Comparative efficiency analysis of Greek bank branches in the light of the financial crisis.

Transcription

1 Comparative efficiency analysis of Greek bank branches in the light of the financial crisis. Abstract Eleftherios Aggelopoulos, Antonios Georgopoulos, Costas Siriopoulos 1 During financial crises, which suppress the profitability of banks, it is a common practice for management to aim to minimize inefficiencies in bank operations. This paper reports on an assessment of the efficiency in operations of homogenous branches of a large Greek bank, before and during the crisis, in two different areas: their efficiency in managing the economic record of the branches and their efficiency in generating profits. We have used the non-parametric method input-oriented data envelopment analysis (DEA) under variable returns to scale, for the different efficiency assessments. The results indicated that the average pure technical efficiency both in the two dimensions reduced during the credit crunch period, and mostly for the average profit efficiency. The differences are statistically significant. Performance comparisons on each dimension allowed us to identify those branches that showed a consistently efficient or inefficient behavior over the total period of analysis and form an overall picture of the performance of bank branches. We found that there is the scope for substantial efficiency improvements in the branch network through cost controls and credit risk management, in order to offset the negative impact on earnings caused by the crisis. Keywords: Banking efficiency, Retail branches, Data envelopment analysis, financial crisis JEL: G01, G21 1 ) First Author s Name : Eleftherios Aggelopoulos, Scientific Cooperator, Department of Accounting, Technological Institute of Patras, M. Alexandrou 1 Koukouli, , Patras, Greece, tel , PhD candidate Department of Business Administration, University of Patras, tel , eaggelopoulos@upatras.gr Second Author s Name : Antonios Georgopoulos, Associate Professor, Department of Business Administration, University of Patras, University Campus, 26504, Rio - Patra, Greece, tel , georgop@upatras.gr, Third Author s Name: Costas Siriopoulos, Professor, Department of Business Administration, University of Patras, University Campus, 26504, Rio - Patra, Greece, tel , siriopoulos@upatras.gr 1

2 1. Introduction To the best of our knowledge, our paper investigates for the first time the impact of this crisis on bank efficiency at the network branch level of a large private bank in Greece. In this context, by employing DEA, we extract the efficiency scores of its retail branches in order to find answers related to the following set of concerns. How efficient are the specific branches? How the efficiency of individual branches changed due to the crisis? Which are the main causes of the efficiency downgrading of several branches in the credit crunch period? We empirically concentrate in pure retail banking, which constitutes a central element of the domestic banking system 2. The bank of our study is a representative unit to observe because is one of the three largest banks in the country in terms of assets, given the high concentration rate of the domestic industry (at the end of 2008, the five largest bank concentration ratio in terms of assets stood at 68%, according to Bank of Greece, 2009). Moreover, we achieve branch homogeneity which is a basic requirement of DEA in order to minimize estimation bias (Chortareas et al., 2009; Coelli et al., 2005), by selecting retail branches of the specific bank with common characteristics such as the age of branch establishment, the branch location, the number of employed personnel and similar services provided towards individuals and small enterprises. Furthermore, we classify the selected branches into two main categories, that is bank branches with good performance both in profit and operational terms, the so-called stars (Boussofiane et al., 1991) and the low operation and low profit branches (LOLP), adopting the relevant classification of Portela and Thanassoulis (2007). Our data set is unique, based on inside information derived from monthly Profit and Loss (P&L) statements manually gathering from the Management Information System of the bank of our study. We find that the crisis decreases the average profit-oriented efficiency of the branch network transforming star branches in less star branches and LOLP branches in even more inefficient ones. In this context, it is worth pointing out, that on approximate one out of five star branches is downgraded to the status of LOLP. The 2 In particular, these branches are responsible for the collection of deposits and the efficient allocation of funds to non-financial firms to finance their investment and to reinforce their liquidity (Fiordelisi and Molyneux, 2010). Especially in the case of Greece, local customers show a continued preference for transactions through retail branches (Gaganis and Pasiouras, 2009). Net interest revenue from pure retail banking contributes almost 70% to the Greek banks total operating income (Hardouvelis, 2006). 2

3 main reason for the reduction in efficiency is the inadequate credit risk management in the period of credit sales boom, which is reflected straightforwardly to the substantial increase of loan loss impairments (provisions). Moreover a supplementary role for lower profit efficiency during the crisis seems to play the observed substantial increase of total costs that some branches present in that period. These findings show an over-optimistic measurement of branch efficiency in the expansion period. Such underestimation of efficiency gap reassures the bank managers to take on time measures of reforming of their institutions and keep off the governing bodies to command more effectively the banking sector in order to avoid general destabilizing trends of the financial system. The rest of this paper is organized as follows: Section two contains a literature review of previous studies on bank branch efficiency. Section three presents the methodology and the data set used in the study. Section four presents the empirical findings, while the last section discusses the results and concludes the paper. 2. Literature review Over recent years a large number of studies that examine the efficiency of branch networks of commercial banks have been published. The majority of them use the non-parametric technique DEA for measuring bank branches efficiency. Fethi and Pasiouras (2010) report 30 studies, over the period , that use DEA techniques to estimate branch efficiency. DEA can be related with three 3 different main perspectives and specifically with the degree at which each branch makes use of its resources (e.g. personnel expenses, operational expenses, etc.) to produce: deposits and loans (production or operational approach), transactions (transaction approach) and profits (profit-oriented or intermediation approach 4 ). Most of the existing studies that extract efficiency scores with DEA, follow the production approach 5 and the profit approach 6. In table 1, we present selected studies on branch 3 In the literature few studies utilize another two approaches, which measure the degree at which each branch use its resources: to attract clients (market efficiency) and to provide service quality to its customers (service efficiency). These approaches require data, such as time spent on transactions and profile of customers, which is difficult for someone to have access on it. 4 The intermediation approach is similar to the profit-oriented approach, where a set of expenses is considered on the input side and a set of revenues on the output side (Giokas, 2008a) 5 Parkan, 1987; Drake and Howcroft, 1994; Athanasopoulos, 1997; Giokas, 2008a; Giokas, 2008b 6 Oral and Yolalan, 1990; Wu et al., 2006; Portela and Thanassoulis, 2007; Noulas et al., 2008; Gaganis et al.,

4 efficiency using DEA, showing the observation set of branches, the applied approach and the used input/output variables. TABLE 1 ABOUT HERE In particular, Parkan (1987) examined the production efficiency of 35 branches of a major commercial bank in Canada. He found that only 11 of the 35 branches to be relatively efficient suggesting that improvements in the use of branch resources could eliminate branch network inefficiency. Oral and Yolalan (1990) assessed the profit and service 7 efficiency of a set of 20 retail branches of a Turkish commercial bank and found a positive relationship between them, with 20% per cent of the total branches to be efficient. Drake and Howcroft (1994) investigated the relative production efficiency of 190 branches of a UK bank and found 83 efficient branches (43.68%). Also, they concluded that the branch age, the number of staff, the competitive environment and the location determine the level of the branch efficiency. Hartman et al. (2001), with a data set from 50 savings bank branches in Sweden, showed how production efficiency differs by the level of service and also that small branches offices, though limited in resources, tend to be the most efficient. Camanho and Dyson (2005) analyzed the production efficiency of 144 branches from a Portuguese commercial bank and found the existence of considerable inefficiencies in the branch network. Pastor et al. (2006) examined the profit efficiency of 573 branch offices of a large European saving bank using DEA and found that efficiency improvements at the worst performing branch branches (some branches could achieve 45% reduction in the usage of their resources), can generate a substantial increase in profit for the bank. Wu et al., (2006) integrated DEA and neural networks (NN), to examine the profit efficiency of 142 branches of a Canadian bank. By comparing the efficiency scores with DEA normal results, they found that the combined system identifies more efficient branches. Portela and Thanassoulis (2007) assessed the transaction, production and profit efficiency of 57 branches of a Portuguese bank and identified benchmark bank branches and also problematic branches. Furthermore, they found positive links between production and profit efficiency and also between 7 In this case, they used as inputs the number of personnel, terminals, saving accounts and credit applications and as outputs the time spent on general services, credits, saving accounts and foreign exchange. 4

5 transaction and production efficiency. In general, the existing studies of bank branch efficiency use data on a small number of branches of non-us commercial banks and mostly employ DEA. As far as Greece is concerned, several studies explored efficiency of retail branch networks by employing DEA under different approaches. Vassiloglou and Giokas (1990) assessed the relative transaction efficiency of 20 branches for the Commercial Bank of Greece and found nine branches to be efficient. The authors suggested that a useful interpretation of DEA results requires the discussion of them with management which knows the special characteristics of the branch network. Athanassopoulos (1997), using 68 branches of a commercial bank, investigated the relationship between production efficiency and profit efficiency using a multivariate modeling technique. He found that branch location, network size, product range, personnel abilities and reliability of services have significant effect on Greek branches efficiency. Giokas (2008a) assessed the production, transaction and profit efficiency in operations of a branch network consisting of 44 branches for the financial year The empirical results indicated that there is a scope for substantial efficiency improvements (average inefficiency of 12%) which can generate a substantial increase in profit for the bank. Moreover in another study, Giokas (2008b) examined the production efficiency of a set of 171 bank branches of a large commercial bank in Greece for the financial year He found an average efficiency of 75 % and also that rural branches tend on average to be more efficient than urban branches. Noulas et al., (2008) evaluated the profit efficiency of 58 branches of a major Greek commercial bank for the period and found average inefficiency in the branch network of 30%. Gaganis et al. (2009) examined the profit efficiency of 458 Greek bank s branches of a commercial bank over the period and found that these branches could have achieved improved overall performance (average inefficiency around 30%) during In addition, they concluded that that risk is an important factor for the estimation of the branches efficiency. The general finding of the current empirical evidence for the various dimensions of Greek branch efficiency indicates that there is a scope for substantial efficiency improvements at a level that often exceeds 20% of current costs, which can generate a substantial increase in profit for the bank. 5

6 3. Methodology and Sample 3.1 Measuring the performance of branches As we saw in the literature review, there are three main approaches in the operational research literature used to calculate branch efficiency: The production, transaction and the profit-oriented approach. Different efficiency scores are obtained from these three dimensions since they consider different type of inputs and outputs, without being commonly accepted efficiency estimation (Berger and Humphrey, 1997). Fethi and Pasiouras (2010) reviewed the existing literature in assessing bank and branch efficiency and concluded that there is a general agreement about the main categories of inputs and outputs, however without that implying that there is consistency with respect to the specific inputs/outputs used in various studies. Concerning our methodology, we employ as branch performance measures the production and profit approach 8 given the bank managers objectives to increase sales and manage the product mix in a way that generate high profitability (Portela and Thanassoulis, 2007). We examine the extent to which each of these objectives is being achieved by bank branches. The production efficiency measure is intended to capture the most efficient level of operating costs for managing the economic record of the branches (Giokas, 2008a). In particular, this approach analyzes how each branch uses its resources (inputs) to produce loans, deposits and different transactions (bank guarantees, commissions from imports, exports etc). In our case, we focus on efficient use of number of personnel, personnel expenses and operational expenses, to produce major products generated by retail branches which are loans, deposits and non interest income (fees) from transactions (Table 2). The nature and the number of input output variables used in this assessment were selected after considering data availability, the inputs and outputs that existing literature employs and potential self- 8 The transaction approach requires data such as number of transactions (credit, debit etc) that unfortunately such information was not available in our case. 6

7 identifiers problems 9. In the case of personnel (number of full time employees), we identify branches where employees are the most productive and where branches are overstaffed thus the productivity of labor can be improved. The personnel and operational expenses are used as proxies for the costs that are necessary to carry out banking operations. Personnel costs include salaries and overtime salary costs, while operational costs cover rents for bank branches space, electricity, communication, information technology (IT) and other costs that are necessary to perform operations Bank overheads are excluded since that input is not consumed directly by the branch. The profit efficiency measure is intended to capture the most efficient level of overall costs for generating profits (Giokas, 2008a). As Drake et al., (2006) note, in the context of the DEA profit-oriented approach the more efficient units will be better at minimizing the various costs incurred in generating the various revenue streams and consequently, will be better at maximizing profits. Also the profit approach is considered to be the appropriate system to capture the diversity of strategic responses by financial firms in the face of dynamic changes in competitive and environmental conditions, such as in the case of the current financial crisis (Berger and Mester, 2003). This approach analyzes how each branch uses its resources (inputs) to generate revenues such as interest income and non-interest income from commissions. In our case, we focus on efficient use of personnel expenses, operational expenses and loan loss provisions to generate interest income and fee income (Table 2). 10 Regarding the input side of profit assessment, we include personnel and operational costs (as in the case of the production approach) and loan loss impairments 11, as a proxy for the quality of branch loan portfolio and the sustainability of an interest revenue stream. In particular, the branches of the specific bank create internally - on a monthly basis - a loan loss provision using criteria which are similar with the requirements of IAS 39. Although loan loss impairments are clearly an output measure, they utilized in our assessment as an input because the need is for minimizing this outcome thus penalizing those branches with higher 9 This problem arises when we have few observations and many inputs and/or outputs and that result in many firms appearing on the DEA frontier (Coelli et al., 2005). Dyson et al. (2001) argues that the number of units should be at least twice the product of the number of inputs and outputs. 10 The majority of studies use the above input and outputs to assess the profit efficiency of a branch network, while few of them use as an additional input, interest costs (the overall cost the branch incurs for obtaining deposits). Data limitation however leads us to not incorporate this variable in our model. 11 Impairment loan loss is an amount which has been created against identified credit losses and is calculated on a monthly basis. The amount of the loss impairment is included in the determination of profit or loss for the month. 7

8 losses. Gaganis et al., (2009) empirically found that the inclusion of loan loss impairments as an input variable increases the efficiency score and consequently is an important factor for the estimation of the branches efficiency. Moreover, Pasiouras (2008) calculated higher efficiency scores for the Greek commercial banking industry when provisions included as an input in the efficiency approach. On output side, revenues of the retail branch are split into two categories: Net interest income from lending and deposit operation which is the major source of branch income and non interest income from fees. The latter is an increasingly important output since competitive pressure on bank profits, have led bank branches to create other sources of revenues through commissions and fees (Portela and Thanassoulis, 2007). TABLE 2 ABOUT HERE 3.2 Data Envelopment Analysis (DEA) In order to calculate efficiency of retail branches according to the two aforementioned approaches and compare branches on each dimension, we use the DEA following the foregoing literature. We conduct an input-oriented analysis, following most of the recent studies (Gaganis et al., 2009; Noulas et al., 2008; Giokas, 2008a; Giokas, 2008b; Wu et al., 2006; Camanho and Dyson, 2005; Hartman et al., 2001), in order to identify sources of inefficiency at the input level, which is under direct control of bank management. As Siriopoulos and Tziogkidis (2009) state, during periods of crises it is a common practice for management to aim to enhance the cost efficiency (reducing expenses without lowering the volume of output), so this gives rise to input-oriented models. Since the focus here is on input minimization, performance is measured through an input- oriented model (Banker et al., 1984), under the assumption that the production process in the branch is characterized by variable returns to scale 12 (VRS), as most of the recent papers follow to estimate DEA models (Fethi and Pasiouras, 2010). We use VRS because it is a less restrictive 12 Banker et al., (1984) suggested the use of VRS that decomposes overall technical efficiency (technical efficiency under CRS) into a product of two components. The first is technical efficiency under VRS or pure technical efficiency (PTE) and relates to the ability of managers to utilize firms given resources. The second is scale efficiency (SE) and refers to exploiting scale economies by operating at a point where the production frontier exhibits CRS. In the present study the calculated efficiency scores indicate the technical efficiency under VRS. 8

9 assumption than the hypothesis of constant returns to scale (CRS), where all branches are operating at an optimal scale. We have no a priori reason to believe that for efficient branches output and inputs are proportional (Portela and Thanassoulis 2007, Coelli et al., 2005). This was supported by statistical tests (Banker, 1996) that showed constant returns to scale (CRS) and VRS cannot be assumed to come from the same distribution. 3.3 Sample The data set includes a branch network that consists of 27 branches of a major commercial bank spread across Greece. Since a relatively homogenous set of branches is required in DEA to minimize estimation bias, as we noted in the introduction, we follow the below stepwise sampling process in order to form the observation set of homogeneous retail branches for this study: The starting point of the sample selection procedure is to determine the type of branch operations. 13 The bank s entire branch network is consisting of over 300 branches with different structure of their business. Specifically, there are branches for individuals and small sized enterprises, branches for medium sized enterprises and branches oriented towards private individuals and large companies. We focus on the branches providing similar services to individuals and small sized enterprises. The main characteristics of this branch category are that the total retail exposure of individuals and enterprises is less than 1 million Euros and the enterprises have a maximum turnover of 2.5 million Euros. So under these requirements we concentrate on a branch network of approximate 250 branches. Bearing on mind and the condition that DEA requires the usage of similar resources (in terms of personnel employed) by the sample branches, we select medium size branches according to the banks definition i.e. branches where the number of employees ranges between 10 and 14. After this adjustment, we are left with 200 branches. The next step is to introduce a sample restriction regarding the branch age establishment, since Drake and Howcroft (1994) found that the branch age determines the level of the branch efficiency. This selection criterion is also critical, because we include as an input variable in the profit approach loan loss impairments. So the age of branch establishment influences substantially the magnitude of this variable. After this adjustment, we concentrate on a branch network of 80 branches. 13 As Oral and Yolalan (1990) note, the DEA models are most meaningful when they are applied to observations sets of units or organizations providing similar services and using similar resources. 9

10 Finally, from that network we select branches that operate in urban areas instead of rural places, since DEA efficiency scores are found to be affected by the branch location (Oral and Yolalan, 1990; Drake and Howcroft, 1994; Athanassopoulos 1998; Noulas et al., 2008; Giokas 2008b). So we conclude with a total number of 80 branches with homogenous characteristics. Since our aim is to conduct an extensive analysis of efficiency in the operations of our branch network by manually gathering a series of monthly P&L statements in their original format, we select through a random sampling technique 27 branches (from a total of 80 branches) spread across the urban areas of Greece covering the 34% of the bank s purely retail branch network. The data used for the assessment of bank branches, includes input and output variables available monthly from the end-month Profit and Loss statements, derived from the Management Information System of the specific bank of our study. The data set covers a period from November 2007 to June 2009, including the period when the financial crisis has affected Greek Banks in September We define endogenously the beginning of the crisis for the Greek banking system in that specific month. In particular, the global crisis that started in August 2007 with turmoil in the global money markets (and erupted following the collapse of Lehman Brothers in September of 2008) has inevitably affected the Greek financial system, 14 with a greater time lag if compared with most of the EU. 15 Βased on our sample branches we use as an indicator of the beginning of crisis, the interest rates of time deposits. This is because volumes of time deposits are substantial and at the same time price-sensitive. Hence, any liquidity shortage in the market is reflected straightforwardly to the cost of this type of deposits. Figure 1 illustrates the development of the monthly average spread of time deposits (difference between return on time deposits and interest paid on them) for our 14 One of its initial effects was the drying up of liquidity in the domestic interbank market. This lack of liquidity created an immediate need for banks to draw capital from the local economy leading to increased deposit rates offered to depositors. The deposit rates boom, in turn, resulted in a proportionate increase in lending rates. Lending rates for housing purchase, that absorbed 35% of the economy s total bank credit in 2007, increased from 5.34% in August 2007 to 5.92% in October 2008 (Bank of Greece, 2008a). Furthermore the implementation of more rigorous criteria and prerequisites reduced the volume of credit significantly. In December 2008 lending to non financial corporations and consumers decreased by 64% and 76% respectively, compared with December 2007 (Bank of Greece, 2008b). 15 The reason is that Greek banks strategy has been focused on retail banking and not on investment banking activities, as in the case of other EU banks, that caused huge marked to market losses related to toxic assets. 10

11 sample retail branches, over the period from January 2006 to June We can observe a notably change in the trend after mid September Moreover, this structural point is econometrically derived from the time series of the specific deposits over the total period of the study, by applying the minimum Lagrange Multiplier (LM) unit root test proposed by Lee and Strazicich (2003). Lee and Strazicich s model (in Appendix A the derivation of LM unit root test is provided) allows for two endogenous structural breaks in the trend of the time series, both under the null and the alternative hypothesis. At first, the model determines the optimal lag length at each combination of two breaks. Then, this searches for the two break points where the unit root test statistic is minimized. In our case the optimal lag length is 3 months, with the structural break point in the trend of the specific deposits taking place in September 2008 (minimized LM unit root t-statistic: computed by using the Gauss programming codes.) FIGURE 1 ABOUT HERE So after the endogenous derivation of the beginning of the financial crisis, we can divide our data set to a sub-sample period from November 2007 until August 2008, representing a period (called Period A) where banks business models were oriented to expansion and to another sub-sample period from September 2008 until June 2009, which includes the consequences of the financial crisis (called Period B). In table 3, we present the descriptive statistics (average, median, standard deviation, minimum, maximum) of the input-output variables (in Euro) included in the efficiency assessments for the period A and period B correspondingly. A brief glance at the input-output profile of the branches shows that after the beginning of the financial crisis, there is a reduction on the average monthly deposit balances and a substantial increase on the average loan loss impairments. TABLE 3 ABOUT HERE 16 Due to data limitation, we couldn t incorporate to our analysis a time period before January We believe that since the Greek banking sector experienced a long period of credit sales boom (after the abolition of various regulatory credit ceilings in the mid 90 s and the adoption of the Euro with the subsequent decline in interest rates) lasting on approximate until the beginning of 2008, the time period studied is quite representative to determine when the crisis begun through structural break in deposit interest rates. 11

12 4. Main results Monthly efficiency results, indicating the technical efficiency under VRS assumption, were produced for each performance dimension for a total of 27 retail bank branches. Following the study of Portela and Thanassoulis (2007), we calculate average efficiency scores for each branch, before (Period A) and during the financial crisis (Period B), since our aim is to identify bank branches that showed a consistently efficient or inefficient behavior during these two periods (we do not present monthly detailed efficiency results). So we conclude to general results on each performance dimensions and the links between them. The Appendix B presents the radar graphs comparing for each branch the average profit efficiency scores in Period A with those of Period B and respectively the average production efficiency scores in Period A with those of Period B. Table 4 presents the average efficiency of sample branches for each approach and the percentage of efficient branches for the two periods. TABLE 4 ABOUT HERE The average technical efficiency under VRS for production efficiency in period A is 90.55% and falls to % in period of the financial crisis. The reduction is clearer for the profit efficiency where the average efficiency was 89.25% in period A and deteriorated in 84.14%, indicating that the financial crisis influenced substantially the profit efficiency of the branch network. To examine whether the differences in average efficiency scores between two periods are statistically significant, we perform the paired t-test 17. The mean efficiency loss (Μ=0.051, SD= 0.058, N=27) is significantly greater than zero only in the case of the profit efficiency, with t (26) = 4.53 and two-tail p-value = 0.000, providing evidence that the crisis affected negatively the average profit efficiency of the branch network. A 95% confidence interval about mean efficiency loss is (0.0278, ).Thus, our sample branches could improve their efficiency by % on average. In other words they could have used only 84.14% of the expenses occurred to produce the same level of 17 This statistical test is commonly used to compare means on the same object in differing circumstances. 12

13 revenues. Moreover, the correlation coefficient (Spearman) between average profit and average production efficiency is high with a value of 0.75 and statistical significant at the 1 % level which means that higher production efficiency tends to be related with higher profit efficiency. Also average efficiency scores obtained from the production and the profit efficiency assessments for each period separately, are plotted in figure 2. We choose a threshold of 90% for good efficiency which is common in the literature (Giokas 2008a; Portela and Thanassoulis, 2007). As Camanho and Dyson (1999) and Athanassopoulos and Thanassoulis (1995) point out the precise boundary positions between quadrants is clearly subjective. What is apparent however is that, no matter where the boundaries are drawn, some branches are close to the ideal performance (1,1). FIGURE 2 ABOUT HERE Bank branches with good performance both in production and profit efficiency, which are classified as stars as we saw in the introduction, are located in upper-right quadrant. The LOLP branches are located in the bottom-left quadrant and action is needed to diagnose the causes of their inefficiencies. Bank branches with good profit efficiency and low production efficiency appeared in the bottom right quadrant, are only two (and for the two periods) but with a marginal good efficiency (very close to 90%). This is logical since branches showing low ability of selling banking products are not likely to be efficient in generating profits. Finally in the upper-left quadrant are located branches that generate good profits, in spite of their low production efficiency (sleeper s branches). According to Camanho and Dyson (1999) this is likely to be a consequence of favorable environment rather than good management of the retail branches. Through production and cost efficiency improvements these branches could achieve greater profits. In our analysis we primarily focus on detailed analysis of star branches and LOLP branches in the two periods. Then we emphasize to branches that were stars in period A and became LOLP branches in period B, trying to identify at the same time the causes of that change. In this framework, we analyze differences of star branches and LOLP branches in their average input/output levels for each period. In the radar graphs shown in Fig. 3 we can compare the profit average input/output levels of star 13

14 branches with those of LOLP branches for period A and period B respectively (note that in the below radar graphs values for each variable are normalized by the values observed for star branches). FIGURE 3 ABOUT HERE In period A, LOLP branches use on average more personnel costs (14%) and operational costs (9 %) as star branches, to produce much less fee income (37%) and interest income (11%) than stars branches. In period B, LOLP branches use on average more personnel costs (38%) and operational costs (16%) than star branches, having the same quality of loan portfolio, to produce 30% less fee income and the same amount of interest income as compared to star branches. We point out that cost inefficiency is the main reason for the low efficiency score of LOLP branches in period B. Thus an improvement in the cost management of LOLP branches will contribute a higher profit to the bank. Also we can observe that the financial crisis affected more the performance of branches with a greater size (in terms of amount of personnel costs). In the radar graphs shown in Fig. 4 we can compare the production efficiency average input/output levels of star branches with those of LOLP, for period A and period B respectively. In period A, LOLP branches use on average about the same number of employees and the same amount of personnel and operational costs as star branches, to produce less loan and deposit volumes ( 15%) and much less fee income (37%). In period B, LOLP branches use on average more personnel costs (37%) and more employees (31%) than star branches, to produce approximate the same loan and deposit volumes as star branches and 30% less fee income. The implication of this finding is that cost efficiency improvements at LOLP branches during the financial crisis can generate a substantial increase in production volumes for the bank. FIGURE 4 ABOUT HERE A further aim of our empirical analysis is to identify branches that were stars in Period A and became LOLP in period B and find the reasons for this change. At the beginning, we identify 5 branches that were considered stars in period A and became LOLP in period B (B1, B3, B5, B18, and B19). In the radar graph shown in Fig 5, we 14

15 can compare the profit average efficiency variables of star branches in period A with the profit average efficiency variables of the same branches in period B (note that in the below radar graphs values for each variable are normalized by the values observed for star branches in period A). We can see that the main reason for this importantly change is the doubling of loan loss impairments (99.61%) in the financial crisis period. Also we observe that fees fall short in percentage of 16% in period B (this is because fee income is related to the granting of new loans where demand for them fall sharply during the financial crisis). So we can conclude that branches that turned LOLP in period B became so, because their good loan portfolio quality was superficial in the first place, meaning that had not made a good screening in the first place, thus assuming more credit risk than acceptable. FIGURE 5 ABOUT HERE 5. Summary and discussion In this paper we assessed the production and profit efficiency of a network of retail branches of a representative private commercial bank before and during the financial crisis. This is the first study where the efficiency of retail banking is examined after the beginning of the financial crisis. We measured the average profit and production efficiency of the branch network of our specific bank before and during the financial crisis and we found statistically significant difference in the profit efficiency (production efficiency differences found statistical insignificant) between the two periods. In addition we found that profit efficiency is positively related with production efficiency. The results indicate that the branch network could improve, during the financial crisis period, its ability to generate revenues by % (average profit efficiency 84.14%). So the branch network my minimizing these inefficiencies can offset the negative impact on earnings caused by the crisis and improve its performance. Comparing the average profit efficiency of our branch network with other studies for Greek retail branches, we observe that Noulas et al., (2008) and Gaganis et al., (2009) found an average profit efficiency of 70% but for examined time periods before In particular, Noulas et al., calculated the profit efficiency for the financial years while Gaganis et al., for the period These periods characterized by 15

16 low efficiency in operations (especially for state-owned banks) due to the low competition among banks. The credit-sales boom period that took place mainly after 2005 in conjunction with the privatization of some Greek state owned banks and the entrance of other small banks, formed a strict competitive framework of bank activities. In that environment, profit margins reduced substantially and the level of operating costs increased due to the need for better quality of services. So bank management increased its focus on improvement of the operating efficiency of retail branches. The latter is applied more in our case where the bank of our study is a pure private bank. So it is logical to expect an increase in the average efficiency as time passes. Moreover by identifying benchmark star branches and also problematic LOLP branches and analyze differences in their average input output profit levels, we concluded that cost inefficiency is the main reason for the low efficiency that LOLP branches in period B present and also that branches with greater size seems to be affected more from the crisis In addition by identifying and analyzing star branches before the crisis that downgraded to LOLP branches immediately after the beginning of crisis, we found that the cause of that substantial change in behavior was the doubling of loan loss provisions during the credit crunch period. That finding indicates a deterioration of lending standards in the period of credit sales boom and inadequate credit risk management, with accounting and management implications. As regarding the accounting implications, the current accounting system seems to have a main weakness to capture on time and effectively the credit risk in the expansion period. In this way, the accounting model for loan loss provisioning and impairment, as it is described in the International Accounting Standard (IAS) 39, should bee revised. This is because there is a need for earlier recognition of loan losses in the expansion period through a new dynamic model that should allow early identification of future losses over the life of the loans. The estimation of these losses should be based in a broader range of available credit information than presently included in the incurred loss model by using leading economic indicators (Basel committee on Banking Supervision, 2009). Furthermore our finding provides useful managerial information to bank managers. In period of the financial crisis, branches with previous good profit and production efficiency (star branches in Period A) should act decisively after the burst of the financial crisis and give a priority to credit risk management. To sum up, the 16

17 main implication of our paper is that branches in order to avoid the realization of great loss efficiency during the crisis period, should focus on an effective proactive credit risk in the credit sales boom period. References Altunbas, Y., Gardener, E., Molyneux, P., Moore, B., Efficiency in European banking. European Economic Review 45, Athanasopoulos, A., Service quality and operating efficiency synergies for management control in the provision of financial services: evidence from Greek bank branches. European Journal of Operational Research 98, Athanassopoulos, A., Thanassoulis, E., Separating market efficiency from profitability and its implications for planning. Journal of the Operational Research Society 46, Athanassopoulos, A., Nonparametric Frontier models for assessing the market and cost efficiency of large-scale bank branch networks. Journal of Money, Credit and Banking 30, Bank of Greece, December 2008a. Banking Sector financing to corporations and households, Press Release 29 January Bank of Greece, December 2008b. Interest rates of loans and deposits, Press Release 9 February Bank of Greece, June Financial Stability Report, Bank of Greece, Athens Banker, R., Hypothesis tests using data envelopment analysis, Journal of Productivity Analysis 7, Banker, R., Charnes, A., Cooper, W., Some models for estimating technical and scale inefficiencies in data envelopment analysis. Management Science 30, Berger A., Humphrey D., Efficiency of Financial Institutions: International Survey and Directions for Future Research, European Journal of Operational Research 98, Berger, A., Mester, L., Explaining the dramatic changes in performance of US banks: technological change, deregulation and dynamic changes in competition. Journal of Financial Intermediation 12, Boussofiane, A., Dyson, R., Thanassoulis, E., Applied data envelopment analysis. European Journal of Operational Research 52, Camanho, A., Dyson, R., Efficiency, size, benchmarks and targets for bank branches: An application of data envelopment analysis. Journal of the Operational Research Society 50, Camanho, A., Dyson, R., Cost efficiency measurements with price uncertainty: A DEA application to bank branch assessments. European Journal of Operational Research 161, Coelli, T., Prasada Rao, D., O Donnell, C., Battese, G., An introduction to Efficiency and Productivity Analysis, 2 nd edition USA Kluwer Academic Publishers. 17

18 Charnes, A., Cooper, W., Rhodes, E., Measuring the efficiency of decision making units, European Journal of Operational Research 2, Chortareas, G., Girardone, C., Ventouri, A., Efficiency and productivity of Greek banks in the EMU era. Applied Financial Economics 16, Demiriguc-Kunt, A., Detragiache, E., Gupta, P., Inside the Crises: An empirical Analysis of Banking Systems in Distress. Journal of International Money and Finance 25, Drake, L., Howcroft, B., Relative efficiency in the branch network of a U.K. bank: an empirical study. OMEGA International Journal of Management Science 22, Drake, L., Maximilian, J., Simper, R., The impact of macroeconomic and regulatory factors on bank efficiency: A non-parametric analysis of Hong Kong s banking system. Journal of Banking and Finance 30, Dyson, R., Allen, R., Camanho, A., Podinovski, V., Sarrico, C., Shale, E., Pitfalls and protocols in DEA. European Journal of Operational Research 132, European Central Bank (ECB) Eurosystem, October EU banking structures report, available at: Emrouznejad A., Parker B., Tavares G., Evaluation of Research in Efficiency and Productivity: A Survey and Analysis of the First 30 Years of Scholarly Literature in DEA. Journal of Socio-Economics Planning Science 42, Fethi, M., Pasiouras, F., Assessing bank efficiency and performance with operational research and artificial intelligence techniques: A survey. European Journal of Operational Research 204, Fiordelisi, F., Molyneux P The determinants of shareholder value in European banking. Journal of Banking and Finance 35, Gaganis, C., Liadaki, A., Doumpos, M., Zopounidis, C., Estimating and analyzing the efficiency and productivity of bank branches: Evidence from Greece. Managerial Finance 35, Gaganis, C., Pasiouras, F Efficiency in the Greek Banking Industry: A Comparison of Foreign and Domestic Banks. International Journal of the Economics of Business 16, Giokas, D., 2008a. Assessing the efficiency in operations of a large Greek bank branch network adopting different economic behaviors. Economic Modeling 25, Giokas, D., 2008b. Cost efficiency impact of bank branch characteristics and location: An illustrative application to Greek bank branches. Managerial Finance 34, Hardouvelis, G. (2006). The Greek Banking System in 2006: Comparative Performance. Annual report on the Greek Banking Sector of Eurobank Research. available at:hhtp:// Hartman, T., Storbeck, J., Byrnes, P., Allocative efficiency in branch banking. European Journal of Operational Research 134, Oral, M., Yolalan, R., An empirical study on measuring operating efficiency and profitability of bank branches. European Journal of Operational Research 46, Lee, J., Strazicich, M., Minimum Lagrange Multiplier Unit Root Test with Two Structural Breaks. The Review of Economics and Statistics 85,

19 Maudos, J., Pastor, J., Perez, F., Competition and efficiency in the Spanish banking sector: The importance of specialization. Applied Financial Economics 12, Noulas, A., Glaveli, N., Kiriakopoulos, I., Investigating cost efficiency in the branch network of a Greek bank: an empirical study. Managerial Finance 34, Parkan, C., Measuring the efficiency of service operations: an application to bank branches, Engineering Costs and Production Economics 12, Pastor, J., Knox Lovell, C., Tulkens, H., Evaluating the financial performance of bank branches. Annals of Operations Research 145, Pasiouras, F., Estimating the technical and scale efficiency of Greek commercial banks: The impact of credit risk, off balance sheet activities and international operations. Research in International Business and Finance 22, Portela, M., Thanassoulis, E., Profitability of a sample of Portuguese bank branches and its decomposition into technical and allocative components. European Journal of Operational Research 162, Portela, M., Thanassoulis, E., Comparative efficiency analysis of Portuguese bank branches. European Journal of Operational Research 177, Siriopoulos, C., Tziogkidis, P., How do Greek banking institutions react after significant events? A DEA approach. The International Journal of Management Science. (in press) Vassiloglou, M., Giokas, D., A study of the relative efficiency of bank branches: an application of DEA. Journal of Operational Research Society 41, Wu, D., Yang, Z., Liang, L., Using DEA neural network approach to evaluate branch efficiency of a large Canadian bank. Expert Systems with Applications 31,

20 TABLES Table 1: Selected studies on branches efficiency using DEA, with analysis of the observation set of branches, the applied approach and the used input/output variables. Author date) (publication Athanassopoulos (1997) 68 branches of a Greek commercial bank Camanho and Dyson (2005) Number of Retail branches 144 branches of a Portuguese commercial bank. Drake and Howcroft (1994) 190 branches of a UK clearing bank Gaganis et al (2009) 458 branches of a Greek commercial bank Giokas (2008a) 171 branches of a Greek commercial bank Giokas (2008b) 44 branches of a Greek commercial bank Hartman et al. (2001) 50 branches of a Swedish commercial bank Noulas et al (2008) 58 branches of a Greek commercial bank Oral and Yolalan (1990) 20 branches of a Turkish commercial bank Applied approach Inputs Outputs Production efficiency number of employees, number of ATM, number of computers number of deposit accounts, number of credit and debit transactions, number of loan applications Profit efficiency interest expenses, fee expenses fee income, volume of loans, number of time deposit, saving and current accounts. Production efficiency number of branch and account managers, number of administrative staff, number of tellers, operational expenses Production efficiency number of ATM, operational costs, space Profit efficiency interest expenses, fee expenses, loan loss provisions Production efficiency personnel expenses, other operating expenses Production efficiency personnel expenses, other operating expenses number of general service transactions value of loans, deposits and insurance products interest income, fee income value of deposits and loans, fee income value of deposits and loans, fee income Transaction efficiency personnel expenses, other number of loan, deposit and operating expenses remaining transactions Profit efficiency interest expenses, fee expenses interest income, fee income Production efficiency number of employees, number value of deposits and loans, of computer terminals, square value of house mortgages, meters of premises number of customers Profit efficiency number of employees, other value of deposits and loans operating expenses Profit efficiency personnel expenses, operating interest income, fee income expenses, interest expenses, depreciation 20