Applied Economics, 2004, 36, 19151924

Financial liberalization and bank eciency: a comparative analysis of India and Pakistan
A L I A T A U L L A H * , T O N Y C O C K E R I L L and HA N G L E y Durham Business School, University of Durham, Mill Hill Lane, Durham DH1 3LB, UK and yDepartment of Economics and Politics, Nottingham Trent University, Nottingham, NG1 4BU

This paper provides a comparative analysis of the evolution of the technical eciency of commercial banks in India and Pakistan during 19881998, a period characterized by far-reaching changes in the banking industry brought about by nancial liberalization. Data Envelopment Analysis is applied to two alternative inputoutput specications to measure technical eciency, and to decompose technical eciency into its two components, pure technical eciency and scale eciency. The consistency of the estimated eciency scores are checked by examining their relationship with three traditional non-frontier measures of bank performance. In addition, the relationship between bank size and technical eciency is examined. It is found that the overall technical eciency of the banking industry of both countries improved gradually over the years, especially after 1995. Unlike public sector banks in India, public sector banks in Pakistan witnessed improvement in scale eciency only. It is also found that banks are relatively more ecient in generating earning assets than in generating income. This is attributed to the presence of high non-performing loans. In addition, it is found that the gap between the pure technical eciency of dierent size groups has declined over the years.

I . I N T R O D U C T IO N After decades of excessive government regulations and restrictions, the implementation of nancial liberalization has brought substantial changes in the banking sector of developing countries: The sector has become relatively less state-directed, more competitive, and open to foreign banks and non-bank nancial institutions.1 While considerable research has gone into the macroeconomic impacts of these changes, only a handful of studies have empirically examined the impact of nancial liberalization on the eciency of banks in developing countries. Until recently, the empirical studies on the eciency of banks have

primarily concentrated on the banking industry of developed countries, especially of the USA (see Berger and Humphrey, 1997; Isik and Hassan, 2003). This paper contributes to the burgeoning literature on the eciency of banks in developing countries by providing a comparative analysis of the evolution of the technical eciency of commercial banks in two South Asian economies, namely India and Pakistan, before and after the implementation of nancial liberalization in the early 1990s. Financial liberalization is an integral element of the ongoing Economic and Structural Reforms (ESRs) in India and Pakistan (see Ahluwalia, 1999; Zaidi, 1999). Financial liberalization includes, inter alia, a gradual deregulation

*Corresponding author. E-mail: ali.ataullah@durham.ac.uk 1 See Fry (1995) for changes in the nancial sector in developing countries after the implementation of nancial liberalization programmes.
Applied Economics ISSN 00036846 print/ISSN 14664283 online # 2004 Taylor & Francis Ltd http://www.tandf.co.uk/journals DOI: 10.1080/000368404200068638 1915

1916
of interest rates and state-directed credit policies, reduction in banks reserve requirements, entry of non-bank nancial institutions, reduced restrictions on entry and operations of private (domestic and foreign) banks, and privatization of public sector banks2 (see SBP, 2000; Arun and Turner, 2002). Prior to the nancial liberalization, the governments of India and Pakistan followed a policy of social control that emphasized controlling banks operations either through state directives or through nationalization. In India, the 14 largest commercial banks were nationalized in 1969, and six more banks in 1980. In Pakistan, all domestic private banks were nationalized and merged during the mid 1970s to form ve large public sector banks. By the late 1980s, these nationalized banks controlled more than 90% of the total deposits and the earning assets of the banking industry (see SBP, 2000; Arun and Turner, 2002). Operations of private sector banks, especially foreign banks, were restricted to a few large cities only. In addition, the governments stipulated lending targets to priority sectors (e.g. agriculture), imposed low ceilings on interest rates on loans and deposits, directed public sector banks to open branches in rural and semi-urban areas, and made it mandatory for banks to hold government securities in their asset portfolios in order to nance growing scal decits (see Sen and Vaidya, 1998; Zaidi, 1999). Under this policy of social control, the governments determined the direction and prices of nancial services provided by the banking industry; banks themselves had little control over their inputs and outputs. This policy of social control, though augmented deposit mobilization and provision of loans to the priority sectors, resulted in a deterioration of banks protability and capital base and in an unsustainable accumulation of non-performing loans in public sector banks asset portfolios. In both countries, on average the share of non-performing loans in total loans and advances in public sector banks was above 20% (see SBP, 2000; Arun and Turner, 2003). In this context, a key objective of the nancial liberalization of the early 1990s was to revive the banking industry by reducing government regulations and restrictions, underpinning on-site and o-site bank supervision, strengthening the capital base of public sector banks, privatizing public sector banks, and elevating competition through entry of new foreign and domestic private banks. These measures are expected to enable and encourage banks to enhance their eciency, i.e. their ability to transform inputs into outputs, which, in turn, is expected to enhance economic growth by increasing the volume of funds intermediated in the economy.

A. Ataullah et al.
Although some studies have examined the performance of commercial banks in India, only a recent study by Kumbhakar and Sarkar (2003), by using econometric technique to measure the total factor productivity of domestic banks, has analysed a time period long enough to shed some light on the impact of nancial liberalization. In the case of Pakistan, only one study has measured the eciency of commercial banks by using a parametric Distribution Free Approach (DFA). These studies are extended by employing non-parametric Data Envelopment Analysis (DEA) to calculate the eciency of commercial banks in India and Pakistan before and after the liberalization. Following Bauer et al. (1998), the consistency of the DEA-based eciency scores are checked by examining their relationship with three traditional nonfrontier based performance indicators. In addition, the relationship between size and the pure technical eciency of banks is examined.The comparative analysis of Indian and Pakistani banking industries suggests that a similar nancial liberalization programme in two developing countries may lead to dierent outcomes in terms of its success in fostering the technical eciency of banks operating in those countries. The rest of the paper is structured as follows. Section II briey reviews some recent studies on nancial liberalization and the eciency and productivity of banks in developing countries. Section III provides an overview of the measurement of technical eciency using DEA. Section IV presents empirical ndings. Section V concludes.

II. FINANCIAL LIBERALIZATION AND THE EFFICIENCY OF BANKS IN DEVELOPING COUNTRIES Although many developing countries initiated nancial liberalization in the early 1980s, only recently have a few studies examined its impact on the eciency and productivity of banks operating in these countries. These studies postulate that nancial liberalization enhances the eciency and productivity of banks by creating a competitive and exible environment in which banks have more control over their operations. For example, nancial liberalization allows banks to set interest rates on their assets and liabilities that were previously determined by the government. The empirical evidence on the impact of nancial liberalization on the eciency of banks is mixed. Leightner and Lovell (1998) measure the total factor productivity growth of Thai banks during 19891994 to evaluate the nancial liberalization of the late 1980s. Using two alternative inputoutput models, one based on commercial banks

2 During the sample period used in this study, the legislative changes in India allowed public sector banks to tap the capital market to the extent of 49% of their total capital (see Bhide et al., 2002). In Pakistan in contrast, a major portion of two public sector banks, Muslim Commercial Bank and Allied Bank of Pakistan, was sold to private investors (see SBP, 2000).

Financial liberalization and bank eciency

objective to generate revenue and the other based on central banks objective to intermediate funds, they construct a Malmquist total factor productivity index for Thai banks. Leightner and Lovell nd that the productivity of banks improved after the liberalization. Using a similar approach, Gilbert and Wilson (1998) also nd that nancial liberalization in Korea had positive impacts on the productivity of the Korean banking industry during the early 1990s. In contrast, Hao et al. (2001) use a parametric Stochastic Frontier Approach (SFA) to measure the eciency of Korean banks, and do not nd any positive relationship between the measured eciency and nancial liberalization. Isik and Hassan (2003) employ DEA to construct a Malmquist total factor productivity index for Turkish banks during 19801990, and suggest that the performance of banks improved after the implementation of nancial liberalization. In contrast, Yildirim (2002) analyses the technical eciency of Turkish banks between 1988 and 1999 using non-parametric DEA, and nds that the Turkish banks did not achieve any sustained eciency gains over the sample period. Although some recent studies have measured the eciency of Indian banks, their analysis is restricted either to the pre-liberalization period (see Bhattacharyya et al., 1997) or to a single year in the post-liberalization period (see Sathye, 2003). Only a recent study by Kumbhakar and Sarkar (2003) investigates the impact of nancial liberalization by calculating growth in the total factor productivity (TFP) of 23 public sector banks and 27 private domestic banks during 19851996 (their study excludes foreign banks). Kumbhakar and Sarkar (2003) measure TFP growth by estimating a translog cost function, and decompose TFP growth into a technological change, a scale, and a miscellaneous component. They nd considerable over-employment of labour in Indian banks and nd little evidence to suggest that the liberalization enhanced the productivity of banks, especially that of public sector banks. Kumbhakar and Sarkar suggest that public sector banks in India have become too dominant to feel the impact of changes in the economic environment brought about by nancial liberalization. Hardy and de Patti (2001) examined the cost and revenue eciency of 33 banks in Pakistan during 19811998 by utilizing DFA. They nd that during the post-liberalization period, both costs and revenues of banks increased, and therefore conclude that the benets of improvements in revenue eciency were transferred to customers, e.g. borrowers and depositors. However, it is submitted here that during the post-liberalization period, the interest rate margin of the banking industry in Pakistan increased considerably (see SBP, 2000). That is, banks
3

1917
charged higher interest rates on their loans, but did not transfer the higher rates to their depositors. Moreover, there has been constant criticism in the domestic media on the quality of services provided by Pakistani banks, especially by public sector banks. Therefore, it may be dicult to justify Hardy and de Pattis conclusion that benets of improvement in banks performance, if any, were transferred to customers.

I I I . M EA S U R E M E N T O F T E C H N I C A L EFFICIENCY USING DEA The technical eciency of a rm refers to its success/ failure in transforming its inputs into outputs. It is a relative concept as its measurement requires a standard of performance against which the success/failure of the rm is assessed. Broadly speaking, the contemporary empirical studies employ parametric or non-parametric frontier techniques to measure the eciency of rms relative to an estimated best-practice frontier that represents the optimal utilization of resources (see Berger and Mester, 1997).3 The parametric approaches usually involve econometric estimation of a prespecied stochastic production, cost or prot function (see Bauer et al., 1998, pp. 9396). In contrast, non-parametric DEA does not require the specication of a particular functional form for the frontier. Instead, the production frontier is constructed through a piecewise linear combination of the actual inputoutput correspondence set that envelops the inputoutput correspondence of all the rms in the sample (see Thanassoulis, 2001). Hence, eciency measurement is not contaminated by a possible misspecication of the production function (see Bauer et al., 1998). The main weakness of the DEA is that measurement error and statistical noise are assumed to be non-existent (Berger and Mester, 1997; Yildirim, 2002). In this paper DEA is employed for two reasons. First, as discussed above, the existing studies have already employed parametric techniques to investigate the impact of nancial liberalization on the performance of banks in India and Pakistan. Therefore, it is pertinent to examine whether the eciency scores obtained through DEA calculation support the conclusions reached by the existing studies. Second, as Bhattacharyya et al. (1997, p. 335) point out, regulations and other market imperfections in developing countries (especially decades of excessive regulation in the banking industry) may distort input/output prices, and, therefore, may complicate the measurement cost and/or prot function using parametric approaches.

Parametric techniques are: Stochastic Frontier Approach, Distribution Free Approach, and Thick Frontier Approach. Non-parametric approaches are: Data Envelopment Analysis, and Free Disposal Hull (see Bauer et al., 1998).

1918
A simple DEA model Consider N decision-making units (DMUs) (commercial banks in the present case) producing J outputs using I inputs (see Thanassoulis, 2001 for details). To measure the eciency of a DMU, Charnes et al. (1978) proposed the use of the maximum of the ratio of weighted outputs to weighted inputs for that DMU, subject to the condition that similar ratios for all other DMUs in the sample be less than or equal to 1. Mathematically, PJ o o j1 wj yj max eo PI o o 1 i1 vi xi subject to PJ o n j1 uj yj PI o n i1 vi xi

A. Ataullah et al.
technical eciency (PTE) and scale eciency (SE) (see Thanassoulis, 2001). PTE refers to managers capability to utilize rms given resources, while SE refers to exploiting scale economies by operating at a point where the production frontier exhibits constant returns to scale. Inputoutput specication and data source The rst step in measuring eciency using DEA is to specify the inputs and outputs of banks.4 Following Leightner and Lovell (1998) two dierent, albeit complementary, inputoutput models for banks in India and Pakistan are specied: Model A (loan-based model) postulates that banks incur operating and interest expenses to produce loans and advances, and investments; Model B (income-based model) postulates that banks incur operating and interest expenses to produce interest and noninterest income. The analysis covers the period from 1988 to 1998. The sample includes all the commercial banks in India and Pakistan for which data for at least three years are available. This will allow, to some extent, one to see whether the eciency of a bank is due to the capability of managers or due to some random factors that cannot be controlled for in the DEA calculations. In the case of both the countries, the commercial banks included in the sample control over 95% of total assets, deposits, and loans of the commercial banking industry. Data for commercial banks in Pakistan are obtained from various issues of Banking Statistics of Pakistan published annually by the State Bank of Pakistan. In the case of India, data from 1990 to 1998 are obtained from the recently uploaded data set on the website of the Reserve Bank of India,5 and data from 1988 and 1989 are obtained from various issues of Financial Analysis of Banks published by the Indian Banks Association. Banks having zero recorded values for one or more outputs or inputs variables in any year are excluded from the sample for that year in recognition of the fact that the DEA is sensitive to outliers (see Yildirim, 2002, p. 2294).

n 1, . . . , N i 1, 2, . . . , I; j 1, 2, . . . , J

v o , uo ! 0 i j

where yn and xn are positive known outputs and inputs, j i respectively, of the nth DMU, and vo , uo are the variable i j weights to be determined by solving linear problem 1. The DMU being measured is indicated by the index o. The optimization is dened for every DMU in the sample. If the eciency score eo 1, the DMUo is 100% ecient within the sample; otherwise it is DEA inecient. Charnes et al. (1978) transformed the above into the following linear programming problem: XJ o o max ho uy 2 j1 j j subject to XI XJ
j1

vo x o i1 i i

uo yn j j

XI

vo xn i1 i i

n 1, . . . , N i 1, 2, . . . , I

vo ! " i

uo ! " j

j 1, 2, . . . , J

" is an arbitrary small positive number introduced in the above problem to ensure that all of the known inputs and outputs have positive weights. When h 1, DMU is DEA ecient; otherwise it is DEA inecient with respect to other DMUs in the sample. The problem is solved N times to obtain an eciency score for each DMU in the sample. The DEA is carried out by assuming either Constant Returns to Scale (CRS) or Variable Returns to Scale (VRS). The estimation with these two assumptions allows the overall technical eciency (OTE) to be decomposed into two collectively exhaustive components: pure
4

I V . E M P IR I C A L F IN D I N G S Trends in the eciency of commercial banks in India and Pakistan Tables 1 and 2 present the average annual eciency scores of commercial banks in India and Pakistan, respectively, using output-oriented DEA calculated separately for

In the case of banks, there is no agreement on the inputs and outputs. This disagreement is due to dual nature of some of the services that banks provide. For example, bank deposits can be regarded as banks inputs as they are the main inputs for loan production. On the other hand, high value added deposits, like integrated saving and checking accounts, can be regarded as banks outputs. See, for example, Berger and Humphrey (1997) for various approaches to specify banks inputs and outputs, especially the intermediation approach and the production approach. 5 Website: http://www.rbi.org.in/annualdata/index.html

Financial liberalization and bank eciency

Table 1. Technical eciency of commercial banks in India 1988 1989 1990 70.2 75.5 67.9 67.2 87.2 93.0 85.2 83.5 80.4 81.2 79.6 80.5 59.3 52.5 58.3 67.0 81.4 83.5 79.3 81.4 72.9 62.9 73.5 82.3 1991 67.0 75.5 61.7 63.7 84.8 90.6 80.9 82.7 78.8 83.3 76.2 77.0 60.6 53.8 61.2 66.9 81.8 82.1 80.1 83.1 74.1 65.5 76.3 80.5 1992 68.7 78.3 67.2 60.6 88.3 92.1 89.8 83.0 77.6 85.0 74.8 73.0 59.6 57.2 57.7 63.8 81.1 84.2 77.6 81.4 73.5 67.9 74.3 78.4 1993 71.0 76.9 61.9 74.2 88.3 91.9 85.0 87.9 80.3 83.7 72.8 84.4 60.5 52.0 61.6 68.0 82.6 84.3 81.4 82.0 73.4 61.7 75.7 82.9 1994 70.5 73.7 68.1 69.7 83.1 90.3 76.6 82.5 85.0 81.6 88.9 84.5 62.6 54.3 64.7 68.8 83.7 86.1 79.6 85.5 75.0 63.1 81.3 80.5 1995 79.1 79.5 75.8 82.1 88.6 92.5 87.3 85.9 89.4 85.9 86.8 95.6 67.9 57.1 64.4 82.3 86.4 87.3 82.3 89.6 78.5 65.4 78.3 91.9 1996 80.4 80.4 74.7 86.1 88.8 92.8 82.5 91.1 90.6 86.6 90.6 94.5 65.6 54.6 70.7 71.4 84.7 89.1 81.9 83.0 77.9 61.3 86.3 86.0 1997 78.0 82.7 76.1 75.1 88.4 93.5 82.7 89.0 88.3 88.5 92.0 84.4 69.4 60.8 70.8 76.7 86.7 89.0 85.5 85.5 80.3 68.3 82.8 89.7 1998 82.1 83.2 79.5 83.4 91.4 94.8 87.2 92.2 89.8 87.8 91.2 90.5 71.6 66.2 72.2 76.5 87.8 90.1 86.0 87.4 81.6 73.5 83.9 87.5 19881991 67.9 74.8 63.8 64.9 85.5 90.8 82.8 82.8 79.3 82.5 77.1 78.4 59.0 52.2 59.0 65.9 80.6 82.0 78.7 81.0 73.3 63.6 74.9 81.3 (15.0) (7.2) (20.0 (7.7) (13.7) (8.5) (17.0) (11.2) (12.1) (7.7) (10.8) (8.1) (14.7) (7.4) (17.3) (8.4) (10.4) (12.7) (7.8) (8.5) (9.3) (8.5) (13.9) (4.5) 19921994 70.1 76.3 65.7 67.9 86.5 91.4 83.6 84.4 80.9 83.4 78.5 80.4 60.9 54.5 61.3 66.8 82.5 84.9 79.5 82.9 74.0 64.2 77.0 80.6 (13.5) (6.1) (17.8) (8.8) (12.6) (8.1) (15.8) (10.4) (10.9) (6.5) (9.2) (7.4) (15.7) (6.1) (14.3) (6.4) (13.2) (9.7) (8.8) (12.6) (13.8) (5.1) (12.3) (8.7) 19951998 79.9 81.4 76.5 81.6 89.3 93.4 84.9 89.5 89.5 87.2 90.1 91.1 68.6 59.5 69.4 76.6 86.4 88.9 83.9 86.3 79.6 67.0 82.8 88.7 (17.8) (6.7) (24.4) (13.9) (17.1) (6.7) (21.2) (14.0) (14.5) (4.9) (18.2) (9.1) (18.4) (5.4) (23.0) (14.9) (15.5) (7.5) (18.9) (14.0) (18.2) (4.1) (18.6) (13.0)

Loan-based model (model A) OTE ABs 67.0 67.3 PSBs 73.3 75.1 DPBs 63.1 62.9 FBs 64.7 64.1 PTE ABs 85.0 84.8 PSBs 89.2 90.2 DPBs 81.3 83.6 FBs 84.6 80.6 SE ABs 78.8 79.3 PSBs 82.1 83.2 DPBs 77.6 75.2 FBs 76.5 79.5 Income-based model (model B) OTE ABs 58.0 58.2 PSBs 49.9 52.5 DPBs 59.3 57.3 FBs 64.9 64.7 PTE ABs 80.1 79.1 PSBs 81.0 81.5 DPBs 79.9 75.7 FBs 79.4 80.2 SE ABs 72.5 73.6 PSBs 61.6 64.4 DPBs 74.2 75.8 FBs 81.7 80.7

Note: Abs All banks; PSBs public sector banks; DPBs domestic private banks. Foreign Banks: OTE overall technical eciency; PTE pure technical eciency; SE scale eciency. Figures in parentheses are standard deviations.

1919

1920

Table 2. Technical eciency of banks in Pakistan 1988 1989 1990 35.6 34.4 n.a. 36.8 86.7 91.2 n.a. 82.1 41.2 37.7 n.a. 44.8 46 31.6 n.a. 60.3 84.3 86.2 n.a. 82.4 54.9 36.6 n.a. 73.2 1991 40.2 38.4 n.a. 42 86.9 89.5 n.a. 84.3 46.4 43 n.a. 49.8 47.4 37.3 n.a. 57.5 84.5 88.5 n.a. 80.6 56.7 42.1 n.a. 71.4 1992 44.1 32.5 48.2 51.7 76 84 65.3 78.7 59.4 38.6 73.8 65.7 52.6 37.4 61.8 58.8 78.5 76.7 80.1 78.7 66.8 48.7 77.1 74.7 1993 35.9 28.4 36.7 42.5 68.1 78.5 51.5 74.3 54.9 36.1 71.3 57.2 52.8 37.7 55.2 65.4 82 79.4 80.3 86.2 64 47.5 68.7 75.9 1994 37.6 25.5 42.5 44.9 74.8 86.1 56.6 81.6 53.2 29.6 75.1 55 48.4 37.2 45.4 62.6 76.8 79.1 69.1 82.1 63 47 65.7 76.3 1995 39.9 33.5 46.4 39.8 71.3 75.4 58.5 79.9 57.8 44.4 79.3 49.8 56.9 37.3 60.3 73.1 79.3 80.5 72.2 85.1 71.9 46.3 83.5 85.9 1996 42.4 41.7 45 40.4 73.3 74.1 65.5 80.2 58.4 56.2 68.7 50.4 63.8 34.5 80.9 75.9 84.9 75.8 88.9 89.9 73.6 45.5 91 84.4 1997 53.9 45.7 55.8 60.3 77.9 76.7 70 87 69.6 59.6 79.7 69.3 64.4 35.3 82.4 75.5 86.4 81.2 86.4 91.5 73.8 43.4 95.4 82.5 1998 57.4 48.6 59.7 63.9 80.1 78.1 72.6 89.6 71.9 62.2 82.3 71.2 65.6 40.9 79.3 76.6 85.2 79.6 87.5 88.4 76.2 51.4 90.6 86.6 19881991 38.6 (17.32) 36.1 (10.45) n.a. (n.a.) 41.1 (14.37) 87.6 (14.65) 91.6 (5.74) n.a. (n.a.) 83.7 (19.21) 44.3 (12.21) 39.4 (9.34) n.a. (n.a.) 49.1 (13.36) 47.7 (17.7) 34.7 (5.2) n.a. (n.a.) 60.5 (17.2) 85.0 (16.7) 87.4 (5.6) n.a. (n.a.) 82.6 (17.8) 56.5 (18.2) 39.8 (19.9) n.a. (n.a.) 73.2 (9.1) 19921994 39.0 28.6 42.2 46.2 72.9 82.8 57.5 78.1 55.8 34.6 73.4 59.1 51.2 37.4 53.7 62.2 79.0 78.4 76.3 82.3 64.6 47.7 70.3 75.6 (15.0) (9.0) (n.a.) (12.4) (12.7) (5.0) (n.a.) (16.6) (10.6) (8.1) (n.a.) (11.5) (18.7) (5.5) (n.a.) (18.1) (17.6) (16.4) (n.a.) (18.9) (19.3) (21.0) (n.a.) (9.6) 19951998 47.8 42.0 51.4 49.9 75.6 76.1 66.4 84.1 64.1 55.2 77.3 59.3 62.6 36.9 75.1 75.2 83.9 79.3 83.5 88.7 73.9 46.6 90.0 84.8 (10.8) (6.5) (n.a.) (8.9) (9.1) (3.6) (n.a.) (11.9) (7.6) (5.8) (n.a.) (8.3) (16.1) (4.8) (n.a.) (15.6) (15.2) (14.1) (n.a.) (16.2) (16.6) (18.1) (n.a.) (8.3)

Loan-based model (model A) OTE ABs 41.1 37.9 PSBs 38.5 33.5 DPBs n.a. n.a. FBs 43.7 42.3 PTE ABs 88.1 88.9 PSBs 92.6 93.2 DPBs n.a. n.a. FBs 83.7 84.6 SE ABs 46.9 43 PSBs 41.6 35.9 DPBs n.a. n.a. FBs 52.2 50 Income-based model (model B) OTE ABs 45.8 51.7 PSBs 33 37.5 DPBs n.a. n.a. FBs 58.5 66 PTE ABs 83.7 87.5 PSBs 85.6 89.3 DPBs n.a. n.a. FBs 81.9 85.7 SE ABs 55 59.5 PSBs 38.6 42 DPBs n.a. n.a. FBs 71.4 77

Note: ABs All banks; PSBs public sector banks; DPBs domestic private banks. Foreign banks: OTE overall technical eciency; PTE pure technical eciency; SE scale eciency. Figures in parentheses are standard deviations. n.a. refers to the time period when domestic private banks were not allowed to operate in Pakistan.

A. Ataullah et al.

Financial liberalization and bank eciency

each country using annual frontiers.6 The banking industry has been divided into three groups according to ownership: public, foreign, and domestic private. The whole period (i.e. 19881998) is divided into three subperiods: 1988 1991 refers to the pre-liberalization period, 19921994 is considered as the transition period, and 19951998 represents the post-liberalization period when the liberalization programme is expected to have some impact on the eciency of banks. The banking industry in both the countries exhibits very low OTE, and witnessed little improvement until 1995. In the case of India, this is consistent with Kumbhakar and Sarkars (2003) ndings. In both the countries, the major source of low OTE was low SE, which has not been examined by the previous empirical studies on Indian and Pakistan banking industry. The low level of SE could be attributed to governments restrictions on private banks to extend their operations, and governments direction to public sector banks to extend their branch network to rural and suburban areas. These policies hindered banks ability to exploit scale economies. The limited improvement in OTE until 1995, it is submitted, suggests that banks adapted slowly and cautiously to the changes brought about by the liberalization (see Bhattacharyya et al., 1997). The average OTE of the Indian banking industry improved from 67.9% (Model A) and 59.0% (Model B) in the pre-liberalization period to 79.9% (Model A) and 68.6% (Model B) in the post-liberalization period. In case of Pakistan, the OTE of the banking industry increased from 38.6% (Model A) and 47.7% (Model B) in the preliberalization period to 47.8% (Model A) and 62.6% (Model B) in the post-liberalization period. Unlike in India, where improvement in the OTE was due to improvement in both PTE and SE, the improvement in the OTE of the banking industry in Pakistan was due only to improvement in SE, especially after 19951996 when the government allowed public sector banks to reduce the number of employees and close unprotable branches in rural areas. In Model A, the average PTE of the banking industry in Pakistan declined from 87.6% during the pre-liberalization period to 75.6% during the post-liberalization period, while in Model B PTE declined from 85.0% (pre-liberalization) to 83.9% (post-liberalization). This decline in PTE in the banking industry was due to a sharp decline in the PTE of public sector banks even when the PTE of foreign banks and private domestic banks improved during this period. The PTE of public sector banks declined from 91.6% (Model A) and 87.4% (Model B) in the preliberalization to 76.6% (Model A) and 79.3% (Model B) in the post-liberalization period. It could be argued that,

1921
unlike in India, the nancial liberalization process in Pakistan failed to encourage the managers of public sector banks to utilize their resources more eciently. This could be due to the fact that although both the countries followed a similar nancial liberalization programme, the economic environment in Pakistan was marred by high political instability during the 1990s. This high political instability could have undermined the Pakistani governments commitment to the liberalization process, and, therefore, failed to encourage public sector banks to enhance their resource utilization. Like Kumbhakar and Sarkar (2003), it is found that, unlike private sector banks, public sector banks in both India and Pakistan were relatively slow in improving their eciency over the years. Following Kumbhakar and Sarkar, it is suggested that this group has become too dominant (controlling more than 90% of the assets of the banking industry) to feel any need to quickly transform itself in the face of competition from smaller foreign and private domestic banks. Also, public sector banks huge non-performing loans and extensive branch-networks might have made them inexible even if they wanted to adapt to the changing environment. However, after 19951996, public sector banks exhibit more improvement in their eciency. This could be due to a slight intensication of competition as a result of adopting new nancial technology (e.g. computerization of bank branches and Automated Teller Machines) and the introduction of new nancial products (e.g. credit cards and car nancing schemes) by private banks, especially foreign banks. Private sector banks, especially foreign banks, in both the countries witnessed improvement in both PTE and SE. Non-performing loans and the gap between the two inputoutput models As non-performing loans (NPLs) are a major problem for the banking industry in India and Pakistan, it is crucial to examine their impact on the evolution of technical eciency. This could be achieved by using NPLs as another input, usually non-discretionary, that banks use (see Berger and Humphrey, 1997). However, as the bank-level data on NPLs are not available for India and Pakistan, an attempt is made to examine their impact by taking a closer look at the dierence between the eciency scores obtained from the two models, i.e. the loan-based model and the income-based model. Model A postulates that banks produce loans, advances and investments with given resources, while Model B postulates that banks produce income with given resources. Consequently, the outputs of Model B (income) depend primarily on the outputs of Model A

For DEA estimation, we use DEAP 2.1 is used, developed by Tim Coelli of the University of New England, Australia.

1922
Table 3. Correlation between frontier and non-frontier based measures India Model A PTE ROA TC/TA TC/TR 0.075* 0.166* 0.075** SE 0.035* 0.060** 0.055 Model B PTE 0.304** 0.088* 0.321** SE 0.050* 0.046 0.095** Pakistan Model A PTE 0.014* 0.019* 0.099** SE 0.002 0.032** 0.055* Model B PTE 0.015** 0.096 0.264** SE

A. Ataullah et al.

0.094* 0.032 0.072**

Note: PTE pure technical eciency; SE scale eciency; ROA return on assets; TC/TA total costs/total assets; TC/TR total costs/total revenue. *Spearman Rank Correlation is statistically signicant at 5% level. **Spearman Rank Correlation is statistically signicant at 1% level.

(loans, advances and investments).7 However, if banks are unable to enhance their income-based eciency even when they are able to improve their loan-based eciency, this could be due to the presence of high NPLs in their portfolios. In the case of Indian public sector banks, at the start of nancial liberalization in 19911992, the NPLs as a percentage of total advances were around 24% (see Bhide et al., 2002). This percentage, however, gradually declined to 16% in 19971998. The gap between the eciency scores obtained from two inputoutput models follows a similar trend: during the early years, public sector banks were much more ecient in generating loans, advances and investments than in generating income. During the postliberalization era, however, this gap gradually declined. In the case of the public sector banks in Pakistan, the level of NPLs increased after the implementation of the nancial liberalization from around 18% of total advances to around 26% (see SBP, 2000). The gap in the eciency scores of Pakistani public sector banks also increased over the years. A similar gap between the eciency scores of private sector banks also exists in the two countries. However, as the level of NPLs of private banks is much lower than in the public sector banks, the gap between the eciency scores obtained from the two models is also lower. This gap in the eciency scores from the two models may reect the impact of the presence of high NPLs. That is, over the years, the presence of NPLs impeded banks ability to generate income even when they were relatively more ecient in generating earning assets.8 It could be argued that if the liberalization programme fails to enhance the eciency of banks to generate income from their resources, it could, in the medium- and long-run, impede

their ability to intermediate between savers and borrowers and to enhance the quality of their services, which, in turn, may negatively inuence the process of economic growth.

Consistency of the DEA eciency scores As suggested by Bauer et al. (1998), for the frontierbased eciency scores to be useful, the estimated scores should be positively correlated with the traditional nonfrontier based measures of performance used by regulators, managers, and industry consultants: Positive rank-order correlations with these measures would give assurance that the frontier measures are not simply articial products of the assumptions made regarding the underlying optimisation concept (Bauer et al., 1998, p. 108). Table 3 presents the Spearman Rank correlations between the PTE and SE of the banking industry in India and Pakistan generated by DEA and three non-frontier based measures of bank performance, namely return on assets (ROA), total operating and interest cost per rupee of assets (TC/TA), and total cost per rupee of revenue (TC/TR). The rst measure is expected to have a positive correlation with the frontierbased eciency scores, while the latter two are expected to have a negative correlation. The results in Table 3 suggest that most of the DEA-based eciency scores are consistent with the three non-frontier based performance measures. Only in case of Pakistan, ROA is not consistent with the loan-based PTE of banks. That is, there is an unexpected negative correlation between loan-based PTE and ROA of banks. This could be due to the increasing NPLs of public sector banks in Pakistan, which suggests that even when banks were becoming more ecient in generating

7 This is especially the case for the commercial banks in developing countries where, unlike in developed countries, fee income is very low for commercial banks, and banks rely on traditional loans and government securities for income. 8 Another possible explanation for this gap between the eciency scores obtained from the two models could be that banks transferred the benets of improvement in their eciency to their customers because though banks produced more loans, advances and investments (i.e. intermediated more funds) with given inputs, they did not extract more income from this intermediation process. However, increasing interest margins in both the countries, coupled with constant criticism in the domestic media about the quality of customer services provided by banks, especially public sector banks, may cast some doubt on this interpretation.

Financial liberalization and bank eciency

loans and advances, the protability of banks (i.e. their ROA) was deteriorating.

1923
banks outperformed the smaller ones. However, over the years, the gap between the largest group and other groups declined, and in case of Pakistan, the gap virtually disappeared. The catching-up of smaller banks could be due to their higher exibility, which allowed them to adapt to changes in the banking industry brought about by the nancial liberalization programme. In contrast, the declining eciency of the largest group, which primarily constitute public sector banks, could be due to their complex and politically-determined bureaucratic organizational structure that impeded their ability to keep up with smaller private domestic and foreign banks, which were quicker to adopt new nancial technology (e.g. Automated Teller Machines) and to introduce new nancial products (e.g. car nancing and credit cards) (see SBP, 2000; RBI, various issues).

Bank size and pure technical eciency The evidence on the relationship between size and PTE of banks is mixed. For example, in the context of Singaporean banking sector, Leong and Dollery (2002) nd that larger banks, due to complexity of their operations, exhibit higher ineciencies. In contrast, Yildirim (2002) nd a positive relationship between size and PTE of Turkish banks. This positive relationship is attributed to larger banks market power and their ability to diversify credit risk in an uncertain macroeconomic environment. Berger and Humphrey (1997) also nd a positive relationship between size and eciency for the US banking industry. To examine the relationship between size and PTE, the banking industry in India and Pakistan was divided into four quartiles according to their size, where the size of each bank is determined by the total assets of that bank as a percentage of the total assets of the whole commercial banking industry. Figure 1 presents the evolution of PTE of dierent size groups. The gure suggests that in both the countries, during the pre-liberalization period, the largest

V. CONCLUSION This paper provides a comparative analysis of the evolution of the technical eciency of the banking industry in India and Pakistan before and after the implementation of the nancial liberalization programme of the early 1990s. Using non-parametric DEA, it is found that the

India Model A
100 100

Model B

% Efficiency

% Efficiency

75 50 25 0 1988

75 50 25 0 1988

1990

1992
Year

1994

1996

1998

1990

1992
Year

1994

1996

1998

Pakistan
100

Model A
100

Model B

% Efficiency

% Efficiency
1990 1992
Year

75 50 25 0 1988

75 50 25 0 1988

1994

1996

1998

1990

1992
Year

1994

1996

1998

Largest 3rd Quartile

2nd Quartile Smallest

Fig. 1. PTE scores by size quartile for commercial banks

1924
overall technical eciency of the banking industry improved following the nancial liberalization, especially after 19951996. In the case of India, eciency increased due to improvement in both pure technical eciency and scale eciency. In Pakistan, however, the increase in overall technical eciency was due primarily to an improvement in scale eciency. The results suggest that the eciency of commercial banks is much higher in Model A, which uses earning assets as outputs, than in Model B, which uses income as output. This gap in eciency scores obtained from the two models could be due to the presence of high nonperforming loans in the asset portfolios of banks in the two countries. It is argued that even when banks are becoming more ecient in increasing the quantity of loans, advances and investments, this eciency is not being translated into higher eciency in generating income. The results also suggest that the implementation of the nancial liberalization closed the eciency gap between large and small banks. The results suggest that there is still room for improvement in the eciency of banks in both the countries. A major problem, however, is the presence of high non-performing loans. It should be noted that in developing countries the non-performing loans accumulate not only due to the ineffectiveness of banks managers but also due to other factors, such as economic downturns, politicians, pressure on banks, managers to provide loans to clients who may not have economically viable projects, or the weakness of legal system to support the recovery of non-performing loans (see, e.g. Bhide et al., 2002 for the limited success of Debt Recovery Tribunals in India). A step forward for the liberalization programme, therefore, is not only to deregulate interest rates and enhance the level of competition but also to strengthen the institutional structure to support good practices in the banking industry. A C KN O W L E DG E ME N T S Helpful comments by Canan Yildirim and the participants of the 4th Annual International Economics and Finance Society Conference, London 2003, are gratefully acknowledged. The usual disclaimer applies.

A. Ataullah et al.
Berger, A. N. and Humphrey, D. B. (1997) Eciency of nancial institutions: international survey and directions for future research, European Journal of Operational Research, 98, 75212. Berger, A. N. and Mester, L. J. (1997) Inside the black box: what explains dierences in the eciencies of nancial institutions, Journal of Banking and Finance, 21, 895947. Bhattacharya, A., Lovell, C. A. K. and Sahay, P. (1997) The impact of liberalization on the productive eciency of Indian commercial banks, European Journal of Operational Research, 98, 33245. Bhide, M. G., Prasad, A. and Ghosh, S. (2002) Banking sector reforms: a critical overview, Economic and Political Weekly, February, 399408. Charnes, A., Cooper, W. W. and Rhodes, E. (1978) Measuring the eciency of decision making unit, European Journal of Operational Research, 2, 42944. Coelli, T., Rao, D. S. P. and Battese, G. E. (1998) An Introduction to Eciency and Productivity Analysis, Kluwer Academic Publisher, USA. Fry, M. J. (1995) Money, Interest, and Banking in Economic Development, 2nd edn, Johns Hopkins University Press, Baltimore. Gilbert, R. A. and Wilson, P. W. (1998) Eects of deregulation on the productivity of Korean banks, Journal of Economics and Business, 46, 3964. Hao, J. Hunter, W. C. and Yang, W. K. (2001) Deregulation and eciency: the case of private Korean banks, Journal of Economics and Business, 53, 23754. Hardy, D. C. and di Patti, E. B. (2001) Bank reform and bank eciency in Pakistan, IMF Working Paper, WP/01/138. IBA (19861992) Financial Analysis of Banks, Indian Banks Association, Bombay. Isik, I. and Hassan, M. K. (2003) Financial deregulation and total factor productivity change: an empirical study of Turkish commercial banks, Journal of Banking and Finance, 27(8), 145585. Kumbhakar, S. C. and Sarkar, S. (2003) Deregulation, ownership, and productivity growth in banking industry: evidence from India, Journal of Money Credit and Banking, 35(3), 40324. Leightner, J. E. and Lovell, C. A. K. (1998) The impact of nancial liberalisation on the performance of Thai banks, Journal of Economics and Business, 50, 11531. Leong, W. H. and Dollery, B. (2002) The productive eciency of Singapore banks, Working Papers Series in Economics, University of New England, website: http://www.une. edu.au/febl/EconStud/wps.htm. RBI (various issues), Trend and Progress in Banking in India, Reserve Bank of India, website: http://www.rbi.org.in. Sathye, M. (2003) Eciency of banks in a developing economy: the case of India, European Journal of Operational Research, 148(3), 66271. SBP (2000) Pakistan: nancial sector assessment 19902000, website: http://www.sbp.org.pk/publications/fsa/index.htm. Sen, K. and Vaidya, R. R. (1998) India, in Financial Reforms in Developing Countries (Eds) J. M. Fanelli and M. Rohinton, International Development Research Centre, Ottawa, Canada, pp. 5789. Thanassoulis, E. (2001) Introduction to the Theory and Application of Data Envelopment Analysis, Kluwer Academic Publisher, USA. Yildirim, C. (2002) Evolution of banking eciency within an unstable macroeconomic environment: the case of Turkish commercial banks, Applied Economics, 34, 2289301. Zaidi, S.A. (1999) Issues in Pakistans Economy, Oxford, University Press, Karachi.

REFERENCES
Ahluwalia, M. S. (1999) Indias economic reforms: an appraisal, in India in the Era of Economic Reforms (Eds) J. D. Sachs, A. Varshney and N. Bajpai, Oxford University Press, Oxford. pp. 2780. Arun, T. G., and Turner, J.D. (2002) Financial sector reforms in developing countries: the Indian experience, World Economy, 25(3), 42945. Bauer, P. W., Berger, A. N., Ferrier, G. D. and Humphrey, D. B. (1998) Consistency conditions for regulator analysis of nancial institutions: a comparison of frontier eciency methods, Journal of Economics and Business, 50, 85114.