IJBF7-marina.indd The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 51 THE INPUT REQUIREMENTS OF CONVENTIONAL AND SHARIAH- COMPLIANT BANKING1 Mariani Abdul Majid Universiti Kebangsaan Malaysia _____________________________________________________ Abstract Islamic banking activities are limited within the scope of shari’ah which is within the scope of socially responsible and ethical banking activities, different from that based on interest-based banking. This paper attempts to measure the input data required by shari’ah-compliant banking in comparison with conventional banking to estimate their relative effi ciencies and economies of and returns to scale. Cost and output distance functions were estimated for a sample of banks in 10 countries which operate both types of banking. The results showed that shari’ah-compliant banking has higher input requirements relative to interest- based banking, but exhibit superior average effi ciency only in Malaysia but inferior average effi ciency in cross-country analysis. There is little evidence of differences in economies/returns to scale between shari’ah and conventional banks. Keywords: Islamic Banking, Stochastic Frontier Analysis, Effi ciency, Frontier Analysis, Conventional Banks. JEL Classifi cation: G21, G28. _____________________________________________________ 1. Introduction The development of modern Islamic banking arose from adherents to Islam becoming conscious about rejection of the interest element in conventional banking. Islamic banks which started to operate in the early 1960s were initially concentrated in the Middle East before spreading to other regions such as Asia and Europe, due to demand mainly from the Muslim communities. Banks IJBF 1Note from editors: This paper is one of three best papers selected by a review panel of three professors at a Symposium held in November, 26-28, 2008 in Melbourne, Australia. The Symposium was funded by the Australian Research Council grant, 2007-2009/10, for research on Islamic Banking and Finance. ht tp :// ijb f.u um .e du .m y 52 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 started to offer this as a choice to bank customers. Although Islamic banking is sometimes perceived as a limiting choice, it is actually broadening the banking choice. Compared to conventional banking, Islamic banking activities are limited within the scope of shari’ah hence, the mechanism involved in Islamic banking is different from that on interest-based banking. This gives bank customers an alternative to interest-based banking. In addition, Islamic banking is not viewed as threatening the existing business. Instead, it opens opportunity for new business as its operation is within the scope of ethical banking activities (Wilson, 2007). Islamic banking services have now been offered by both full-fl edged Islamic banks, as well as conventional banks that choose to operate Islamic banking windows, and they can either be foreign- or domestic-owned fi rms. As Islamic banking has been in operation for over 47 years and is viewed as an alternative to interest-based banking, the performance of Islamic banking needs to be assessed. Moreover, as Islamic banking is part of a country’s banking system, the performance of Islamic banks may affect the soundness and stability of the banking system. Furthermore, Islamic banking infl uences the performance of conventional banks, if they choose to operate Islamic banking windows in addition to conventional windows. Hence, determination of the relative performance of Islamic and conventional banks will help policy makers to devise policies in order to improve the performance of a country’s banking system as well as to provide some guidelines for managers of conventional banks with Islamic banking windows to improve bank performance. In addition, the rising number of Islamic banks has increased the competition between full- fl edged Islamic banks and conventional banks. Therefore, the determination of their relative performances will encourage both full-fl edged Islamic and conventional bank managers to improve their performance in order to compete with each other. Given the above issues, the aim of this paper is to measure the effi ciency of Islamic banking and compare it with conventional banks, concentrating on the impact of operational characteristics. Specifi cally, the fi rst objective is to compare the effi ciency of Islamic banking relative to conventional banks in Malaysia, focussing on the impact of operating characteristics. The second objective of the paper is to compare the effi ciency of Islamic banks relative to conventional banks in countries operating Islamic banking. 2. Methodology A. The Econometric Specifi cation In achieving the fi rst objective, translog cost and output distance functions were applied to study the commercial banks to measure their effi ciency. The measured effi ciency of a fi rm is calculated as the difference between its observed ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 53 input and output levels and the corresponding optimal values (given a country’s fi tted frontier). Given that Islamic banks cannot charge or pay interest and are hence, likely to face higher capital costs2 and meet objectives other than profi t maximization, a cost function has been employed which allows the potential higher costs of capital faced by Islamic banks to be controlled. Furthermore, if the non-profi t oriented activities of Islamic banks are carefully controlled for, it is reasonable to assume that Islamic banks will try to minimize their costs of operation. In specifying the model, the intermediation approach, which has been widely employed in most bank studies (Brown and Skully, 2003; Hassan, 2003; Saaid, Saiful, Mansor and Naziruddin., 2003, Yudistira, 2004), and Islamic and conventional bank studies (e.g., Alshammari, 2003; El-Gamal and Inanoglu, 2005) was employed in this paper: see Ariff, Badar, Shamser and Taufi q. (2008). Given this discussion, stochastic frontier analysis (SFA) will be employed in order to estimate a total cost function for Malaysian commercial banks. A single-equation stochastic cost function model can be described as: InC n,t = (Y n,t ,W n,t ,Z n,t )+ n,t (1) where C n,t, ,Y n,t, , W n,t , and Z n,t are the observed total cost of production for the n-th fi rm at time t, a vector of outputs is Y n,t , an input price vector is W n,t , and an exogenous factor vector is Z n,t . The assumption of the composed error term is as below (Aigneret, Knox and Peter 1977),  n,t = v n,t + u n,t (2) where v n,t represents random uncontrollable error and is assumed to be normally distributed with zero mean and variance, and is drawn from a one-sided distribution that is assumed to capture ineffi ciency. Similar to many previous studies, u n,t is assumed to be drawn from a half-normal distribution with mean zero and variance (Berger and Mester, 1997, Mester, 1996). v n,t and u n,t are independently distributed. Given this assumption, the log likelihood for ineffi ciency is expressed in terms of the two variance parameters, which captures the variance of composed error and, which is a measure of the amount of variation originating from ineffi ciency relative to statistical noise (Jondrow, Knox Lovell, Ivan and Peter, 1982). 2 v . 0 ,u tn 222 uv 2 As they cannot issue or hold interest-bearing loans or securities but use alternative contract arrangements such as musharaka (Karim, 2001). However, as the available investment avenues using contracts are very limited, and most of them concentrate on short term investments, they may yield lower returns. In contrast, interest-based banks have wide choice of both short- and long-term investments thus potentially yield higher returns. ht tp :// ijb f.u um .e du .m y 54 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 Maximum-likelihood estimates (MLE) are obtained by estimating a translog cost function as below, after including environmental variables, imposing the standard assumption of homogeneity in input prices, and allowing for the composed error terms: (3) where, P k,n,t = W k,n,t W k,n,t and k=1,…,K, and s=1,…,K are indices for input prices; m=1,…M and j=1,…,M are indices for output prices; h=1,…,H is an index for environmental variables; while the Greek letters (except v and u) represent unknown parameters to be estimated. Output and input variables follows the existing literature (Allen and Rai, 1996; Casu and Girardone, 2002; Mester, 1996), are normalized around their means and the values are in real 2000 MYR. Total costs (C) are defi ned as operating and fi nancial costs and are calculated as the sum of labour expenses, physical capital expenses, and either income paid to depositors for Islamic banks or interest expense for conventional banks. Input prices W1, W2, and W3 are the price of labour, price of fi nancial capital and price of physical capital, respectively. W1 is labour expenses divided by the number of full time workers, W2 is the amount of income paid to depositors divided by total deposits, and W3 is the physical capital expenses divided by the fi xed assets. Bank outputs, are defi ned as the sum of total loans (Y1), and total other earning assets (Y2). The latter comprise deposits with other banks, securities and equity investments. Standard symmetry is imposed to the second order parameters: and  ks =  sk and  mj =  jm . Given the above model specifi cation and assumptions, a measure of cost effi ciency can be derived as the ratio of observed costs to predicted effi cient costs, which is theoretically equivalent to: CE n,t = exp(u n,t ) (4) These relative effi ciency measures range from one to infi nity with a score of one indicating full effi ciency. However, CE n,t relies on the unobservable ineffi ciency, u n,t . Following Jondrow, et al. (1982), the conditional expectation of u n,t given tnKtntn WCC ,,,, ~ 1 1 sk, 1 1 1 1k k PlnPlnPlnln tn,s,tn,k,.50tn,k,+, ~ K s K k K tnC M 1 ,,,,jm, M 1m M 1m ,,m lnlnlnY 0.5 j tnjtnmtnm YY 1 11 , 1 1 ,,,,,, lnlnln K k k M m mk K k tPYP tnktnmtnk tttY M m m tnm 2 111 1 5.0,,ln tntn H h tn,h,h uvZ ,, 1 ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 55 the observed value of the overall composed error term,  n,t , can be expressed as: (5) where,  is the standard normal cumulative distribution function and ø is the standard normal density function. Failure to account for differences between bank groups may yield inappropriate conclusions about bank performance (Bos and Kool, 2006). Therefore, differences in operating characteristics that may affect the effi cient level of costs or output in this paper have been controlled for, by including environmental factors directly in the function, hence the resulting effi ciency scores are net of the impact of environmental infl uences on effi cient input requirements. As a result, these effi ciency measures permit one to predict the ranking of fi rms under the assumption that fi rms operate in an equivalent environment. However, these exogenous factors are possibly an indicator of differences in effi ciency rather than differences in effi cient costs or outputs. The estimated economies of scale enable banks to identify potential costs savings if they change the operation scale which can be obtained by fi rst calculating the M output elasticities: (6) From which a scale elasticity can be calculated as: (7)  Scale,n,t >1 ,  Scale,n,t = 1,  Scale,n,t <1, indicates economies of scale, constant returns to scale and diseconomies of scale respectively. Banks that produce at constant returns to scale realise the lowest average costs in which any increase (decrease) in output will increase (decrease) costs proportionately. Cost effi ciency, as mentioned above, measures how effi cient banks minimise inputs, given outputs. As effi ciency estimates obtained from different approaches should generate consistent estimates of effi ciency and effi ciency rankings, as well as give consistent results over time, an alternative method employing an output-oriented distance function, which estimates how effi cient banks transform inputs into outputs, is also applied. By using this function, the paper has also the benefi t of employing a quantity measure to identify bank inputs and outputs, thus avoiding possible problems leading to distorted and inaccurate price estimates that might occur given divergences in asset classifi cation among Islamic and conventional banks, hence, potentially resulting in unreliable estimates of cost effi ciency. Moreover, this function does not call for the strong behavioural assumptions of a profi t maximisation tn tntn tn tn uE ,,, , , 2 11 tmmY C tnm K k tnkmktnj M j jmtnm tn PY 1 1 ,,,,, 1 ,,, , ~ lnlnln ln ,, M i tnm tnScale 1 ,, 1 ,, . ht tp :// ijb f.u um .e du .m y 56 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 or cost minimisation approach and is therefore appropriate for Islamic banks as they have dual objectives of fulfi lling non-profi t obligations for the society and profi t or revenue maximisation for the depositors and shareholders. Moreover, if behavioural objectives between Islamic and conventional banks differ, the weaker behavioural assumptions of the output distance function approach may allow more consistent estimates of relative effi ciency. A production technology that transforms inputs into outputs can be represented by the technology set, which is the technically feasible combination of inputs and outputs (Coelli, Prasada and George, 1998; Cuesta and Orea, 2002; Fare and Primont, 1957). If the vector of K inputs, indexed by k is denoted by X=(X 1 ,X 2 ,…,X K ) and the vector of M outputs, indexed by m , is denoted by Y=(Y 1 ,Y 2 ,…,Y M ), the technology set can be defi ned as: (8) where and are the sets of non-negative, real K and M-tuples respectively. For each input vector, X, let P(X) be the set of producible output vectors, Y, that are obtainable from the input vector X: P(X) = {Y: (X,Y) T)} (9) The output distance function can then be defi ned in terms of the output set, P(X) as: (10) The output distance function is defi ned as the maximum feasible expansion of the output vector given the input vector which is non-decreasing, positively linearly homogeneous and increasing in Y, and decreasing in X (Cuesta and Orea, 2002). Given an output distance function with two outputs and a given input vector, X, the production possibility set is the area bounded by the production possibility frontier (PPF), which indicates the maximum feasible output given X, and the Y 1 and Y 2 axes. If the output vector, Y, is an element of the feasible production set, P(X), Do(X,Y)≤1, fi rms which produce on the PPF, D 0 (X,Y) = =, thereby indicating technical effi ciency. In contrast, for a fi rm operating inside the PPF, D 0 (X,Y) = <1, thereby indicating the proportion by which output is below potential output. This also illustrates that Farrell is (1957) output-oriented measure of technical effi ciency, defi ned as the maximum producible radial expansion of the output vector, and can be represented as: OE 0 = 1/ D 0 (X,Y) (11) YproducecanXRYRXYXT MK ,,:), .)(:0.min, XP Y YXDo OB OA KR MR ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 57 OE 0 increases with ineffi ciency and lies between one and infi nity. If Y is located on the outer boundary of the production possibility set, OE 0 = 1, indicating effi ciency. On the other hand, if Y is in the interior of the production possibility set, OE 0 >1 indicating ineffi ciency. Following Fare and Primont (1957) and Cuesta and Orea (2002), and allowing for exogenous factors, the general form of a stochastic output distance function can be shown as follows: 1 = D 0 (Y n,t ,X n,t , Z n,t ,) h( n,t ) (12) where h( n,t ) = exp (u n,t + v n,t ), Y n,t is a vector of outputs, X n,t is an input vector, Z n,t is an exogenous factor vector and  is a vector of parameters. Ineffi ciency is accommodated in the specifi cation of h (.) as n,t is a composed error term comprised v n,t which represents random uncontrollable error that affects the n-th fi rm at time t, and u n,t , which is assumed to be attributable to technical ineffi ciency. In order to facilitate estimation, the author followed the standard practice of imposing homogeneity of degree one in outputs on the distance function, which implies that D 0 (Z,X, Y) = D 0 (Z,X,Y),  > 0. By arbitrarily choosing the M-th output, the author can then defi ne and write: (13) From Equation 13 and after assumingY* n,t = (Y 1,n,t /Y ,M,n,t ,Y 2,n,t /Y M,n,t , ...,Y M-1,n,t / Y M,n,t ). and rearranging terms yields the general form: (14) Finally after assuming the standard translog functional form3 to represent the technology, the output distance can be represented as: (15) MY 1 M o M o Y YXZD Y Y XZD ,, ,, tntntntno tnM hZXYD y ,,, * , ,, ,,, 1 3 In the literature, the translog function is preferred in estimating a parametric distance function because it is fl exible, easy to calculate and permits the imposition of homogeneity (Fuentes, Emili and Sergio, 2001). tnstnk K s sk K k tnm M m mtnk K k kotnM XXYXY ,,,, 1 , 1 * ,, 1 1 ,, 1 ,, lnln5.0lnlnln * ,,,, 1 1 , 1 * ,, * ,, 1 1 1 1 , lnlnlnln5.0 tnmtnk M m mk K k tnjtnm M m M j jm YXYY ht tp :// ijb f.u um .e du .m y 58 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 where, Y* m,n,t = Y m,n,t / Y M,n,t , k=1,2,..K and s=1,2,..K are indices for inputs; m=1,2,…M and j=1,2,..M are indices for output. The selection of the input and output variables follows the existing literature (Cuesta and Orea, 2002; Cuesta and Zofío, 2005; Iqbal, Kizhanathan, and Aigbe, 1999). The outputs Y 1 , Y 2 are loans and total other earning assets, and the inputs X 1 , X 2 , X 3 are labour deposits and capital (fi xed assets), respectively. X 1 is the number of full time workers, X 2 is total deposits including customer funding and short term funding, and X 3 is the total expenses on fi xed assets. It is noted that linear homogeneity in outputs is imposed using Y 2 as a numeraire and these variables have been mean-corrected prior to estimation. The approach of Jondrow, Lovell, Materov, and Schmidt (1982) is followed to derive the log likelihood which is expressed in terms of the two variance parameters, Following from Equation 12, and given current model assumptions, an estimate of output distance can be derived as D 0 (Y n,t ,X n,t ,Z n,t, ) = exp(–  . Equivalently an estimate of Farrell output oriented effi ciency is obtainable as: (16) However, OE n,t relies on the unobservable ineffi ciency, u n,t ,. Following Jondrow et al. (1982), the conditional expectation of u n,t given the observed value of overall composed error term,  n,t can be expressed as: (17) where, ,  (.) is the standard normal cumulative distribution function and  (.) is the standard normal density function. Given the estimated model, estimated scale elasticity can be calculated as the negative of the sum of the input elasticities (Cuesta and Orea, 2002): (18) SCALE n,t > 1, SCALE n,t < 1, and SCALE n,t = 1, are when a bank is operating with increasing returns to scale (IRS), decreasing returns to scale (DRS) and constant returns to scale (CRS) respectively. The second objective of how Islamic banks perform relative to conventional banks internationally is examined by employing a translog output distance function. The relative effi ciency and returns to scale of Islamic and conventional banks have been investigated in countries that operate Islamic banking namely Malaysia, Sudan, Bangladesh, Tunisia, Jordan, Lebanon, Yemen, Indonesia, Bahrain and Iran. Except for Sudan and Iran which only 222 uv and 222 / uvu . )exp( ,,, ,,, 1 0 , tntntn ZXYD OE tn A tn Atntn Atn Atn uE , ,, , , 1 21A K k tnk tnktnmo tn X XYD SCALE 1 ,, ,,,, , ln ,ln ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 59 operate Islamic banking, banks from other countries operate both Islamic and conventional banking. A common frontier with country-specifi c environmental variables is estimated after allowing for country specifi c differences in estimated ineffi ciency and the analysis puts emphasis on the impact of operating characteristics, including Islamic banking and country-specifi c conditions on the relative outputs of banks. The frontier is controlled for variations in economic and regulatory environments between countries that may justify differences in effi ciency, by including country-specifi c variables directly in the distance function, and also allowed country dummies to directly infl uence output ineffi ciency. These country dummy variables simultaneously capture other country-specifi c environmental conditions and determine relative effi ciency between countries. This implies that the resulting effi ciency scores are net of the impact of controlled for environmental infl uences on effi cient input requirement, and the differences in these scores are directly infl uenced by country-specifi c ineffi ciency distributions. As a result, these effi ciency measures enable one to determine how fi rms are ranked under the assumption that fi rms operate in an equivalent environment, while at the same time measuring how bank effi ciency in one country differs from another. Employing the output distance function approach, the cross-country analysis specifi es Battese and Coelli's (1995) truncated normal SFA model with the mean of the truncated normal distribution made an explicit function of country dummy variables. The illustration is in Equation 18 and the formulation of the model detailed in (Coelli, 1996) is followed. Previous studies that employ the intermediation approach found that equity is signifi cant in defi ning bank output but many (Girardone, Philip and Edward, 2004; Kasman and Yildirim, 2006) include it either as an environmental variable or a netput (fully interactive with input and output). Nevertheless, in fi nancing the operation of banks, equity capital is an alternative to deposits and inter-bank borrowings (Bonaccorsi di Patti and Hardy, 2005). Furthermore, Islamic banks that apply an equity participation principle rely heavily on their equity to fi nance loans (Metwally, 1997). Therefore, it is appropriate that equity is considered as part of bank inputs for studies employing the intermediation approach. The author therefore, included equity as an input, because of both its role in Islamic banking and because all banks can potentially raise funds to fi nance their loans through equity, rather than deposits. The specifi cation therefore extends the standard intermediation model by including two outputs, Y 1 , Y 2 representing loans and total other earning assets, and three inputs, X 1 , X 2 , X 3 representing total operating expense, deposits, measured by total deposits including customer funding and short term funding, and equity, measured by total equity. Thus, v n,t is assumed to be normally distributed with zero mean and variance and independently distributed of the u n,t , where u n,t ≥ 0 is assumed to be drawn from a truncation (at zero) of the normal distribution with mean, EM n,t and variance ,   is a parameter to be estimated, f=1,2...,F is an index for countries, and C is a country dummy. Hence, given the absence of a constant 2 v 2 u ht tp :// ijb f.u um .e du .m y 60 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 in Equation 18, each country  is estimated to have ineffi ciency drawn from a distribution with mean   , that is truncated at zero. The parameters in the translog function as defi ned in Equation 14, the composed error parameters , and the estimated means of the country specifi c ineffi ciency distributions (  ) specifi ed in Equation 18 are estimated simultaneously using maximum likelihood estimation (MLE) techniques. (18) Given the model assumptions and following Equation 18, an estimate of output distance can be derived as D 0 (Y n,t ,X n,t ,Z n,t ,  ) = exp() . Equivalently, an estimate of Farrell (1957)’s output oriented effi ciency is obtainable as: (19) OE n,t relies on the unobservable ineffi ciency, u n,t , hence following the approach of Battese and Coelli (1995) and Frame and Coelli (2001) to estimate the unobservable ineffi ciency, u n,t , the conditional expectation of u n,t given the observed value of the overall composed error term,  n,t , can be expressed as: (20) Following Cuesta and Orea (2002), returns of scale for the banks in the sample can be estimated using the estimated scale elasticity which is calculated as the negative of the sum of the input elasticities: (21) where if SCALE n,t > 1, and SCALE n,t < 1, SCALE n,t = 1, when a bank is operating at IRS, DRS and CRS respectively. B. Data and Input Variables For Malaysian banks, data on 33 banks were drawn from Bureau van Dijk’s (BvD’s) BankScope database for the period 1996-2002 and were verifi ed against the banks’ annual reports. The data are expressed in Malaysian Ringgit (MYR) and are adjusted for infl ation using the Malaysion GDP Defl ator, which was extracted from IMF (2004). The number of full-time workers and ownership information is taken from the Central Bank of Malaysia (2002) and Association of Banks in Malaysia (Various Years). As some banks have incomplete information, this has resulted in an unbalanced panel of 168 observations. tnff F f tn CEM ,, 1 , )exp( ,,, , ,,, 1 0 , tn tntntn ZXYD OE tn *,, **,, 2 *,, ,, 1 15.01exp exp tntn tntntntn tntn m mm uE K k tnk tnktnmo tn X XYD SCALE 1 ,, ,,,, , ln ,ln 222 uv and 222 / uvu ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 61 Mergers during the sample period have caused a marked reduction in the number of Malaysian commercial banks. Over this period, ten mergers and acquisitions took place: two in 1999, one in 2000, six in 2001 (involving 14 banks) and one in 2002. Given these trends, each pre-merger commercial bank is included as a separate bank and these banks are assumed to have merged into one of the pre- merger banks. For cross-country analysis, data on 23 Islamic and 88 conventional banks from 10 countries that operate Islamic banking were drawn from the BankScope database for the period 1996-2002 resulting in an unbalanced panel of 558 observations expressed in constant 2000 US dollars.4,5 C. Environmental environments Focussing on Malaysian banks while applying the cost function, the fi rst operating environment variable is an indicator of loan quality, and is proxied by the ratio of the non-performing loans (NPL)-to-total loans (Berger and Mester, 1997; Clark, 1996, Girardone et al., 2004; Mester, 1996; Williams and Nguyen, 2005). The second operating environment variable is measured by the equity-to- total assets ratio (Berger and Mester, 1997; Clark, 1996; Girardone, et al., 2004; Mester, 1996; Williams and Nguyen, 2005). The remaining environmental variables are dummy variables that are designed to capture potential differences in bank characteristics, and operating environment that may infl uence costs. The dummy variable indicating full- fl edged Islamic banks, is to control for the potential impact of full-fl edged Islamic banking on bank costs. As changes in bank scale should be captured through the impact of output growth on estimated costs, the impact of mergers will be net of the impact of changes in bank scale attributable to the merger. A dummy for observations in 1998 is included to control for the East Asian fi nancial crisis. The author considered including a foreign-owned dummy, for banks with more than 50% foreign ownership. However, while almost all domestic banks operate an IBS window relatively few foreign banks do. The author therefore, chose to interact a foreign dummy variable with a dummy variable for conventional banks that operate IBS windows and include the resulting set of dummy variables. Therefore, the model includes dummy variables for foreign banks without IBS, foreign banks with IBS, domestic banks with IBS, and leaves domestic banks 4In the estimation, all input and output variables were normalized around their means and the linear homogeneity in outputs was imposed using the output Y2 as a numeraire. 5All data employed in this analysis were converted into constant international dollars according to the purchasing power parity hypothesis (Lozano-Vivas, Jesus and Jose, 2002). ht tp :// ijb f.u um .e du .m y 62 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 without IBS as the base case measured in the constant.6 Finally, Z 9 provides a dummy variable indicating public ownership, and is expected to have a positive sign indicating higher costs.7 Turning to the operating environments for Malaysian banks employing output distance function, the fi rst operating environment variable is loan quality, the dummy variable indicating full-fl edged Islamic banks is to control for the potential impact of full-fl edged Islamic banking on bank output. The model also includes a dummy variable for foreign banks, foreign banks with IBS and all banks with IBS, leaving conventional domestic banks without IBS as the base case measured in the constant, where banks with IBS are conventional banks offering Islamic banking products through a separate Islamic banking window. A dummy variable for observations in 1998 is included to control for the East Asian fi nancial crisis. Finally, given that some banks have gone through mergers, one can control for this effect by using a merger dummy variable. The author also tests for the potential effects of individual mergers, fi nding that the dummy is signifi cant for three individual mergers, namely merger 1, merger 2 and merger 3.8 For cross-country analysis, in order to identify a common frontier, variables describing distinctive features of the economy, the banking industry as well as the geography of each country were identifi ed. These variables are grouped into three categories. The fi rst category includes macroeconomic conditions, and consists of a measure of population density, per capita income, density of demand (deposits per kilometer squared) and real GDP growth. These indicators explain the macro conditions under which banks operate. Population density is measured by the ratio of inhabitants per square kilometre, and it is expected that with high population density, the retail distribution of banking services becomes less costly. High per capita income, measured by Gross National Income (GNI) per inhabitant, is usually associated with countries having a mature banking environment, and thus, competitive interest rates and profi t margins which lower banking costs and increase bank outputs. Density of demand is measured as total deposits per square kilometre. A less concentrated demand for banking services is costly because demand is more dispersed. As a result, bank customers are less 6As all Islamic banks in the sample are domestically owned, and by defi nition are not conventional banks, the impact of Islamic banking on costs is also relative to the base case of a domestic bank that does not operate IBS. 7Publicly-owned banks are defi ned as banks with more than 50 percent government ownership through its agencies such as the Employees Provident Fund (EPF) and Permodalan Nasional Berhad (PNB). By defi nition, no foreign banks are included in the publicly owned category. 8Merger 1, 2, 3 refer to mergers between Oriental Bank and EON Bank, between Chung Khiaw Bank and UOB Bank, and between International Bank Malaysia, Sabah Bank and Multi-Purpose Bank respectively. ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 63 informed and banks tend to achieve lower output.9 Finally, real GDP growth is expected to increase bank outputs due to increasing economic activities. The second group of environmental variables identifi es differences in banking structure and therefore provides measures of both banking concentration and the intermediation ratio. The concentration ratio is defi ned as the ratio of the total assets of the fi rst three largest banks in a country to total banking assets. Higher concentration may be associated with higher or lower output. If higher concentration of banks is a result of market power, then the banks may become ineffi cient in producing outputs (Leibenstein, 1966). On the other hand, if higher concentration is a result of effi ciency, then bank costs are reduced and bank outputs increase (Demsetz, 1973). In order to control for differences in regulation or allow factors that may affect the ability to convert deposits to loans among banking industries, the intermediation ratio, as measured by the loan-to- deposits ratio is employed. It is expected that the higher the intermediation ratio, the higher bank outputs will be. Thus, the fi rst two groups of variables follow closely those of Dietsch and Lozano-Vivas (2000), and Carvallo and Kasman (2005). The fi nal group of environmental variables includes proxies for accessibility of banking services. The proxy variables are roads paved and telephone lines per 100 inhabitants. Roads paved is the percentage of road being paved in total roads, and is expected to positively impact bank outputs. Finally, the author expects that easier access to telephone lines will also increase potential bank outputs. One fi nal control variable is a dummy variable indicating whether a bank is an Islamic bank. Inclusion of this variable allows the author to test whether full-fl edged Islamic banks have a different operating environment from conventional banks. Therefore, a dummy variable is included in the model to capture for this difference, but no a priori assumption is made due to mixed results in the literature on the direction of the infl uences of Islamic banking on ineffi ciency (Al-Jarrah and Molyneux, 2005; El-Gamal and Inanoglu, 2005; Mokhtar, Naziruddin and Syed M. Al-Habshi, 2006) and none has assumed Islamic banking to infl uence potential bank output. The author also noted that while this modelling assumption maintains the assumption that adherence to shariah causes a shift in potential output obtainable from given inputs, it could also be argued that any difference in output between conventional and Islamic banks is evidence of differences in effi ciency. However, the author adopts this approach because it is believed that the restrictions imposed by shariah require Islamic banks to operate a modifi ed banking technology that is not equivalent to that of conventional banks. 9Countries with population concentrated in small habitable area(s) warrant careful judgement with regard to these results. ht tp :// ijb f.u um .e du .m y 64 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 3. Results and Interpretation A. Efi ciency Table 1 Maximum Likelihood Estimates for Parameters of the Environmental Factors:1996-2002 Coeffi cient Parameters Estimated value a Std. Error A. Costs function for Malaysian banks ζ 1 Loan Quality 0.309*** 0.103 ζ 2 Equity/Asset Ratio -0.736*** 0.229 ζ 3 Islamic Bank 0.150*** 0.041 Ζ 4 Foreign without IBS -0.218*** 0.028 ζ 5 Financial Crisis -0.048** 0.023 ζ 6 Merged Bank 0.108*** 0.026  Lambda 1.501*** 0.439 Σ Sigma 0.096*** 0.014 B. Output distance function for Malaysian banks ζ 1 Loan Quality 0.380*** 0.048 ζ 2 Islamic Bank 0.066*** 0.021 ζ 3 Foreign Owned Bank -0.140*** 0.027 ζ 4 Foreign with IBS 0.118*** 0.031 ζ 5 Financial Crisis -0.027*** 0.012 ζ 6 Merged Bank 1 0.083*** 0.035 ζ 7 Merged Bank 2 0.097*** 0.034 ζ 8 Merged Bank 3 0.063* 0.038 Σ 2 Sigma-squared 0.005 0.001 Γ Gamma 0.826*** 0.143 (continued) ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 65 For Malaysian banks employing cost function, recalling that = u /  v , Table 1 shows the highly signifi cant estimate of 1.501 implies that estimated deviation from the frontier is due mainly to ineffi ciency rather than statistical noise. Loan quality (Z 1 ) is positive as predicted and indicates that the lower output quality (higher the NPL-to-loan ratio), the higher the cost incurred by banks, which may refl ect higher monitoring costs. The equity-to-asset ratio (Z 2 ) has a negative relationship with costs, indicating that as the equity-to-asset ratio increases, costs are lower relative to those banks that depend more on deposits. The positive coeffi cient for the Islamic bank dummy (Z 3 ) indicates that full-fl edged Islamic banks are found to have costs that ceteris paribus are 15.0% higher than for other banks. This may result from constrained opportunities in terms of investments and limited expertise in Islamic banking. Merged banks Coeffi cient Parameters Estimated value a Std. Error C. Output distance function for banks internationally ζ 1 Islamic Bank 0.141*** 0.022 ζ 2 Density of Population 2.82x10-4*** 7.83x10-5 ζ 3 Density of Demand -0.035*** 0.008 ζ 4 Telephone lines 0.015*** 0.003 δ 1 Malaysia -0.541*** 0.096 δ 2 Sudan 0.537*** 0.082 δ 3 Bangladesh -0.366*** 0.097 δ 4 Tunisia 0.210*** 0.047 δ 5 Jordan -0.047 0.095 δ 6 Lebanon 0.112*** 0.041 δ 7 Yemen 0.412*** 0.083 δ 8 Indonesia 0.212*** 0.053 δ 9 Bahrain -0.353*** 0.121 δ 10 Iran -0.987* 0.555 σ2 Sigma-squared 0.029*** 0.002 γ Gamma 0.491*** 0.076 a *,**,*** Signifi cance at 90, 95 and 99% confi dence levels, respectively. Source: Extracted from (Abdul-Majid, 2008) Table 1 Maximum Likelihood Estimates for Parameters of the Environmental Factors:1996-2002 ht tp :// ijb f.u um .e du .m y 66 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 (Z 4 ) are found to have costs that are 10.8 percent higher, after controlling for other variables.10 The dummy variable for the fi nancial crisis (Z 5 ) is positive, indicating that costs fell by 4.8% in 1998 after controlling for other variables. Finally, foreign banks without IBS windows (Z 6 ) are found to have costs that are 21.8% lower than the combined group of all domestic banks, publicly owned banks, and foreign owned banks with IBS windows. Focussing on Table 2, the cost effi ciency of Malaysian commercial banks is on average 1.066, and ranges from 1.019 to 1.217. The yearly average as well as the range of the effi ciency scores has increased. The trend in effi ciency suggests a decline in average effi ciency over the sample period, but also the presence of a group of fi rms that were steadily slipping further away from the cost frontier. Thus, average effi ciency deteriorated from 1.064 in 1996 to 1.075 in 2002 and the maximum effi ciency score increased from 1.142 in 1996 to 1.206 in 2002. This may indicate that there are high gains achieved by best- practice banks but declines in effi ciency as other banks struggle to keep up with best practice. The effi ciency scores is judged against an effi cient frontier, which for example allows full-fl edged Islamic banks to have 15% higher costs and requires foreign banks without IBS windows to have 21.8% lower costs. These results can be compared to the previous literature: Islamic banks are found to have no difference with conventional banks in Malaysia (Abdul-Majid, Mariani and Fathin, 2005; Mokhtar, et al., 2006), but are equall if not more effi cient in Turkey (El-Gamal and Inanoglu, 2005), are more effi cient in Arabian countries (Al-Jarrah and Molyneux, 2005) and in GCC countries (Alshammari, 2003), when compared to conventional banks. These differences may potentially be due to the absence of environmental variables in some previous studies employing the intermediation approach, different input and output specifi cations, and cross-country differences in Islamic banking that may infl uence relative cost effi ciency.11 Concentrating on the estimated output distance function parameters for Malaysian banks as reported in Table 1B, recalling that , the highly signifi cant estimate of 0.826 for this parameter suggests that the portion of technical ineffi ciency in total variance is high. Thus, the estimated deviation from the frontier is mainly due to ineffi ciency rather than statistical noise. The estimated coeffi cients of all variables have the expected signs. Loan quality (1) is positive as predicted, and indicates that lower output quality (higher NPL- to-loan ratio) reduces output, thereby refl ecting the higher input requirement needed to monitor default loans. 222 / uvu 10Berger and Humphrey (1997) noted that some mergers improve cost effi ciency whereas others worsen their performance. Orea (2002) found that merged banks have negative effi ciency change in contrast to the unmerged banks in the initial period of merger activities. 11For example, Islamic banks in other countries may employ more equity-based fi nancing rather than debt-like fi nancing which is more common in Malaysia. ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 67 B. Malaysian banks using output distance function Descriptive Statistics: All Banks Average 1.042 1.061 1.054 1.052 1.060 1.050 1.060 1.055 Standard Deviation 0.023 0.027 0.034 0.026 0.052 0.037 0.044 0.036 Minimum 1.016 1.015 1.014 1.015 1.011 1.015 1.015 1.011 Maximum 1.104 1.109 1.161 1.123 1.220 1.144 1.211 1.220 Average effi ciency of conventional, conventional with IBS and Islamic banks All Banks 1.042 1.061 1.054 1.052 1.060 1.050 1.060 1.055 Without IBS 1.043 1.062 1.056 1.055 1.069 1.063 1.069 1.060 With IBS 1.041 1.060 1.055 1.052 1.054 1.041 1.052 1.052 Islamic 1.037 1.066 1.017 1.028 1.061 1.062 1.086 1.057 C. Banks in 10 countries using output distance function Descriptive Statistics: All Banks Average 1.087 1.106 1.102 1.106 1.102 1.120 1.112 1.105 Standard Deviation 0.121 0.158 0.173 0.159 0.151 0.173 0.167 0.158 Minimum 1.014 1.012 1.011 1.010 1.014 1.014 1.019 1.010 Maximum 1.756 1.949 2.352 1.918 1.743 1.882 2.114 2.352 Average Effi ciency of conventional and Islamic banks Conventional banks 1.076 1.076 1.081 1.081 1.076 1.096 1.094 1.082 Islamic banks 1.187 1.289 1.195 1.204 1.214 1.215 1.200 1.215 1996 1997 1998 1999 2000 2001 2002 All Years A. Malaysian banks using cost function Descriptive Statistics: All Banks Average 1.064 1.057 1.064 1.071 1.075 1.056 1.075 1.066 Standard Deviation 0.029 0.026 0.033 0.039 0.048 0.036 0.041 0.037 Minimum 1.033 1.022 1.025 1.026 1.02 1.019 1.024 1.019 Maximum 1.142 1.124 1.155 1.181 1.217 1.157 1.206 1.217 Average effi ciency of conventional, conventional with IBS and Islamic banks All Banks 1.064 1.057 1.064 1.071 1.075 1.056 1.075 1.066 Without IBS 1.071 1.057 1.066 1.082 1.078 1.057 1.083 1.071 With IBS 1.061 1.057 1.062 1.068 1.076 1.057 1.072 1.065 Islamic 1.058 1.056 1.072 1.061 1.062 1.042 1.059 1.057 Table 2 Effi ciency Estimates for Banks and by Types, 1996-2002 Source: Extracted from (Abdul-Majid, 2008). ht tp :// ijb f.u um .e du .m y 68 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 The positive estimate for 2 implies that full-fl edged Islamic banks are found to have outputs that ceteris paribus are 6.6% lower than other banks and this may be due to constrained opportunities in terms of investments and limited expertise in Islamic banking. The coeffi cient for foreign-owned banks is negative, indicating that output increases by 14.0% relative to domestic banks. However, foreign-owned banks with IBS (Z 4 ) are found to have potential output that is 11.8% lower than foreign banks without IBS. The coeffi cient for the fi nancial crisis dummy variable (Z 5 ) is negative, indicating that output increased by 2.7% in 1998 after controlling for other variables. This fi nding is consistent with the reactions of banks towards the fi nancial crisis, which was to lay off substantial number of workers and to cut other operating expenses. The individual mergers (Z 6 , Z 7 , Z 8 ) are found to be associated with output that is 8.3%, 9.7% and 6.3% lower respectively, after controlling for other variables. The effi ciency estimates of Malaysian commercial banks using output distance function as shown in Table 2B is on average 1.055, and ranges from 1.011 to 1.220, hence on average, banks only produce 94.8%12 of the output they could produce if they operated on the effi cient frontier. The effi ciency scores demonstrate that while there is little variation in the estimated effi ciency once differences in the environmental variables are controlled for. In other words, if effi ciency is judged against an effi cient frontier, which for example, allows full-fl edged Islamic banks to have 6.6 percent lower output, it should be expected that the resulting effi ciency scores exhibit small difference across bank types. The yearly average and the range of the effi ciency scores have risen and it implies a deteriorating in average effi ciency over the sample period, but also the existence of a group of banks that were steadily deviating from the output frontier. Hence, average effi ciency worsened from 1.042 in 1996 to 1.060 in 2002 and the maximum effi ciency score deteriorated from 1.104 in 1996 to 1.211 in 2002. The author fi nally compares the output distance function with the results the cost function in order to check the consistency of results. With the cost function approach, slightly higher average ineffi ciency estimates of 1.066 percent are found as compared to 1.055 when using an output distance function. On balance however, the author believes that an output distance function approach is a better method because the behavioural assumptions being made with the output distance function are less likely to create biases when jointly evaluating Islamic and conventional banks, and this approach also allows the author to avoid the further potential pitfall associated with price endogeneity. Concentrating on the estimated output distance function parameters for cross-country analysis as reported in Table 1C,13 recalling that, , the highly signifi cant estimate of 0.491 for this parameter, suggests that the 12OE=(1/ 1.055)100. 13The author notes that a log likelihood ratio test for the joint signifi cance of the 6 parameters related to equity is 17.98, thus the author can reject the null hypothesis that these parameters are jointly insignifi cant at the 99 percent confi dence level. 222 / uvu ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 69 14Bank specifi c loan quality and merger dummy variables were also found to be statistically insignifi cant when they were included in the distance function. 15The fi nding is consistent with cost function studies in which higher population density contributes to an increase in banking costs in France and Spain (Dietsch and Lozano- Vivas, 2000), and Latin American and Caribbean countries (Carvallo and Kasman, 2005). 16All countries in the sample are developing economies except for Bahrain (World Bank, 2007). estimated deviation from the frontier is equally due to both ineffi ciency and statistical noise. Besides the statistically signifi cant Islamic bank dummy variable, the only signifi cant country-specifi c environmental variables are density of population, density of demand, and telephone lines per 100 inhabitants. Many country-specifi c variables become insignifi cant when country dummy variables are included in the model, thereby suggesting that these factors serve as proxies for cross country differences in bank effi ciency, rather than legitimate determinants of potential output.14 The Islamic bank dummy (Z 1 ) has a positive coeffi cient, indicating that full-fl edged Islamic banks are found to have potential effi cient outputs that ceteris paribus are 14.1 % lower than other banks. Therefore the results suggest a systematic reduction in potential output that can be attributed to Islamic banking, which may result from constrained opportunities in terms of investments and limited expertise in Islamic banking. However, because the estimated model effectively assumes that the reduced outputs associated with Islamic banking result from legitimate differences in operating environment that reduce potential output, the effi ciency scores reported below for Islamic banks must be carefully interpreted as they net out this impact. In contrast to expectations, the sign of the coeffi cient of the population density variable (Z 2 ) is positive indicating that, ceteris paribus countries with high population density have lower bank output.15 A possible explanation for this fi nding is that in non-price bank competition, banks may open branches in large cities, in which real estate and labour costs are high, for strategic reasons, and thereby reduce their potential outputs (Dietsch and Lozano-Vivas, 2000). As expected, lower density of demand (Z 3 ), tends to increase expenses thereby, limiting potential output. The fi nding of reducing potential output is consistent with (Dietsch and Lozano-Vivas, 2000) and (Carvallo and Kasman, 2005), which found that lower density of demand raises bank costs, and hence reduces effi ciency. Finally, in contrast to the a priori assumption, the positive sign of telephone lines per 100 inhabitants’ variable (Z 4 ) indicates that greater availability of telephone lines decreases bank outputs. This is possibly because most countries in the sample are developing economies16 in which electronic communications including phone- and internet-banking are not fully developed. Hence, telephone usage may raise relative bank costs within the sample of countries. ht tp :// ijb f.u um .e du .m y 70 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 Table 1C demonstrates that the country dummy variables illustrate systematic and signifi cantly differences in the relative ineffi ciency of banks across countries. Thus, for example,  Jordan is found to be insignifi cantly different from zero, thereby suggesting that ineffi ciency for Jordanian banks is drawn from a standard half-normal distribution. However, banks in Malaysia, Bangladesh, Bahrain17 and Iran are found to have   <0 and hence, ineffi ciency in these countries is estimated as being drawn from truncated normal distributions with lower expected ineffi ciency than in a half normal distribution. In contrast, Sudan, Tunisia, Lebanon, Yemen, and Indonesia all have   >0, and hence are estimated to have higher expected ineffi ciency than that drawn from a half- normal distribution, with given variance . Furthermore, Table 1C suggests that while Iranian banks have on average the best output performance, Sudanese banks experience the worst output performance. This is consistent with two previous DEA studies, which fi nd that Iranian banks are among the most effi cient banks (Brown, 2003; Brown and Skully, 2003) and Sudanese banks are among the least effi cient banks (Brown, 2003).18 The parameters suggest a clear hierarchy of estimated effi ciency across countries, with higher indicating greater ineffi ciency. Table 2C reports the estimated effi ciency of all, conventional and Islamic banks on average for cross country analysis, respectively. The effi ciency of all banks is on average 1.105, and ranges from 1.010 to 2.352. The yearly average as well as the range of the average effi ciency scores, has only slightly increased over time. Thus, average effi ciency deteriorated from 1.087 in 1996 to 1.112 in 2002. The trend in both conventional and Islamic banks suggests only a slight decline in average effi ciency over the sample period. Hence, the conventional bank average effi ciency score increased from 1.076 in 1996 to 1.094 in 2002 and the Islamic bank average effi ciency score increased from 1.187 in 1996 to 1.200 in 2002. Across all countries, the average conventional and Islamic bank effi ciency measures are 1.082 and 1.215, respectively. This suggests that on average, even after having netting out the 14.1% lower output associated with Islamic banking, potential output of conventional banks is only 8.2 % higher than actual output, while for Islamic banks this difference is 21.5%. Sudan and Yemen, which have only Islamic banks in the sample, have extremely low average estimated effi ciency, even after netting out the impact of the statistically signifi cant environmental characteristics and Islamic banking. Put differently, while the results do clearly demonstrate a signifi cant 14.1% decrease in potential output attributable to Islamic banking, the further particularly poor performance 17Al-Jarrah and Molyneux (2005) also found that Bahrain is relatively effi cient when compared to Jordanian banks. 18 Even within Sudanese banks, wide ineffi ciency difference exists (Hussein, 2004). 2 u ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 71 of Islamic banks in Sudan and Yemen must be attributed to country specifi c banking ineffi ciency.19 The author fi nally emphasized that because the methodology assumes that differences in operating environment infl uence potential output rather than effi ciency, the resulting effi ciency estimates should in principle be interpreted as allowing for legitimate difference in potential output associated with compliance with shari'ah. Therefore, as argued by (Coelli, Sergio and Elliot, 1999), as this approach nets out the impact of operating environments, it provides a measure of managerial effi ciency. Thus, based on this argument, Islamic banks are substantially more effi cient in Tunisia and marginally more effi cient in Malaysia, but less effi cient in all other countries where both Islamic and conventional banks operate. However, this interpretation is dependent on the assumption that all of the reduced output of Islamic banks is attributable to differences in technology rather than systematically greater ineffi ciency amongst Islamic banks. These results can be compared to the previous literature that does not allow for exogenous variables in either the frontier or as an infl uence on ineffi ciency: Islamic banks are found to be no different with conventional banks in Malaysia (Abdul-Majid, Mariani, Nor Ghani and Fathin, 2005; Mokhtar, et al., 2006), and equally if not more cost effi cient in Turkey (El-Gamal and Inanoglu, 2005). Modelling for bank types of the Islamic bank, commercial, investment banks, country dummy, assets, liquidity, concentration ratio, and market share to directly infl uence ineffi ciency effects in Arabian countries, Islamic banks are found to be more cost effi cient (Al-Jarrah and Molyneux, 2005). Controlling for loan quality and capital in the function and modelling for bank type, country dummy, assets, liquidity, concentration ratio, and market share to directly infl uence ineffi ciency effects in Arabian countries using profi t function, Islamic banks are also more effi cient (Al-Jarrah and Molyneux, 2005). Alshammari (2003) also found relatively effi cient Islamic banks in GCC countries when loan quality and capital are included in the function, and bank type and country dummies are assumed to directly infl uence ineffi ciency. The differences in results may potentially be due to different environmental variables in the function, different input and output specifi cations, and cross-country differences in Islamic banking operation that may infl uence relative effi ciency.20 B. Returns to scale Table 3 provides fi rm specifi c scale economy estimates for all banks and by bank types using the three methodology. Using cost function, Table 3A demonstrates the range of the estimated scale economies is between 0.911 and 1.218 and this is consistent with the previous literature (Carvallo and Kasman, 2005; Orea, 2002). On average, these estimated scale economies have declined from 1.066 19Effi ciency estimates by country is available from the author by request. 20For example, Islamic banks in countries other than Malaysia may have a higher percentage of equity-based fi nancing which has been controlled for in this study. ht tp :// ijb f.u um .e du .m y 72 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 1996 1997 1998 1999 2000 2001 2002 All Years A. Malaysian banks using cost function Descriptive Statistics: All Banks Average 1.066 1.061 1.059 1.042 1.026 1.026 1.025 1.043 Standard Deviation 0.036 0.042 0.041 0.040 0.053 0.039 0.049 0.048 Minimum 0.990 0.973 0.965 0.944 0.925 0.936 0.911 0.911 Maximum 1.115 1.140 1.150 1.166 1.218 1.084 1.104 1.218 Average return to scale of conventional, conventional with IBS and Islamic banks All banks 1.066 1.061 1.059 1.042 1.026 1.026 1.025 1.043 Without IBS 1.070 1.080 1.073 1.054 1.032 1.013 1.015 1.045 With IBS 1.064 1.056 1.054 1.038 1.027 1.038 1.038 1.045 Islamic 1.051 1.045 1.056 1.023 0.992 0.992 0.980 1.010 B. Malaysian banks using output distance function Descriptive Statistics: All Banks Average 1.018 1.017 1.004 0.989 0.974 0.969 0.967 0.990 Standard Deviation 0.035 0.032 0.038 0.035 0.047 0.040 0.046 0.044 Minimum 0.943 0.945 0.912 0.894 0.869 0.880 0.856 0.856 Maximum 1.062 1.061 1.081 1.067 1.092 1.034 1.051 1.092 Average return to scale of conventional, conventional with IBS and Islamic banks All Banks 1.018 1.017 1.004 0.989 0.974 0.969 0.967 0.990 Without IBS 1.015 1.030 1.006 0.988 0.968 0.957 0.957 0.985 With IBS 1.020 1.012 1.003 0.990 0.978 0.977 0.975 0.993 Islamic 1.016 1.013 1.004 0.989 0.972 0.963 0.945 0.978 C. Banks in 10 countries using output distance function Descriptive statistics: All Banks Average 1.045 1.044 1.040 1.032 1.025 1.023 1.022 1.034 Standard Deviation 0.021 0.022 0.023 0.025 0.020 0.023 0.023 0.024 Minimum 0.989 0.995 0.996 0.945 0.983 0.984 0.981 0.945 Maximum 1.093 1.117 1.128 1.106 1.097 1.103 1.096 1.128 Average return to scale of conventional and Islamic banks Conventional banks 1.044 1.040 1.035 1.027 1.021 1.018 1.019 1.030 Islamic banks 1.061 1.066 1.065 1.054 1.040 1.040 1.036 1.052 Table 3 Return to Scale for all, Islamic and Conventional Banks Note: a: No mergers between Islamic banks have occurred during the sample period. If return to scale >,< or =1, there are increasing return to scale; decreasing return to scale or constant returns to scale respectively. Source: Extracted from (Abdul-Majid, 2008) ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 73 in 1996 to 1.025 in 2002. Similarly, within all of the bank types summarized, very moderate economies of scale and a slight downward trend in estimated scale economies is evident. Thus, there is little evidence for a difference in scale economies across the groups identifi ed in Table 3A. Moreover, even though full- fl edged Islamic banks are the only type with average economies of scale less than one in any year, this result is also consistent with the broader fi nding that most banks in the sample appear to operate at or near CRS.21 Table 3B illustrates the average estimated return to scale is 0.990 for Malaysian banks using output distance function, thereby indicating the presence of mild decreasing return to scale. The range of estimated returns to scale is between 0.856 and 1.092, and is consistent with the previous output-oriented literature (Cuesta and Orea, 2002). On average, this estimated scale elasticity has decreased from 1.018 in 1996 to 0.967 in 2002. Likewise, within all bank categories summarised in Table 3B, very mild decreasing returns to scale and a slight downward trend in estimates is observed. Thus, there is little evidence for a difference in returns to scale across the bank types identifi ed in Table 3B.22 Using both methods however, banks experience almost constant returns to scale. Table 3C provides fi rm specifi c returns to scale estimates cross –country analysis for all, conventional and Islamic banks on average. Estimated returns to scale averages 1.034 for all banks, ranges between 0.945 and 1.128, and is consistent with the previous literature (Abd Karim, 2001; Carvallo and Kasman, 2005; Cavallo and Rossi, 2001). On average, these estimated returns to scale have declined from 1.045 in 1996 to 1.022 in 2002. The average estimated returns to scale for conventional banks is lower (1.030) than for Islamic banks (1.052) and this applies to all countries except for Malaysia and Jordan. This suggests that generally a larger scale of operation will be useful if Islamic banks wish to eliminate disadvantages attributable to their relatively small size. However, there is little evidence of substantial returns to scale to be gained, nor is there substantial difference in potential returns to scale between conventional and Islamic banks.23 The trend for both conventional and Islamic banks also suggests a decline in average returns to scale over the sample period. Hence, conventional bank average returns to scale declined from 1.044 in 1996 to 1.019 in 2002 and Islamic bank average returns to scale declined from 1.061 in 1996 to 1.036 in 2002. Compared to other countries, Sudanese banks exhibit relatively strong returns to scale, which is consistent with the very small bank size in this country. This is consistent with Kasman (2005) who found economies of scale in small-sized banks in Poland and the Czech Republic. 21Yudistira (2004) found that small and medium-sized Islamic banks in most countries have diseconomies of scale but Alshammari (2003) found that bank type has no effect of economies of scale in GCC countries. 22Yudistira (2004) found that small and medium-sized Islamic banks in most countries have diseconomies of scale but Alshammari (2003) found that bank type has no effect of economies of scale in GCC countries. 23Alshammari (2003) found almost constant returns to scale in banks (including Islamic banks) in GCC countries and no difference across bank types. However, Yudistira (2004) found that small and medium-sized Islamic banks in most countries have diseconomies of scale. ht tp :// ijb f.u um .e du .m y 74 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 4. Conclusions The aim of this paper is to examine the effi ciency and economies of scale of Islamic banks relative to conventional banks using SFA. Operating characteristics such as shari’ah compliant banking could capture validated differences in costs or systematic differences in effi ciency. Similar to cost function approach, using output distance function on Malaysian banks, higher input requirements for full-fl edged Islamic banks relative to average banks have been found. In cross- country analysis, having netted out the 14.1 % lower output, the potential output of conventional banks is only 8.2 % higher than actual output, while for Islamic banks this difference is 21.5 %. However, as these effi ciency estimates are net of the measured effect of Islamic banking, the inferior average performance of Islamic banks must be in part attributed to the low country-specifi c effi ciency scores for certain countries. Furthermore, it has demonstrated that country effects play a signifi cant role in explaining effi ciency distributions between countries, even after controlling for country-specifi c environment conditions, including Islamic banking. The paper has however concluded that bank compliance with shari’ah which operates ethical banking activities has higher input requirements and it is possible that the reduced potential output is proof of systematic ineffi ciency. Although studies on Malaysian banks found banks to operate at almost constant return to scale, the cross-country analysis demonstrates that the average estimated returns to scale for conventional banks are lower than those for Islamic banks, with the exception of Malaysia and Jordan. Therefore, moderate benefi ts will be realized even if Islamic banks attempt to increase their scale size. Finally, the main conclusion derived from the paper, in which Islamic banks have relatively higher input requirements compared to conventional banks should however, motivate policy makers involved in Islamic banking and Islamic bank managers to identify and overcome factors leading to these higher input requirements. In addition, they should aggressively work to create a more encouraging banking environment for Islamic banking, if they plan to further expand Islamic banking. In order to expand our knowledge and understanding concerning the investigated issues, more research especially applying advanced techniques needs to be carried out in different time and country settings and the investigation should also extend to foreign-owned Islamic banks. Author Information: Mariani Abdul Majid is a teaching staff, School of Economics, Faculty of Economics and Management, Universiti Kebangsaan Malaysia, 43600 UKM Bangi, Selangor, Malaysia. Tel: +(60-3)- 8921-5780; Fax: +(60-3)-8921-5789. Email: mariani@ukm.my. ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 75 References Aigner, D. J., C. A. Knox Lovell, and Peter Schmidt (1977). Formulation and estimation of stochastic frontier production function models. Journal of Econometrics, 6, 21-37. Al-Jarrah, Idries, and Philip Molyneux, (2005). Effi ciency in Arabian Banking. In Munawar Iqbal and Rodney Wilson (Eds.), Islamic perspectives on wealth creation ( pp.97-117). Edinburgh: Edinburgh University Press. Allen, Linda, and Anoop Rai (1996). Operational effi ciency in banking: An international comparison. Journal of Banking & Finance, 20(4), 655-72. Alshammari, S.H., (2003). Structure-conduct-performance and effi ciency in Gulf Co-Operation Council. University of Wales, UK. Ariff, M., M Bader, Shamsher, M. and Taufi q Hassan (2008). Cost, revenue, and profi t effi ciency of Islamic versus conventional banks: International evidence using data envelopment analysis. Islamic Economic Studies, 15(2), 23-76. Association of Banks in Malaysia. ABM bankers directory. Kuala Lumpur, Various Years. Battese, G.E., and T.J. Coelli, (1995). A model for technical ineffi ciency effects in a stochastic frontier production function for panel data. Empirical Economics, 20( 2), 325. Berger, Allen N., and David B. Humphrey (1997). Effi ciency of fi nancial institutions: International survey and directions for future research. European Journal of Operational Research, 98, 175-212. Berger, Allen N., and Loretta J. Mester (1997). Inside the Black Box: What Explains Differences in the Effi ciencies of Financial Institutions? Journal of Banking and Finance, 21, 895-947. Bonaccorsi di Patti, Emilia, and Daniel C. Hardy (2005). Financial sector liberalization, bank privatization, and effi ciency: Evidence from Pakistan. Journal of Banking & Finance, 29(8-9),2381-406. Bos, J.W.B., and C.J.M. Kool, (2006). Bank effi ciency: The role of bank strategy and local market conditions. Journal of Banking & Finance, 30(7), 1953-74. Brown, Kym (2003). Islamic banking comparative analysis. The Arab Bank Review , 5( 2), 43-50. Brown, Kym, and Michael Skully (2003, September). A comparative analysis of Islamic bank effi ciency. Paper presented at the International Banking Conference, Prato, Italy. Carvallo, Oscar, and Adnan Kasman (2005). Cost effi ciency in the Latin American and Caribbean banking systems. Journal of International Financial Markets, Institutions and Money, 15(1), 55-72. Casu, Barbara, and Claudia Girardone (2002). A comparative study of the cost effi ciency of Italian bank conglomerates. Managerial Finance, 28(9), 3-24. Cavallo, Laura, and Stefania P S. Rossi (2001). Scale and scope economies in the European banking systems. Journal of Multinational Financial Management, 11( 4-5),515-31. ht tp :// ijb f.u um .e du .m y 76 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 Central Bank of Malaysia (2002). Annual Report. Malaysia. Clark, Jeffrey A. (1996). Economic cost, scale effi ciency, and competitive viability in banking. Journal of Money, Credit and Banking, 28(3), 342- 64. Coelli, Tim, (1996). A guide to Frontier 4.1: A computer program for stochastic frontier production and cost function estimation. Retrieved from http://www. uq.edu.au/economics/cepa/frontie and htm.www.uq.edu.au/economics/ cepa. Coelli, Tim, D. S. Prasada Rao, and George E. Battese (1998). An introduction to effi ciency and productivity analysis. London: Kluwer Academic Publishers. Coelli, Tim, Sergio Perelman, and Elliot Romano (1999). Accounting for environmental infl uences in Stochastic frontier models: With application to International Airlines. Journal of Productivity Analysis, 11 (3), 251-73. Cuesta, Rafael A., and José L. Zofío (2005). Hyperbolic effi ciency and parametric distance functions: With application to Spanish savings banks. Journal of Productivity Analysis, 24(1), 31. Cuesta, Rafael A., and Luis Orea (2002). Mergers and technical effi ciency in Spanish savings banks: A stochastic distance function approach. Journal of Banking and Finance, 26(12), 2231. Demsetz, H., (1973). Industry structure, market rivalry, and public policy. Journal of Law and Economics, 16, 1-9. Dietsch, Michel, and Ana Lozano-Vivas (2000). How the environment determines banking effi ciency: A comparison between French and Spanish industries. Journal of Banking and Finance, 24, 985-1004. El-Gamal, Mahmoud A., and Hulusi Inanoglu (2005). Ineffi ciency and heterogeneity in Turkish banking: 1990-2000. Journal of Applied Econometrics, 20(5), 641-65. Fare, Rolf, and Daniel Primont (1957). Multi-Output Production and Duality: Theory and Applications. London: Kluwer Academic Publishers. Farrell, M. J., (1957). The Measurement of Productive Effi ciency. Journal of the Royal Statistical Society, 120(Series A Part III), 253-81. Frame, W. Scott, and Tim J. Coelli (2001). U.S. Financial Services Consolidation: The case of Corporate Credit Unions. Review of Industrial Organization, 18(2), 229. Fuentes, Hugo J., Emili Grifell-Tatjé, and Sergio Perelman (2001). A parametric distance function approach for Malmquist productivity index estimation. Journal of Productivity Analysis, 15(2), 79. Girardone, Claudia, Philip Molyneux, and Edward P. M. Gardener (2004). Analysing the determinants of bank effi ciency: The case of Italian banks. Applied Economics, 36(3), 215-27. Hassan, M. Kabir (2003). Cost, profi t and X-effi ciency of Islamic banks in Pakistan, Iran and Sudan. Paper presented at International Conference on Islamic Banking: Risk Management, Regulation and Supervision. Jakarta, Indonesia. ht tp :// ijb f.u um .e du .m y The input requirements of conventional and shariah-compliant banking: 51-78 77 Hussein, Khaled A., (2004). Operational effi ciency in Islamic banking: The Sudanese experience. Islamic Research and Training Institute, Islamic Development Bank. Retrieved June, 30, from http:// www.irti.org/ Sudan%20-Banks.pdf. IMF (2004). International fi nancial statistics. Retrieved June, 9, from http:// imfstatistics.org/im/logon.aspx Iqbal, Zahid, Kizhanathan V Ramaswamy, and Aigbe Akhigbe (1999). The output effi ciency of minority-owned banks in the United States. International Review of Economics & Finance, 8(1), 105. Jondrow, James, C.A. Knox Lovell, Ivan S. Materov, and Peter Schmidt (1982). On the estimation of technical ineffi ciency in the Stochastic Frontier production function model. Journal of Econometrics, 19, 233-38. Karim, Rifaat Ahmed Abdel (2001). International accounting harmonization, Banking regulation, and Islamic banks. The International Journal of Accounting, 36(2), 169-93. Kasman, Adnan (2005). Effi ciency and scale economies in transition economies. Emerging Markets Finance & Trade, 41(2),60-81. Kasman, Adnan, and Canan Yildirim (2006). Cost and profi t effi ciencies in transition banking: The case of new Eu members. Applied Economics, 38(9), 1079. Leibenstein, H. (1966). Allocative effi ciency vs. X-effi ciency. American Economic Review, 56, 392-415. Lozano-Vivas, Ana, Jesus T Pastor, and Jose M Pastor (2002). An effi ciency comparison of European banking systems operating under different environmental conditions. Journal of Productivity Analysis, 18(1),59. Mariani Abdul-Majid, , David, S, and Giuliana , B. (2008). Effi ciency of Islamic and conventional banks. Aston Business School Working Paper Series RP0811. Aston University. Mariani Abdul-Majid, Nor Ghani, M. N. and Fathin, F. S. (2005). Effi ciency of Islamic banks in Malaysia. In Munawar Iqbal and Ausaf Ahmad (Eds.), Islamic Finance and Economic Development (pp.94-115). New York: Palgrave Macmillan. Mariani Abdul-Majid. (2008). Effi ciency of Islamic and conventional banks (Unpublished doctoral dissertation). Aston Business School. Aston University. Mester, Loretta J., (1996). A Study of bank effi ciency taking into account risk- preferences. Journal of Banking and Finance, 20, 1025-45. Metwally, M. M., (1997). Differences between the fi nancial characteristics of interest-free banks and conventional banks. European Business Review, 97(2). Mohd Zaini Abd Karim. (2001). Comparative bank effi ciency across selected Asean countries. ASEAN Economic Bulletin, 18(3), 289-304. Mokhtar, Hamim S. Ahmad, Naziruddin Abdullah, and Syed M. Al-Habshi (2006). Effi ciency of Islamic banking in Malaysia: A Stochastic Frontier approach. Journal of Economic Cooperation, 27(2),37-70. ht tp :// ijb f.u um .e du .m y 78 The International Journal of Banking and Finance, Vol. 7. Number 1: 2010: 51-78 Orea, Luis, (2002). Parametric Decomposition of a generalized Malmquist productivity index. Journal of Productivity Analysis, 18(1), 5-22. Saaid, Abd elrhman Elzahi, Saiful Azhar Rosly, Mansor H. Ibrahim, and Naziruddin Abdullah (2003). The X-effi ciency of the Sudanese Islamic banks. IIUM Journal of Economics and Management, 11(2), 123-41. Williams, Jonathan, and Nghia Nguyen (2005). Financial liberalisation, crisis, and restructuring: A comparative study of bank performance and bank governance in South East Asia. Journal of Banking and Finance, 29(8- 9), 2119-54. Wilson, Rodney (2007-8). Islamic fi nance in Europe. European University Institute. World Bank (2007). Country groups. Retrieved June 13, from http:// www. worldbanl.org/data/countryclass/classgroups.htm. Yudistira, Donsah (2004). Effi ciency in Islamic banking: An empirical analysis of eighteen banks. Islamic Economic Studies, 12(1). ht tp :// ijb f.u um .e du .m y