The Effect of Post-2000 Banking Reforms on Efficiency of Turkish Banks

Transcription

1 The Effect of Post-2000 Banking Reforms on Efficiency of Turkish Banks Msc. In Finance Master Thesis Student name : İkram Altındağ ANR : Department : Finance Faculty : School of Economics and Management Date of Graduation : The Board of Examination : Prof. Dr. Steven Ongena Dr. Fabiana Penas

2 Table of Contents I. Introduction... 1 II. Summary of Turkish Economy and Banking System... 2 A. Pre and Post 1980 Economy... 2 B. Banking Sector in Post-1980 Period... 4 C. Restructuring the Turkish Banking Sector Banks Owned by State Banks Transferred to SDIF Private Banks... 6 III. Efficiency Concept and Methods Used for Measuring Efficiency in Banks... 7 A. Financial Ratio Analysis... 8 B. Parametric Approach Stochastic Frontier Approach Thick Frontier Approach Distribution-Free Approach C. Non-Parametric Approaches (Data Envelopment Analysis) Charnes, Cooper, Rhodes Model of DEA (CCR) Banker, Charnes, Cooper Model of DEA (BCC) IV. Literature Review V. An Empirical Study on the Efficiency of Turkish Banks after post-2000 reforms A. Inputs and Outputs Used In the Study B. Results VI. Conclusion Appendix References... 33

3 I. Introduction Being the intermediary institutions, banks have a vital importance for the economy. Their main function is to create a bridge between the individuals/institutions which have fund surplus and the ones which have fund needs. In other words, they provide a chance for entrepreneurs to use investors` funds. Another macroeconomic importance of banks is that they allocate the available funds to the most productive and effective sectors/businesses via detailed credit allocation processes for the customers (Yigitbas, 2012, pp ). This leads flow of the funds to the individuals/institutions which have the least default risk, and the highest profitability. As well as being financial intermediaries, banks play an important role in transmission of monetary policies. In their study, Kashyap and Stein (1997) conclude that the independence of banks from firms or consumers, and the ability of the banks to offset deposit outflows, caused by monetary policies, are main determinants of the significance of their role, within the European Monetary Union (p. 13). On the other hand, the default risk of loans given and withdrawal risk of the funds deposited make the banks face risks related to both assets and liabilities (O hara, 1983, p. 127). This makes banks more delicate to economic developments. In Turkey, banks form the majority of the financial system. According to the financial stability report, published by Turkish Central Bank in May 2013, the banking system forms 87.4% of the financial sector (TCMB, 2013, p. 35). This means the majority of financial sources are being collected and distributed by banks. In such a financial environment, efficiency of the banks becomes more and more important. Due to this fact, an inefficient banking sector in a country limit the progress in its economy. In the Turkish Banking System, many important developments have taken place in the last 15 years. With the letter of intent in December 1999, the International Monetary Fund (IMF) and the Turkish government reached an agreement of decreasing the inflation, enhancing the growth and stabilizing the economy ( Letter of Intent of the Government of Turkey, 1999). This agreement had been applied until the November 2000 crisis. However, after a new currency crisis occurred in February 2001, The Transition Program for Strengthening Economy (more disciplinary economic program) had been applied in May This program included three main pillars, which are Restructuring Financial Sectors, Financial Transparency in Government and Increasing Competition and Efficiency in Economy ( Guclu Ekonomiye Gecis Programi Transition to Strenghtening the Economy, 2001, p. 13). The main points related with the banking sector, stated in this official announcement are (1) to restructure and increase the efficiencies of banks owned by the state, (2) to solve financial problems 1

4 caused by the banks, which had been transferred to SDIF, and (3) to make necessary regulations to improve financial situations of private banks, which had been negatively-affected by crises. This study aims to show how the efficiency of Turkish Banks has changed after introducing The Transition Program for Strengthening Economy. All operating commercial (deposit) banks have been included in the study. The reason of taking only commercial banks into the study is the difference of financial structures between commercial and investment banks. An important difference is that the investment and development banks collect funds instead of deposits, which causes higher maturities for liabilities, and also different cost of capital. These differences make it inappropriate to compare investment/development banks with commercial banks. Finally, although the strengthening program started in 2001, the data for the last 5 year pre-reform period is also included in this study in order to see the difference between pre- and post-reform efficiency better. In the second part of the study, a brief history of Turkish economy will be introduced, and will be followed by more detailed information about the Banking Sector Restructuring Program as an extension of Strengthening Turkish Economy: Turkish Transition Program under supervision of the IMF. In the third part, efficiency measurement methodologies and related researches are explained. The fourth part comprises of literature review. The fifth part is separated for the study of the efficiency of Turkish Banks in the last 15 years. And finally, the sixth part is dedicated to the conclusion of the study. II. Summary of Turkish Economy and Banking System A. Pre and Post 1980 Economy Turkey had followed state-oriented economic policies since its establishment in 1923 until In the late 1970s, high inflation rate, budget and trade deficits and difficulties in debt-repayments forced the government to take on new policies in the economy. The stand-by agreements with the IMF, signed in 1978 and 1979 failed to improve the economy and the Treasury were having difficulties with paying the principal and interest of the previously issued bonds (Esen, 2010, p. 78). In 1980, a new stability program, named January 24th Decisions was introduced. This program aimed at a transition from a state-oriented economy to a free market economy, and the integration of the Turkish economy to global economy (TCMB, 2002, pp. 5-19). The introduction of Capital Market Law in 1981 and foundation of Istanbul stock Exchange 1984 were the main steps of this program. The main rationale of January 24 th Decisions was to minimize the effect of the state on economy. It was aimed at letting the free market economy create micro and macroeconomic balances instead of the decisions made by the state. In the working paper, published by Turkish Central Bank (TCMB, 2002) the main targets of these decisions are defined in Table 1. 2

5 On the other hand, in later decades, the Turkish economy faced three major financial crises, the first of which was in The main reasons of this crisis were budget deficit and inefficiently managed state economic enterprises (Isik&Kabir, 2002, pp ). The government tried to solve these two macroeconomic problems with monetization (increasing money supply by printing money), instead of bond issuing. However, this strategy caused speculative attacks to foreign currency, and resulted in massive depreciation in the Turkish currency. This situation made it difficult to get fund from internal markets. When the government decided to turn into international market, the worldwide credit rating institutions had already decreased ratings for Turkey. In addition, the Gulf War in 1990 and 1991, and the Europe Monetary Crisis in made it harder for Turkey to find foreign funds. This finally caused one of the most severe liquidity crises in the history of Turkey, and the interest rates which had been around 70-80% at the end of 1993, were around 400% in the middle of 1994 (Isik&Kabir, 2002, pp ). In 1999, the Turkish economy was giving positive signals for the future. In November, Moody`s changed the outlook for the economy from stable to positive ( Moody`s Changes The Outlook On Turkey's Country Ceilings To Positive From Stable, 1999). In December, the IMF and Turkey signed a letter of intent regarding the stabilization and disinflation ( Letter of Intent of the Government of Turkey, 1999). In Helsinki European Council in 10 th - 11 th December 1999, Turkey was to be considered for a membership for future enlargement ( Helsinki European Council 10 And 11 December 1999 Presidency Conclusions, 1999). In spite of all these positive appearances, Turkey was still struggling with its chronic economic problems. The summary of key macroeconomic data from 1994 to 2001 is shown in Table 2. By the end of November 2000, these negative signals put down all the positive signals, explained previously, and Turkey experienced its second severe liquidity crisis in the liberalization period. Ghoshal (2006) gives three major factors for the crisis of November 2000, which are a rising current account deficit, a loss of credibility regarding government policies, and a criminal probe into ten failed banks (p. 182). The number of banks transferred to Saving Deposit Insurance Fund (SDIF) was six in 1999, three in 2000, and eight in 2001, most of which were due to liquidity and fraud problems (BDDK, 2003, pp ). In such an economic environment, foreign investors began to shut down lines of credit to domestic banks, and interest rates began to rise. Banks, which were strong enough to have foreign credit lines, were still in trouble because of increasing interest rates (Ghoshal, 2006, p. 184). Facing the liquidity crisis, the central bank started to inject reserves into financial system, and made heavy purchase of government securities. During the month of November the central bank lost $5 billion in foreign exchange reserves (Ekinci&Erturk, 2007, p. 32). The weekly changes in the central bank reserves are shown in Table 3. 3

6 The liquidity problems in November 2000 continued until February On February 19, 2001, a political feud erupted between the president and the prime minister of Turkey, which created an immediate reflection on the markets. As a result, the stock market fell by 18%, the interest rates on one-month Treasury Bill jumped to 144% in February 20, while overnight-interest rates suddenly climbed to 4000% (Ghoshal, 2006, p. 184). USD appreciated by 76% against TL in five days, and the currency peg became too painful to maintain. Economy became even more severe than it was in November Some related data for February 2001 is shown in Table 4. After February 2001 crisis, the currency peg was left and the floating rate policy was accepted for the Turkish currency. Due to high increases in interest rates, the 2000 and 2001 crises caused severe decreases in the prices of government bonds, which were composing of 12% of total assets of commercial banks by the end of Moreover, due to introduction of the floating rate policy instead of a fixed rate, the debt burden on banks became even heavier, because the total assets were 75% of total liabilities in foreign currency ( The Banks Assosiation of Turkey, 2000). As a result, profitability decreased, and some banks experienced capital decreases. The profitability ratios of banks during this period are summarized in Table 5. In May 2001, Turkey entered another agreement with the IMF, which provided $19 billion to assist the recovery, and a new stabilization program, which was named Turkey`s Transition Program by the Turkish Central Bank ( IMF Approves Augmentation of Turkey's Stand-By Credit to US$19 Billion, 2001). B. Banking Sector in Post-1980 Period After the January 24 th Decisions in Turkey, the banking sector was one of the most-affected sectors by liberalization and free market system. The major goals of the January 24 th Decisions on liberalization of banking sectors is shown in Table 6 (Cankaya&Oz, 2001, p. 16) In spite of the developments in 1980s, in 1990s some structural problems were still detaining the banking sector from its major intermediary duty. Based on the Restructuring Program Progress Report, which was published by BDDK in July 2002, these problems are summarized in Table 7 (p. 4). The instability in the economy, such as unreasonable fluctuations in interest rates, chronic inflation and budget deficits, balance of payment crisis etc., caused high level of credit, currency and country risks. This instability also shortened the maturities of liabilities in banks` balance sheets (Tunaboylu, 2001, p. 42). In addition, in the post-1980 period, banks started to focus on collecting deposits with reasonable interest and financing the government with high interest rates, instead of financing the real sector (Alper, Berument & Malatyali, 2001)). This situation clearly shows that in that period banks had not carried out their natural operations of being a financial intermediary between deposit holders and real sectors. 4

7 At this stage, the introduction of the Banking Code in June 18 th, 1999 was a touchstone in the financial history of Turkey. One of the most important novelties of this code is the establishment of a Regulatory and Supervision Agency (BDDK) to make decisions and controls over all the operational processes of the banks, from establishment to abrogation. In the code, the major mission of the agency is stated as to increase the efficiency and competition ability of banks, to gain permanent trust to the sector, to minimize its possible damages on the economy, to raise it strength, and to protect depositholders` rights. With the code, the establishment of the banks has become more difficult (e.g. the needed qualifications of the founders and high level managers, and capital requirements increased), new limitations were created against loans to be given by banks (e.g. narrower limitations were introduced for a single customer or risk group, for employees, and stockholders), new enforcements for banks were implemented to increase their capital quality, and risk monitoring and management (e.g. efficient internal auditing and risk management systems became obligatory) (Banking Code-4389, 1999). C. Restructuring the Turkish Banking Sector As stated earlier in this study, in May 2001, The Transition Program for Strengthening Economy was introduced. According to the publication of BDDK (2001), one of the major goals of this program was to find and implement structural solutions to the structural problems in the banking sector, and raise its strength in order to be able to compete with worldwide actors (p. 9). In the same publication, the major goals of The Restructuring Program of Banking sector under the transition program were stated as (1) restructuring and increasing the efficiency of banks owned by the state, (2) solving financial problems caused by the banks, which had been transferred to the SDIF, and (3) making necessary regulations to increase financial situations of private banks, which had been negatively-affected by crises. 1. Banks Owned by State The banks owned by the state, undertook many duties given by the government beside their regular banking operations. For example, Ziraat Bankasi was an intermediary bank in the agriculture sector, while Halkbank was a bank to support artisans and small enterprises. Bumin (2009) states that due to the default of the majority of their specific classes of customers, these banks had faced dramatic losses, which were called mission losses in their financial statements. By the end of the 1990s, these mission losses comprised of a high proportion of their total asset. For example, in 2000 the mission loss of Ziraat Bankasi was 50% of its total assets, while the same ratio was 65% in Halkbank (pp ). 5

8 According to The Program of Restructuring Banking Sector, the first action to be taken regarding the banks owned by the state is the purification of these mission losses. The other steps are to increase their capital, to decrease the number of employees, to shut down unproductive branches, and to make mergers/acquisitions to increase their efficiencies (BDDK, 2001, pp ). During 2001, 23 quadrillion Turkish Lira (USD16 billion by the-year-end exchange rate) valued bond had been issued to cover mission losses of the state banks, and these losses were totally cleared out from their balance sheets. In the same year, more than 100 cabinet decisions, which were causing losses, were cancelled in order to prevent new ones. Another action regarding the financial health of state banks were to decrease their short-term liabilities, which were causing liquidity problems and high cost of capital. In this term, overnight repurchase agreements were banned for state banks. The possible short-term liquidity problems were planned to be solved by bond issuance. By the end of October 2001, overnight repurchase agreements were completely cleared out from their short-term liabilities. Another important problem of state banks was their low capital level. By the end of August 2001, total capital of state banks increased by 609%, compared to December The operational structure of the state banks was also a pillar of the restructuring program. First, organizational structures of the banks had changed to be more customer-oriented. Second, one important acquisition took place to increase the efficiency. Third, thirteen thousand employees transferred to other governmental institutions. Finally, the banks started to have external audits for a better assurance and faster technical improvements (BDDK, 2001). 2. Banks Transferred to SDIF At the date of the introduction of the Restructuring Program, there were 13 banks under the control of SDIF. Their total assets were 13.5 billion TL, which makes 13% of total assets of the whole banking sector (BDDK&TBB, 2001). The strategy of authorities on banks which were transferred to the SDIF includes four options: (1) purification of the banks via mergers/acquisitions or direct sale; (2) financial restructuring; (3) operational restructuring; and (4) direct management by the SDIF (BDDK, 2001). The number of banks under the control of the SDIF during period is shown in Table Private Banks Beside state-owned banks and the ones which transferred to SDIF, private banks were also included among the list of actions in the restructuring program. The main goal of the program in terms of private banks is to increase their financial health. More detailed information on these goals is shown in Table 9 (BDDK, 2001). 6

9 As the first step to achieve these goals, necessary changes had been made in laws, especially in banking code, including strengthening capital structure with their own resources, tax advantages on mergers&acquisitions, establishment of a standardized internal control systems, risk monitoring and management systems. Considering the trend in critical numbers and ratios in the banking sector, one can say that the transition program, started in 2001, had a quite positive effect on the sector. These positive effects are shown in Charts 1-6. III. Efficiency Concept and Methods Used for Measuring Efficiency in Banks The efficiency of the banking sector has always been interrogated after almost all financial crises. Especially, from the early 1980s to early 1990s, many banks had been transferred to Saving Deposit Insurance Fund (SDIF) in US. Moreover, the Asian crisis in 1997 showed how default of banks caused a regional macroeconomic crisis (Lee, 2002). And finally, the worldwide financial crisis in 2008 raised questions concerning the reliability of banks efficiency. Academic studies on the efficiency of financial institutions shed light on policies and strategies, regarding how financial goals are achieved, and whether or not to change these policies and strategies in case of failures. However, measuring the efficiency of banks or their branches is not an easy task, since there are too many quantitative indicators (variables) of performance of banks. Before going into the details in the efficiency concept, understanding the difference between efficiency and productivity is crucial. Daraio and Simar (2007) define productivity as the ratio between output and the factors that made it possible and efficiency as the distance between the quantity of input and output, and the quantity of input and output that defines a frontier, the best possible frontier for a firm in its clusters (industry). It can be seen that relativity is the main idea behind efficiency analysis which makes it differ to productivity. For two consecutive periods, a decrease in efficiency can be observed while productivity increases, or vice versa. The first case can happen if the positive move of best-practicing level (efficient frontier) is higher than the positive move of the practicing level of an institution/enterprise (Decision Making Unit - DMU) (Daraio&Simar, 2007). In other words, although a DMU is operating better than its performance in previous period, its efficiency can still decrease in case the other DMUs show a better progress. Efficiency answers such a question: Does a DMU show a better improvement than its rivals?, while productivity calculation does not take the performance of the rivals into account. Three of the earliest studies regarding microeconomic efficiency measurement are Koopman`s An Analysis of Production As an Efficient Combination of Activities in 1951, Debreu`s The Coefficient of Resource Utilization in 1951, and Farrell`s The Measurement of Productive Efficiency in In his paper, Farrell (1957) shows how to find a single efficiency value for an enterprise, which has 7

10 multiple inputs and outputs. Farrell divides the economic efficiency of an enterprise into two components, which are technical and price (allocative) efficiencies. Technical efficiency stands for the ability of producing maximum amount of output with a given amount of inputs, while price efficiency is the ability to create an optimal combination of inputs and outputs in the light of their prices. In other words, price efficiency is the best allocation of the budget among inputs. These two efficiencies, together, comprise cost efficiency of the enterprise (Farrell, 1957). On the other hand, Harvey Liebenstein (1966) contributes to the literature by developing Xefficiency, which stands for how each decision-making-unit (DMU-the enterprises in our case) is close to the efficiency frontier, which is drawn by using observed data. Efficiency frontier is a line where all best-practising DMUs are on and will be discussed later in this study. Different from Farrell`s study, Liebenstein uses frontier analysis to assess DMUs` efficiency, under the idea of the closer to the frontier, the more efficient the DMU. Berger and Mester (1997) analyze three efficiency concepts in their study, which are cost, standard profit and alternative profit. Cost efficiency is the distance of a DMU to the best practising one, considering the cost of having same amount of inputs. Standard profit efficiency is the distance of a DMU to the best practising one in terms of profit and assuming that the prices of inputs and outputs are same for each DMU. Finally, alternative profit efficiency takes amounts, instead of prices of inputs and outputs, unlike standard approach. Beside these different types of efficiencies, scale efficiency is a measure of how the DMU is close to its optimum scale. Sanchez (2009) analyzes technical efficiency into two groups, which are pure technical efficiency and scale efficiency. The idea is that the inefficiency of a DMU may arise from either inefficient implementation of a production plan (pure technical inefficiency) or divergence of DMU from the most productive scale (scale inefficiency). He also states that scale efficiency can be calculated as the ratio of CCR efficiency to BCC efficiency, which will be covered later in this study (p. 2). In the banking sector, efficiency analysis can be divided into three groups, which are ratio analysis, parametric analysis, and non parametric analysis (Seyrek, 2010, p. 69). A. Financial Ratio Analysis Financial Statement Analysis has always been one of the major analysis methods in assessing the performance of an enterprise. As to be briefly explained, financial ratio analysis is to follow the trend in a ratio of an output to an input. It is quite easy, but inefficient model, because most of the times it can be misleading to interpret the efficiency of a company which has multiple inputs and outputs, by looking at a single ratio. In order to avoid this inefficiency, multiple ratios can be analyzed at the same time. However, in this case, it may not be possible to gather all independent ratios into a 8

11 meaningful group and asses them at once in order to answer the questions addressed in this study. Regarding to Turkish Banking sector, one of the most detailed studies was done by Aktas&Kargin (2007), in which they categorize financial ratios into five groups, which are: Capital Requirement Ratios This shows capital structure and financial strength of the banks Asset Quality Ratios This shows the probability of payback (default risk) of the loans given Liquidity Ratios This shows the ease of converting assets into cash or equivalents, which is very important in case of unexpected happenings, such as the 2000 and 2001 crises Profitability Ratios This is the most important output, not only for banks but also for all enterprises Expense/Income Ratios This shows the main sources of income, and it is quite useful for the banks to see their strengths and weaknesses Aktas & Kargin (2007) analyze these ratios for domestic and foreign banks in Turkey for the period between 2003 and 2006, and find that liquidity and capital requirement ratios of foreign banks are significantly higher than the ones in domestic banks, and there is no significant difference in terms of asset quality and profitability. B. Parametric Approach In parametric approaches, the best performance is assumed to be on a regression line and the DMUs which do not deviate from this regression line are assumed to be efficient and the ones which deviate are inefficient. This approach always assumes a random error even for efficient DMUs. In such a case, it becomes hard to differentiate the random error from the deviation that is observed in inefficient DMUs (Weill, 2003). In other words, it is hard to say whether the error is random or real inefficiency. In the literature of Efficiency Analysis, there are three main approaches under the category of parametric approach, which are stochastic frontier approach, thick frontier approach, and distributionfree approach. The main difference among these three approaches is the assumption of the distributions of inefficient DMUs, and error terms (Weill, 2003, p. 579). 1. Stochastic Frontier Approach Berger&Humprey (1997) state that the stochastic Frontier Approach (sometimes called economic frontier approach) sets up a functional form for the cost, profit, or production relationship among explanatory variables such as inputs, outputs and environmental factors, under the assumption of 9

12 asymmetric (e.g. half-normal) distribution of inefficiency error, and standard normal distribution of random error. They also explain the reason of asymmetric (truncated) distribution of inefficiency error is because it is impossible to observe a negative inefficiency (pp. 6-8). Although the accuracy of assumption of distribution is the main challenge of SFA (like other parametric approaches) it does not affect the ranking of DMU in terms of efficiency, because the effect of the assumptions will be same for all DMUs (Bauer et al., 1998, p. 12). In other words, for given prices for inputs and outputs, and cost function, the DMUs which produce with less cost will always be more efficient than the others. 2. Thick Frontier Approach Different from SFA, Thick Frontier Approach (TFA) does not require an assumption for distribution of random error and inefficiency error. In this approach, the production units are classified according to their performance (quartiles). The error terms within the highest and lowest performance in quartiles of observations represent random error, while the differences between the highest and lowest quartiles represent inefficiencies (Berger&Humprey, 1997). If efficiency is measured for the banking sector in an economy, based on TFA, first the banks need to be classified according to their performance. TFA assumes that the banks in the best performing group are efficient and form the thick frontier for them. The deviation of DMUs within the best and worst groups represents random error, while the deviation between the best and the worst groups corresponds to inefficiency value. 3. Distribution-Free Approach The critics on SFA about the uncertain distribution of error terms are the origin of the Distribution- Free Approach (DFA). Unlike the SFA, the DFA does not assume the distributions of inefficiencies or random error, which means the errors and inefficiencies may follow any distribution. The DFA also does not claim that the deviation within the best and the worst DMU groups is random error, and the deviation between the best and the worst performing ones is inefficiency, unlike the TFA. Instead, the DFA assumes the efficiency of each firm is stable, and random error tends to average out to zero over time. The inefficiency is determined as the difference between a firm`s average residual and the average residual of the firm on the frontier (Berger&Humprey, 1997). C. Non-Parametric Approaches (Data Envelopment Analysis) Non-Parametric Approach (also named as Data Envelopment Analysis) is a data-oriented approach to measure the efficiencies of DMUs, similar production units that use multiple inputs to produce multiple outputs. Hospitals, universities, courts, enterprises, banks and even countries can be taken as DMUs. Compared with other techniques, DEA is more useful, since it requires much less assumptions 10

13 and it can create clear conclusions even for the DMUs, which have complex inputs and outputs (Wade&John, 2005, p. 1). Data Envelopment Analysis is mainly based on finding the distance between a DMU`s performance to efficient frontier, which is found by using linear programming tools (Cooper et al, 2006, pp. 2-3). Charnes, Cooper and Rhoades (1978) and Banker, Charnes, Cooper (1984) versions are two of the most commonly-used techniques in DEA. 1. Charnes, Cooper, Rhodes Model of DEA (CCR) In their study in 1978, Charnes, Cooper and Rhoades define efficiency of a DMU as maximum ratio of weighted outputs to weighted inputs, under the constraint that this ratio is not above 1 for all DMUs. This definition can be formulated as follows (Charnes et al, 1978): Max h 0 = s r=1 u r y r0 m v i i=1 x i0 (1a) Subject to: s r=1 u r y rj m v i i=1 x ij 1 ; j = 1,., n, (1b) v r, u i 0 ; r = 1,., s ; i = 1,., m (1c) * J=1,2,3..n represents DMUs * h 0 represents the efficiency of the DMU, which is being analyzed. * i=1,2,3..m represent inputs * r=1,2,3..s represent outputs * u r, v i represent the weights of inputs and outputs *y rj represents the output (r) of DMU (j), while x ij represents input (i) of DMU (j) This model has been designed to assess the relative performance of one of decision-making units, DMU 0, among j=1,2,3.,n DMUs. The efficiency of DMU 0 is also included among the constraints, while the goal of the model is to maximize it. The rationale of its being among constraints, and maximized to 1 is to set the maximum efficiency value to be between 0 and 1 (0 < h 0 < 1). In the function that has been shown in (1), m s r=1 u r y r0 represents the amount of output, while i=1 v i x i0 is the amount of inputs. Besides, h * 0 = 1 means DMU 0 is efficient, while h * 0 < 1 suggest that inefficiency exists for DMU 0. With the solution of this maximization problem, u * r and v * i values are determined, which are called virtual multipliers. Based on these virtual weights, virtual input ( m i=1 m i=1 v i x i0 ), virtual output u r y r0 ), and virtual efficiency (output/input) are found. And finally, the efficiency of DMU 0 is 11

14 compared to this virtual efficiency. The model is based on the idea that no other values for u * * r and v i can result in better efficiency, considering all DMUs. In addition, (1-h * 0 ) value represents the level of inefficiency (slack) (Charnes et al., 1978). The fractional model, shown in (1), can be converted into a linear programming (LP) model. The matrix notations for linear programming form are as follows: Maximize z 0 = u T Y 0 (2a) Subject to v T X j = 1 u T Y j v T X j 0 v T, u T 0 (2b) (2c) (2d) The rationale of this form is to fix weighted inputs to 1 (2b), and maximize output level (2a). The inequality, shown in (2c) is a linear representation of inequality in (1b). The dual version can be stated as follows: Minimize f = θ ε(e T s + + e T s - ) Subject to Yλ s + = Y j Xλ + s - = θx j λ, s +, s - 0 λ = (λ 1, λ 2,., λ n ) 0 represents the vector assigned individual DMUs, while s + and s - are vectors of addition input and output variables. In addition, e T = (1, 1, 1,., 1) and ε is non-archimedean constant, which is greater than zero (usually 10-6 or 10-8 ). Xλ and Yλ represent virtual inputs and outputs in the fractional model, respectively. The difference between virtual inputs and the inputs of DMU j is s -, which means DMU j is using s - amount of inputs in excess to reach its current output level. In the same way, s + is the difference in terms of output, which means DMU j can produce s + amount of more output with given inputs used (Vincona, 2005, p. 25). s + and s - are also called slack in the literature. If DMU j is efficient, then the value of θ is 1, and s + and s - are 0. There are two ways to run the CCR model, which are input-oriented and output-oriented versions. Martic et al. (2009) describe output oriented CCR as fixing the weighted outputs to 1 and measuring the amount of inputs each DMU has used. In this version, the DMUs which reach the same amount of outputs with a least amount of inputs are considered to be efficient and they act as an envelope for the inefficient ones. All the inefficient DMUs lie above the efficiency frontier. The input-oriented CCR is 12

15 exactly the opposite. In this version, the weighted inputs are equalized to 1 and efficiency is calculated based on the amount of outputs produced. In this case, the efficient DMUs form a concave frontier and the inefficient ones lie below the frontier (pp ). The charts below show two hypothetical cases for input and output-oriented CCR. For simplicity, one input and two outputs are included in the first graph, while two inputs and one output are chosen for the second one: Chart 7: Efficient Frontiers in Input and Output-Oriented CCR 10 8 Input-Oriented CCR Efficiency Frontier 10 8 Output-Oriented CCR Efficiency Frontier Output Input Output1 Input1 In input-oriented CCR, the efficiency of each DMU can be calculated by dividing its distance to origin to the distance of closer point in efficiency frontier to origin. On the other hand, in outputoriented CCR, the efficiency is the ratio of the distance of the closer point on frontier to origin to the distance of the DMU to origin. a. A Brief Example for Implementation of the CCR Method Assume that there are eight supermarkets in the town T, and we are analyzing their efficiency based on output-oriented CCR. We have data on two inputs (number of employees, and the width of the store) and one output (sale). Also assume that the data on inputs and output are as follows: S.market A B C D E F G H # of employees (x 1 ) Witdh of store (x 2 ) Sale First, we can reform the table to let the sale be equal to one, dividing inputs and output by the output value for each supermarket. S.market A B C D E F G H # of employees (x 1 ) Witdh of store (x 2 ) Sale

16 The data of supermarkets can be shown in the following chart: Chart 8: Efficiency Frontier in Supermarket Example witdh of the store number of employees A B C D E F G H According to output-based CCR, the units that use minimum amount of inputs to produce 1 unit of output are efficient compared to the others. Thus, the line combining A, G and H corresponds to the efficiency frontier. These supermarkets (A, G, and H) are considered to be efficient, while the others are not. The efficiency of inefficient units can be calculated by looking at the ratio of the distance of closer point in efficient frontier to origin divided by distant of DMU to origin. For instance, in the chart if we assume that OE line intersects at the point P, then the efficiency of E is OP/OE. By solving the maximization problem, we come up with the results below: Supermarket A B C D E F G H Efficiency 100% 83% 80% 91% 91% 92% 100% 100% To summarize, DEA is based on an idea to allocate weights for inputs and outputs in order to reach the maximum efficiency. For this reason, the inputs or outputs that decrease the efficiency have the minimum values. b. Implementation of CCR Model to Turkish Banks for year-2008 In this section, a sample empirical study will be run to calculate CCR efficiency and to show efficiency frontier for Turkish Banks in For simplicity of showing efficiency frontier in twodimensional plane, two inputs (external sources and capital) and one output (net income) is included in the study. The inputs and output values for 31 operating banks in 2008 is shown in the table

17 Chart 9: Efficiency Frontier in Banks-2008 Example Capital/Net Income Efficient Frontier External Sources/Net Income Three banks on the efficient frontier (3, 21, 29) operates with 100% efficiency, which means they generate a certain amount of net income by using the least (best) combination of external sources and capital. If we assume a line from origin to DMU, the efficiency of an inefficient DMU can be calculated by the distance of the intersection point in efficient frontier to origin, divided by distance of DMU to origin. 2. Banker, Charnes, Cooper Model of DEA (BCC) Another main model for DEA was introduced by Banker, Charnes and Cooper in The main difference between these two models is their assumptions. The CCR model assumes constant rate of return for inputs and outputs, while the BCC model considers the rate of return to be not only constant but also variable (increasing or decreasing). A variable rate of return can be added to the formula as a condition for convexity, which can be shown as e T λ = 1 (Vincona, 2005). Thus, the LP model of BCC can be written as: Minimize f = θ ε(e T s + + e T s - ) (3a) Subject to Yλ s + = Y j (3b) Xλ + s - = θx j (3c) e T λ = 1 (3d) λ, s +, s - 0 (3e) Constant rate of return, used in CCR model, assumes that a proportional change in inputs leads to same proportion of change in outputs. On the other hand, variable rate of return in BCC posits a 15

18 proportional change in inputs can result in not only constant but also decreasing or increasing change in outputs. The linear version of BCC model can also be shown as follows: Min h 0 = m i=1 v i x i0 + v 0 s u r r=1 y r0 (4a) Subject to: m i=1 v i x ij +v 0 s u r r=1 y rj 1 ; j = 1,., n, (4b) v r, u i 0 ; r = 1,., s ; i = 1,., m (4c) Introducing v 0 to linear equation (4a) makes the rate of return variable. v 0 is not limited by sign: v 0 = 0 - constant rate of returns to scale (CRS) v 0 > 0 - increasing returns to scale (IRS) v 0 < 0 - decreasing returns to scale (DRS) If v 0 > 0; then outputs will increase more than λ, under λ increases in inputs, and vice versa (Hahn et al, 2009). Introducing this new constraint (3d or 4b) removes the scale efficiency assumption of CCR, and allows model to measure only technical efficiency. In other words, if a DMU is efficient in CCR model, then it is both scale and technical efficient, but if it is efficient in BCC model, then it is only technical efficient (Varadi, 2004). Thus, an efficient DMU in CCR is also efficient in BCC, but the opposite is not always true. And similarly, the scale efficiency can be calculated by dividing CCR efficiency to BCC efficiency. IV. Literature Review In academic literature, there are numerous studies on the topic of efficiency analysis of banks. For example, Kablan (2007) investigates the technical and cost efficiencies of West African Economic Monetary Union banks, by using data, and finds out that during the period the efficiency of the banks in the region increases except for the Ivory Coast and Burkina Faso. Kablan (2007) also concludes that local banks with private capital are the most efficient ones, followed by foreign banks subsidiaries and state-owned banks. Fukuyama (2006) studied productivity and efficiency of Japanese banks using the non-parametric frontier approach, by using data (which corresponds to the bubble economy in Japan), and finds that efficiency decreases during this period also causes productivity losses, while productivity gains are caused by technological progress, but not efficiency increases. 16

19 Another study by Maudos, Pastor and Perez (2002) is about the competition and efficiency in Spanish banking sector for According to the results, size, qualification of productive factors, and productive specialization plays an important role in the efficiency of banks. Although the number of studies which analyses efficiency of Turkish Banks is quite appreciable, most of them concern the effect of liberalization after 1980s, and there are not many studies for post-2000 reforms. As one of them, Percin and Ayan (2006) studied the efficiency of commercial banks in Turkey by using DEA for the years and found a slight increase in efficiency. Oral and Yolalan (1990) concludes that Data Envelopment Analysis is a useful approach in reallocating resources between branches, after analyzing branch efficiency by using different combinations of inputs and outputs. Zaim (1995) analyzes the effects of liberalization policies on Turkish Banks, by using data. He finds an increase in both technical and allocative efficiency of the banks, and in the number of efficient banks after the liberalization process. Denizer, Dinc, and Tarimcilar (2000) studied the comparison of pre and post liberalization by using the same method with data. According to their result, after liberalization, banking efficiency increased until 1984, the relative efficiency of Turkish banking sector began to fluctuate after Isik and Hassan (2003) studied the same effect by using the Malmquist Productivity Index, using data of and find a positive and significant correlation between liberalization and efficiency (and also productivity) of the banks. V. An Empirical Study on the Efficiency of Turkish Banks after post-2000 reforms In this section of the study, the changes in efficiency of Turkish commercial banks will be analyzed, considering the economic reforms starting with The Transition Program for Strengthening Economy in May 2001, after the crises of 2000 and The selected inputs, outputs, and variables are among the most commonly used ones in the literature, and they have been selected after having considered the intermediary role of the banks. Off-balance sheet accounts of the banks, such as letters of guaranty, and letters of credits, have also been used as an output since their proportion has increased considerably after the integration of the Turkish economy to the world economy, especially after 1990s. In this study, software program DEAOS has been used. This program has been developed by Behin- Cara Co. Ltd. and is mostly being used in analyzing the efficiencies of big companies and high-budget projects. 17

20 As a first step of the study, the operating banks in the last 15 years have been detected, considering their ownership structure: state, domestic-private or foreign-private. The number of the operating banks during the period is shown in Chart 10. A. Inputs and Outputs Used In the Study Banks are intermediary institutions, which produce more than one output by using again more than one input. They supply a variety of services, such as collecting deposits, giving loans, selling unit trusts, offering foreign exchange services, surety services (letters of guaranties, letter of credits), intermediary services in security issues, etc. This product diversity makes it difficult to understand which one performs better. In such a case, selection of inputs and outputs becomes more vital for the study. Basically, there are three main approaches to select inputs and outputs, which are production, intermediation and modern approaches. Frexias & Rochet (1999) define the production approach as dividing operational costs (not including currency conversion losses and interest expenses) with the number of accounts within the bank. In this way, we can find the cost of operating one single account. On the other hand, they define intermediation approach as dividing operational costs (including currency conversion losses and interest expenses) with the amount of deposit or asset. In this way, we can find the cost of operating 1 Turkish Lira (TL) of deposit or asset (pp ). Since the production approach does not consider conversion losses and interest expenses and ignores the intermediation characteristic of banks, the intermediation approach is superior to the former one. In addition to these two approaches, Hughes&Mester (1993 and 1994) developed the modern approach, which also considers conflict of interest between executive managers and shareholders, and the risk aversion level of banks executive managers. This approach is quite useful, but also difficult to obtain since it may not be possible to measure risk aversion level of all executive managers during the period. Thus, in this study, the intermediation approach has been used for selecting inputs and outputs. Number of employees, external sources (liabilities, excluding suspense accounts) and capital are selected as inputs while loans, other interest-bearing assets, net income, and off-balance sheet accounts (including letters of guaranties and letters of credits) are outputs. The reason for this selection is that these are the main variables of what banks produce (outputs) and what they use to produce (inputs). The first input used in the study is the amount of external sources of finance, mostly composed of deposits, loans from the interbank money market and the central bank, liabilities of repurchase agreements, and bonds issued. These are the sources of finance that a bank is trying to transfer to the market by giving loans, or purchasing investment instruments. The second one is the internal source of finance, which is capital. And finally, number of employees has been chosen as final input for the 18

21 study, since it was clearly stated as a major action plan in the Banking Sector Restructuring Program. The first output used is loans given, which composes major items in assets and has a big importance since it is the main interest income source for a bank. The second one is net income, which is one of the major key performance measurements not only for banks, but also for all profit organizations. The third input is other interest-bearing assets except loans, such as deposits in other banks and other financial institutions, receivables from repurchase agreements, securities invested. And finally, the fourth one is off-balance sheet accounts; mostly consisting of non-cash loans, such as letters of guaranty, and letters of credits, which creates non-interest income for the banks. One can argue about why deposit accounts, number of branches etc have not been included in the inputs, and interest income, non-interest income, total assets as outputs. Deposit accounts is already included in liabilities and are not only sources of external sources. Loans from interbank markets and the central bank, securities issued, and syndicated loans are also important items in external sources (liabilities) of the balance sheet. Besides, number of branches is directly related with the banks strategy, and would distort the effect of number of employees if used. Since net income includes both interest and non-interest income, they have not been included seperately. Finally, total assets is not taken into consideration, because it is a natural result of the sum capital and liabilities. In other words, a bank which has higher capital and liabilities will automatically have higher total assets, no matter the level of its productivity and efficiency performance. In this study, the CCR and BCC models, two major models of data envelopment analysis have been used. In the first phase, technical efficiency has been calculated by using the CCR, then secondly, the BCC for local technical efficiency (excluding scale efficiency). Another important criterion for an effective efficiency analysis is that all decision-making units need to be homogeneous. For this reason, not all types of banks, but only commercial banks are included in this study. Since the investment and development banks have different balance-sheet structures and operational activities, they have not been included in this study. For instance, these banks do not have permission to collect deposits, while commercial banks do. This discrepancy causes a noteworthy difference in cost of finance and would create an unbalanced comparison, and lead to incorrect results, if investment and development banks were added to the study. At the end of 2012, assets of commercial banks composed of 96% of the total assets in the Turkish banking sector. Taking this fact into account, this sample is big enough to obtain general view of the Turkish banking sector. 19