PRODUCTIVITY GROWTH IN THE GREEK BANKING INDUSTRY: A NON-PARAMETRIC APPROACH

I. Introduction

The purpose of this paper is to investigate productivity growth in the Greek banking industry for the period 1982-1997. Furthermore, this paper compares and contrasts productivity growth results between the sub-periods 1982-1992 and 1993-1997, since after 1992, with the enactment

of the Second Banking Directive, the process of deregulation and liberalization of the Greek banking sector accelerated. Dramatic and substantial changes have taken place that encouraged competition in the industry by enacting the abolition of credit rationing, deregulating interest rates, eliminating the requirement of banking institutions to invest a significant portion of their reserves in public bonds and permitting free entry to banking institutions from abroad. These factors, together with the development and adoption of information technology, and the provision of new products and services are considered to have caused changes in the productivity and efficiency of the Greek banking industry.

Until the mid-1980s, the Greek banking industry operated in an environment heavily controlled and regulated by the Central Bank, which had gradually caused significant distortions and great inefficiency in the Greek financial system. The types of regulation included barriers to the development of new financial products, the regulation of interest rates, specific asset-holding and branching as well as complex credit rules that determined the interest rates on business loans. These financial regulations aimed at achieving economic policy priorities set by the Greek government, such as the financing of state-owned firms, the development of small and medium-sized enterprises, the expansion of exports, etc. The operation of the banking industry was under such complete dominance of the monetary authorities that it caused the absence of competition in the industry. In particular, the Central Bank of Greece and two major state-owned banks, the National Bank of Greece and the Commercial Bank of Greece, almost completely dominated the banking industry. As a result, banks abstained from adopting advanced technology due to the absence of competition in the industry. Towards the end of the 1980s, the Greek banking industry gradually moved towards a more deregulated system due to international developments and the need to participate in the Single European Market and the EMU. In the 1990s, the Greek banking industry was affected by the harmonization of national regulations within the EU and mainly by the enactment of the Second Banking Directive in 1992, which sought to facilitate the liberalization of financial markets and to enable banks and other financial institutions to operate throughout the European Union (EU) under a single banking license.

In this paper, the investigation of productivity growth is achieved by applying a non-parametric method developed by Fare et al. (1989) which computes total factor productivity (TFP) growth by using a Malmquist index of productivity change. Within this framework, productivity growth may occur due to a combination of industry-wide technological change, i.e. a shift in production surface, and a change in technical efficiency at the level of the operating unit, i.e. movement towards or away from the production surface. The Malmquist index can be decomposed to capture these two components, i.e. technological change and change in technical efficiency. Furthermore, the efficiency component can be decomposed into a pure technical and a scale efficiency change component. This paper also presents technical efficiency measures for the entire period 1982-1997 and for the sub-periods 1982-1992 and 1993-1997, which illustrate how closely an operating unit functions in relation to the production frontier. Technical efficiency indicates the degree to which the operating unit produces the maximum feasible output for a given level of inputs, or uses the minimum amount of feasible inputs to produce a given level of output. Higher efficiency from one period to another does not necessarily suggest that the operating unit achieves higher productivity since technology may have changed.

There are a large number of non-parametric studies that examine the banking industries around the world. The papers by Berger and Humphrey (1997) and Goddard et al. (2001) provide an extensive review of the literature on the efficiency and productivity of financial institutions. There is, however, a small number of non-parametric studies that have investigated the efficiency and productivity growth of the Greek banking industry. The most recent non-parametric studies include work by Noulas (1997), Giokas (1991) and Vassiloglou and Giokas (1990), but these studies are limited to a small time horizon. In particular, the first study refers to the 1991-1992 period while the second and third to the years 1988 and 1987, respectively. In addition, the studies by Giokas (1991) and Vassiloglou and Giokas (1990) examine the relative efficiency of the branches of only one bank, the Commercial Bank of Greece. It should also be noted that there are few parametric studies that examine the efficiency and productivity of the Greek banking sector. Among this research are studies by Apergis and Rezitis (2004), and Christopoulos and Tsionas (2001). The first study examines the cost structure, technical change and productivity in the banking industry for the period 1982-1997 using a traditional translog cost function approach. However, this study does not measure the degree of efficiency of the Greek banking industry due to the inability of the approach used to do so. The other study examines the economic efficiency of Greek banking during the period 1993-1998 using a stochastic frontier approach.

The paper is organized as follows: Section II describes the Malmquist TFP index while Section III presents the data. Section IV discusses the empirical results and Section V provides the conclusions.

II. Methodology

A. Productivity Change

Productivity change over time is an indicator of the performance of an industry. This study calculates the Malmquist productivity index as a measure of total factor productivity change. The Malmquist index approach has been used in a variety of studies related to the financial sector to measure productivity change. In particular, this approach has been applied in studies such as Berg et al. (1992) to examine the productivity of the Norwegian banking sector during the deregulation of the 1980s; Fukuyama (1995) to measure efficiency and productivity growth in the Japanese banking industry during the 1989-1991 period; Noulas (1997) to compare efficiency and productivity differences among state and private banks in Greece during the 1991-1992 period; Leightner and Lovell (1998) to construct productivity indices for Thai banks for the period 1989-1994; Gilbert and Wilson (1998) to study the effects of deregulation on the productivity of Korean banks during the period 1991-1994; Mukherjee et al. (2001) to explore productivity growth for a group of large US commercial banks over the initial post-regulation period 1984-1990; Canhoto and Dermine (2003) to examine banking efficiency and productivity in Portugal during the deregulation period 1990-1995; andCasuetal. (2003) to investigate productivity change in European banking during the 1994-2000 period.

The advantages of the Malmquist productivity index are that it does not make assumptions about the optimizing behavior of the producers and it allows for inefficiency (Fare et al. 1994). Furthermore, the Malmquist index does not rely on econometric estimation, but instead it uses a nonparamctric approach similar to (hat used by data envelopment analysis (DEA). The advantages of using a nonparametric approach are that it avoids imposing a parametric specification for ihe underlying technology as well as for the distributional assumption of the inefficiency term. However, there are some weaknesses associated with a nonparametric approach. First, since a nonparametric method is deterministic and attributes all the variation from the frontier to inefficiency, a frontier estimated by it is likely to be sensitive to measurement errors or other noise in the data. In oilier words, it does not deal with stochastic noise. Another weakness of a nonparametric method is that it does not permit statistical tests and hypotheses to pertain to production structure and the degree of inefficiency. In this paper, a nonparametric approach is used because it is relatively less data demanding, i.e., it works quite well with a small sample size, compared to a parametric approach. Thus, the small sample size of this study, which contains only 6 banks, is conducive to the use of a nonparametric approach.

Coelli et al. (1997) discuss several ways in which environmental variables can be accommodated in a DEA analysis. The term "environmental variables" is usually used to describe factors which could influence the efficiency of a firm. In this case, such factors are not traditional inputs and are assumed to be outside the control of the manager. The two-stage method used in this paper involves the solution of a DEA problem in a first-stage analysis which comprises only the traditional outputs and inputs. In the second stage, the efficiency scores obtained from the first stage are regressed on the environmental variables. The main disadvantage of this method is that if the variables used in the first stage are highly correlated with the variables used in the second stage, then the results could be biased. Although there are a number of alternative approaches to dealing with environmental variables, in most cases Coelli et al. (1997) recommend the use of the two-stage approach because of its several advantages. The main advantages of the two-stage approach are that it can accommodate more than one variable; it can accommodate both continuous and categorical variables; it makes no prior assumption regarding the direction of the effect of the categorical variable; hypothesis tests can be performed to test if the variables have any significant effect on efficiency; and the method is simple and easy to calculate.

III. Data

In the banking literature, there is some debate about what constitutes inputs and outputs for banks. Most banking studies have tended to adopt one of the two main approaches to the input and output specification, i.e., the intermediation approach and the production approach. The intermediation approach considers banks as financial intermediaries that convert deposits and purchased funds into loans and financial investments. This approach treats loans as outputs, while deposits and other liabilities are treated as inputs. Outputs are measured in value terms and costs include both interest expenses and production costs. On the other hand, the production approach considers banks as producers of loan services and deposit accounts using capital and labor as inputs. Outputs are measured in terms of the number of accounts serviced and costs include production costs but not interest expenses. This study uses the intermediation approach, which is the approach most commonly used in the literature (Favero and Papi 1995, Fukuyama 1995, among others).

Data used in this analysis are those used in Apergis and Rezitis (2004). More specifically, the data were obtained from the annual reports of six individual banks for the period 1982 to 1997. Four of the banks used in the sample were state banks, while the other two were private. State banks controlled approximately 80% of the market and dominated the Greek banking industry during the period under consideration. Private banks actually began to enter the banking sector just after 1987. The Greek financial-credit system is characterized by one of the smallest number of credit institutions in the EU. Its characteristics include the prevailing role of a few large state-owned banks, a small share of foreign-owned banks and the presence of a few Greek banks with international operation. The state banks included in the sample are the National Bank of Greece, the Commercial Bank of Greece, the Ionian Bank, and the Bank of Macedonia and Thrace. The private banks are the Alpha Credit Bank and the Ergo Bank.

This study specifies two output and three input variables (Noulas 1997). Variables in values are defined in real terms where 1992 is the base year. In particular, the two output variables (y^sub 1^ and y^sub 2^) are defined as the value of loans and advances (y^sub 1^), which includes short and long term loans and advances to industry and customers, and the value of investment assets (y^sub 2^), which includes shares and other variable-income securities, participation in companies, investments in fixed income securities and government securities. The three input variables (x^sub 1^, x^sub 2^ and ^sub 3^) are defined as labor (x^sub 1^), which is the total number of full-time employees, capital expenses (x^sub 2^), which is defined as fixed assets, including tangible fixed assets (such as buildings, lots, land, furniture, office equipment, etc. net of depreciation) and intangible fixed assets (such as research and development expenses, goodwill, software, underwriting expenses, restructuring expenses, etc.), and, finally, the value of deposits (x^sub 3^), which includes bank bonds and site, saving and time deposits. Table 1 reports the descriptive statistics of the variables used in the analysis for the period 1982-1997. The Data Envelopment Analysis Program (DEAP) of Coelli (1997) is used to compute productivity and efficiency measures presented in this paper. The multi-stage DEA is used to compute the efficiency measures such as overall technical efficiency (OTE), pure technical efficiency (PTE) and scale efficiency (SCE) measures.

IV. Empirical Results

A. Technical Efficiency

Table 2 presents the mean level of the various efficiency measures, i.e. overall (OTE), pure (PTE) and scale efficiency (SCE) for each year of the period 1982-1997, for the whole period and for the two sub-periods 1982-1992 and 1993-1997.1 It is worth noting that these two sub-periods are considered because, as discussed in the Introduction, substantial changes took place in the Greek banking industry after 1992. The results indicate that the mean level of overall efficiency is 0.913 for the whole period and for the two sub-periods. This implies that banks could have increased outputs by 8.7%, on average while still using the same level of inputs.

The average values of the two technical efficiency components, i.e. PTE and SCE, for the period 1982-1997 (sub-periods: 1982-1992 and 1993-1997) are 0.982 (0.977, 0.994) for pure technical efficiency and 0.929 (0.934,0.918) for scale efficiency. This indicates that pure technical inefficiency constitutes a smaller source of inefficiency than scale inefficiency for the banks in the sample under consideration. In other words, the major loss of efficiency for the sample banks is identified as improper scale operation. A comparison of the technical efficiency components of the two sub-periods shows that the sub-period 1982-1992 has a lower (higher) mean level of pure (scale) efficiency than that of the sub-period 1993-1997. The empirical finding that the pure technical efficiency is higher in the second subperiod than in the first one, is attributed to the increased competition and internationalization of the Greek banking system, which happened in the second sub-period due to the accelerated liberalization and deregulation of the financial system. These factors, together with the fast adoption of information technology by the banks, have caused major structural changes in the Greek banking industry during the second sub-period, which may have moved the banks away from an optimal scale of operation. This is indicated by the finding that scale efficiency is lower in the second sub-period than in the first. The increased competition of the Greek banking sector during the second sub-period is supported by the study of Hondroyiannis et al. (1999) which provides evidence that the Greek banking industry has decreased its oligopolistic character and has moved towards conditions of monopolistic competition during the period 1993-1995.

Table 2 also presents the frequency distribution of bank returns to scale for each year of the period 1982-1997. It also shows the cumulative frequency distribution for the whole period and for the two sub-periods. The results indicate that while most of the banks operated under CRS in the first sub-period, they moved to IRS in the second one. This finding further accords with the aforementioned argument that the use of information technology during the second sub-period has moved the banks away from an optimal scale of operation. The finding that banks operated under CRS in the first sub-period is supported by the study of Kalafocas and Mantakas ( 1996), which indicates that scale economies did not exist in the Greek banking industry during the 1980-1989 period.

Table 3 presents Tobit results of pure (PTE) and scale (SCE) efficiency scores on the bank specific factors according to models (12) and (13). It should be stated that most of the estimated coefficients in both models are statistically significant and indicate that the models fit the data well. Furthermore, the high values of the RSquared, i.e. 0.83 for the PTE and 0.86 for the SCE model, show that the explanatory power of the Tobit models are significant. The results show that market share has a positive impact on both PTE and SCE scores at the 0.05 significance level. That is, banks exploit economics of scale as their sizes expand. A positive relationship exists between the banks' service concentration and both PTE and SCE scores at the 0.1 and 0.05 significance level, respectively. In other words, banks with a higher service concentration enjoy higher PTE and SCE due to the existence of gains from specialization. The effect of the time trend variable is positive on both PTE and SCE scores at the 0.01 significant level, implying that PTE and SCE increase with time. However, the impact of the time trend squared variable is negative on both efficiency scores at the 0.05 significant level, indicating that the rate of change of both PTE and SCE decreases with time. The coefficients of the bank specific dummies (d^sub 2^-d^sub 6^) show that there is significant variation of both efficiency scores throughout the sample banks. Finally, the coefficient of the d^sub 1993^ dummy variable has a positive effect on the PTE score at the 0.05 significant level, but a negative effect on the SCE score at the same significant level. This implies that banks in the 1993-1997 sub-period have higher (lower) PTE (SCE) scores than those in the 1982-1992 sub-period.

B. Productivity change

Table 4 presents the Malmquist productivity index, i.e., total factor productivity change (TFP), and its components, technical efficiency change (FCH), technical change (TCH), pure efficiency change (PCH) and scale efficiency change (SCH), for the period 1982-1997 and for the two sub-periods 1982-1992 and 1993-1997.2 If the value of the Malmquist productivity index or any of its components is less (greater) than one, it denotes deterioration (improvement) in performance. The results indicate that total factor productivity (TFP) increased at an average rate of 2.4% per year over the entire 1982-1997 period. On average, this improvement is ascribed to a technical progress (TCH) of 1.2% and to an efficiency improvement (FCH) of 1.2%. The latter, in turn, is attributed to a scale efficiency improvement (SCH) of 0.8% and to a smaller pure efficiency improvement (PCH) of 0.4%.

Comparing the two sub-periods, 1982-1992 and 1993-1997, the second sub-period has a higher TFP than the first one, with an average rate of 4.4% versus 1.7% per year. The finding that TFP, on average, is higher in the second sub-period could be attributed to the increased competition and internntionalization of the Greek banking system, which took place in the second sub-period due to the accelerated liberalization and deregulation of the financial system. It is worth mentioning again the study of Hondroyiannis et al. (1999), which provides evidence of increased competition in the Greek banking sector during the second sub-period.

It is evident that the source of the total factor productivity growth, in the case of the second sub-period, comes entirely from an average technical progress of 5.3% per year, since there is an average deterioration in efficiency of -0.8% per year. The aforementioned deterioration, in turn, is ascribed to a pure efficiency deterioration of -0.5% and to a scale efficiency deterioration of -0.2%. Note that the opposite is true for the first sub-period in which the source of the total factor productivity growth comes exclusively from an average improvement in efficiency of 2% per year, since there is an average technical regress of -0.3% per year. This efficiency improvement is due to a slight pure efficiency improvement of 0.7% and to a strong scale efficiency improvement of 1.2%.

The empirical finding that total factor productivity growth, which originates exclusively from technical change, is higher in the second sub-period than in the first, is attributed to the rapid adoption of new information technology by Greek banks. The deterioration in efficiency observed during the second sub-period could be attributed to the presence of adjustment costs related to the use of this new technology. As for the first sub-period, given the empirical finding of technical regress, banks used the existing technology as efficiently as possible and, for this reason, total factor productivity growth during this sub-period resulted solely from improvements in efficiency.

V. Conclusions

This study examines productivity growth and technical efficiency in the Greek banking industry for the period 1982-1997. Furthermore, it compares productivity growth before and after 1992, since after 1992 the Greek banking industry has experienced a rapid acceleration of liberalization and deregulation. This paper uses the Malmquist productivity index to measure and decompose the total factor productivity growth, as well as the DEA method to measure technical efficiency. It should be noted that one of the main limitations of the DEA method is the presence of outliers which may influence the empirical results, especially in the present study, since the sample used consists of only six banks. However, the results of the present study, in terms of bank level efficiency and productivity measures, do not show big discrepancies among banks. This indicates an absence of outliers in the sample.

The results indicate that productivity growth increased on average by 2.4% per year over the entire period. The findings of increased productivity growth after deregulation in the present study are in accordance with banking industry results obtained in other studies. For instance, the paper by Casu et al. (2003) suggests clear productivity growth for Italian and Spanish banks during deregulation, a growth mainly ascribed to technical progress. Additional papers indicating improvement in productivity due to deregulation are the studies by Mukherjee et al. (2001) for US banks, Gilbert and Wilson (1998) for Korean banks, and Leightner and Lovell (1998) for Thai banks. Most of the aforementioned studies attribute their findings of accelerated productivity to technical progress. There is, however, a number of empirical studies which does not support the claim that deregulation increases productivity in the banking sector, i.e., the studies by Humphrey and Pulley ( 1997) for US banks, and Casu et al. (2003) for French and German banks.

The empirical results show that the average level of overall technical efficiency is 91.3%, suggesting that banks could have increased outputs by 8.7% with the existing level of inputs. The high overall technical efficiency scores of the present study are in line with banking industry results obtained in other studies. For example, high technical efficiency scores are presented by Christopoulos and Tsionas (2001) for Greek banks, Favero and Papi (1995) for Italian banks, and Elyasiani and Mehdian (1995) for US banks. The finding of the present paper that the mean overall technical efficiency is the same for the two sub-periods agrees with the study of Berger and Humphrey (1997) suggesting that the conventional wisdom which implies that deregulation improves efficiency is not always supported by empirical studies. For instance, among the studies indicating that efficiency was relatively unchanged by deregulation are the studies by Elyasiani and Mehdian (1995) for US banks and Hao et al. (2001) for Korean banks. Furthermore, research by Khumbakar et al. (2001) on Spanish banks reported that efficiency was diminished after deregulation. On the other hand, among the studies reporting improvements in efficiency after deregulation are the studies by Berg et al. (1992) of Norwegian and Australian banks, and Canhoto and Dermine (2003) of Portuguese banks.