American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) ISSN (Print) 2313-4410, ISSN (Online) 2313-4402 © Global Society of Scientific Research and Researchers http://asrjetsjournal.org/ The Effect of Bank Size and Bank Capital on the Bank Lending Channel for Turkish Banks Gökhan Meral* Koç University, Ilıca Mah. Sarisin Sk. No:2/9 Narlidere, Izmir 35320, Turkey Email: gokmeral@gmail.com Abstract This study investigates the existence of a bank lending channel in Turkey over the estimation period 2002Q4- 2008Q4. Consistent with the bank lending channel, I search for evidence that small, undercapitalized banks are more sensitive to changes in monetary policy than large, well-capitalized banks. Results suggest that small, undercapitalized banks find it harder to raise time deposits than large, well-capitalized banks which support the existence of a bank lending channel however the insignificant relationship between the change in the monetary policy indicator (Central Bank overnight rates) and the growth rate of bank loans does not support the existence of a bank lending channel. Keywords: Monetary Transmission Mechanism; Credit Channel, Bank Lending Channel; Capital Leverage Ratio; Contractionary Monetary Policy 1. Introduction A major concern in macroeconomics has been the determination of real economic activity, including its dimensions like output, employment, income, and spending. To be able to understand the movements in real economy and its dimensions, monetary policy makers must accurately assess the timing and effect of their policies on the economy. In order to make this assessment, policymakers should understand the monetary transmission mechanisms through which monetary policy affects the economy. Economic research goes in two directions in the literature studying monetary transmission mechanisms. The first direction followed is the money view (or the interest rate channel), and the second direction is the credit view (or the credit channel) [1]. ------------------------------------------------------------------------ * Corresponding author. E-mail address: gokmeral@gmail.com. 1 http://asrjetsjournal.org/ mailto:example@yahoo.com American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 This study focuses on the credit view (credit channel) of monetary transmission mechanism, and henceforth attention will be focused on this view. First, to accurately define the credit view, one should understand the ‘money view’ of neoclassical economics and the differences between the two competing views of monetary policy in detail. The traditional ‘money view’ of the monetary transmission mechanism is based on the so-called money or interest rate channel, stemming from the standard Keynesian IS-LM framework. According to this framework, a contractionary monetary policy leading to a rise in real interest rates, which in turn increases the cost of capital, causes a decline in investment spending, thereby leads to a decrease in aggregate demand and a decrease in output [1]. The interest rate channel operates through the liability side of the banks’ balance sheets. When the central bank drains reserves from the banking system, this action lowers the demand deposits (DD) of banks and the money supply. The increase in interest rates and the reduction in money supply depress the economy, and if bank loans fall it is due only to a reduction in loan demand triggered by the economic depression. This is how the money or interest channel works [2]. The money view sees banks as passive intermediaries for funds between a central bank and borrowers. In contrast, the credit channel puts special emphasis on the role of banks in the aggregate economic activity. Assuming the existence of bank-dependent borrowers, the credit view states that a change in monetary policy (either an increase or a decrease in open-market interest rates) tends to change the external cost of financing in the same direction for bank-dependent borrowers. The money channel tries to explain the effect of policy through changes in the money supply-which change interest rates and spending in turn-, however if there exist i) substantial information-related frictions in financial markets, and ii) there exist borrowers those are primarily bank-dependent, then the credit channel provides a more plausible explanation for the transmission of monetary policy through banks to the real economy. According to the credit view, monetary policy shocks may affect the real economic activity in two ways; first, a policy shock can influence the net worth of borrower firms (increasing their liabilities and decreasing their assets) and affect their ability to borrow. Monetary policy can affect firms’ financial positions both directly and indirectly. Policy may have two direct affects; first it may increase the interest payments on firms’ outstanding debt, second it may influence the value of their collateral assets. Indirectly, policy may affect the spending of firms’ customers and cause a decline in firm revenue. This mechanism of the credit channel is called the borrower net worth channel (BNWC) or the balance - sheet channel. The second mechanism through which the credit channel operates is the bank lending channel (BLC). BLC works as follows; contractionary policy decreases reserves and the decrease in reserves decreases demand/reservable deposits (DD), if some banks can not offset this decline in DD by increasing non- reservable/large time deposits (TD) or by decreasing securities holdings, then the contractionary policy will decrease their loan supply. Decrease in loan supply will affect bank-dependent borrowers and reduce their investment and consumption expenditures [3]. Existence of the BLC require three conditions: (1) existence of bank-dependent, high information cost borrowers, (2) existence of at least some banks which can not cover a decrease in demand deposits frictionlessly by raising large time deposits or by reducing securities (in other words monetary policy can shift the loan 2 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 supplies of at least some banks), and (3) prices must be sticky, so that monetary policy can have real effects on the economy without being neutralized. The last condition is necessary for both the money and the credit views. Depending on banks’ cross-bank characteristic differences (for example asset size or liquidity), their sensitivity to monetary policy may be different. This study tries to find evidence on the existence of a bank lending channel in Turkey between 2002Q4:2008Q4, dividing banks both by asset size and by capital leverage ratio (equity capital/assets). The hypothesis to be tested is; small, undercapitalized banks are more responsive to monetary policy than large, well-capitalized banks. The intuition behind this hypothesis can be summarized as follows: small, undercapitalized banks find it more difficult to raise time deposits during periods of contractionary policy compared to large, well-capitalized banks, hence they may not be able to offset a drain in reserves as easily as large, well-capitalized banks can. This argument is supportive for the BLC [2]. This study contributes to the credit view literature in Turkey in two ways. First, to my knowledge, this is the first study that considers both asset sizes and capital leverage ratios of Turkish banks while investigating the existence of a bank lending channel in Turkey. Second, results including the effect of bank capital on the transmission of monetary policy underline the importance of regulatory policy on banking during economic crisis. The following section discusses the BLC literature that motivates my study. 2. Literature Review Bernanke and Blinder [4] developed a simple model of aggregate demand, which is a variant of the textbook IS/LM model, aimed to compare the effects of money and credit (bank loans) on real economic activity. They find that money-demand shocks became more important relative to credit-demand shocks in the 1980s, and they suggest a more symmetric treatment of money and credit. Bernanke and Blinder [5] use aggregate bank data from 1959 to 1978 to identify a decline in bank loans after contractionary monetary policy. However, loans respond with a significant lag (6-9 months) and interestingly this response roughly corresponds to the response of unemployment. They conclude that the co-movement of unemployment and bank loans following a change in the funds rate is consistent with the credit view, hence monetary policy works at least in part through the credit channel as well as through the interest rate channel. There are studies favoring the money view and studies favoring the credit view. Romer, Romer, Goldfeld and Friedman [6] find two types of evidence supporting the money view; first, banks are able to raise funds with little cost during tightening periods, so the impact of restrictive policy on bank lending will be small. Second; because reserve requirements on bank deposits are high, contractionary policy is more likely to operate through bank liabilities (deposits) rather than bank assets (bank lending). Bernanke and Gertler [7] do not see the credit channel as a distinct alternative to the interest rate channel. Instead they see it as a complementary tool to the conventional interest rate effects which help explaining the magnitude, timing and composition of policy effects on the economy in a more reasonable way. Using vector auto-regressions (VARs), they try to identify the effects of policy on economy by looking at output, demand and spending responses to monetary policy shocks. They point out shortcomings of the conventional money view in explaining these responses. Bernanke and Gertler [7] also investigate the housing market in the U.S., focusing on the borrowing and spending decisions of households, particularly their spending on costly durable items such as houses. They suggest that the high 3 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 sensitivity of housing investment to monetary policy shocks may be explained in part by the workings of both the bank lending and the balance sheet channels. They argue that the bank lending channel may have been in effect especially during the Regulation Q period when there were funding constraints for banks because of the imposed deposit rate ceilings. Their argument is that the deposit funding constraint for banks may have played a role in the decline of housing investment (which accounts for the largest fraction of the decline in economic activity) by disrupting the flow of mortgage loans. They also argue that the balance sheet channel may be another plausible explanation for the decrease in housing investment such that, consumer balance sheets having been deteriorated by increasing down-payment requirements and transactions costs may have caused a decline in housing demand and hence a decline in demand for mortgage loans. The BLC literature includes studies focusing on different credit aggregates, or on balance sheets of firms. It is hard to distinguish between loan supply and loan demand effects looking at the aggregate data. Kashyap, Stein and Wilcox [8] use a mix of bank loans and commercial papers-issued by firms- to provide evidence of a credit channel. They argue that if the monetary policy operates solely through the money channel and if the bank loans fall after a tightening due only to a decrease in credit demand, then the demand for non-bank sources of credit- for example, the commercial paper demand- should decline as well. On the other hand, if the monetary policy operates through the credit channel, then a tightening reduces the supply of bank credit while it increases commercial paper issuance, to the extent that businesses have some ability to substitute between the two sources of finance. They find that shifts in monetary policy change the mix of loans and commercial paper where commercial paper issuance rises and bank loans fall. In addition, those shifts in loan supply seem to affect investment even controlling for interest rates and output. These results support the existence of a bank lending channel. Morgan [9] contributes to the literature on the different credit aggregates by comparing commitment and non-commitment loans where a loan commitment is defined as a contract which obligates banks to lend to a firm up to some limit, for some length of time, at specified terms. He finds that tight policy slows the growth of non-commitment loans relative to commitment loans. This finding coincides with reports of tighter credit by lenders and by smaller firms suggesting the divergence between commitment and non-commitment loans may reflect either a lending channel or a balance sheet channel of policy. Disentangling declines in bank loans resulting from reduced loan demand from declines resulting from reduced loan supply has been the major concern in the BLC literature. Using micro-bank level-data is an attempt to overcome this problem in aggregate data. There are papers including Kashyap and Stein [10] that look for balance sheet effects by comparing the impact of monetary policy on the loan supply of small and large banks. Kashyap and Stein [10] separate banks by both asset size and liquidity (securities to asset ratio). Using quarterly data of U.S. commercial banks from 1976 to 1993; they find that small, less liquid banks are most responsive to monetary policy. Although they find evidence of a lending channel, the results do not allow them to make precise statements about the quantitative importance of the BLC in real economy. The “credit crunch” studies in the monetary transmission mechanism literature focuses on the effect of bank capital on loan and output growth. Peek and Rosengren [11] find that poorly capitalized banks contracted more during New England banks’ credit crunch. Peek and Rosengren [12] provide evidence that undercapitalized banks are more engaged with derivatives activity and eager to take speculative positions which makes them 4 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 more vulnerable to a monetary policy shock. Berger and Udell [13] examine how risk-based capital (RBC) regulations affected the reallocation of U.S. commercial bank credit from loans to securities. Motivated by the above credit crunch literature, Kishan and Opiela [2] choose bank capital among the alternative bank balance sheet characteristics-that affect bank loan supplies- as a constraint on banks’ ability to fund loans. They separate banks according to asset size and capital leverage ratio and provide evidence that the loan growth of small, undercapitalized banks; small, adequately capitalized banks and small, well-capitalized banks is significantly affected by monetary policy. They contribute to the literature in three ways; first, they suggest that policy has distributional effects on loan supplies of banks with different asset sizes and capital leverage ratios. Second, they find that the growth in large time deposits for small undercapitalized banks is unresponsive to policy which they use as evidence favoring the BLC against bank net worth channel (BNWC). In their more recent paper, Kishan and Opiela [14] investigate loan responses of low and high capital banks in expansionary and contractionary periods separately. They find that contractionary policy decreases the loans of small low-capital banks, and expansionary policy is not able to increase the loan growth of the low-capital banks relative to the high capital banks. Erdoğan, Beşballı [15], Çavuşoğlu [3] and Işık [16] are some of the studies searching for evidence of a bank lending channel in Turkey. Çavuşoğlu [3] uses dynamic panel data modeling with Generalized Method of Moments estimations to empirically test the existence of a bank lending channel in Turkey. He does not find evidence of a bank lending channel in Turkey over the period 1988-1999, this outcome is reflected in the lack of a significant relationship between the change in the monetary policy indicator and the growth rate of the loan supply in the estimated models. Separating banks by asset size does not improve his results. Erdoğan, Beşballı [15] use VAR methodology to empirically analyze the BLC between 1996 and 2006, according to their findings the credit channel operates partially in Turkey. Işık [16] investigates the bank loan supply behavior of Turkish banks over the 1998Q1-2007Q3 period. She divides banks by both asset size and liquidity ratio and finds that small banks with less liquid balance sheets are more sensitive to monetary policy shock and takes this as evidence of a broad credit channel. I follow the above literature and use bank-level data in order to identify loan supply movements of banks distinguished by their asset-size and capital leverage ratio. I expect that there is evidence of a bank lending channel in the case that following a monetary tightening, the loan supply of small and/or less capitalized banks shrinks more than the loan supply of large and/or well-capitalized banks. However, in Turkey it is hard for banks in general to find external sources of funding that is why it may not be surprising if the loan supply movements of small banks do not differ significantly from the loan supply movements of large banks. This fact suggests that the results of this study will be conservative. This study is mainly motivated by Kishan and Opiela [2] paper, and it expands the methodology used by them with panel pooled OLS fixed effects models. First, I divide banks by both asset size and leverage ratio and form four groups. Second, I divide banks by asset size and leverage ratio separately forming two groups for each criteria. Finally, I (1) consider a regression for the pooled sample adding dummies for capital and asset size. (2) I then consider a panel regression to control for bank-specific effects on the supply of bank loans. I look for evidence of a credit channel in general, and a bank lending channel in particular. To my knowledge, this study is the first to investigate the existence of a BLC in Turkey that distinguishes banks by both asset size and capital leverage ratio. Before moving on to the details of the methodology used, next section provides general 5 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 information about the Turkish banking sector over the estimation period 2002-2008. 3. Turkish Banking Sector 2002-2008 Turkish economy experienced a stable and high growth performance between 2002 and 2008, except for a rapid deceleration in the last quarter of 2008. Gross domestic product (GDP) followed an increasing trend in general. Total consumption expenditure followed a similar pattern with GDP. Figure 1 and Figure 2 show quarterly values of these two macro indicators and Figure 3 shows annual change in investment and consumption over the period 2002Q4-2008Q4. Figure 1: Gross Domestic Product (2002-2008) Figure 2: Consumption Expenditure (2002-2008). Data Source: Central Bank of Turkey 0 50000 100000 150000 200000 250000 02 /4 03 /1 03 /2 03 /3 03 /4 04 /1 04 /2 04 /3 04 /4 05 /1 05 /2 05 /3 05 /4 06 /1 06 /2 06 /3 06 /4 07 /1 07 /2 07 /3 07 /4 08 /1 08 /2 08 /3 08 /4 M ill io n € Quarters GDP 0 20000 40000 60000 80000 100000 120000 140000 160000 180000 02 /4 03 /1 03 /2 03 /3 03 /4 04 /1 04 /2 04 /3 04 /4 05 /1 05 /2 05 /3 05 /4 06 /1 06 /2 06 /3 06 /4 07 /1 07 /2 07 /3 07 /4 08 /1 08 /2 08 /3 08 /4 M ill io n € Quarters Consumption Expenditure 6 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 Figure 3: Annual change in Consumption and Investment (in constant price) As banks are the major source of funding and investment, economy’s expansionary pattern between 2002 and 2008 brought about an increase in both the bank loans to GDP and the bank deposits to GDP ratios. Figure 4 presents these two ratios annually. Figure 4: Annual Loans/GDP and Deposits/GDP ratios. Source: "Banks in Turkey 2008", The Banking Association of Turkey 7 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 Total number of banks in Turkey was 45 as the end of December, 2008. Out of 45 banks, 32 were deposit banks, and 13 were development and investment banks. There was also one bank owned by Saving Deposits and Insurance Fund (SDIF). When the balance sheet components of banks are analyzed, we see an upward trend in total deposits, securities and loans. Figure 5 shows quarterly movements in total deposits and securities over the period 2002Q4-2008Q4. Figure 5: Total Deposits and Securities. Data Source: Turkish Banks Association 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 02 /4 03 /1 03 /2 03 /3 03 /4 04 /1 04 /2 04 /3 04 /4 05 /1 05 /2 05 /3 05 /4 06 /1 06 /2 06 /3 06 /4 07 /1 07 /2 07 /3 07 /4 08 /1 08 /2 08 /3 08 /4 M ill io n € Quarters Total Deposits 0 500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 02 /4 03 /1 03 /2 03 /3 03 /4 04 /1 04 /2 04 /3 04 /4 05 /1 05 /2 05 /3 05 /4 06 /1 06 /2 06 /3 06 /4 07 /1 07 /2 07 /3 07 /4 08 /1 08 /2 08 /3 08 /4 M ill io n € Quarters Securities 8 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 Figure 6: Bank Loans versus Overnight Rates. Data Source: Turkish Banks Association Figure 6 shows quarterly movements of bank loans and Central Bank of Turkey’s overnight rates. When we analyze the movements of bank loans and overnight rates; during the first five quarters, there is a decline in the overnight rates; on the other hand, loans follow an upward trend during the period and in the following five quarters. When we look at the third quarter of 2006, we see a jump in the overnight rate, however loans do not decline in that contractionary period, in contrast they follow an upward trend. In general, loans follow an upward trend through the 25 quarters from 2002Q4 to 2008Q4 and overnight rates follow a downward trend. 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 02 /4 03 /1 03 /2 03 /3 03 /4 04 /1 04 /2 04 /3 04 /4 05 /1 05 /2 05 /3 05 /4 06 /1 06 /2 06 /3 06 /4 07 /1 07 /2 07 /3 07 /4 08 /1 08 /2 08 /3 08 /4 M ill io n € Quarters Total Loans 0 10 20 30 40 50 60 02 /4 03 /1 03 /2 03 /3 03 /4 04 /1 04 /2 04 /3 04 /4 05 /1 05 /2 05 /3 05 /4 06 /1 06 /2 06 /3 06 /4 07 /1 07 /2 07 /3 07 /4 08 /1 08 /2 08 /3 08 /4 % Quarters Overnight Rates 9 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 Table 1: Bank mean balance sheet ratios, 2002:Q4-2008:Q4 (I) (II) Bank K/A Ratio Category Small banks(Assets<8 billion €) Large banks(Assets>8 billion €) Mean-All banks Number of Banks 25 7 As a Fraction of Total Assets 1. Securities 0.533 0.396 2. Total loans 0.387 0.371 3. Total Deposits 0.478 0.680 4. Demand Deposits 0.085 0.108 5. Large Time Deposits 0.402 0.573 6. Capital 0.214 0.118 7. Loan delinquency rate 1.638 0.103 8. Correlation coefficient between delinquency rate and capital/asset 0.714 -0.381 Mean-K/A<0.13 Number of banks 10 5 As a Fraction of Total Assets 9. Securities 0.412 0.369 10 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 10. Total loans 0.481 0.362 11. Total Deposits 0.596 0.705 12. Demand Deposits 0.099 0.106 13. Large Time Deposits 0.497 0.599 14. Capital 0.109 0.105 15. Loan delinquency rate 0.035 0.126 16. Correlation coefficient between delinquency rate and capital/asset -0.323 0.048 Mean-K/A>0.13 Number of banks 15 2 As a Fraction of Total Assets 18. Securities 0.614 0.464 19. Total loans 0.293 0.394 20. Total Deposits 0.400 0.618 21. Demand Deposits 0.075 0.112 22. Large Time Deposits 0.338 0.507 23. Capital 0.283 0.150 24. Loan delinquency rate 2.783 0.047 11 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 25. Correlation coefficient between delinquency rate and capital/asset 0.737 - To analyze the cross-sectional differences in bank lending decisions, banks are divided into two asset-size categories and two capital leverage ratio groups. Banks with an asset size larger than 8 billion € are considered large banks where the others are considered small banks. Meanwhile banks with an equity capital to assets ratio greater than 13% are considered as well-capitalized, the others are considered as undercapitalized. This methodology is taken from Kishan and Opiela [2] who divide banks into six asset-size categories and three capital leverage ratio groups. This study divides banks into smaller number of categories because the number of banks available for analysis is far less than the number of banks used by Kishan and Opiela [2]. Table 1 shows the mean value of some basic balance sheet variables for each asset-size category for all banks and for the same size categories for banks divided into two leverage ratios. There are several points worth emphasizing: (1) Securities to assets ratio decreases significantly when we move from the small banks to the large banks in both of the leverage ratio groups, (compare ninth row first and second columns and compare eighteenth row first and second columns) (2) Loans to assets ratio is close for small and large banks in the whole sample, (second row column one and two) however for the undercapitalized group, loans/assets ratio decreases, (compare tenth row first and second columns) whereas for the well-capitalized group this ratio increases (compare nineteenth row first and second columns) while moving from the small banks to the large banks, (3) Small banks depend less on deposits funding than large banks (third row first and second columns) and well- capitalized banks depend less on deposits funding than undercapitalized banks, (compare first column eleventh and twentieth rows, compare second column eleventh and twentieth rows) (4) Demand deposits (DD) has the least importance in funding loans and DD/assets ratio is quite similar for all asset-size and leverage ratio groups, (look at twelfth row first and second columns, and twenty-first row first and second columns) and well-capitalized banks are less dependent on time deposits (TD) (compare first column thirteenth and twenty-second rows, and compare second column thirteenth and twenty-second rows) (2) is not compatible with the bank characteristics of Kishan and Opiela (2000) sample. Regardless of leverage ratio, loans to assets rise with bank size in their sample. (3) is not compatible with Kishan and Opiela (2000) bank sample, either. Their implication is that larger and better capitalized banks are less dependent to deposits than smaller and less-capitalized banks. Kishan and Opiela (2000) bank sample implies that large, well-capitalized banks depend less on DD and more on TD however in this sample large, undercapitalized banks depend on TD more [implication (4)]. Having observed the general outlook of the economy, the banking sector and the bank balance sheet 12 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 characteristics in Turkey over the estimation period 2002Q4-2008Q4, now we are ready to move to the methodology section of the study. 4. Methodology and Data This study is mainly motivated by the methodology used by Kishan and Opiela [2]. Consistent with the bank lending channel, my hypothesis is that an increase (decline) in overnight rates should decrease (increase) the loan supply of banks and this decline should be more pronounced for the small and less-capitalized banks. Although an increase in overnight rates is expected to increase time deposits, loans will still fall because the increase in time deposits will not be high enough to cover the decrease in demand deposits. The testable hypotheses are; ∂LN/∂rON<0 and ∂TD/∂rON>0; where LN, TD and rON stand for loans, time deposits and Central Bank of Turkey overnight rates in order. The data is obtained from the Turkish Banking Association and from the Central Bank of Turkey. The data set consists of the overnight interest rates, and the bank level variables; total loans, demand deposits, large time deposits, securities and capital for 25 quarters over the period 2002Q4-2008Q4. This section is divided into two sub-sections. In the first sub-section, the methodology used by Kishan and Opiela [2] is introduced and it is estimated with a few modifications. The second sub-section, utilizes panel pooled OLS estimation model with the fixed effects estimation technique. 4.1. Ordinary Least Squares Model Banks are separated into four groups by both asset-size and capital leverage ratio, and then they are divided into two groups first by asset-size and then by capital leverage ratio separately. The variables used in the first step regressions of the methodology include; grLN (growth rate of loans), grLNt−1,..., grLNt−4 (four lagged values of grLN), chratet−1,.., chratet−4 (four lagged values of the change in overnight rates), grTD (current period growth in large time deposits), grSEC (current period growth in securities), plus three seasonal dummy variables (D1, D2, D3). The econometric model is; 𝑔𝑟𝐿𝑁 = 𝛼_𝑖𝑡 + 𝛽_1 (�𝑔𝑟𝐿𝑁�_(𝑡 − 1) ) + 𝛽_2 (�𝑔𝑟𝐿𝑁�_(𝑡 − 2) ) + 𝛽_3 (�𝑔𝑟𝐿𝑁�_(𝑡 − 3) ) + 𝛽_4 (�𝑔𝑟𝐿𝑁�_(𝑡 − 4) ) + 𝛽_5 (�𝑐ℎ𝑟𝑎𝑡𝑒�_(𝑡 − 1) ) + 𝛽_6 (�𝑐ℎ𝑟𝑎𝑡𝑒�_(𝑡 − 2) ) + 𝛽_7 (�𝑐ℎ𝑟𝑎𝑡𝑒�_(𝑡 − 3) ) + 𝛽_8 (�𝑐ℎ𝑟𝑎𝑡𝑒�_(𝑡 − 4) ) + 𝛽_9 (𝑔𝑟𝑇𝐷) + 𝛽_10 (𝑔𝑟𝑆𝐸𝐶) + 𝛽_11 (𝐷_1 ) + 𝛽_12 (𝐷_2 ) + 𝛽_13 (𝐷_3 ) + 𝜀_𝑖𝑡 (1) Using the Ordinary Least Squares (OLS) estimation technique, the growth rate of loans is regressed on the above mentioned variables. First, pooled OLS estimation technique is used to capture the difference in the sum of coefficients of the (lagged) change in overnight rate variables for different capital leverage and asset size groups. Second, this estimation technique is used to see the change in policy response for only asset size classes. Third, the same estimation technique is used to see the change in policy response for only leverage ratio classes. αit captures the bank level fixed effects. Both αit and εit are assumed to be independently and identically distributed. For stationarity, growth rate of the variables are used. Time deposits and securities are included to 13 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 capture the ‘funding’ effects on loans. For each group of banks, the OLS estimations are run separately and one coefficient is reported per group in the results. The main interest is on the sum of the coefficients of 𝑐ℎ𝑟𝑎𝑡𝑒𝑡−1,.., 𝑐ℎ𝑟𝑎𝑡𝑒𝑡−4 (𝛽5+𝛽6+𝛽7+𝛽8) and on the coefficient of 𝑔𝑟𝑇𝐷, 𝛽9. Consistent with my hypothesis, a negative sign associated with the overnight rate coefficient will imply that an increase in overnight rates will cause a decline in loans and a positive sign associated with the 𝑔𝑟𝑇𝐷 coefficient will imply that an increase in time deposits will affect loans positively. For small and undercapitalized banks, the sum of the coefficients of the lagged values of overnight rates is expected to be negative and the coefficient of the growth in time deposits is expected to be positive. For large and well-capitalized banks, sum of the coefficients are expected to be positive (less negative) and the coefficient of 𝑔𝑟𝑇𝐷 is expected to be positive and more significant. The response of securities to changes in the overnight rates is indeterminate. Banks may choose to sell securities to fund loans or time deposits may be used for funding loans and securities may increase to balance asset risk. Hence, 𝛽10 may be either negative or positive [2]. 4.2. Panel Ordinary Least Squares Model In this section, I use a panel regression approach with pooled OLS estimation technique and fixed effects models. Unlike the previous sub-section, banks are not grouped and all banks are regressed at once. The aim of this method is to get more accurate econometric results with the small data set used (which consists of 30 banks). In order to observe the policy responses of different asset-size and capital leverage ratio classes, slope dummies are added to the regressions. By clustering banks, panel robust standard errors are obtained. Three different models are used; (i) first-difference model, (ii) log model, and (iii) log difference model. The first difference model is; 𝐿𝑁𝑡 − 𝐿𝑁𝑡−1 = 𝛽1[(𝐿𝑁𝑡 − 𝐿𝑁𝑡−1)𝑡−1] + 𝛽2[(𝐿𝑁𝑡 − 𝐿𝑁𝑡−1)𝑡−2] + 𝛽3[(𝐿𝑁𝑡 − 𝐿𝑁𝑡−1)𝑡−3] + 𝛽4[(𝐿𝑁𝑡 − 𝐿𝑁𝑡−1)𝑡−4] + 𝛽5[(𝑟𝑡 − 𝑟𝑡−1)𝑡−1] + 𝛽6[(𝑟𝑡 − 𝑟𝑡−1)𝑡−2] + 𝛽7[(𝑟𝑡 − 𝑟𝑡−1)𝑡−3] + 𝛽8[(𝑟𝑡 − 𝑟𝑡−1)𝑡−4] + 𝛽9(𝑇𝐷𝑡 − 𝑇𝐷𝑡−1) + 𝛽10(𝑆𝐸𝐶𝑡 − 𝑆𝐸𝐶𝑡−1) + 𝛽11(𝐷1) + 𝛽12(𝐷2) + 𝛽13(𝐷3) + 𝜀𝑖𝑡 (2) In the first difference model, change in loans (LNt − LNt−1) is regressed on four lagged values of itself, four lagged values of the change in overnight rates (rt − rt−1) , change in TD (TDt − TDt−1) , change in SEC (SECt − SECt−1), plus three seasonal dummy variables. The log model is; 𝑙𝑛 𝐿𝑁𝑡 = 𝛽1[(𝑙𝑛 𝐿𝑁𝑡)𝑡−1] + 𝛽2[(𝑙𝑛 𝐿𝑁𝑡)𝑡−2] + 𝛽3[(𝑙𝑛 𝐿𝑁𝑡)𝑡−3] + 𝛽4[(𝑙𝑛 𝐿𝑁𝑡)𝑡−4] + 𝛽5[(𝑟𝑂𝑁)𝑡−1] + 𝛽6[(𝑟𝑂𝑁)𝑡−2] + 𝛽7[(𝑟𝑂𝑁)𝑡−3] + 𝛽8[(𝑟𝑂𝑁)𝑡−4] + 𝛽9(𝑙𝑛 𝑇𝐷𝑡) + 𝛽10(𝑙𝑛 𝑆𝐸𝐶𝑡) + 𝛽11(𝐷1) + 𝛽12(𝐷2) + 𝛽13(𝐷3) + 𝛽14(𝐷𝑦1) + 𝛽15(𝐷𝑦2) + 𝛽16(𝐷𝑦3) + 𝛽17(𝐷𝑦4) + 𝛽18(𝐷𝑦5) + 𝜀𝑖𝑡 (3) In the log model, log of loans (ln LNt) is regressed on four lagged values of itself, four lagged values of the overnight rates (rON), log of TD (ln TDt), log of SEC (ln SECt), three seasonal dummies and also on five yearly dummies for the 6 year estimation period. The final model is the log difference model; 14 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 𝑙𝑛 𝐿𝑁𝑡 − 𝑙𝑛 𝐿𝑁𝑡−1 = 𝛽1[(𝑙𝑛 𝐿𝑁𝑡 − 𝑙𝑛 𝐿𝑁𝑡−1)𝑡−1] + 𝛽2[(𝑙𝑛 𝐿𝑁𝑡 − 𝑙𝑛 𝐿𝑁𝑡−1)𝑡−2] + 𝛽3[(𝑙𝑛 𝐿𝑁𝑡 − 𝑙𝑛 𝐿𝑁𝑡−1)𝑡−3] + 𝛽4[(𝑙𝑛 𝐿𝑁𝑡 − 𝑙𝑛 𝐿𝑁𝑡−1)𝑡−4] + 𝛽5[(𝑟𝑡 − 𝑟𝑡−1)𝑡−1] + 𝛽6[(𝑟𝑡 − 𝑟𝑡−1)𝑡−2] + 𝛽7[(𝑟𝑡 − 𝑟𝑡−1)𝑡−3] + 𝛽8[(𝑟𝑡 − 𝑟𝑡−1)𝑡−4] + 𝛽9(𝑙𝑛 𝑇𝐷𝑡 − 𝑙𝑛 𝑇𝐷𝑡−1) + 𝛽10(𝑙𝑛 𝑆𝐸𝐶𝑡 − 𝑙𝑛 𝑆𝐸𝐶𝑡−1) + 𝛽11(𝐷1) + 𝛽12(𝐷2) + 𝛽13(𝐷3) + 𝜀𝑖𝑡 (4) In the log difference model, log change in loans (ln LNt − ln LNt−1) is regressed on four lagged values of itself, four lagged values of the change in overnight rates (rt − rt−1), log change in TD (ln TDt − ln TDt−1), log change in SEC (ln SECt − ln SECt−1) and three seasonal dummy variables. Sum of the coefficients of lagged values of overnight rates (𝛽5+𝛽6+𝛽7+𝛽8) is expected to be negative and the coefficient of time deposits (𝛽9) is expected to be positive in all three models. When banks are divided into asset size and capital leverage ratio categories, the overnight rate coefficient associated with the small bank group is expected to be negative and greater in absolute value than the large bank group. Similarly, the overnight rate coefficient associated with the undercapitalized bank group is expected to be negative and greater in absolute value than the well-capitalized bank group. In addition, the time deposits coefficient is expected to be less positive for the small and the undercapitalized bank groups which will reflect the severity of funding loans by raising time deposits for small and undercapitalized banks. If these coefficients end up being significant with the predicted signs, then results will provide evidence of a bank lending channel. There are some important points to mention about these models. First, the first-difference and the log-difference models drop the bank-specific effects variable (αit). This helps to relieve the strong assumption in the previous sub-section that αit is i.i.d. Second, by estimating the system as a panel dataset, all models will assume that the error term may change between different banks. Stationarity is still assumed as a time series property of the data. Finally, in the log model yearly dummies are added to control for the annual economic fluctuations. There are two effects of this modification; it increases the R-squared values in estimation results significantly, however it decreases the degrees of freedom. In the difference models-both first difference and log difference- the yearly dummies drop out in the same way the bank-level fixed effects do. 5. Emprical Results This section is divided into two parts. The first sub-section shows the results of the OLS estimation for four groups of banks (divided by both asset-size and leverage ratio), and 2 groups of banks divided by asset-size and 2 groups of banks divided by leverage ratio. The second sub-section presents the results for panel pooled OLS (fixed effects) estimation for the first-difference, the log, and the log difference models. Table 2 below shows the number of banks grouped by both asset size and capital leverage ratio. Number of banks in each group of the 4-group regression is as follows; 10 banks in the small, undercapitalized category; 5 banks in the large, undercapitalized category; 13 banks in the small, well-capitalized category and 2 banks in the large, well-capitalized category. Number of banks in each group of the 2-group asset-size based regression is as follows; 23 banks in the small category and 7 banks in the large category. Number of banks in each group of the 2-group leverage ratio based regression is as follows; 15 banks in the undercapitalized category and 15 banks in 15 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 the well-capitalized category. Table 2: Number of banks in each group Bank K/A Ratio Category All banks Small banks Large banks (Assets<8 billion €) (Assets>8 billion €) All banks 23 7 Undercapitalized banks 15 10 5 (K/A<0.13) Well-capitalized banks 15 13 2 (K/A>0.13) Table 3: The effect of the change in overnight rates on the growth rate of loans1 4 GROUPS Banks with Capital/Asset Ratio<0.13 Bank Size Variable Small Banks Large Banks ∑ Change in ON rates4 1 -0,0101 -0,0728 (0.955) (0.482) Change in Securities 0.0560 -0.0354 (0.210) (0.518) Change in Time Deposits -0.0016 0.6864 (0.000)*** (0.000)*** R2 0.3521 0.6279 No of observations: 200 100 Banks with Capital/Asset Ratio>0.13 Bank Size Variable Small Banks Large Banks ∑ Change in ON rates4 1 2,6792 -0,2379 (0.389) (0.066)* Change in Securities -0.0437 0.2003 (0.804) (0.177) Change in Time Deposits 0.0347 0.4898 16 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 (0.056)* (0.003)** R2 0.0581 0.7498 No of observations: 260 40 NOTES: P-values in paranthesis 1 Each specification also includes four lagged values of the dependent variable and three seasonal dummy variables. * = Significant at 10 percent level. ** = Significant at 5 percent level.*** = Significant at 1 percent level. 5.1. Results of Ordinary Least Squares Regression Table 3 shows the estimates of the effect of policy on loans. The sums of the coefficients associated with changes in the overnight rates are negative for small, undercapitalized; large, undercapitalized and large, well- capitalized banks. However the overnight rate coefficients reported for undercapitalized banks are insignificant and smaller in absolute value than those of the large, well-capitalized banks. The sum of the coefficients of the change in overnight rates is negative and significant at 10 percent level for the large, well-capitalized banks group. The overnight rate coefficient of the small, well-capitalized group is positive and large, but it is not significant. The coefficient associated with the overnight rates for small, undercapitalized bank category was expected to be negative, significant and greater in absolute value than the overnight rate coefficient associated with large and well-capitalized banks. The overnight rate coefficient of the small, undercapitalized bank group is negative but not significant. Consequently, the overnight rate coefficient values in Table 3 do not provide evidence consistent with our hypothesis that small, undercapitalized banks are most responsive to a monetary policy shock. The coefficients of the change in securities are insignificant for all bank groups. However, the coefficients of the change in time deposits are significant and positive for all groups except the small, undercapitalized group. This result may provide evidence for the existence of a bank lending channel in the sense that the small or undercapitalized banks may find it difficult to raise time deposits during contractionary periods, compared to the large or well-capitalized banks. But this alone is not enough to provide evidence for a lending channel, since the effect of this severity of financing for small, undercapitalized banks is not reflected on the sum of the coefficients associated with changes in the overnight rates. Dividing banks separately first by asset-size and then by capital leverage ratio may be helpful in order to distinguish the effects of asset-size and capital leverage ratio on the sensitivity of bank loans to changes in the overnight rates. Table 4: The effect of the change in overnight rates on the growth rate of loans2 2 GROUPS (Asset-based) Bank Size Variable Small banks(<8 billion €) Large banks(>8 billion €) ∑ Change in ON rates4 1 1.4902 -0.1009 17 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2015) Volume 12, No 1, pp 1-28 (0.404) (0.215) Change in Securities -0.0452 -0.0233 (0.777) (0.646) Change in Time Deposits -0.0008 0.6851 (0.462) (0.000)*** R2 0.0351 0.6315 No of observations 460 140 NOTES: P-values in paranthesis 2Each specification also includes four lagged values of the dependent variable and three seasonal dummy variables. * = Significant at 10 percent level. ** = Significant at 5 percent level. *** = Significant at 1 percent level. Table 5: The effect of the change in overnight rates on the growth rate of loans2 2 GROUPS(Capital-based) Capital/Asset Ratio Variable K/A<0.13 0.13