JBICI Working Paper Efficiency in the Pakistani Banking Industry: Empirical Evidence after the Structural Reform in the Late 1990s Atsushi Iimi NO. 8 December 2002 JBIC Institute (JBICI)
The JBICI Working Papers are based on the research done by staffs of Japan Bank for International Cooperation (JBIC) and published by the JBIC Institute. The views expressed in this paper are those of the author and do not necessarily represent the official position of JBIC.
Efficiency in the Pakistani Banking Industry: Empirical Evidence after the Structural Reform in the Late 1990s December 10, 2002 By Atsushi Iimi 1 Development Policy Research Division Japan Bank for International Cooperation (JBIC) Institute 4-1 Ohtemachi 1-Chome Chiyodaku, Tokyo 100-8144, JAPAN Tel: +81-3-5218-9744 Fax: +81-3-5218-9846 E-mail: a-iimi@jbic.go.jp 1 I thank Harumi Ito and seminar participants at JBICI for helpful comments. The paper is benefited from the data provision of the State Bank of Pakistan and many suggestions by the SBP staff. Any remaining errors are the responsibility of the author.
Abstract I examine the change in technical (in)efficiency of the Pakistani banking industry after the structural reform started in the late 1990s. With international assistance, the Pakistani government has undertaken the restructuring and preparation for privatization of national commercial and development banks, of which the main goal is the improvement of the efficiency in financial markets. Despite the small sample size, the estimated stochastic production frontier implies that both employees and branches are statistically productive in a concave fashion. This is an example counter to the common view that in a less developed banking industry, employees are too often idle and are not productive at all. It is also shown that the efficiency performance of the structural adjustment programs is in marked contrast among banks. Some banks were improving their technical efficiency during the reform period: 1997 to 2001, while the efficiency improvement of others was ambiguous. Key Words: Efficiency, Stochastic Frontier Analysis, Pakistani Banking Industry, Banking Sector Adjustment Loan. 2
1 Introduction The productivity in the banking industry has been of research interest for industrial economists, because the banking sector had been a typically regulated industry, but was becoming more deregulated in the 1990s. Thus, by comparing the estimated productivity before and after the structural reform including deregulation and privatization, it is possible to empirically examine the reform effects. This type of investigation into bank efficiency is quite important from the viewpoint of macroeconomics as well, since the adequate development of financial markets is essential for stabilizing the macro-economy and accelerating economic growth. In fact, there are many previous studies discussing the efficiency and economies of scale in the banking industry. For example, Berger, Leusner and Mingo (1997) investigate the branch efficiency of U.S. large commercial banks from 1989 to 1991, by separately estimating frontier-flexible and translog cost functions for several years. Their evidence shows that banks are likely to over-branch twice as many as the possible cost minimizing level, and technical inefficiency, namely X-inefficiency, amounts to about 20% of their operation costs. Berger and Hannan (1998) also in part examine the U.S. bank efficiency, concluding that the efficiency cost (i.e. X-inefficiency) resulting from a lack of market discipline is much larger than the deadweight welfare loss due to misallocation by monopoly power. 1 Battese, Heshmati and Hjalmarsson (1998) examine the efficiency of labor utilization in the Swedish banking industry, using the stochastic frontier analysis (SFA). Regressing the labor input on the outputs of financial services such as loans, guarantees, and 1 Moreover, Berger, Saunders, Scalise and Udell (1998) partly examine the bank efficiency change. Focusing on the great number of merger and acquisition (M&A) activities following the U.S. deregulation since the late 1970s, they estimate the impact of such bank M&A on small business lending. 3
deposits, and the quasi-fixed input such as branches, given one-sided stochastic inefficiency and idiosyncratic noise, they show that technical inefficiency of the banks in their use of labor is on the average 12% above the stochastic frontier. Further, the technical inefficiency increased immediately after the reform in the banking industry in the mid-1980s, and then has decreased due to the reform effect since 1991. Adams, Berger and Sickles (1999) perform a stochastic panel distance frontier estimation, using the data of over 2500 U.S. banks over 10 years. The estimation of the Cobb-Douglas production functions indicates that technical efficiency scores normalized by the most efficient bank are quite small and range from 53.5% to 54.3%. In this paper, I examine the change in technical (in)efficiency of the Pakistani banking industry, where the government has undertaken the restructuring and preparation for privatization of national commercial banks (NCBs) and development financial institutes (DFIs) since the late 1990s. Historically, the banking industry in Pakistan has been one of the most inefficient sectors for religious and political reasons. In the Islamic banking system, the exaction of debts might be difficult. Moreover, the Pakistani government has owned many public enterprises and large agriculture-based manufacturers, which are traditionally short of funds for new investment, operation and maintenance. Thus, the government s instruction to finance such state-owned enterprises (SOEs) may be politically justifiable for securing the basic life of the public, and it seems to be irresistible for the NCBs and DFIs. However, as a result of such discretionary governmental intervention, rather than market competition, the Pakistani banking industry has accumulated a considerable amount of non-performing loans, which approximately amount to a half of the total credit. 2 2 Inthesampleofthefive major state-owned banks from 1997 to 2001, the ratio of non-performing loans ranges from 16.7% to 87.8%. 4
Accordingly, the banking industry comes to play only a minor role in financial intermediation services. Even innovative business firms cannot have the access to bank finance, and no consumers are willing to deposit money in such inefficient banks. This inefficient financial system in Pakistan may be a crucial restriction on its economic growth. As Khan and Senhadji (2000) summarize in previous empirical studies, there is a statistically significant relationship between economic growth and financial development. Using two macro-variables: GDP per capita as a proxy of economic growth and the ratio of domestic credit to GDP as a common indicator for financial depth, the relationship in major Asian developing countries is illustrated in Figure 1. It implies that the Pakistani low economic growth is related to (but not necessarily caused by) its immature financial system. For further examination in the sense of macroeconomics, the development of financial depth in Pakistan is depicted in Figures 2 and 3. In the last two decades, there was no clear relationship between economic growth and financial development. This may be caused by a measurement error in which the domestic credit includes non-performing loans. Given these situations, the Pakistani government started the macro-economic and financial sector restructuring program under guidance of the International Monetary Fund (IMF) in 1996. The World Bank and Japanese government also co-financed the banking sector adjustment loan (BSAL) to support this Pakistani government s effort, of which the main goal is to improve the efficiency in financial markets through separating ownership and management, and strengthening the accountability mechanism. It was expected that the more efficient financial system would lead to mobilization of human resources in the banking industry, efficient allocation of domestic capital resources, and 5
improvement in the access of the poor people to financial services. All effects are conducive to stable economic growth and poverty reduction. The first phase program in 1997-1998 mainly consisted of: (i) preventing a further increase of bad loans, (ii) recovery of non-performing loans through the Incentive Scheme, invented by the State Bank of Pakistan (SBP), (iii) retrenchment of surplus staff through the Golden Handshake Scheme and closure of over-extended branches, (iv) preparation for privatizing three NCBs and two DFIs, (v) introduction of international accounting standards and strengthening prudential regulation, and (vi) establishment of banking courts for enhancement of the solution of disputes related to non-performing loans through legal procedures. Furthermore, the World Bank continued to support this structural adjustment in the banking sector, and approved the second phase credit in October 2001. The second phase program included: (i) restructuring of the cost structure, (ii) complete privatization of partially privatized banks, (iii) liberalization of bank branching policy, (iv) facilitation of loan collateral foreclosure to reduce the default cost, (v) reform of national savings schemes to integrate the financial market, (vi) discontinuance of the mandatory placement of foreign currency deposits, and (vii) strengthening the central bank to play amoreeffective role as a regulator of the banking sector. 3 Thus, the examination of the recent efficiency improvement in the Pakistani banking industry is directly related to the evaluation of these structural adjustment programs. Recall that the effect of the structural adjustment in the late 1990s on financial development and economic growth is difficult to capture on the macroeconomic level. (See 3 For the details of these programs, see Press Release: March 20, 1998, published by the Overseas Economic Cooperation Fund, Japan (currently Japan Bank of International Cooperation), and News Release No.98/1563SAS by the World Bank. Also see News Release No. 2002/113/SAS for the World Bank s second phase assistance. 6
Figures 2 and 3.) Instead, in this paper, using the micro-data of loans, labor and the number of branches, I examine the effectiveness of the programs by estimating a stochastic production frontier of the banking industry over the reform period: 1997 to 2000. Although the number of observations is quite limited, it is shown that some NCBs are continuously improving technical efficiency, and others not. This paper is organized as follows. In Section 2, I describe the data and econometric specification based on the standard SFA. In Section 3, I show the estimation results and discuss some implications and potential issues. 2 Data and Econometrics I use the panel data of loans, employment and the number of branches of the five major state-owned banks, Habib Bank Limited (HBL), United Bank Limited (UBL), National Bank of Pakistan (NBP), Industrial Development Bank of Pakistan (IDBP) and National Development Financial Corporation (NDFC), over the period: December 1997 to March 2001. 4 These banks are dominating the domestic banking services in Pakistan. I treat the amount of performing loans, which is defined by the total credit minus non-performing loans, as an output service of the banking sector. loans are one of the important products of the banking industry. Obviously, bank As Adams, Berger and Sickles (1999) point out, there is no clear consensus about what assets or liabilities constitute outputs or inputs in the banking industry. One possible idea is that demand deposits, time and savings deposits, real estimate loans, commercial loans, and 4 HBL, UBL and NBP are grouped in the NCBs and IDBP and NDFC belong to the DFIs. The State Bank of Pakistan mentioned that the DFIs are not functioning on commercial basis but playing specially assigned roles, such as financing priority sectors. Thus, as I will mention in the econometric part, their production function might be different than that of the national commercial banks. 7
installment loans are considered as outputs, while labor, physical capital and branches are inputs to produce the outputs. 5 When accounting for the fact that the Pakistani financial industry has started taking off, it is reasonable to assume that the increase in bank loans is a primary performance measurement of the financial depth. Of course, the more balance of total credit does not necessarily mean that the banks are better performing and providing the financial intermediation services more efficiently, since the total credit amount may include nonperforming loans. As I have mentioned, this does really matter, in the case of the Pakistani banking industry where the ratio of non-performing loans reaches 50% of total credit. Therefore, in order to partly account for this measurement error, I use the performing loan amount as an output proxy, deducting non-performing loans from the total credit. 6 For producing performing loans, I consider two necessary inputs: labor and branches. It is reasonable to assume that the capital requirement such as information systems does not yet to play a significant role in the Pakistani banking industry. In Table 1, I show the summary statistics of the five banks over the period: 1997 to 2001. Following Fan, Li and Weersink (1996), I estimate a simple stochastic production frontier with the two-part-composed-error term: one-sided stochastic technical inefficiency and two-sided idiosyncratic error. Letting the amount of performing loans of bank i at period t be Y it, and the corresponding inputs such as employees and branches 5 See Berger and Hannan (1998). 6 There are many alternative output and input variables. According to the SBP, the main activity of employees in the branch network in rural areas is to mobilize deposits, and thus the deposit amount may be one of reasonable output measures. Furthermore, one may consider the amount of newly loaned money to be a better output, which may be approximated by differentials of total credit. However, during the structural adjustment period, the total credit is greatly influenced by loan recovery, rather than new loans. My data do not allow for reducing the number of observations by taking the differentials. Therefore, in this paper, I use the gross amount of performing loans, but not new credit, as an output. 8
be X it, the production function can be written as: y it = g(x it )+² it for i =1,,N t=1,,t (1) where y it ln Y it, x it ln X it. For a functional form of production, I employ the Cobb-Douglas function: g(x it )=x 0 itβ (2) Although a balanced panel data is used in the current paper, I do not rely on a time-variant stochastic frontier analysis (SFA), in which the error term is assumed to be decomposed as follows. 7 ² it = v it u it (3) and u it = u i exp( η(t T )) (4) where u it is a non-negative random variable associated with technical inefficiency representing the deviation from the efficient production frontier, and v it is a statistic noise. Note that η is a parameter to be estimated, which will indicate a time-trend of the improvement of technical inefficiency over the period; if η is positive (negative), the technical inefficiency decreases (increases) over time, and if η is not statistically different from zero, the technical inefficiency tends to be unchanged over time. Despite Equation (4), in order to save the degree of freedom, relative to a limited number of observations in my data, I pool the panel data and make a direct assumption 7 For the details of the time-variant SFA, see Battese and Coelli (1992), Coelli (1996), Filippini (2001), and Battese, Rao and Walujadi (2001). In particular, Battese and Coelli (1992) provide a complete form of the log-likelihood function to be maximized, and Coelli (1996) proposes a computer program for that maximum likelihood estimation. 9
on inefficiency term, u j. I denote the pooled observation by j. u j is assumed to be independently and identically distributed (i.i.d.) according to a half normal distribution N(0, σ u ), andv j is i.i.d. according to a standard normal distribution N(0, σ v ). Thus, Ihave: y j = x 0 jβ u j + v j for j =1,,NT (5) As Weinstein (1964) proves, the probability density function of the sum of normal and truncated normal distributions, ² j = v j u j,is: f(²) = 2 σ φ ³ ² σ µ λ² 1 Φ σ for <²< (6) where φ( ) and Φ( ) are the normal probability and cumulative density functions respectively. Note that in Equation (6), I re-parameterize σ 2 = σ 2 u + σ 2 v and λ = σ u σ v, as usual. Recall that ² j = y j x 0 j β. Thus, as mentioned by Fan, Li and Weersink (1998), the log-likelihood function is: ln l(y λ, σ, β) = n 2 ln( 2 π ) n ln σ + nx j=1 µ λ²j ln 1 Φ 1 σ 2σ 2 nx ² 2 j (7) j=1 where n is the number of pooled observations. The firm-specific efficiency estimate can be provided by the conditional expectation of u j given ² j : E [u j ² j ]= " σλ λ² φ( j σ ) # 1+λ 2 1 Φ( λ² j σ ) λ² j σ (8) 10
3 Estimation Results In Table 2, I show the estimation results from maximizing the log-likelihood function given in Equation (7). There is a significantly positive productivity of employees and branches. Thus, although a serious concern in the Pakistani banking industry is that employees and branches are utilized in excess and are idle, the evidence does not indicate such a case. This evidence is plausible, when considering the fact that the Pakistani banking industry remains at the initial stage of financial development and thus continues to be a labor-intensive industry. Therefore, as more employees (or branches) are inputted, the output of the performing loan increases in a concave fashion. In the case of regressing on both labor and branches, the estimation significance is lost. This is mainly due to the small sample size. In addition, the SFA characteristic parameter λ is not significant at all (due to the small sample, again). The estimated λ and σ (i.e. σ 2 u + σ2 v ' 0.6 and σu σ v ' 0) imply that the technical inefficiency term in u j is almost negligible in the estimation. When σ u =0, the SFA model in Equations (1) through (3) becomes equivalent to the standard ordinary least squares (OLS) estimation. 8 Thus, y j = x 0 j β + ² j (9) In Table 2, I also show the estimation results from the OLS estimation. Not surprisingly, the OLS results are not different from the corresponding SFA estimates. In the case of the OLS estimation, the technical (in)efficiency may be defined by simple residuals e j,ols = y j x 0 j ˆβ, which possibly include not only technical (in)efficiency but also 8 One element of x j is assumed to be a constant term so that the inefficiency term u j can be included. 11
statistical noise. However, if Equation (9) is reinterpreted as another deterministic model, corrected OLS (COLS), which has the non-positive error restriction, the predicted residuals can be considered to be consistent estimates of technical inefficiency. In the COLS estimation, a log-linear production function is given by: y j = x 0 j β + ² j with ² j < 0 (10) Due to this non-positive disturbance restriction, E[²] < 0, only the constant included in x j becomes inconsistent. The COLS estimation in Equation (10) provides the same coefficients but constant as the above OLS in Equation (9). 9 The COLS residuals can be obtained by: e j,cols = e j,ols max e j,ols j Then, all the COLS residuals, e j,cols, satisfy the theoretical non-positive disturbance restriction and can predict technical inefficiency, rather than idiosyncratic noise. I depict the residuals in Figure 4, which shows that HBL, NBP and IDBP are improving the technical efficiency over the period, while the technical inefficiency of the other banks tends to be increasing or fluctuating. UBL deteriorated in efficiency at the initial stage, and then began improving since 1999. Therefore, significantly, the effects of the structural adjustment programs started in 1997 on the bank efficiency vary across banks. This may reflect the difference in the initial conditions of individual banks and 9 The consistency of the COLS estimation requires: plim max{² j} =0 Equivalently, given ² j = u j, is needed. plim min{u j} =0 12
their effort levels in reforming their internal organizations during the reform time. On the above estimation results, there are several limitations. First of all, the sample size is quite small, and thus the results may not be efficient. 10 This comes from the data availability, and more data would improve the estimation inefficiency. Secondly, there are alternative inputs and outputs of the banking industry, as I have mentioned. In addition, they may vary across banks. Related to the second point, in the above argument, I implicitly assume a single commonproductionfunctionforallbanks. 11 However, the production frontier may be different among banks, since banks usually differentiate their financial services and provide vertically different services to different types of firms. Some finance large manufacturers at a low interest rate, while others provide only small business loans at a relatively high interest rate. If this is the case, the labor (or branch) productivity may take different forms across banks. Nevertheless, the five banks sampled in the current paperseemtoengageinthesimilarfinancial services. Although it is not the case in this paper, one justification in economics against the single production frontier assumption is that a hull frontier in the long-run exists and can be estimated with many observations, even if firms adopted technologies are heterogeneous. Finally, the most serious limitation of this paper is this; all the estimation results are relative values within the Pakistani banking industry. If all the banks are equally inefficient, then the measured efficiency might be over-estimated. Thus, the estimation 10 This inefficiency will also cause a considerable amount of deviation of point estimates, e.g. in Figure 4 and the following Table 3. 11 The SBP mentioned that during the period, the production function might be significantly shifted by political shocks, e.g. the nuclear denotation in May 1998, resulting in the frozen foreign currency deposits, and the military takeover in October 1999. Moreover, the estimation results should be adjusted by seasonal factors and other institutional scheme, e.g. the export finance scheme. Nevertheless, the given data is not sufficient to account for these issues. 13
results should be interpreted more carefully; some of the Pakistani banks are more efficient than others, and all the Pakistani banks might operate far below the standard production frontier in the world. There are several ways of partly comparing the above estimated bank (in)efficiency in Pakistan with some earlier studies in other countries. One possibility may be based on the ratio of technical inefficiency in the overuse of labor to actual employment. 12 Using the COLS regression result on labor, the factor demand conditional on output level is given by: X j = h Y j exp( β b i 1 cβ 0 ) 1 where b β 0 and b β 1 are a constant term and the coefficient of labor in Equation (10). Note that under the COLS estimation, all observations must be below the production frontier, and thus all banks, except for the best-practicing firms, more or less overuse production factors such as labor. In Table 3, I calculate a point-estimate of the ratio of actual employment to the efficient factor demand in labor conditional on output level for each bank. The reason for some predictions significantly departing from the estimated frontier is that although non-performing loans may pop up at the time of re-evaluating bank assets, the increase or decrease in labor is quite gradual. In general, over-employment in the Pakistani banking sector has been significantly large. Technical inefficiency in their overuse of labor is on the average more than fivetimesasmuchaswouldbeefficient. The inefficiency of even the best-practicing firm observed in Pakistan, NBP, is on average 50.2% of the production frontier. In Table 3, it is shown that over-employment is 12 This ratio of overuse of labor is adopted by Battese, Heshmati and Hjalmarsson (1998), in which their estimate indicates that the Swedish banking industry involves on average 12% overuse of labor, relative to the stochastic frontier. 14
roughly declining over the reform period, implying that the targeted banks have mainly responded to the structural adjustment programs by staff retrenchment. Another possibility is to estimate the relative value of bank (in)efficiencies. 13 Suppose that given the COLS framework in Equation (10), technical efficiency is defined by: TE j =exp(² j,cols ) The ratio of the estimator of the technical efficiencies, exp(² j,cols )/ exp(² k,cols ),can be consistently estimated by exp(e j,ols )/ exp(e k,ols ), which is the same as the case of using the COLS residuals, since only the constant term is shifted. In Table 4, I show the estimated relative bank efficiency. The differential in technical efficiencies ranges from 51.9% to 79.5% and seems to be decreasing over time, implying that there appear to be striking differences between banks which undergo internal restructuring and banks which are failing to do so. 4 Conclusion In this paper, I examine the change in technical (in)efficiency of the Pakistani banking industry, by estimating a stochastic production frontier. As the banking industry is one of the most inefficient sectors in Pakistan, the government has undertaken the restructuring and preparation for privatization of the major state-owned banks since the late 1990s. The main goal is to improve the efficiency in financial markets through separating ownership and management, and strengthening the accountability mechanism. 13 This is corresponding to the estimation by Adam, Berger and Sickles (1999); their estimation indicates that the relative technical efficiency in the U.S. banking industry is about 53%, relative to the best-efficient production experience. 15
Since it is known that the financial development is correlated with economic growth, the international and bilateral organizations have aggressively supported the Pakistani government s reform efforts by providing the structural adjustment program loans. By using the stochastic frontier analysis and the COLS estimation, I found that employees and branches are significantly productive in a statistical sense. Thus, a serious concern that employees and branches are over-inputted and too often idle in the Pakistani banking industry may not be the case. As more employees (or branches) are inputted, the output of performing loans increases in a concave fashion. However, when taking a point-estimate of optimal factor demand conditional on output level, it is shown that the actual employment is far above the production frontier. Therefore, although employees are somewhat productive, the plausible situation is that too many employees are sharing a small amount of work with each other. This in part reflects the fact that the Pakistani banking industry remains quite labor intensive. Based on the OLS estimation, which provides statistically equivalent results to the SFA, it is also shown that HBL, NBP and IDBP are improving the technical efficiency overtheperiod,whilethetechnicalefficiency of UBL and NDFC is ambiguous. Thus, the efficiency performance of the former banks after the structural adjustment programs is in marked contrast to that of the latter. One policy implication of these estimates is that the Pakistani government is able to encourage relatively inefficient banks to engage more aggressively in restructuring their internal organizations and establishing an efficient corporate structure, since some banks are responding to the structural reform, and others not. Thepossiblewaystotheefficient production, of which the effect is directly shown in this paper, are certainly retrenchment of staff and closing of branches, as intended by 16
the programs. In addition, there exist other ways for banks to improve efficiency; they can provide job training for their employees to become more skilled in terms of financial knowledge. Furthermore, banks may be able to replace unskilled labor with information technological systems and skilled labor with advanced knowledge for operating them. They can also build up more effective internal incentive schemes for employees to improve their moral and encourage them to work harder, even without managers monitoring. Since the existing over-employment of the Pakistani state-owned banks partly results from the governmental employment measures, those strategies aiming at enhancing the existing labor abilities might be more effective in banks progressing restructuring, rather than the simple retrenchment of staff. Of course, since new internal training and incentive systems are difficult to introduce, one key reform is the replacement of top management with outsiders with sufficient experience and advanced knowledge, which constitutes the World Bank assisted programs. One remaining important thing is that the evidence in this paper does not imply the absolute efficiency improvement in the Pakistani banking industry. It is still possible that all the Pakistani banks would be operating inefficiently, compared with the average of the international efficiency level. Thus, the Pakistani government might have to continue restructuring the entire banking industry and engage in making financial development conducive to stable economic growth and poverty reduction. Related to this point, for future investigation following this paper, richer data would allow for estimating a production frontier more efficiently and providing more specific policy implications. If detailed data of the financial and managerial structures, particularly including foreign banks in Pakistan, are available, this would show whether domestic state-owned banks operate less efficiently than the international standard level, and 17
if so, what causes domestic bank inefficiency. Therefore, more concrete restructuring plans could be provided for the Pakistani banks. 18
References [1] Adams, Berger, and Sickles, R. 1999. Semiparametric Approaches to Stochastic Panel Frontiers with Applications in the Banking Industry. Journal of Business and Economic Statistics, 17, 349-358. [2] Battese, and Coelli, T.J. 1992. Frontier Production Functions, Technical Efficiency and Panel Data: With Application to Paddy Farmers in India. The Journal of Productivity Analysis, 3, 153-169. [3] Battese, Heshmati, and Hjalmarsson, L. 1998. Efficiency of Labour Use in the Swedish Banking Industry: A Stochastic Frontier Approach. CEPA Working Paper No. 6/98, Department of Econometrics, University of New England, Armidale. [4] Berger, Leusner, Mingo, J. 1997. The Efficiency of Bank Branches. Journal of Monetary Economics, 40, 141-162. [5] Berger, Saunders, Scalise, and Udell, G. 1998. The Effects of Bank Mergers and Acquisitions on Small Business Lending. Journal of Financial Economics, 50, 187-229. [6] Berger, and Hannan, T. 1998. The Efficiency Cost of Market Power in the Banking Industry: A Test of the Quiet Life and Related Hypotheses. Review of Economics and Statistics, 80, 454-465. [7] Coelli, T.J. 1996. A Guide to FRONTIER version 4.1: A Computer Program for Stochastic Frontier Production and Cost Function Estimation. Mimeo, Department of Econometrics, University of New England, Armidale. [8] Coelli, T.J. 1996. Measurement and Sources of Technical Efficiency in Australian Coal-fired Electricity Generation. CEPA Working Paper No. 1/96, Department of Econometrics, University of New England, Armidale. [9] Filippini, M. 2001. Economies of Scale in the Swiss Nursing Home Industry. Applied Economics Letters, 8, 43-46. [10] Filippini, Wild, and Kuenzle, M. 2001. Scale and Cost Efficiency in the Swiss Electricity Distribution Industry: Evidence from a Frontier Cost Approach. CEPE Working Paper No. 8. [11] Greene, W. 1997. Econometric Analysis 3rd Edition. Prentice-Hall, Inc. [12] Greene, W. 1997. Frontier Production Functions. In Pesaran, and Schmidt, P. eds., Handbook of Applied Econometrics, vol II: Microeconometrics. Blackwell Publishers, MA. [13] Hamermesh, D. 1986. The Demand for Labor in the Long Run. In Ashenfelter, and Layard, R. eds., Handbook of Labor Economics, vol I. Elsevier Science Publishers, BV. [14] Khan, and Senhadji, A. 2000. Financial Development and Economic Growth: An Overview. IMF Working Paper No. WP/00/209. 19
[15] La Polta, and López-de-Silanes. The Benefits of Privatization: Evidence form Mexico. NBER Working Paper No. 6215. [16] López-de-Silanes, Shleifer, and Vishny, R. Privatization in the United States. The Rand Journal of Economics, 28, 447-471. [17] Weinstein, M. The Sum of Values From a Normal and a Truncated Normal Distribution. Technometrics, 6, 104-105. 20
Table 1: Summary Statistics Variable Obs Mean Std. Dev. Min Max Performing Loan (mil. Rs.) 25 48219.1 42141.6 2711.0 112604.0 Employees 25 11012.2 9040.6 510.0 23599.0 Branches 25 908.4 768.9 14.0 1748.0 Note that performing loan is defined by the total credit minus non-performing loan amount. Table 2: Estimation Results of SFA and OLS SFA SFA SFA OLS OLS OLS ln (Employees) 0.5969 0.3262 0.5969 0.3262 (0.0725) (0.4403) (0.0756) (0.4694) ln (Branches) 0.4914 0.2262 0.4914 0.2262 (0.0599) (0.3629) (0.0625) (0.3869) constant 5.2047 7.4572 6.2084 5.2068 7.4587 6.2094 (3.2735) (3.9880) (4.9175) (0.6504) (0.3760) (1.8376) sigma 0.5932 0.5950 0.5886 (0.0846) (0.0847) (0.0836) lambda -0.0042-0.0031-0.0022 (6.7901) (8.3657) (9.8066) obs. 25 25 25 25 25 25 log likelihood -31.080-31.159-30.887 R-squared 0.731 0.729 0.735 The dependent variable is the logarithm of performing loan amount. Note that the standard errors are shown in parentheses. Table 3: Estimated Ratio of Overuse in Labor HBL UBL NBP IDBP NDFC Average 199712 2.505 4.731 2.986 4.631 2.976 3.566 199806 2.217 16.901 1.231 1.261 2.525 4.827 199812 2.014 34.410 1.186 1.124 9.945 9.736 199906 1.519 7.940 1.000 1.183 3.497 3.028 200003 1.395 5.348 1.106 1.304 22.805 6.392 Average 1.930 13.866 1.502 1.900 8.350 5.510 Note that the best-practicing firm observed in all samples is used as a baseline. The ratio of overuse in labor is defined by the rate of actual employment to predicted labor demand conditional on correspondent output level. 21
Table 4: Estimated Relative Bank Efficiency Mean Std. Dev. Min Max 199712 0.795 0.123 0.684 0.902 199806 0.637 0.321 0.209 0.985 199812 0.519 0.387 0.130 0.968 199906 0.612 0.281 0.290 0.905 200003 0.583 0.365 0.164 0.907 Note that the best-practicing firm observed in each period is used as a baseline. The relative efficiency is defined by exp(ej)/exp(ei). Figure 1: Financial Deepness and GDP per capita 3.00 Malaysia 2.50 Domestic Credit/GDP 2.00 1.50 1.00 China Thai India Philippines Indonesia Nepal Pakistan 0.50 Sri Lanka Kyrgyz Vietnam 0.00 Cambodia Lao PDR Kazakhstan 0 200 400 600 800 1000 1200 1400 1600 GDP per capita in USD 22
Figure 2: Trend of Financial Deepness in Pakistan 350 0.70 300 0.60 250 0.50 GDP per capita in USD 200 150 0.40 0.30 Domestic Credit/GDP 100 0.20 50 0.10 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 0.00 Figure 3: Cycle of Economic Growth and Financial Depth 0.59 0.57 1986 1991 0.55 Domestic Credit/GDP 0.53 0.51 0.49 1989 1994 1999 1997 1980 0.47 0.45 200 220 240 260 280 300 320 340 GDP per capita in USD 23
Table 4: COLS Residuals on ln(employees) 0 199712 199806 199812 199906 200003-0.5 Technical Inefficiency -1-1.5 HBL UBL NBP IDBP NDFC -2-2.5 24