Bank Efficiency in South-Eastern Europe: The Role of Ownership, Market Power, and Institutional Development

Transcription

1 Fordham University From the SelectedWorks of katherin marton Summer July, 2011 Bank Efficiency in South-Eastern Europe: The Role of Ownership, Market Power, and Institutional Development katherin marton Available at:

2 Economics of Transition Volume 19(3) 2011, DOI: /j x Bank efficiency in South-Eastern Europe The role of ownership, market power and institutional development 1 Yiwei Fang*, Iftekhar Hasan** and Katherin Marton*** *Rensselaer Polytechnic Institute, New York, USA. fangy2@rpi.edu **Rensselaer Polytechnic Institute and Bank of Finland, New York, USA. hasan@rpi.edu ***Fordham University, New York, USA. marton@fordham.edu Abstract This study examines the cost and profit efficiency of banking sectors in six transition countries of South-Eastern Europe over the period Using a stochastic frontier approach, our analysis reveals that the average cost efficiency of South-Eastern European banks is percent, and the average profit efficiency is percent. Regressions on the determinants of bank efficiency show that foreign banks are characterized by higher profit efficiency but lower cost efficiency, and government-owned banks are associated with lower profit efficiency than domestic private banks. However, the efficiency gap between foreign-, domestic private- and government-owned banks narrows over time. We also find that the market power of a bank has a positive association with both cost and profit efficiency. Institutional development, proxied by progress in banking regulatory reforms, privatization and enterprise corporate governance restructuring, also has a positive impact on bank efficiency. Received: January 31, 2010; Acceptance: March 27, This research received partial funding from the US State Department Bureau of Intelligence and Research Office of Outreach under 22USC (Program of Research and Training for Eastern Europe and the Former Soviet Union) Title VIII, We are grateful to the editor (Isabel Schnabel) and two anonymous reviewers for very helpful suggestions.. Published by Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main St, Malden, MA 02148, USA

3 496 Fang, Hasan and Marton JEL classifications: G21, P30, P34, P52. Keywords: Bank efficiency, foreign ownership, institutional development, market power, transition economies. 1. Introduction During the last decade, banking sectors in the transition economies of South-Eastern Europe (SEE) went through dramatic changes because of liberalization, consolidation and privatization. At the end of the second decade of transition, banking sectors have been transformed from the former socialist mono-bank systems, or socialist decentralized system in former Yugoslavia, to market-oriented, privatelyowned sectors. The restructuring process in transition economies received considerable interest, both from academics and policymakers. Most studies focused on countries of Central and Eastern Europe (CEE), and less attention has been devoted to the SEE region where political events delayed major economic reforms. Owing to the relatively short span of transition, it is only recently that sufficient reliable data became available to permit a systematic assessment of the banking sector. This study seeks to undertake this assessment by examining banking efficiency in six SEE countries (Albania, Bulgaria, Croatia, Macedonia, Romania and Serbia). 2 We use financial data from Bankscope and calculate cost and profit efficiency of 171 commercial banks for The existing literature suggests that bank efficiency in CEE countries has been significantly influenced by bank ownership, market power and institutional development (e.g. Bonin et al., 2005a; Brissimis et al., 2008; Poghosyan and Poghosyan, 2010). Our study tests whether the main explanatory factors of bank efficiency in CEE countries are also relevant for SEE banks. Our efficiency estimation shows that with an average cost efficiency of percent and average profit efficiency of percent, there is significant room for improvement, although this varies substantially among the six countries. We also compare efficiency levels of SEE banking sectors with eight CEE countries that joined the European Union (EU) in Our results show that SEE banks, on average, operate at lower cost efficiency than banks in CEE, but profit efficiency levels in the two regions are not significantly different. Our analyses on the determinants of SEE bank efficiency obtain striking results. Controlling for various bank characteristics, we find that, compared with domestic ownership, foreign ownership is associated with significantly higher profit efficiency and moderately lower cost efficiency. There is no significant cost efficiency difference between domestic private- and government-owned banks; the latter, however, exhibit significantly lower profit efficiency. The higher profit effi- 2 We omit Bosnia Herzegovina and Montenegro from our study because we could not obtain sufficient data.

4 Bank Efficiency in South-Eastern Europe 497 ciency of foreign ownership is in line with former research findings that foreign banks in developing countries obtain higher returns on their investment (Claessens et al., 2001). The lower cost efficiency of foreign banks can be explained by initial cost disadvantages related to their limited local knowledge in new markets and high investment in local operations for the modernization and expansion of branch networks. This liability of foreignness appears to be higher during the early years of operations (Green et al., 2004; Lensink et al., 2008; Zajc, 2006), since, when we interact ownership with a time trend, we find that cost efficiency of foreign banks improves and that profit efficiency decreases over time. Starting from relatively low levels, in the later years of transition, government-owned banks improve their profit efficiency. Secondly, we examine the effect of individual banks market power on bank efficiency. We measure market power by the Lerner index and find that banks with a higher Lerner index are associated both with higher cost and profit efficiency. This implies that higher market power allows banks to reduce costs and gain monopoly rents. We also employ an instrumental variable (IV) approach to address the existing endogeneity problem and our findings confirm the positive role of market power. Thirdly, we investigate the role of institutional development by regressing efficiency scores on indicators of banking reform, enterprise restructuring and privatization. Our findings in general support the positive impact of these economic reforms on both cost and profit efficiency. Liberalization and deregulation of banking sectors are particularly crucial for banks gains in cost efficiency through economies of scope and scale. Improved corporate governance of enterprises plays a more important role in enhancing profit efficiency. Privatization progress promotes both cost and profit efficiency. The rest of the article is structured as follows. Section 2 briefly reviews the literature on banking efficiency in transition economies and the background of banking industries in SEE countries. Section 3 presents the methodology, data and summary statistics. Section 4 reports the results of the efficiency estimation and the regression analysis studying the determinants of cost and profit efficiency of SEE banks. We end with concluding remarks in Section Literature review and background 2.1 The impact of ownership, market power and institutional development on bank performance The findings of the literature on the relationship between bank ownership and performance in transition economies are contradictory. Most early studies of single country experience find that foreign banks perform better, followed by domestic private- and government-owned banks (see Hasan and Marton, 2003 for

5 498 Fang, Hasan and Marton Hungary; Kraft et al., 2006; Jemric and Vujcic, 2002 for Croatia; Weill, 2003 for Czech Republic and Poland). Similar findings are also supported by a number of cross-country studies (see Bonin et al., 2005a, 2005b; Kasman and Yildirim, 2006). These studies attribute the superior performance of foreign banks to their management skills, advanced technology, access to lower cost funds from the parent company, lack of legacy costs (such as non-performing loans from former periods) and differences in clientele (such as a larger share of foreign-owned companies). It is also argued that foreign banks tend to cherry pick and acquire the most efficient local banks. Despite having such benefits, foreign banks may experience informational asymmetries in new markets, inadequate knowledge of local conditions and difficulties in establishing relational networks. These factors tend to increase the costs of foreign banks and reduce their inherent advantages vis-à-vis domestic banks (Berger et al., 2000; Buch, 2003). Consistent with this point of view, a number of studies find that foreign banks are not more cost efficient than domestically owned banks (e.g. Green et al., 2004; Mamatzakis et al., 2008; Zajc, 2006). Some studies further suggest that foreign banks are likely to incur higher costs subsequent to their acquisition of large government-owned banks (Havrylchyk and Jurzyk, 2011; Poghosyan and Poghosyan, 2010). However, over time with more experience and improved institutional environment, the initial high costs of foreign banks decline, and consequently their cost efficiency improves (Lensink et al., 2008; Poghosyan and Poghosyan, 2010). Regarding the relationship between market power and bank performance, there exist two different theoretical approaches. On the one hand, the quiet life hypothesis assumes that when a firm has monopoly power, managers reduce their efforts and gain monopoly rents through discretionary expenses (Hicks, 1935). Under this hypothesis, market power affects both cost and profit efficiency negatively. The structure conduct performance paradigm, on the other hand, postulates that strong market power allows banks to extract monopoly rents through offering low deposit rates and charging high borrowing rates (Bain, 1956). Under this assumption, market power and profit efficiency are related positively. Strong market power is also found to allow banks to take advantage of economies of scale in monitoring borrowers and to operate at higher cost efficiency (Diamond, 1984). Empirical research on the role of banking competition and market power on efficiency in transition economies presents inconsistent findings. This may partly be related to different countries and the various approaches that were used to measure competition, for example, industry level concentration measures, number of firms in the industry and individual bank market power (Brissimis et al., 2008; Kosak et al., 2009; Pruteanu-Podpiera et al., 2008; Yildirim and Philippatos, 2007). The close relationship between bank performance and institutional settings is well established (Haselmann and Wachtel, 2010; Williams, 1998, 2003). This is

6 Bank Efficiency in South-Eastern Europe 499 also the case in transition economies where major structural reforms have significantly changed the institutional framework. Bonin et al. (2005b) show that the method and timing of privatization significantly influence bank efficiency. Poghosyan and De Haan (2010) find that institutional conditions can be crucial in determining the performance of cross-border bank acquisitions in transition countries. In analysing the impact of banking reforms on cost efficiency, Fries and Taci (2005) show that costs decrease during the early stages of the transition, but over time, and with the implementation of reforms, costs rise. The study of Brissimis et al. (2008) looks at transition economies that joined the EU and finds that banking reforms exert a positive impact on profit efficiency. Grigorian and Manole (2002), however, show that different regulatory measures affect cost efficiency differently. Higher minimum capital requirements for banks, for example, improve cost efficiency, whereas limits on exposure to a single borrower have no significant impact. 2.2 Background of SEE banking sectors Although countries of SEE are heterogeneous in terms of their legacy, the transition process of their banking sectors shows commonalities. Political events delayed the implementation of major economic reforms, and inadequate bank regulation, supervision and accumulation of large non-performing loans from stateowned enterprises contributed to the financial crisis that all countries experienced in the late-1990s. Subsequent to such crises, governments changed their former policies, speeded up the privatization of banks and encouraged foreign banks to participate in that process. The European Bank for Reconstruction and Development (EBRD) and the International Monetary Fund assisted the countries during the financial crises and encouraged governments to adopt policies to encourage foreign bank entry. By around 2000, several countries initiated accession negotiations with the EU and made commitments to align their financial sector regulatory and supervisory norms with those of the EU, and to open up the banking sector to foreign investors. These measures improved significantly the investment climate and encouraged foreign banks to acquire state-owned banks, and to a lesser extent, set up de novo operations. Most of the foreign banks are based in neighbouring European countries, notably in Austria, Italy, Greece, and some in France and Turkey. Several of the banks had already made major acquisitions during the privatization of CEE banks in the 1990s and gained experience of operating in the region. Banking sectors that emerged in SEE countries are largely foreign-owned, and the share of state- and domestic-owned private banks is very small. A summary review of bank ownership of assets, and changes during is presented in Table 1.

7 500 Fang, Hasan and Marton Table 1. Key characteristics of banking sectors in SEE, Total no. of banks No. of foreignowned banks % Assets share of foreignowned banks % Assets share of stateowned banks EBRD index of banking sector reform GDP level in = Albania na 162 Bulgaria na 107 Croatia na 111 Macedonia na 96 Romania na 120 Serbia na 68 Source: EBRD Transition Report, various issues. 3. Methodology and data 3.1 Estimation of cost and profit efficiency In line with Berger and Mester (1997), we measure cost efficiency by how close a bank s actual cost is to what a best-practice bank s cost would be to produce the same bundle of outputs. Banks that are cost inefficient are either wasting some of their inputs (technical inefficiency) or are using the wrong combination of inputs to produce outputs (allocative inefficiency), or both (Mester, 2005). Similarly, profit efficiency is measured by how close a bank s profit is compared with what the bestpractice bank would produce given the same input conditions. The concept of profit efficiency is derived mostly from the revenue side of the banking business. Although it is affected by costs, it allows banks to offset their additional costs to achieve higher service levels. Hence, for profitability and firm value considerations, profit efficiency is a better concept because it also takes the quality of the outputs into account (Mester, 2005). Following many recent studies on banking efficiency in transition countries (e.g. Bonin et al., 2005a; Karas et al., 2010; Yildirim and Philippatos, 2007), our article employs the stochastic frontier analysis (SFA) developed by Aigner et al. (1977) and Meeusen and van den Broeck (1977). The advantage of SFA is that it can disentangle the inefficiency term from the residual. Using a Cobb Douglas translog functional form, our cost efficiency model is specified as follows:

8 Bank Efficiency in South-Eastern Europe 501 lnðtoc=w 2 Þ¼a 0 þ a 1 lnðw 1 =w 2 Þþ 1 2 a 2 lnðw 1 =w 2 Þ lnðw 1 =w 2 Þþ X3 b k lnðy k Þ þ 1 2 þ 1 2 þ 1 2 X 3 X 3 k¼1 k¼1 X 3 X 2 k¼1 l¼1 X 2 l¼1 b kk lnðy k Þ lnðy k Þþ 1 2 d kl lnðy k Þ lnðz l Þ X 3 k¼1 k¼1 c k lnðy k Þ lnðw 1 =w 2 Þ g l lnðw 1 =w 2 Þ lnðz l Þþyear dummies þ region dummy þ GDP growth þ inflation + m þ l; where the input prices and outputs are specified following the intermediation approach. 3 The two input prices are the price of borrowed funds (w 1 = interest expense divided by total borrowed funds) and the price of capital per fixed asset (w 2 = non-interest expense divided by fixed assets). 4 The three outputs are total loans (y 1 ), securities (y 2 ) and other earning assets (y 3 ). Total equity (z 1 ) and nonperforming loans (z 2 ) are used as fixed inputs in the estimation to account for the banks endogenous choices of risk (Hughes et al., 2001). In the estimation, we normalize the equation by one input price (w 2 ) to impose linear homogeneity of input prices (Kuenzle, 2005). Finally, m represents the random noise which incorporates both measurement error and luck, and l is the inefficiency term that increases banks costs and is assumed to have a half-normal distribution with positive value. Note that we include banks from CEE countries in the frontier analysis. This allows us to use CEE bank efficiency levels as a benchmark for SEE countries and also to compare the two regions. As we apply a common frontier for all countries, we control for country heterogeneity by using gross domestic product (GDP) growth and inflation as proxies for economic development and stability, respectively (Berger and Mester, 1997; Kosak and Zorić, 2010). We also include a region dummy to account for differences between CEE and SEE countries (Grigorian and Manole, 2002). Kosak and Zorić (2010) find that the inclusion of country-specific environmental variables in the cost frontier function does not change the estimated 3 The intermediation approach considers banks as mediators that collect deposits and transform them to loans and other earning assets using labour and fixed capital (Sealey and Lindley, 1977). It has been used, for example, by Berger and Mester (1997), Weill (2003) and Karas et al. (2010). 4 Some studies use the personnel price as a separate price term. In our case, however, there are many missing values for personnel expenses for Bulgaria and Romania before To account for this issue, we use total non-interest expense divided by fixed asset as a proxy for the price of total capital. The personnel price is taken into account in this price term, as non-interest expenses include personnel expenses.

9 502 Fang, Hasan and Marton cost efficiency levels. They attribute this finding to the similar transition paths and environmental conditions in the new EU member states. Therefore, we do not include institutional variables in the efficiency estimation, but apply them in the second-stage analysis to explain differences in efficiency (Bonin et al., 2005a; Hasan et al., 2009; Yildirim and Philippatos, 2007). Year dummies are included to control for time fixed effects. To estimate profit efficiency, we apply the non-standard profit function, which maximizes profits given output levels rather than output prices assuming that banks can adjust output prices and input quantities to maximize profit (Berger and Humphrey, 1997; Berger and Mester, 1997). 5 The specification of the non-standard profit function is the same as the Cobb Douglas translog functional form of the cost function except that it replaces total cost with total profit as the dependent variable. Also, the error term becomes m ) l because the inefficiency term reduces profit. Additionally, to avoid taking logarithms of negative profit, consistent with the literature, we add a constant amount to all observations that equals the absolute value of the minimum profit plus one. Financial data for all banks for come from BankScope. We edit the data to take into account reported shortcomings of the database (Bonin et al., 2005a). 6 We allow for failures, mergers and de novo entry of banks. To be included in our sample, a bank has to have a minimum of 3 years of continuous data to obtain reliable efficiency estimates (Fries and Taci, 2005). The selection process yields an unbalanced panel with 171 banks (943 observations) in six SEE countries (Albania, Bulgaria, Croatia, Macedonia, Romania and Serbia), and 209 banks (1,032 observations) in eight CEE countries (Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Slovakia and Slovenia). Table 2 presents the country distribution and mean values of key financial variables of banks in each country. Table 3 presents the summary statistics of the input and output variables for all banks in our sample. 5 According to Berger and Mester (1997), the non-standard profit efficiency approach may provide useful information under the following conditions: (i) there are substantial unmeasured differences in the quality of the output; (ii) a bank cannot achieve every output scale and product mix; (iii) banks have some market power over the prices they charge; and (iv) output prices are not accurately measured. Kasman and Yildirim (2006) also highlights that it is appropriate to use a non-standard profit function when comparing countries with different levels of competition. 6 We make a careful examination of multiple entries for the same bank because they are not completely duplicate observations. First of all, we choose the unconsolidated financial reports of commercial banks because this gives the financial data for the bank rather than the holding company. Then, we check the accounting standards. International Accounting Standards data are used wherever available; otherwise, inflation-adjusted local accounting standards data are used. All bank-level data are inflation-adjusted and reported in US dollars.

10 Bank Efficiency in South-Eastern Europe 503 Table 2. Sample distribution by country and the mean values of key bank financial ratios of each country Country name N Total assets ($mil) ROA Interest margin Loan ratio Deposit ratio Expense ratio Capital ratio Panel A: SEE countries Albania Bulgaria Croatia 298 1, Macedonia Romania 177 1, Serbia Panel B: CEE countries Czech Republic 139 2, Estonia 48 2, Hungary 172 2, Latvia 119 1, Lithuania 89 1, Poland 228 2, Slovakia 108 3, Slovenia 129 2, Note: Variable definitions can be found in the Appendix. 3.2 Measuring market power The Lerner index is a well-established indicator of market power at the individual bank level. This index is defined as the mark-up of the price over marginal cost, divided by the price (Lerner, 1934). It captures the power of individual banks to set prices above marginal cost. A higher value of the Lerner index denotes higher market power of the bank. The Lerner index is calculated as follows, Lerner ¼ p it mc it p it : In line with previous studies estimating the Lerner index in banking (Brissimis et al., 2008; Koetter et al., 2008; Maudos and de Guevara, 2007), we define the price as total revenue (REV) divided by total earning assets (Y), where Y equals the sum of loans (y 1 ), securities (y 2 ), and other earning assets (y 3 ). Marginal cost is calculated from the estimation of a translog cost function and is specified as follows:

11 504 Fang, Hasan and Marton Table 3. Summary statistics of input and output variables of CEE and SEE banks Variable description Variable Observations Mean SD Min. Max. Total cost ($mil) TOC 1, , Total profit ($mil) PBT 1, ) Output Loans ($mil) y 1 1, , , Securities ($mil) y 2 1, , Other earning assets ($mil) y 3 1, , , Input Price of borrowed funds w 1 1, ($/unit) Price of capital ($/unit) w 2 1, Fixed input Equity ($mil) z 1 1, , Non-performing loans ($mil) z 2 1, lnðtoc=w 2 Þ¼a 0 þ a 1 lnðw 1 =w 2 Þþ 1 2 a 2 lnðw 1 =w 2 Þ lnðw 1 =w 2 Þþb 1 lnðyþ 1 þ b 2 2 lnðyþ lnðyþþ1 2 lnðyþ lnðw 1=w 2 Þ X 2 X 2 þ 1 lnðyþ lnðz l Þþ 1 g 2 2 l lnðw 1 =w 2 Þ lnðz l Þþyear dummies l¼1 l¼1 þ region dummy þ GDP growth þ inflation + m þ l: We use the SFA to estimate this equation and obtain the coefficients. Marginal cost follows directly from the estimation by taking derivatives with respect to Y. As the Lerner index is the mark-up of the price over marginal cost divided by the price, it has an upper bound of 1 when the bank has monopoly power and operates with zero marginal cost. It has a value of zero when the price is equal to marginal cost and the bank operates in a perfectly competitive market. The Lerner index can also be negative. This happens when the price is below marginal cost and the bank is in serious trouble. Table 4 (column 4) reports the Lerner index of SEE banks from 1998 to Bank ownership The data on bank ownership are drawn from BankScope. As many banks in transition economies changed ownership several times over the past two decades, it is important to have yearly ownership data so that all ownership changes in our

12 Bank Efficiency in South-Eastern Europe 505 Table 4. Mean values of bank ownership, market power and institutional reforms in SEE countries Year Percentage of foreign banks Percentage of domestic private banks Percentage of government banks Market power (Lerner index) Banking reforms (EBRD indicator) Enterprise restructure (EBRD indicator) Privatization (EBRD indicator) Note: Numbers of ownership percentage and market power are the mean values of the sample banks in each year. sample period are identified (De Haas and Van Lelyveld, 2006). A limitation of Bank- Scope is, however, that it only provides shareholder information for the year when the database was last updated. Therefore, we take separate editions of the database (1999, 2001, 2003, 2005 and 2007) and fill-in the years in between with data from the previous year, if available. We also return to 1998 using the same ownership as To achieve higher accuracy we also search bank homepages and business publications. We group shareholders into three categories: foreign, domestic private and government. We then calculate the aggregated shares held by each group, and construct three ownership dummy variables for each bank in each year according to the type of majority shareholders. In Table 4 (first three columns), we show the evolution of ownership structure in SEE countries. The numbers reported represent the percentage of foreign-, domestic private-, and government-owned banks from 1998 to With progress in privatization, our sample shows that the share of foreign banks increased dramatically over the years. By the end of 2008, foreign banks gained a dominant market share (about 67 percent). Domestic private- and government-owned banks accounted for about 29 and 4 percent of the sample, respectively. 3.4 Institutional development Our indicator of institutional development information is based on data from the EBRD. The EBRD measures banking reform progress in the following three areas:

13 506 Fang, Hasan and Marton (i) creation of a two-tier banking sector and interest rate liberalization; (ii) establishment of regulatory norms for prudential regulation and supervision and full liberalization of credit and interest rates; and (iii) significant progress towards the implementation of the core principles of the Basle Committee (EBRD, 2006). The value of the indicator ranges from 1 to 4.3, a higher value denoting more progress. Our second institutional variable, measuring the progress of enterprise restructuring, captures a country s adoption of modern corporate governance, tighter credit and subsidy policies, the enforcement of bankruptcy legislation, actions that foster competition and market discipline, strengthening of financial discipline, and hardening of budget constraints. The ultimate goal of enterprise restructuring is to foster market-driven restructuring and improve corporate control via financial institutions and markets. The EBRD provides an indicator of progress in these areas. The indicator has a value ranging from 1.0 to 4.3, a higher value indicating significant improvement in corporate governance and more effective corporate control exercised through domestic financial institutions and markets, which foster market-driven restructuring (EBRD, 2005). Our third measure of institutional development is related to privatization. The EBRD compiles two indices to measure this (large-scale privatization and smallscale privatization). The indicator for large-scale privatization ranges from 1.0 to 4.3, with 1.0 indicating very limited private ownership, whereas an index of 4.3 indicates that more than 75 percent of enterprise assets are in private ownership, with corporate governance and performance close to that of advanced industrial economies. Small-scale privatization refers to the privatization of small- and medium-sized firms. The indicator ranges from 1.0 to 4.3, higher values indicating progress in privatization of small companies (EBRD, 2005). We use the mean of the two indicators to obtain the average degree of privatization of all enterprises. Table 4 (last three columns) summarizes the pattern of institutional development from 1998 to Results 4.1 Bank efficiency in SEE countries Table 5 presents the estimated cost and profit efficiency scores of six SEE banking sectors from 1998 to 2008 (3 years of efficiency scores are unavailable for Albania because of insufficient data). As can be seen in Panel A, with an efficiency score of percent, the Croatian banking sector exhibits the highest cost efficiency (76.95 percent), followed by Albania (76.15 percent) and Macedonia (73.74 percent). Serbia exhibits the lowest cost efficiency (54.75 percent), with a substantial decline from 2002 to Bulgaria and Romania have average cost efficiency of and 67 percent, respectively, and in both countries, cost efficiency increased over time.

14 Bank Efficiency in South-Eastern Europe 507 Table 5. Mean scores of cost and profit efficiency in SEE countries during Year Albania Bulgaria Croatia Macedonia Romania Serbia Panel A: Cost efficiency Overall Panel B: Profit efficiency Overall Panel B of Table 5 reports profit efficiency scores. Bulgaria has the highest level of profit efficiency, followed by Macedonia, Albania, Croatia and Serbia. Prior to 2003, Albania had a high level of profit efficiency which was largely because of the high ownership of government bonds (up to 75 percent of assets) that paid high interest rates. Croatia exhibits a stable level of profit efficiency, whereas in Serbia profit efficiency drops greatly from 2006 to Romania has the lowest profit efficiency in our estimation. Prior to the banking crisis of 1999, the efficiency level was 63 percent, but since then, has decreased in most years.

15 508 Fang, Hasan and Marton Table 6. Cost and profit efficiency comparison of SEE and CEE during Year Cost efficiency Profit efficiency SEE CEE Difference SEE CEE Difference ) )0.0581** * ) ) ) ) )0.0358* ) ) ) ) )0.0634** ) Overall )0.0218*** Note: *, ** and *** represent significance levels of 10, 5 and 1 percent, respectively. In Table 6, cost and profit efficiency scores are presented for and scores of SEE and CEE regions compared. The analysis shows that the average cost efficiency of banks in SEE countries is percent, 2.18 percentage points lower than the average of CEE banks. Lower average cost efficiency in SEE may reflect the relatively late start of the banking sector transition compared with CEE countries where this transition started about a decade earlier. Although SEE banks, on average, operate at lower cost efficiency than those in CEE, we do not find statistically significant differences in their profit efficiency. Our findings of the relatively high profit efficiency in SEE are also reflected in the high interest rate margins that characterized countries of the SEE region during the period In 2005, we note a divergence of trend in profit efficiency scores in the two regions, and scores start to decline in SEE countries and increase in CEE countries. It is also noteworthy that during the financial crisis of 2008, both cost and profit efficiency drop significantly for CEE banks, which indicates that a financial crisis has a larger impact on the CEE banking market than on that of SEE. 4.2 Regression analysis on the determinants of bank efficiency in SEE countries Empirical model The second part of our analysis tests the relationship between bank efficiency (cost and profit efficiency) and our set of independent variables (ownership, market

16 Bank Efficiency in South-Eastern Europe 509 power and institutional reforms) in SEE countries. The baseline equation is specified as: P i;t ¼ a 0 þ a 1 foreign dummy i;t þ a 2 government dummy i;t þ a 3 Lerner index i;t þ a 4 banking reform j;t þ a 5 enterprise restructuring ðresidþ j;t þ a 6 privatization ðresidþ j;t þ Rb controls i;t þ e i;t ; where i indexes individual banks, j indexes countries and t indexes years. Our dependent variables (P i,t ) represent cost and profit efficiency of bank i at time t. The key independent variables include majority ownership (as proxied by two indicator variables: foreign dummy and government dummy), individual banks market power (as measured by the Lerner index), and three institutional reform indicators. Note that the three institutional variables are highly correlated to an extent that it would create a multicollinearity problem if we control all three simultaneously in a single model. One way to solve this problem is to use orthogonal measures (Martin and Mauer, 2005). In particular, we regress privatization on banking reform and enterprise restructuring, and enterprise restructuring on banking reform and predict the residuals from each model. The residuals are labelled as privatization (resid) and enterprise restructuring (resid), respectively. 7 After the transformation, three institutional variables, as measured by banking reform, enterprise restructuring (resid) and privatization (resid), are uncorrelated. The equation includes controls for bank-specific financial conditions, time trend and macroeconomic conditions and uses the logarithm of total assets as a proxy for bank size. As Hughes and Mester (1993) and Mester (1996) demonstrate convincingly that managers risk preference may have important implications for efficiency, we include the loan-, capital- and non-performing loan ratio to capture the proportion of loan size to total assets, capital adequacy and credit risk, respectively. To avoid a potential reverse causality between these variables and bank efficiency, we use lagged levels of the previous year. We also use the ROA of the previous year as proxy for managerial efficiency following Lensink et al. (2008). To control for time effects, we include a time trend, calculated as the present year less the benchmark year (1997). We also assume that time may have different effects under different ownership structures, and therefore include the interaction terms. As shown in Table 7 (cost efficiency) and Table 8 (profit efficiency), we estimate the baseline equation using four models. Model 1 is ordinary least squares (OLS) with robust standard errors clustered at the country level. Model 2 is OLS with country-fixed effect and robust standard errors. Model 3 is the IV approach using competition policy and security market development as instruments for the 7 This approach implicitly assumes that the joint variations of the three institutional variables are attributable to banking reform, and that the joint variations of privatization and enterprise restructuring are attributable to enterprise restructuring.

17 510 Fang, Hasan and Marton Table 7. Determinants of cost efficiency in SEE banking sectors Dependent variable: cost efficiency Model 1 Model 2 Model 3 Model 4 Foreign dummy ) )0.0654* )0.0694* )0.0680* ()1.4352) ()1.6758) ()1.7808) ()1.6531) Government dummy (1.2799) (1.3014) (1.2852) (1.5508) Lerner index * *** ** ** (2.4902) (2.8883) (1.9933) (2.0031) Banking reform *** * *** *** (5.5844) (1.8935) (3.0867) (3.3408) Privatization (resid) ** *** *** *** (3.2265) (2.7166) (4.1905) (4.2895) Enterprise restructuring (resid) (1.0899) (0.4561) (1.4285) (0.9832) Time trend )0.0116** )0.0137** )0.0125*** )0.0152*** ()3.9100) ()2.5021) ()4.5637) ()5.3809) Foreign dummy time trend ** ** ** (1.5332) (2.0290) (2.2979) (2.2488) Government dummy time trend ) ) ) )0.0161* ()1.1405) ()1.2827) ()1.6307) ()1.8998) Logarithm of total assets ) )0.0111** )0.0166*** )0.0199*** ()1.9622) ()2.4913) ()3.9347) ()4.5787) Loan ratio * *** *** *** (2.5301) (5.7079) (4.6503) (4.2879) Capital ratio ) )0.1292** )0.1763*** )0.2099*** ()1.5770) ()2.1583) ()2.9983) ()3.4099) Non-performing loan ratio ) )0.3131** ) ) ()1.3073) ()2.1385) ()1.2440) ()0.5695) ROA ) ) )1.1418* ) ()0.8335) ()1.4330) ()1.7401) ()1.6085) Constant *** *** *** *** (6.2673) (4.3274) ( ) ( ) Observations P-value of Hansen J statistic

18 Bank Efficiency in South-Eastern Europe 511 Table 7. (cont) Determinants of cost efficiency in SEE banking sectors Dependent variable: cost efficiency Model 1 Model 2 Model 3 Model 4 First-stage F-statistic R-squared Adjusted R-squared Notes: The dependent variable is cost efficiency. Main variables of interest include bank ownership, market power (Lerner index), and institutional reforms. Model 1 is OLS with robust standard errors clustered by country. Model 2 is OLS adding country dummies. Model 3 is two-stage IV estimation. The two instrumental variables used for the Lerner index are competition policy and security market development. Model 4 is the same as Model 3 but drops the financial crisis year of All model specifications control for time trend and various bank financial characteristics of the previous year. The t-statistics are presented in brackets. *, ** and *** represent significance level of 10, 5 and 1 percent, respectively. Lerner index. We describe this approach in detail in Section Model 4 is a robustness check for Model 3 after dropping the financial crisis year of The role of bank ownership Results in cost efficiency regressions generally find that foreign banks are associated with moderately lower efficiency compared with the benchmark group of domestic private banks (except in Model 1). As shown in Model 2, the coefficient of foreign dummy is ) (P < 0.10), which means that cost efficiency of foreign banks is lower than that of domestic private banks. One explanation of our finding may relate to the high initial investment costs of foreign banks in branch modernization and training subsequent to their acquisition of state-owned banks (Havrylchyk and Jurzyk, 2011; Poghosyan and Poghosyan, 2010). The initial unfamiliarity of foreign banks with the local environment may increase their costs of gathering and processing locally based relationship information. Difficulties in breaking into existing relational networks may also increase initial costs of operation (Buch, 2003; Green et al., 2004; Lensink et al., 2008; Mamatzakis et al., 2008; Zajc, 2006). The lack of significance of Model 1 reflects the fact that the effect of foreign ownership on cost efficiency is not robust with clustering by country, and it is more pronounced with country-fixed effects and with the IV approach. Examining the interaction terms between ownership and time trend, we find that over time, cost efficiency increases for foreign banks but decreases for domestic private banks. 8 These effects are reflected by the positive coefficient of foreign dummy time trend and negative coefficient of time trend, which captures the time 8 Another perspective of time trend is that institutional development has created an environment for bank consolidation. In our article, time trend does not explicitly capture this consolidation movement.

19 512 Fang, Hasan and Marton Table 8. Determinants of profit efficiency in SEE banking sectors Dependent variable: profit efficiency Model 1 Model 2 Model 3 Model 4 Foreign dummy ** *** ** * (3.3853) (3.6120) (2.1873) (1.8115) Government dummy )0.2710** )0.2529*** )0.2114** ) ()2.9007) ()2.8212) ()2.2589) ()1.0824) Lerner index *** *** *** *** ( ) (8.4484) (6.3221) (6.3404) Banking reform * ** ) ) (2.3461) (2.1098) ()0.5333) ()1.4316) Privatization (resid) ** *** ** (2.8019) (2.7189) (2.3495) (1.2068) Enterprise restructuring (resid) * *** *** ** (2.4208) (4.1236) (2.8268) (2.1113) Time trend ) (0.8512) (1.3333) (0.0703) ()0.3880) Foreign dummy time trend )0.0148* )0.0155*** )0.0111** )0.0100* ()2.5421) ()3.4944) ()2.1401) ()1.8114) Government dummy time trend * ** ** (2.4479) (2.5247) (2.1033) (0.7930) Logarithm of total assets ) ) ) ) ()1.0599) ()1.2562) ()1.4815) ()1.0981) Loan ratio ) ) ) ) ()0.9846) ()1.5390) ()1.1216) ()1.0683) Capital ratio ) ) ) ) ()0.6456) ()0.9918) ()1.0352) ()0.2703) Non-performing loan ratio (1.6857) (0.9138) (0.8298) (0.5149) ROA ) )1.6882** (0.6127) (1.3359) ()1.1602) ()1.9610) Constant ** *** *** (1.9785) (2.0044) (2.8288) (2.7345) Observations P-value of Hansen J statistic

20 Bank Efficiency in South-Eastern Europe 513 Table 8. (cont) Determinants of profit efficiency in SEE banking sectors Dependent variable: cost efficiency Model 1 Model 2 Model 3 Model 4 First-stage F-statistic R-squared Adjusted R-squared Notes: The dependent variable is profit efficiency. Main variables of interest include bank ownership, market power (Lerner index), and institutional reforms. Model 1 is OLS with robust standard errors clustered by country. Model 2 is OLS adding country dummies. Model 3 is two-stage IV estimation. The two instrumental variables used for the Lerner index are competition policy and security market development. Model 4 is the same as Model 3 but drops the financial crisis year of All model specifications control for time trend and various bank financial characteristics of the previous year. The t-statistics are presented in brackets. *, ** and *** represent significance level of 10, 5 and 1 percent, respectively. effect for domestic private banks. We do not find a significant time effect for government banks in the first three models, although in Model 4, when we leave out the crisis year, it shows that government banks have decreasing cost efficiency over time. Regarding the other variables, we find that banks with greater loans and less non-performing loans are more efficient. In contrast to our expectations, we also find that larger and highly capitalized banks with good-earning performance in the past year are associated with lower cost efficiency. These results partly reflect the fact that those banks are not efficient in minimizing their operating costs compared with their best-practice peers. In the unreported results, we also find that it needs two more years for them to utilize their advantage in scale and scope economies to reduce costs. Table 8 shows that in all model specifications, foreign ownership is associated with higher profit efficiency. The coefficients are statistically significant and economically meaningful. In particular, as shown in Model 1, profit efficiency of foreign banks is higher than that of domestic private banks (the benchmark group). The higher profit efficiency for foreign banks could mostly be related to the revenues side, reflecting superior management skills, easy access to international markets and their introduction of new products with premium pricing. Our findings on the interaction term of foreign dummy time trend in all four models show significant and negative coefficients of this variable, which implies that profit efficiency of foreign banks has declined over time and that it is hard to maintain early competitive advantages. We also find that government banks are associated with lower profit efficiency. Taking the estimation of Model 1 as an example, government banks are operating at a lower profit efficiency than domestic private banks. The lower profit efficiency of government banks is consistent with previous findings (Hasan and Marton, 2003; Weill, 2003). We find, however, that over time government banks improve profit efficiency (except Model 4). Overall, our

21 514 Fang, Hasan and Marton results indicate that the efficiency gap between foreign-, domestic private- and government-owned banks declined over time. This is partly attributable to the improvement of corporate governance and increased competitiveness of domestic banks over the recent years (Karas et al., 2010; Kraft et al., 2006). In some countries, government-owned banks may have an advantage in obtaining lower deposit rates and thus may improve profit efficiency (Karas et al., 2010; Mamatzakis et al., 2008) The role of market power We first consider the results from the OLS regressions. As shown in Model 1 of Table 7, the Lerner index has a positive and significant coefficient with cost efficiency, indicating that banks with the highest Lerner margins are most cost efficient. We find similar results for the effect of market power on profit efficiency in Table 8, suggesting that banks with greater market power enjoy higher profits through a higher interest margin. These findings are robust when we control for country-fixed effects (Model 2). As both cost efficiency and the Lerner index are derived from the same production function of total operating cost, there is a potential endogeneity problem. Moreover, there could be reverse causality in that higher efficiency leads to higher market power (Demsetz, 1973). These factors may obscure the true effects of the Lerner index. To address these econometric problems, we employ two-step IV estimation and use EBRD indicators of competition policy and security market development as two instruments for the Lerner index. Competition policy captures the enforcement actions that were implemented to reduce abuse of market power and promote competitive environment. It is very intuitive that competition policy has a direct impact on individual banks market power. The security market development indicator measures the progress in establishing securities exchanges, issuance of securities by private enterprises and laws of investor protection. Consequently, the development of the security market inherently influences competition in the banking market. Therefore, our choice of instruments is theoretically valid as both of them should be correlated with individual banks market power (endogenous variable) and they only affect bank performance through the channel of market competition. The results of the IV estimations are reported in Model 3 for the whole time period and in Model 4 for the non-financial-crisis period. Across all model specifications, the predicted Lerner index is positively and significantly associated with cost and profit efficiency. A comparison with the OLS regression further reveals that after adjusting the endogenous relation between the two measures, the effect of market power on efficiency increases substantially. Hence, we confirm the positive role of bank market power in enhancing bank efficiency. As we have one endogeneous variable, but two instruments, we are able to use Hansen s J statistic to test for over-identifying restrictions. This is a joint test of the null hypothesis that the model is correct and the correlation between the instruments and the error term is zero. As shown in the regression tables, the null hypothesis of the Hansen J test is not rejected, which confirms exogeneity of the instruments. We also show that