The Effect of US Unconventional Monetary Policy on Cross-Border Bank Loans: Evidence from an Emerging Market

Transcription

1 The Effect of US Unconventional Monetary Policy on Cross-Border Bank Loans: Evidence from an Emerging Market Koray Alper Central Bank of the Republic of Turkey Fatih Altunok Central Bank of the Republic of Turkey Tanju Çapacıoğlu Central Bank of the Republic of Turkey Steven Ongena University of Zurich, Swiss Finance Institute and CEPR Abstract: We analyze the impact of Fed s QE on the cross-border loans of Turkey and asses the existence of global bank lending and borrowing channels for Turkey by using a very unique and comprehensive data set. We find that cross-border bank loans in Turkey have increased significantly as a result of Fed s QE. Furthermore, Turkish banks have borrowed at lower interest rates with longer maturities. These effects are significantly stronger for less-capitalized and liquidity-constrained borrower and lender banks. Less-capitalized and illiquid lender banks expand more loans with lower interest rates and longer maturities. Similarly, less-capitalized and illiquid borrower banks acquire more loans with more favorable terms. Therefore, we find strong evidence for the existence of global bank lending and borrowing channels. We also find that results above still hold for different types of maturity and currency. Moreover, we find strong evidence that not only U.S. banks but also European banks do respond to QE in terms of flow of cross-border loans to Turkey. Nevertheless, the effect of QE is stronger for U.S. banks compared to European banks. Keywords: bank lending channel; bank borrowing channel; monetary transmission; quantitative easing (QE); cross-border bank loans, micro-level data JEL classification: E44; E52; F42; G15; G21 Authors addresses: koray.alper@tcmb.gov.tr; fatih.altunok@tcmb.gov.tr; tancu.capacioglu@tcmb.gov.tr; steven.ongena@bf.uzh.ch

2 1. Introduction Global banks and financial institutions have significantly increased their international activities especially over the last twenty years. Thus, financial integration have deepened and gained strength globally. With the financial integration, especially financial linkages as well as trade linkages started to play a crucial role in the contagion of crisis. Moreover, impacts of central bank policies of developed countries reached far beyond national borders via these financial linkages. Global banks or financial institutions transmit these monetary effects across local and international financial markets, and in turn real side of the economy. There are many studies about the impacts of monetary policy on local financial markets (Bernanke and Blinder (1992), Kashyap and Stein (2000), Jiménez, Ongena, Peydró and Saurina (2012), Jiménez, Ongena, Peydró and Saurina (2014)). Moreover, recent empirical studies have focused on the impacts of monetary policy on the international financial markets (Cetorelli and Goldberg (2012), Morais, Peydró and Ruiz (2015), Temesvary, Ongena and Owen(2015)). Central banks of several advanced economies initially cut their interest rates to mitigate effects of the 2008 global financial crisis. However, those policies proved inadequate, and hence they started to pursue unconventional monetary policies (UMP). As a leading actor of this period, the Federal Reserve (Fed) employed several rounds of quantitative easing (QE) with various asset purchase programs which resulted in abundant amount of liquidity in global as well as domestic markets,. After these steps, the balance sheet size of the Fed - an indicator of monetary expansion - reached record-high levels. It is believed that abundant global liquidity boosted capital inflows via cross-border borrowing of banks and portfolio investment for especially emerging markets. This raised the question of international spillovers of UMP especially through cross-border loans of financial institutions which may have crucial effects on the financial stability of those countries. A complete understanding of international spillovers of a domestic monetary policy requires estimating two separate channels, global bank lending and borrowing channels, simultaneously. Existing literature provides increasing evidence on the global bank lending channel (Cetorelli Goldberg (2012), Cerutti, Claessens and Ratnovski (2014), Ioannidou, Ongena and Peydró (2014), Coleman, Correa, Feler and Goldrosen (2014), Morais et al. (2015), Ongena, Schindele and Vonnák (2015)). However, in this paper we identify not only global lending but also borrowing channel for an emerging market, focusing on the case of Turkey, by examining the transmission of Fed s QE through the cross-border loans of Turkish banks. We exploit a comprehensive and novel data set which consists of detailed information on cross-border loans of all Turkish Banks that are borrowed from 143 countries and 1046 banks or institutions between December 2002 and December

3 To identify the global bank lending and borrowing channels of monetary policy (QE???), we analyze loan-level data on a monthly basis, and control demand and supply sides separately by using borrower bank*time and lender bank*time fixed effects, respectively. This allows us to control demand and supply sides on the monthly basis Moreover; our identification strategy is based on the proposition that less-capitalized and illiquid banks exhibit a stronger response to changes in domestic liquidity conditions than their well-capitalized and liquid peers as in Kashyap and Stein (2000). We show that this hold for lender banks as well as borrower banks. We find the following robust results. Cross-border bank loans in Turkey have increased significantly as a result of Fed s QE. Furthermore, interest rate and maturity of loans are also significantly affected. In particular, Turkish banks have borrowed at lower interest rates with longer maturities. These effects are significantly stronger for less-capitalized and liquidity-constrained borrower and lender banks. Less-capitalized and illiquid lender banks expand more loans with lower interest rates and longer maturities. Similarly, less-capitalized and illiquid borrower banks acquire more loans with more favorable terms. Therefore, we find strong evidence the existence of global bank lending and borrowing channels. We also analyze how our results may vary depending on the maturity and currency types of cross-border loans. We find that results above still hold for different types of maturity and currency. Moreover, we also explore the direct and indirect effects of QE by separating the lender banks located in Euro Area and U.S. We find strong evidence that not only U.S. banks but also European banks do respond to QE in terms of flow of cross-border loans to Turkey. Nevertheless, the effect of QE is stronger for U.S. banks compared to European banks. Our paper contributes to the literature in three main ways. First, our results give us strong and significant evidence on the existence of global borrowing channel as well as lending channel. Since our data set includes both demand and supply sides, we explicitly control changes in the demand and supply at the monthly basis with the exhaustive set of fixed effects, thereby providing a clearer identification of the global bank lending and borrowing channels, respectively. To our knowledge our work is the first to document the existence of the global borrowing channel through an emerging market s flow of cross-border loans. By exploring the demand side, our results complement the findings of the studies that show the existence of global bank lending channel via internal capital markets (Cetorelli and Goldberg (2012)) and external capital markets (Temesvary et al.(2015)). Second, we have information on loan terms such as maturity, currency types and interest rate which are usually not available in most loan-level data sets used in the similar studies. However, thanks to the detailed and unique structure of our data set, we can test whether the findings are robust to various specifications, loan terms (maturity, interest rate), sub-periods (full period vs. post crisis period), and subsamples (short-term vs. long-term, USD vs. non-usd). Third, since we 2

4 have the information on lender banks, we also explore the direct and indirect effects of Fed s QE across regions. The rest of the paper is organized as follows. Section 2 specifies the empirical methodology and describes the structure of used datasets. Section 3 summarizes the empirical results of the estimations. Section 4 concludes. 2. Data Set and Methodology We use three novel data sets for our empirical analysis: loan-level data on borrowing of Turkish banks from international banks, bank-level balance sheet data of Turkish banks, and banklevel balance sheet data of lender international banks. We obtain the first two data sets from the Central Bank of the Republic of Turkey (CBRT), and the last one from the Fitch. All data sets cover the period of December 2002 and December While the first two data sets have monthly frequency, the last one has quarterly frequency. At the macro level, we use data on macro variables both for Turkey and lender countries as well as global liquidity variables. Turkey macro variables include industrial production index, domestic interest rates, inflation, and reel exchange rate. Lender country variables comprise reel GDP growth, inflation, policy rate of the related central bank, and reel exchange rate. Global liquidity variables include VIX, US real policy rate, real credit growth, and total M2 growth rate of four finance centers (US, ECB, UK and Japan). These variables allow us to control for the business cycles and monetary policy stance in Turkey and lender countries, and to better isolate changes in QE from other changes in economic activity or monetary conditions. The definitions of the variables, data sources, and summary statistics are given in Table 1. We merge loan-level data on borrowing from international banks to Turkish banks with the bank-level balance sheet data sets of Turkish banks and lender banks. This data set consists of detailed information on the cross-border bank loans that are borrowed from 143 countries and 1046 banks or institutions in the various types of loans such as credit, deposit, credit for foreign trade finance, syndicated loans, securitization, repo and subordinated loans. In addition to the lender country and lender bank or institution, the dataset includes information on the volume, type, currency, interest rate, beginning, and maturity date of a loan. The country of a direct lender as well as its headquarter and if available, the country of a guarantor bank are also available in the dataset. The volume of loans can also be acquired as flow or stock. Lender country information is provided according to the ISO-Swift BIC Directory. However, there is no standardization (in terms of a unique identifier) related to the lender banks or institutions since we have only their names (a bank's name might be entered in many different ways) as a string variable. Therefore, we explored identities of 3

5 lender banks manually. During this process, we proceed very carefully in order to prevent possible mistakes. Therefore, we think that our dataset is very unique in terms of characteristics of loans and supply side information as well as extensive amount of hand work. Our main objective is to understand how Fed s QE is transmitted to Turkish banks via crossborder loans. To do so, we use the following model: L ialcmf,t = β uuuu kk=1 β k (QQQQ) tt kk + 12 uuuu kk=1 γ k (QQQQ) tt kk CC jj,tt kk + 12 kk=1 δ k (CC) jj,tt kk + ξξ 1 (BBBBBBBB) bbbbbbbbbbbbbbbb ii,tt 1 + ξξ 2 (BBBBBBBB) llllllllllll ll,tt 1 + ξξ 3 (CCCCCCCCCCCCCC) bbbbbbbbbbbbbbbb tt 1 + ξξ 4 (CCCCCCCCCCCCCC) llllllllllll cc,tt 1 + ξξ 5 (GGGGGGGGGGGG) tt 1 + λ i + α aa + ɳ l + θ cc + µ m + ζ f + ε i L ialcmf,t denotes the yearly change in the natural logarithm of Turkish banks' (i) stock crossborder loans borrowed from country a and lender bank l with loan type c, maturity m and currency type f at time t. The QE variable is the monthly change in the natural logarithm of the securities held by the Fed form time t-1 to t. Moreover, CC denotes the lender or borrower bank s net worth or liquidity ratio defined as capital ratio or liquid assets over total assets ratio, respectively. We include twelve lags of the QE, net worth or liquidity ratio and their interactions in the model. We also include the bank-specific variables for borrower and lender banks, the macroeconomic indicators related to Turkish economy and lender countries, and global liquidity indicators that have the potential to affect cross-border bank loans. λ i, α aa, ɳ l, θ cc, µ m and ζ f denote the fixed effects for borrower bank i, lender country a, lender bank l, loan type c, maturity m, and currency type f. Since we aim to identify the global bank lending and borrowing channels of QE by controlling demand and supply sides separately, we also add borrower bank*time for the global bank lending channel and lender bank*time fixed effects for the global bank borrowing channel to the model.. Borrower bank*time fixed effects allow us to examine whether for the same borrower bank in the same month, the loans offered by different lender banks depend on the QE. In this case, we control exhaustively for unobserved time-varying borrower bank fundamentals (such as creditworthiness, balance sheet characteristics etc.) This will restrict our sample to the borrower banks with at least two lender-bank relationships. Similarly, lender bank*time fixed effects allow us to explore whether for the same lender bank in the same month differentiate the loans borrowed by different borrower banks with the QE. We also control exhaustively for unobserved time-varying lender bank fundamentals (such as risk appetite, balance sheet characteristics etc.). Moreover, we saturate our the regressions including the double interactions of all macro variables (Turkey, lender country and global liquidity variables) with the net worth or liquidity ratio variables of lender or borrower banks depending on the channel investigated. 4

6 3. Estimation Results Table 2 provides results how Fed s QE affect Turkish Banks cross-border loans. In all our specifications, we include fixed effects for borrower and lender banks, lender countries, loan types, maturities and currency types to control non-monetary shocks and unobservable factors. In Column 2, as well as the QE variable, we also include the bank-specific variables for borrower and lender banks, the macroeconomic indicators related to Turkish economy and lender countries, and global liquidity indicators. After Lehman failure in 2008, central banks of advanced economies pursued QE to mitigate the effects of the global financial crisis. Therefore, we also repeat the same exercises for the period of post-2008 crisis. While Columns 1 and 2 represent the full period, Columns 3 through 10 represent the period of post-2008 crisis using different maturities, currency types and lender banks regions. Columns 1 through 4 show that cross-border bank loans in Turkey increased significantly with QE in all specifications. In other words, a 1 standard deviation expansion in the securities held by the Fed causes a cumulative 6.82 to 8.16 percentage increase in cross-border loans of Turkish banks. This effect is also statistically significant at a 1 percent level. In Column 4 and 5, we examine how the effects of QE on cross-border bank loans vary in terms of the maturity of cross-border flows. Since short-term flows are more inclined to adopt liquidity conditions than long-term flows, we expect that monetary policies, regardless of being conventional or unconventional, have a stronger impact on short-term flows. Results indicate that short-term cross-border loans respond more to QE. For example, 1 standard deviation expansion in the securities held by the Fed causes a cumulative 8.20 percentage increase in short-term cross-border bank loans. In Column 6 and 7, we examine how the effects of QE may vary depending on the currency types. Since Fed s expansionary policies have loosened the liquidity conditions in dollar funding, one can expect that US dollar denominated crossborder loans have been affected more by these policies. However, Columns 6 and 7 show that there is no significant difference between US dollar and other currency types, and cross-border bank loans have increased regardless of currency types. Although we include the exhaustive set of fixed effects and control groups for banks and countries, there still may be additional variation in the characteristics of lender countries that explains the direct and indirect effects of QE. We take onestep towards exploring that variation by separating the lender banks located in Euro Area and United States (U.S.). We can conclude that cross-border loans of Turkish banking sector have increased not only through U.S. banks, but also through European banks. However, the effect of U.S. banks on cross-border bank loans is stronger than European banks such that a 1 standard deviation increase in 5

7 QE causes a cumulative percentage increase cross-border loans through U.S. lender banks, but percentage increase through European banks. Table 3 shows the channels through which the Fed s QE affect the lender banks with different capital by including the interaction of the QE with the lender bank s capital ratio. We control demand side by including borrower bank-time fixed effects to identify the supply side and the existence of global bank lending channel. Columns 5 & 7 also contain the interaction of lender bank s capital ratio with all macroeconomic indicators in order to control for the business cycles or monetary conditions of Turkey and lending countries. Columns 1 through 5 represent the full period and Columns 6 through 13 represent the period of post-2008 crisis using different maturities, currency types and lender banks regions. Table 3 provides us a strong evidence of the global bank lending channel in cross-border bank loans. Results show that less-capitalized lender banks exhibit a stronger response to the QE than their well-capitalized counterparts. For example, the impact of a 1 standard deviation increase in QE is 7.07 to 8.25 percentage higher for less-capitalized banks (at the 25th percentile of capital ratio) compared to well-capitalized banks (at the 75th percentile of capital ratio). Columns 8 through 13 shows the results how global bank lending channel may vary with different types of maturity and currency, and lender banks region. Columns 8 and 9 show that less-capitalized banks offer more loans by 9.45 percentage with the 1 standard deviation increase in QE compared to their well-capitalized counterparts. Regarding the dollar denominated cross border loans, Columns 10 and 11 show that less-capitalized banks extend more loans by percentage as a respond to a 1 standard deviation increase in QE compared to their well-capitalized counterparts. The impact of QE across regions seem much larger for the global lending channel such that that the difference between less-capitalized and high capitalized banks is percentage for the U.S. banks while it is percentage for European banks as a respond to a 1 standard deviation increase QE (Columns 12 and 13). Overall, less-capitalized lender banks exhibit a stronger response to the Fed s QE policies than their well-capitalized counterparts. Moreover, global lending channel of QE have stronger impact on short-term flows, dollar-denominated flows, and works stronger through US banks than European banks. In Table 4, we repeat the same specifications as in Table 3 using the liquid assets over assets ratio as our measure of bank liquidity. Table 4 also provides us a strong evidence of the global bank lending channel in cross-border bank loans. In other words, liquidity constrained lender banks exhibit a stronger response to the QE than their liquidity-abundant counterparts. For example, the impact of a 1 standard deviation increase in QE is 2.28 to 2.83 percentage higher for illiquid banks (at the 25th percentile of liquidity distribution) compared to liquidity-abundant banks (at the 75th percentile of liquidity). Importantly, the impact of QE on international activities of lender banks is less in magnitude than those we obtained using the capital to asset ratio as the liquidity measure. 6

8 Moreover, depending on the specifications, a 1 standard deviation increase in QE causes a 4.24 to 5.97 percentage greater increase in cross border flows by less-capitalized banks than illiquid ones. Columns 8 through 13 show the results how global bank lending channel may vary with different types of maturity and currency, and lender banks region. Looking by maturity, we observe that illiquid banks offer more loans by percentage more to a 1 standard deviation increase in QE compared to their liquid counterparts. Regarding currency types, we find that illiquid banks extend more loans by percentage as a respond to a 1 standard deviation increase in QE compared to their liquid counterparts, and this effect decreases to 8.08 percentage for other currency types. The impact of QE across regions seem much larger for the global lending channel such that the difference between illiquid and liquid banks is percentage for the U.S. banks while it is percentage for European banks as a respond to a 1 standard deviation increase QE (Columns 12 and 13). Overall, illiquid lender banks exhibit a stronger response to the QE than their liquidity-abundant counterparts. Moreover, global lending channel of QE have stronger impact on short-term flows, dollar-denominated flows, and works stronger through US banks than European banks. Our analysis thus far has focused on the supply side of cross border lending, and we explore the existence of global bank lending channel. However, our data set also enables us to extend our study for demand side. Therefore, we take one step towards exploring the demand side or the existence of global borrowing channel by controlling supply side with lender bank-time fixed effect. The results that liquidity constrained and less-capitalized lender banks increase their cross-border activities, apparent through Table 3 and 4, is consistent with the findings of Temesvary (2014) and Ongena et al. (2015). However, to our knowledge our work is the first to document the working of the global borrowing channel through an emerging market s cross-border bank flows. By exploring demand side, our results will complement the findings of aforementioned studies. Table 5 shows the channels through which the Fed s QE affect the borrower banks with different capital by including the interaction of the QE with the borrower bank s capital ratio. To this end, we control supply side by including lender bank-time fixed effects to identify the demand side and the existence of global bank borrowing channel. The results provide us a strong evidence of the global bank borrowing channel in cross-border bank loans. We find that less-capitalized borrower banks exhibit a stronger response to the QE than their well-capitalized peers. For example, the impact of a 1 standard deviation increase in QE is 7.06 to percentage higher for less-capitalized banks compared to well-capitalized ones. Columns 7 and 8 show that less-capitalized banks borrow more short-term loans by percentage with the 1 standard deviation increase in QE compared to their well-capitalized counterparts, and this effect decreases to percentage for the long-term flows. Looking by currency types, Columns 10 and 11 show that less-capitalized banks borrow more US 7

9 dollar denominated loans by percentage as a respond to a 1 standard deviation increase in QE compared to their well-capitalized counterparts. In Table 6, we repeat the same specifications as in Table 5 using the liquid assets over assets ratio as our measure of bank liquidity. The results also provide us a strong evidence of the global bank borrowing channel in cross-border bank loans. In other words, liquidity constrained borrower banks exhibit a stronger response to the QE than their liquidity-abundant counterparts. For example, the impact of a 1 standard deviation increase in QE is to percentage higher for illiquid borrower banks compared to liquid ones. Notably, the impact of QE on the cross border loans of borrower banks is greater in magnitude than those we obtained using the capital to asset ratio as the liquidity measure. Looking by maturity, we observe that illiquid banks borrow more loans by percentage more to a 1 standard deviation increase in QE compared to their liquid counterparts. Regarding currency types, we find that illiquid banks borrow more loans by percentage as a respond to a 1 standard deviation increase in QE compared to their liquid counterparts. Our analysis so far has shown that the QE policies are associated with changes in the amount of cross-border bank loans, especially for illiquid and less-capitalized borrower and lender banks. Results give us a strong evidence of a global lending and borrowing channels in cross-border bank loans. However, our data set also enables us to extend our study for the interest rate and the maturity of cross-border loans. Therefore, we first extend our study for the interest rate, and replace our dependent variable as the natural logarithm of spread between the policy rate of lender countries and the interest rate of Turkish Banks cross-border borrowing across countries, lender banks with different loan types, maturities and currency types. Then, we replicate the specifications in Table 2 through 6 including the maturity and amount of cross-border loans as explanatory variable in order to avoid omitted variable bias problem. Since most of the cross-border loans have floating interest rate, we use stock and flow loan data set in Table 7a and 7b, respectively. Table 7a and 7b provide results how expansionary policies affect the interest rate spread using various fixed effects and control groups. The results indicate that Turkish banks have borrowed at lower interest rates as a result of Fed s QE. In fact, a 1 standard deviation increase in QE causes a cumulative to percentage decrease in the interest rate spread. Moreover, this effect is greater in magnitude in new cross-border loans. These effects are also statistically significant at a 1 percent level. Columns 3 through 6 in Table 7a and 7b show the channels through which the Fed s QE affect the lender banks with different capital and liquidity ratios in terms of interest rate of cross border bank loans. The results provide us a strong evidence of a global lending channel in terms of the interest rate of cross-border bank loans. We find that liquidity constrained and less capitalized lender banks have lent at lower interest rates than their liquidity abundant and well capitalized 8

10 counterparts with QE. For example, the impact of a 1 standard deviation increase in QE on interest rate spread is 2.29 to 4.90 percentage higher for less-capitalized banks compared to well-capitalized banks. Moreover, the percentage change in the spread following a 1 standard deviation increase in QE is to percentage higher for illiquid banks than liquid peers. Importantly, the impact of QE policies on the cross border loans of lender banks is greater in magnitude than those we obtained using the capital to asset ratio as the liquidity measure. Moreover, depending on the specifications, a 1 standard deviation increase in QE causes a 5.56 to percentage greater increase in interest rate spread by illiquid banks than well capitalized ones. These effects are also greater in magnitude in new cross-border loans. Columns 7 through 10 in Table 7a and 7b show the channels through which the Fed s QE policies affect the borrower banks with different capital and liquidity ratios in terms of interest rate. The results provide us a strong evidence of a global borrowing channel in terms of the interest rate of cross-border bank loans. We find that liquidity constrained and less capitalized borrower banks have borrowed at lower interest rates than their liquidity abundant and well capitalized counterparts with QE. For example, the impact of a 1 standard deviation increase in QE on interest rate spread is 2.68 to 2.87 percentage higher for less-capitalized banks compared to well-capitalized banks. Moreover, the percentage change in the spread following a 1 standard deviation increase in QE is 7.05 to percentage higher for illiquid banks than liquid banks. Notably, the impact of QE policies on the cross border loans of lender banks is greater in magnitude than those we obtained using the capital to asset ratio as the liquidity measure. As mentioned above, we also extent our study for the maturity of cross-border loans, and replace our dependent variable as the natural logarithm of maturity of Turkish Banks new crossborder borrowing. Then, we replicate the specifications in Table 2 through 6 by replacing dependent variable in this way and adding the interest rate and amount of cross-border loans as explanatory variable in order to avoid omitted variable bias problem. Column 1 & 2 in Table 8 indicate that Turkish banks have borrowed with longer maturity with QE. In fact, a 1 standard deviation increase in QE causes a cumulative 4.21 to percentage increase in the maturity. This effect is also statistically significant at a 1 percent level. Columns 3 through 6 in Table 8 give us a strong evidence of a global lending channel in terms of the maturity of cross-border bank loans. The results show that liquidity constrained and lesscapitalized lender banks have lent with longer maturities than their liquidity abundant and wellcapitalized counterparts with QE. For example, the impact of a 1 standard deviation increase in QE on interest rate spread is 3.45 to 4.17 percentage higher for less-capitalized banks compared to wellcapitalized banks. Moreover, the percentage change in the spread following a 1 standard deviation increase in QE is 0.99 to 1.37 percentage higher for illiquid banks than liquid banks. Importantly, the 9

11 impact of QE policies on the cross border loans of lender banks is less in magnitude than those we obtained using the capital to asset ratio as the liquidity measure. Columns 7 through 10 in Table 8 provide us a strong evidence of a global borrowing channel in terms of the maturity of cross-border bank loans. The results show that liquidity constrained and less capitalized borrower banks have borrowed with longer maturities than their liquidity abundant and well capitalized counterparts with QE. For example, the impact of a 1 standard deviation increase in QE on interest rate spread is 6.07 to 7.65 percentage higher for less-capitalized banks compared to well-capitalized banks. Moreover, the percentage change in the spread following a 1 standard deviation increase in QE is 5.11 to percentage higher for illiquid banks than liquid banks. 4. Conclusion In this paper, we trace the impact of Fed s QE on banking system of an emerging market, focusing on the case of Turkey, through cross-border bank loans, and analyze the existence of global bank lending and borrowing channels of these unconventional monetary policies. Our identification strategy is based on the hypothesis that less-capitalized and illiquid banks exhibit a stronger response to changes in liquidity conditions than their counterparts. We find that cross-border bank loans in Turkey have increased significantly as a result of Fed s QE. Furthermore, interest rate and maturity of loans are also significantly affected. In particular, Turkish banks have borrowed at lower interest rates with longer maturities. These effects are significantly stronger for less-capitalized and liquidityconstrained borrower and lender banks. Less-capitalized and illiquid lender banks expand more loans with lower interest rates and longer maturities. Similarly, less-capitalized and illiquid borrower banks acquire more loans with more favorable terms. Therefore, we find strong evidence the existence of global bank lending and borrowing channels. We also analyze how our results may vary depending on the maturity and currency types of cross-border loans. We find that results above still hold for different types of maturity and currency. Moreover, we also explore the direct and indirect effects of QE by separating the lender banks located in Euro Area and U.S.. We find strong evidence that not only U.S. banks but also European banks do respond to QE in terms of flow of cross-border loans to Turkey. Nevertheless, the effect of QE is stronger for U.S. banks compared to European banks. Our paper contributes to the literature in three main ways. First, our results give us strong and significant evidence on the existence of global borrowing channel as well as lending channel. Since our data set includes both demand and supply sides, we explicitly control changes in the demand and supply at the monthly basis with the exhaustive set of fixed effects, thereby providing a clearer identification of the global bank lending and borrowing channels, respectively. To our 10

12 knowledge our work is the first to document the existence of the global borrowing channel through an emerging market s flow of cross-border loans. By exploring the demand side, our results complement the findings of the studies that show the existence of global bank lending channel. Second, with the help of detailed and unique structure of our data set, we can test whether the findings are robust to various specifications, loan terms, sub-periods, subsamples. Third, since we have the information on lender banks, we also explore the direct and indirect effects of Fed s QE across regions. 11

13 References Bernanke, B. S., and A. S. Blinder, 1992, "The Federal Funds Rate and the Channels of Monetary Transmission," American Economic Review 82, Cerutti, E., S. Claessens, and L. Ratnovski, 2014, Global Liquidity and Drivers of Cross Border Bank Flows, International Monetary Fund, Washington DC. Cetorelli, N., and L. S. Goldberg, 2012a, "Banking Globalization and Monetary Transmission," Journal of Finance 67, Coleman, N., R. Correa, L. Feler, and J. Goldrosen, 2014, Unconventional Monetary Policy Spillovers and Local Credit Provision, Federal Reserve Board, Washington DC. Ioannidou, V. P., S. Ongena, and J. L. Peydró, 2014, "Monetary Policy, Risk Taking and Pricing: Evidence from a Quasi Natural Experiment," Review of Finance Forthcoming. Jiménez, G., S. Ongena, J. L. Peydró, and J. Saurina, 2012, "Credit Supply and Monetary Policy: Identifying the Bank Balance Sheet Channel with Loan Applications," American Economic Review 102, , 2014, "Hazardous Times for Monetary Policy: What Do Twenty Three Million Bank Loans Say about the Effects of Monetary Policy on Credit Risk Taking?," Econometrica Kashyap, A. K., and J. C. Stein, 2000, "What Do A Million Observations on Banks Say About the Transmission of Monetary Policy?," American Economic Review 90, Morais, B., J. L. Peydró, and C. Ruiz, 2015, The International Bank Lending Channel of Monetary Policy Rates and Quantitative Easing: Credit Supply, Reach for Yield, and Real Effects, World Bank, Washington DC. Ongena, S., I. Schindele, and D. Vonnák, 2015, In Lands of Foreign Currency Credit, Bank Lending Channels Run Through? The Effects of Monetary Policy at Home and Abroad on the Currency Denomination of the Supply of Credit, University of Zurich, Zurich. Temesvary, J., 2014, "The Determinants of U.S. Banks International Activities," Journal of Banking and Finance 44, Temesvary, J., Ongena, S., & Owen, A. L. (2015). A global lending channel unplugged? Does US monetary policy affect cross-border and affiliate lending by global US banks? (No. 511). CFS Working Paper Series. 12

14 Table 1 Summary Statistics Variable Names Definition Source N Mean SD Min. 10% 25% 50% 75% 90% Max. Dependent Variables Yearly Change in Cross-border Bank Loans Maturity of Cross-border Loans (day) Spread between Interest Rate of Cross-border Bank Loans and policy rate of lender countries Independent Variables Δ US Federal Reserve`s Assets Maturity of Cross-border Loans (day) Interest Rate of Cross-border Bank Loans The Amount of Cross-border Bank Loans Borrower Bank Variables The change in the natural logarithm of Turkish banks' stock cross-border loans borrowed from country a and lender bank (l) with loan type c, maturity m and currency type f at time t The natural logarithm of maturity of Turkish banks' new cross-border loans borrowed from country a and lender bank (l) with loan type c at time t The natural logarithm of spread between policy rate of lender countries and interest rate of Turkish banks' new cross-border loans borrowed from country a and lender bank (l) with loan type c at time t Monthly change in the natural logarithm of the Federal Reserve`s asset size The natural logarithm of maturity of Turkish banks' new cross-border loans borrowed from country a and lender bank (l) with loan type c at time t The natural logarithm of interest rate of Turkish banks' new cross-border loans borrowed from country a and lender bank (l) with loan type c at time t The natural logarithm of Turkish banks' stock crossborder loans borrowed from country a and lender bank (l) with loan type c and maturity m at time t CBRT 141, CBRT 151, CBRT 173, FED 262, CBRT 151, CBRT 541, CBRT 555, Borrower Bank Capital Ratio Capital divided by total assets CBRT 270, Borrower Bank Liquidity Ratio Selected FX liquid assets divided by total assets (Selected FX liquid assets = cash + foreign banks(free) + receivables from CBRT, interbank money market, reverse repo transactions) CBRT 270, Borrower Bank Total Assets The natural logarithm of real total bank assets CBRT 264, Borrower Bank Credit Ratio Total loans divided by total assets CBRT 270, Borrower Bank Deposit Ratio Total deposits divided by total assets CBRT 270, Borrower Bank ROA Ratio Bank net profit divided by total assets CBRT 259, Borrower Bank NPL Ratio Bank non-performing loans divided by bank total loans CBRT 270,

15 Table 1 (continued) Variable Names Definition Source N Mean SD Min. 10% 25% 50% 75% 90% Max. Lender Bank Variables Lender Bank Capital Ratio Capital divided by total assets Fitch 123, Lender Bank Liquidity Ratio Bank liquid assets divided by total assets (liquid assets = trading securities and at FV through income + loans & advances < 3 months + loans & advances to banks < 3 Fitch 123, months) Lender Bank Total Assets The natural logarithm of total bank assets Fitch 122, Lender Bank Credit Ratio Total loans divided by total assets Fitch 122, Lender Bank Deposit Ratio Total deposits divided by total assets Fitch 122, Lender Bank ROA Ratio Bank net profit divided by total assets Fitch 123, Lender Bank NPL Ratio Bank non-performing loans divided by bank total loans Fitch 123, Turkey(TR) Macro Variables Δ Industrial Production Monthly change in industrial production index (used instead of GDP due to discrepancy of frequencies) TurkStat 262, Inflation Rate Monthly change in consumer price index TurkStat 264, Δ BIST o/n Interest Rate Monthly change in Istanbul Stock Exchange(BIST) over/night interest rate TurkStat 264, Δ REER Monthly change in real effective exchange rate based on consumer price index TurkStat 263, Lender Country Variables Real GDP Growth Quarterly change in real GDP IMF 156, Inflation Rate Monthly change in consumer price index IMF 165, Δ Policy Rate Quarterly change in policy interest rate IMF 153, Δ REER Quarterly change in real effective exchange rate IMF 143, Global Liquidity Variables Δ VIX Monthly change in CBOE S&P500 volatility index Bloomberg 264, Δ US real policy rate Monthly change in US real effective federal funds rate Bloomberg 264, US real credit growth Real private credit growth in US Bloomberg 264, Total M2 growth Total M2 growth rate of four financial centers(us, ECB, UK and Japan) Bloomberg 264,

16 Table 2 Yearly change in Turkish banks' cross-border borrowing (from countries and lender banks with different loan types, maturities and currency types) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Period Full Period Post-2008 Crisis Included Maturities All All All All 1 Year > 1 Year All All All All Currency Types All All All All All All USD non-usd All All Region of Lender Banks All All All All All All All All US EA ΣΔ US Federal Reserve`s Asset Size{t-1 to t-12} [0.606]** [1.073]** [0.751]*** [1.299]*** [1.508]*** [2.652] [2.012]* [1.737]** [2.796]*** [2.128]*** Constant [-5.560]*** [-3.250]*** [-0.520] [-0.160] [-0.540] [-0.310] [-0.960] [-0.090] [-0.430] [-1.970]** Lender Bank Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Lender Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes - - Borrower Bank Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Loan Type Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Loan Maturity Fixed Effects Yes Yes Yes Yes - - Yes Yes Yes Yes Currency Type Fixed Effects Yes Yes Yes Yes Yes Yes - - Yes Yes TR Macro Variables Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Global Liquidity Variables Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Borrower Bank Variables Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Lender Bank Variables No Yes No Yes Yes Yes Yes Yes Yes Yes Lender Country Variables No Yes No Yes Yes Yes Yes Yes Yes Yes R-squared Number of Observations ,186 37,029 26,375 10,654 15,258 21,771 7,083 15,138 Percentage change in Turkish Banks' cross-border loans following a 1 std. dev. increase in the securities held by the Fed: Note. -- The table reports estimates from ordinary least squares regressions. The dependent variable is the yearly change in Turkish banks' cross-border borrowing (from countries and lender banks with different loan types, maturities and currency types). Table 1 contains the definition of all variables and the summary statistics for each included variable. Borrower Bank Variables include the lagged values of Bank Total Assets, Capital Ratio, Liquidity Ratio, Credit Ratio, Deposit Ratio, ROA and NPL Ratio. Lender Bank Variables include the lagged values of Bank Total Assets, Capital Ratio, Liquidity Ratio, Credit Ratio, Deposit Ratio and ROA Ratio. Turkey Macro Variables are monthly change in industrial production index, inflation rate, monthly change in BIST o/n interest rate and monthly change in reel effective exchange rate. Lender Country Variables are real GDP growth, inflation rate, quarterly change in policy rate and quarterly change in real effective exchange rate. Global Liquidity Variables are monthly change in VIX, monthly change in US real policy rate, real credit growth rate of US and total M2 growth rate of four financial centers (US, UK, EA, Japan). Full period covers the period of 2002:M :M12. Post-2008 Crisis period covers the period of 2008:M :M12. Coefficients are listed in the first row, robust standard errors are reported in the row below, and the corresponding significance levels are placed adjacently. Σ indicates sum of the twelve coefficients on the indicated lag terms (and corresponding standard errors and significance level). "Yes" indicates set of characteristics or fixed effects. "No" indicates set of characteristics or fixed effects is not included. *** Significant at 1%, ** significant at 5%, * significant at 10%. 15

17 Table 3 Yearly change in Turkish banks' cross-border borrowing (from countries and lender banks with different loan types, maturities and currency types) for lender banks with different capital ratios (Supply Side) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) Period Full Period Post-2008 Crisis Included Maturities All All All All All All All 1 Year > 1 Year All All All All Currency Types All All All All All All All All All USD non-usd All All Region of Lender Banks All All All All All All All All All All All US EA Σ Lender Bank Capital Ratio {t-1 to t-12} [0.659]*** [0.659]*** [0.711]*** [0.773]*** [1.39]*** [0.736]*** [1.478]*** [1.517]*** [2.977] [2.163]*** [1.97] [5.552]* [3.247]*** ΣΔ US Federal Reserve`s Asset Size{t-1 to t-12} * Σ Lender Bank's Capital Ratio {t-1 to t-12} [31.315]** [31.315]** [32.149]** [47.806]* [54.59]* [34.112]*** [56.039]** [59.377]* [ ] [92.322]*** [72.238] [ ]*** [ ]*** Lender Bank Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Lender Country Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Loan Type Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Loan Maturity Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Currency Type Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes (Borrower Bank*Month) Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes TR Macro Variables No No Global Liquidity Variables No No Borrower Bank Variables No Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes Yes Lender Bank Variables No No Yes Yes Yes No Yes Yes Yes Yes Yes Yes Yes Lender Country Variables No No No Yes Yes No Yes Yes Yes Yes Yes Yes Yes TR Macro Variables*Lender Bank's Capital Ratio No No No No Yes No Yes Yes Yes Yes Yes Yes Yes Lender Country Variables*Lender Bank's Capital Ratio No No No No Yes No Yes Yes Yes Yes Yes Yes Yes Global Liquidity Variables*Lender Bank's Capital Ratio No No No No Yes No Yes Yes Yes Yes Yes Yes Yes R-squared Number of Observations 50,333 50,333 50,013 38,943 38,943 46,905 36,158 34,718 3,962 16,231 22,478 7,464 16,055 Percentage change in Turkish Banks' cross-border loans following a 1 std. dev. increase in the securities held by the Fed by lower(25%) versus higher(75%) capitalized lender banks: Note. -- The table reports estimates from ordinary least squares regressions. The dependent variable is the yearly change in Turkish banks' cross-border borrowing (from countries and lender banks with different loan types, maturities and currency types) for lender banks with different capital ratios. Table 1 contains the definition of all variables and the summary statistics for each included variable. Borrower Bank Variables include the lagged values of Bank Total Assets, Capital Ratio, Liquidity Ratio, Credit Ratio, Deposit Ratio, ROA and NPL Ratio. Lender Bank Variables include the lagged values of Bank Total Assets, Capital Ratio, Liquidity Ratio, Credit Ratio, Deposit Ratio and ROA Ratio. Turkey Macro Variables are monthly change in industrial production index, inflation rate, monthly change in BIST o/n interest rate and monthly change in reel effective exchange rate. Lender Country Variables are real GDP growth, inflation rate, quarterly change in policy rate and quarterly change in real effective exchange rate. Global Liquidity Variables are monthly change in VIX, monthly change in US real policy rate, real credit growth rate of US and total M2 growth rate of four financial centers (US, UK, EA, Japan). Full period covers the period of 2002:M :M12. Post Crisis period covers the period of 2008:M :M12. Coefficients are listed in the first row, robust standard errors are reported in the row below, and the corresponding significance levels are placed adjacently. Σ indicates sum of the twelve coefficients on the indicated lag terms (and corresponding standard errors and significance level). "Yes" indicates set of characteristics or fixed effects. "No" indicates set of characteristics or fixed effects is not included. *** Significant at 1%, ** significant at 5%, * significant at 10%. 16