




30
References
Ahmed, A. M., & Khababa, N. (1999). Performance of the banking sector in Saudi Arabia.
Journal of Financial Management & Analysis, 12(2), 30-36.

Altunbas, Y., E.P.M. Gardener, P. Molyneux and B. Moore, (2001). Efficiency in European
banking. European Economic Review 45, 1931-1955.

Amato, Lousi., and Wilder, Ronald., (1985)"The Effect of Firm Size on Profit Rates in U.S.
Manufacturing," Southern Economic Journal, V. 52, July 1985.

Barros, Carlos Pestana; Ferreira, Candida; and Williams, Jonathan; (2007), Analysing the
determinants of performance of best and worst European banks: A mixed logit approach,
Journal of Banking & Finance, Vol. 31, pp. 2189-2203.

Boyd, J.H., R. Levine and B.D. Smith, (2001). The impact of inflation on financial sector
performance. Journal of Monetary Economics 47, 221-248.

Brock, P.L. and L.R. Suarez, (2000). Understanding the behavior of bank spreads in Latin
America. Journal of Development Economics 63, 113-134.

Caves, R., and Porter, M., (1977) "From Entry Barriers to Mobility Barriers: Conjectural
Decisions and Contrived Differences to New Competition," Quarterly Journal of Economics,
May.

Comanor, William S., and Thomas A. Wilson, (1967) "Advertising Market Structure and
Performance," The Review of Economics and Statistics, November.

Demirg-Kunt and Huizinga., (1999) Determinants of Commercial Bank Interest Margins and
Profitability: Some International Evidence World Bank Economic Review. V. 13: 379-408.

Essayyad, M., Madani, H. (2003). Investigating bank structure of an open petroleum economy:
The case of Saudi Arabia. Managerial Finance, 29(11), 73-92.

Islam, M. (2003). Development and performance of domestic and foreign banks in the GCC
countries. Managerial Finance, 29(2/3), 42-72.
Gale, B., and Branch, B., (1982) " Concentration vs. Market Share: Which Determines
Performance and Does it Matter?" Antitrust Bulletin, Spring.

Grant, R., Jammine, A., and Thomas, H., (1988) "Diversity, Diversification, and Profitability
among British Manufacturing Companies," Academy of Management Journal, V. 23.

Grinyer, P., Mckiernan, P., and Yasai-Ardekani, M., (1988)"Market, Managerial , and


31
Organizational correlates of Economic Performance in the U.K. electrical Engineering Industry,"
Strategic Management Journal, V. 9.

Hall, M., and Weiss. L., (1976) "Firm Size and Profitability," The Review of Economics and
Statistics, November.

Kosmidou, Kyriaki., (2008) The determinants of banks profits in Greece during the period of
EU financial integration, Managerial Finance, Vol. 34, No. 3, pp.146-159.

Kosmidou, Kyriaki; Pasiouras, Fotios; and Tsaklanganos, Angelos. (2007), Domestic and
multinational determinants of foreign bank profits: The case of Greek banks operating abroad
Journal of Multinational Financial Management, Vol. 17, No 1, pp. 1-15.

Lenz, R., (1981) "Determinants of Organizational Performance: An Interdisciplinary Review,"
Strategic Management Journal, V. 2.
Maudos, J. and J.F. de Guevara, (2004). Factors explaining the interest margin in the banking
sectors of the European Union. Journal of Banking and Finance 28, 22592281.

McDougall, F. and Round, D., (1984) "A Comparison of Diversifying and Nondiversifying
Australian Industries Firms," Academy of Management Journal, V. 27.

Mueller, William, F. and Richard, R., (1980) "The Role of Advertising Intensity and Diversity,"
Journal of Industrial Economics, April 1980.

Porter, M., (1979) "The Structure Within Industries and Companies Performance," The Review
of Economics and Statistics, May.

Ramakrishnan Ramanathan (2007), "Performance of banks in countries of the Gulf Cooperation
Council", International Journal of Productivity and Performance Management, Vol.56, No. 2,
pp137-154.

Ramanatham, Ramakrishnan. (2008), Performance of banks in countries of the Gulf
Cooperation Council, International Journal of Productivity and Performance Management, Vol.
56, No 2, pp137-154.

Ramanujam, V. and Varadarjan, P., (1989) "Research on Corporate Diversification: A
Synthesis," Strategic Management Journal, V. 10.

Ravenscraft, D., (1983) "Structure-Profit Relationships at the Line of Business and Industry
Level," Review of Economics and Statistics, V. 65, February.

Saunders, A. and L. Schumacher, (2000). The determinants of bank interest margins: An
international study. Journal of International Money and Finance 19, 813-832.


32

Shepherd, W., (1972) "Elements of Market structure: An Inter-Industry Analysis," Southern
Economics Journal, April.

Tarawneh, Mdhat., (2006), A Comparison of financial performance in the banking sector:
Some evidence from Omani commercial Banks International Research Journal of Finance and
Economics, Issue 3, pp. 101-112.

White, R., (1986) "Generic Business Strategies, Organizational Context and Performance: An
Empirical Investigation," Strategic Management Journal, V. 7.

Valverde ,Santiago Carbo., and Fernandez, Francisco Rodrguez., (2007) The determinants of
bank margins in European banking, Journal of Banking & Finance, Vol. 31, pp. 2043-2063.

Vander Vennet, R.,( 2002). Cost and profit efficiency of financial conglomerates and universal
banks in Europe. Journal of Money, Credit and Banking, 34, 254-282.


33
Table 1Description of Data Sample
Number of banks in sample by country and year, and by region and type


Panel A: Sample Distribution by Country and Year

Country Year
2000 2001 2002 2003 2004 2005 2006 2007 Totals
Bahrain 7 7 9 9 9 9 9 0 59
Egypt 2 3 5 7 7 7 4 0 35
Jordan 2 3 6 8 8 10 9 0 46
Kuwait 6 6 6 6 6 6 6 4 46
Lebanon 1 2 6 8 12 12 11 0 52
Morocco 0 0 0 0 6 6 5 0 17
Oman 3 3 4 4 4 4 4 4 30
Qatar 2 4 5 5 5 4 4 1 30
Saudi Arabia 7 8 9 10 9 9 8 7 67
Tunisia 0 0 1 2 2 4 3 0 12
United Arab Emirates 5 7 12 12 11 11 11 7 76

Totals 35 43 63 71 79 82 74 23 470


Panel B: Sample Distribution by Region and by Type of Bank


2000 2001 2002 2003 2004 2005 2006 2007 Totals
GCC (Oil) 30 35 45 46 44 43 42 23 308
Non-GCC (Non-Oil) 5 8 18 25 35 39 32 0 162
Totals 35 45 63 71 79 82 74 23 470

Islamic 10 13 16 17 16 15 14 8 109
Conventional 25 30 47 54 63 67 60 15 361
Totals 35 45 63 71 79 82 74 23 470


34
Table 2
Definitions of Variables


Bank Profitability Ratios
1. ROA = return on assets = net income / (average?) total assets.
2. ROE = return on equity = net income / stockholders equity.
3. NIM = net interest margin = (interest income income expense)/ total assets
4. NNIM = net noninterest margin = (noninterest income noninterest expense)/ total assets
Bank Internal Characteristics
5. SIZE = natural log of total assets
6. LOANS = loan specialization ratio = total net loans / total assets
7. SECUR = security specialization ratio = other interest bearing assets (non-loans)/ total assets
8. DEPLIAB = deposit specialization ratio = total deposits/ total liabilities
Bank Efficiency Measures
9. INEFF = inefficiency ratio = operating expenses/ gross income
10. OVER = ratio of overhead (depreciation plus other expenses) to total assets
11. NIBA = ratio of non-interest bearing assets to total assets, where non-interest bearing assets
include cash, fixed assets, and amount due
12. LCI = labor cost to income = personnel expenses/gross income
Bank Risk Measures
13. RISK = default risk as measured by the debt-equity ratio = long-term debt/total assets
14. CRISK = credit risk = loan loss provisions/ net loans
15. CAPSTR = capital strength = total equity/ total assets
External Variables
16. CGDP = year to year % change in country gross domestic product (GDP) deposits
17. INFL = annual country inflation rate in %
18. CONC = concentration ratio = ratio of a banks total assets to the total assets of all banks in our
sample for that country
19. GCC = dummy variable equal to one if the bank is in a GCC country, zero otherwise
20. TYPE = dummy variable equal to one if the bank is Islamic, zero for conventional banks
____________________________________________________________________________________

???Averages for any variable are the beginning of period value plus the end of period value
divided by 2. They are defined the same way for both conventional and Islamic banks.

Net income for Islamic banks is conventional net income before taxes, plus Zakat.

Interest income and expenses are replaced by commission income and expenses for Islamic
banks. Similarly, investments in Mudaraba, Murabaha, and Musharka are equivalent to loans
and advances.


35

Table 3
Descriptive Statistics for the Financial Ratios
(in %, except for SIZE which is log of total assets)



t-test for equality of means

Variable

All
banks
GCC
Non-
GCC
Islamic
Convent
ional
GCC vs.
Non-GCC
Conventional
vs. Islamic
ROA 1.92 2.37 1.12 2.77 1.67
ROE 14.99 17.12 11.19 20.32 13.38
NIM 2.35 2.51 2.05 2.55 2.28
NNIM 2.27 2.97 1.03 4.24 1.68
SIZE 15.64 15.90 15.19 16.59 15.36
LOANS 46.57 51.19 38.34 50.58 45.35
SECUR 28.33 26.98 30.34 34.23 26.55
DEPLIAB 75.52 73.19 79.67 79.66 74.27
INEFF 22.78 21.49 25.08 22.77 22.78
OVER 1.19 1.12 1.31 1.14 1.20
NIBA 25.10 21.84 30.93 15.18 28.10
LCI 9.94 9.91 10.00 10.37 9.81
RISK 21.11 22.90 17.93 17.43 22.22
CRISK 1.48 1.13 2.10 1.13 1.59
CAPSTR 12.89 14.39 10.23 14.52 12.40
CGDP 6.02 6.76 4.72 5.44 6.20
INFL 2.75 2.61 2.99 1.77 3.04
CONC 16.71 16.90 16.36 11.59 18.25
DMS 15.40 14.91 16.29 11.40 16.61
PWR 15.41 14.91 16.30 11.37 16.63

The t-test for equality of means is based on the mean for GCC minus non-GCC banks and for the
mean of Islamic minus conventional banks for each financial ratio. The test is calculated
assuming unequal sample variances.
* denotes significance at the 10% level
** denotes significance at the 5% level











36
Table 4
Panel Regressions for the Determinants of Profitability Ratios
(t-statistics are in parenthesis below each coefficient)

Independent Variable Profitability ratio
ROA ROE NIM NNIM
SIZE 0.006
(2.32)
LOANS 0.012 0.072 0.016 -0.064
(3.00) (3.27) (4.87) (-2.54)
SECUR -0.102
(-3.94)
DEPLIAB 0.014
(5.87)
INEFF -0.014 -0.167 -0.104 -0.159
(-2.15) (-4.14) (-13.86) (-6.21)
OVER 0.169 1.388 0.999
(2.93) (3.91) (18.55)
LCI 0.094
(6.41)
CRISK 0.034
(1.89)
CAPSTR 0.119 0.040 0.147
(12.29) (5.29) (3.16)
INFL 0.0001 0.004
(4.56) (3.12)
CONC -0.042
(-2.92)
GCC (1=GCC) 0.012 0.034
(3.60) (2.94)
TYPE (1=Islamic) 0.006 0.040 -0.005 0.033
(2.81) (3.56) (-2.97) (4.25)
Constant(s) Multiple .002 Multiple 0.091
(0.05) (4.61)

Hausman statistic 19.12 11.31 21.63 6.99
significance (.008) (.185) (.006) (.221)
Model selected Fixed Random Fixed Random
F statistic 8.78 6.76 16.46 2.28
Significance (.000) (.000) (.000) (.000)
Adjusted R
2
51.88% 44.39% 68.52% 14.64%


(All variables are significant at 5%, except for the coefficient on CRISK in the NIM regression. It is
significant at 6% level. The adjusted R
2
and F statistics for the random effects model are not reported in
RATS 7.0, so approximate values are reported using the fixed effects estimator.)


37
Table 5
Maximum Likelihood estimates of the translog cost and alternative profit functions

Regressors Cost Function Profit Function
Parameter Variable Coefficient t-statistic Coefficient t-statistic

0
constant

2.655 2.36 25.258 8.90

1
ln p
1
0.521 8.37 0.511 7.35

2
ln p
2
0.150 3.88 0.178 4.44

11
ln p
1
ln p
1
0.106 8.86 0.058 4.69

12
ln p
1
ln p
2
-0.069 -8.88 -0.017 4.64

22
ln p
2
ln p
2
0.078 9.18 0.020 2.22

1
ln y
1
0.551 3.20 -0.344 -0.79

2


ln y
2
0.427 2.32 0.041 0.08

11
ln y
1
ln y
1
0.202 4.10 -0.167 -1.86

12
ln y
1
ln y
2
-0.111 -3.52 -0.147 -2.54

22
ln y
2
ln y
2
0.010 6.57 0.018 0.66

11
ln p
1
ln y
1
-0.019 -1.58 -0.018 -1.57

21
ln p
2
ln y
1
-0.019 -2.54 -0.019 -2.57

12
ln p
1
ln y
2
0.055 7.30 0.060 8.36

22
ln p
2
ln y
2
-0.023 -4.96 -0.022 -5.26

E
ln E

-0.590 -1.83 -2.353 -9.43

EE


ln E ln E

0.130 1.23 -0.317 -1.84

E1
ln E ln y
1
-0.102 -1.43 0.360 2.87

E2
ln E ln y
2
0.015 0.48 0.140 2.15

E1
ln E ln p
1
-0.382 -2.81 -0.042 -2.94

E2
ln E ln p
2
0.051 5.66 0.047 5.13

LLF = 485.73 LLF = 206.29