Capital Structure of Banks and their Borrowers: an Empirical Analysis

Transcription

1 Capital Structure of Banks and their Borrowers: an Empirical Analysis Valeriia Dzhamalova * Abstract The paper performs empirical analysis of capital structure of banks and their borrowers for a sample of financial and non-financial companies around the world. We find a positive effect of lenders leverage on the leverage of their borrowers. A one standard deviation increase in lenders leverage corresponds to 6.2% increase in the leverage of their borrowers. This effect is smaller for the borrowers with low level of financial distress. We find that borrowers leverage is not important in determining their lenders capital structure; but lenders characteristics, such as a proportion of loans on the balance sheet, has positive and significant effect on the capital structure. JEL classification: Keywords: Capital structure, Capital regulation, Bank, Borrowers * Department of Economics, Lund University, Box 7082 S-22007, Lund, Sweden; Valeriia.Dzhamalova@nek.lu.se. 1

2 1. Introduction This paper provides empirical evidence on the relationship between the capital structure of lenders and their borrowers. Given that banks provide debt for companies and debt capacity of banks depends on indebtedness and solvency of their borrowers, the capital structure of banks and borrowers may be related. Most of the researches on capital structure ignore interaction among different economic agents. Recent empirical studies incorporate the characteristics of peer firms in capital structure models (Leary and Roberts (2014)), but empirical research lacks the evidence on the relationship between the capital structure decisions of lenders and their borrowers. Theoretical research relates banks and borrowers capital structure by modelling the essential functions of banks. Diamond and Rajan (2000) model optimal capital structure, using the interaction between depositors, equity (debt) holders and borrowers of a bank. They argue banks capital structure determines the nature of banks customers because different customers rely to a different extent on liquidity and credit. Gornall and Strebulaev (2015) develop a model of capital structure decisions by modeling the interaction between a bank s debt decisions and the debt decisions of that bank s borrowers. Our study is the first performing an empirical analysis of capital structure decisions of banks and their borrowers. The paper provides new evidence on capital structure determinants of financial and non-financial companies. We also contribute to the discussion on the effect of banks capital regulation on the real economy. By limiting banks leverage, regulators place a stricter limit on the relative level of debt that banks can use to finance their assets. If the leverage of borrowers decreases together with the leverage of banks, capital regulation may lead to less indebtedness and vulnerability of the economy. To relate borrowers and lenders we use syndicated loan contracts from DealScan. Using a total amount of a contract and allocation of each lender within a syndicate, we determine to which extent borrowers and lenders are related. The repayment schedule of a contract allows us to track the changes in borrower-lender relationship over time. The advantage of using syndicated loan contracts is that they relate a borrower and multiple lenders, which most realistically models the real world relationship. We obtain the borrower specific information from Compustat and lender-specific information from Capital IQ. Our sample of borrowers consists of around 1000 borrowers and 1200 lenders around the world (North America, Asia, Europe). The sample covers the period of 20 years, We estimate two models: 1) a linear fixed effect regression of 2

3 borrower s leverage on the weighted average of its lenders leverage; 2) a linear fixed effect regression of lender s leverage on the weighted average of its borrowers leverage. Controlling for size, profitability, tangibility, growth and risk we find that leverage of borrowers is positively related to the average leverage of their lenders. A one standard deviation increase in the average of lenders leverage is associated with 6.2% increase in their borrower s leverage. This effect differs depending on the level of borrowers distress: a one standard deviation increase in the average of lenders leverage is associated with only 3.2 % increase in the leverage of borrowers with small probability of bankruptcy. As borrowers with low probability of bankruptcy have lower leverage, they receive less tax benefits from debt and affected by lenders leverage to a lesser extent. We do not find any significant effect of borrowers leverage on the leverage of their lenders. The coefficient on lenders leverage has expected sign in some of the specifications, but the coefficient is economically small. One of the explanations for statistical insignificance is regression attenuation due to the measurement error in regressor. In particular, our weighted average of borrowers leverage captures only part of lender-borrowers relationship, while true relationship is not observable. Imprecise measurement of weighted average of borrowers leverage can lead to underestimation of absolute value of its coefficient. 3

4 2. Capital Structure of Financial and Non-Financial Companies: Related Literature 2.1. Capital Structure of Non-Financial Companies Two important capital structure theories are tradeoff theory (Kraus and Litzenberger (1973)) and pecking order theory (Myers (1984)). According to the trade-off theory, debt financing provides tax advantage comparing to equity financing, but at the same time high level of debt increases the probability of bankruptcy. The tradeoff between tax-savings from debt and financial cost of bankruptcy determines the capital structure of a company. The pecking order theory suggests that companies would prefer internal funds (retained earnings or initial equity) for financing their investments. If a company lacks internal financing, it would prefer to issue debt first and equity only at the last resort. Empirical tests of pecking order and trade-off theories provide the evidence on important determinants of leverage. For example, Hovakimian et al (2001) analyze the optimal choice of debt to equity ratio for a large sample of the U.S companies and find that past profits and stock prices play important role in the companies decision to issue debt or equity. Jandik and Makhija (2001) examine firm-specific determinants of leverage for a sample of pooled time-series cross-sectional data for the single industry (electric and gas utilities) for period They conclude that bankruptcy costs, growth, non-debt tax shields, collateral profitability, size and risk are important determinants of leverage; even with the risk having a positive sign, contrary to both pecking order and trade-off theories. Fama and French (2002) conclude that pecking-order and trade-off theories share the same predictions regarding the effect of investments, size, nondebt tax shield and share opposite predictions regarding the effect of profitability on leverage. Several studies extend the models of capital structure by macroeconomic and industry-level variables. Korajcsyk and Levy (2003) model the capital structure as a function of macroeconomic conditions and company-specific variables for the samples of constrained and unconstrained firms. They find leverage is counter-cyclical for the relatively unconstrained sample, but pro-cyclical for the relatively constrained sample. MacKay and Phillips (2005) investigate the effect of industry on companies capital structure and find that industry s effect accounts for around 13% of variation in capital structure, but capital structure also depends on firm s position within its industry. Leary and Roberts (2014) further investigate the effect of industry on capital 4

5 structure. They show that companies financing decisions are responses to the financing decisions and characteristics of the peer firms within the industry Capital Structure of Financial Companies Traditionally, banks provide loans to the customers with a shortage of funds by borrowing from the customers with excessive funds. In other words, banks fulfill the role of intermediary between the companies and investors by granting loans and receiving deposits. The intermediary role allows banks to finance their activity with high level of debt and low level of equity. High proportion of deposits in banks liabilities allows leverage (total liabilities to total assets) of banks to be very high. Figure 1illustrates that the leverage ratio of American banks during is 87%-95% with proportion of deposits between 65%-93%.The leverage of the European banks is also high, for example the ratio of total liabilities to total assets in Germany 2 for period is 94%-96%. On the one hand, the core of the banking activities (attract deposits and grant loans) explains the high level of banks leverage. On the other hand, financing by deposits is risky because depositors are subject to a collective action problem, so called bank runs, when a large number of customers withdraw their deposits from banks at the same time. Why do banks lever up despite the high riskiness of leverage? Existing research on banks capital structure searches for explanation for high banks leverage, but results are still inconclusive. Determinants of capital structure from pecking order or trade-off theories explain some variation in banks capital structure (see for example Gropp and Heider (2010)), but both theories ignore important characteristics of banking industries (deposits, deposits insurance and government guarantees). Some studies argue that government guaranties and deposit insurance have a positive effect on bank leverage (see for example Juks (2010)). But Gropp and Heider (2010) find that mispriced deposit insurance and capital regulation has a second-order importance in determining the capital structure. They find that only the bank fixed-effects are important determinants of banks capital structure and the leverage converges to bank specific, time-invariant targets. In other words, Gropp and Heider (2010) do not find any effect of regulation and deposit insurance on banks capital structure and more theoretical and empirical evidence will shed a light on capital structure puzzle. 1 Similar to Gornall and Strebulaev (2015), we estimate historical averages using the data of Federal Deposit Insurance Corporation: 2 We estimate the averages for the German banks using the data from German Central Bank s web-page: 5

6 Figure 1 1 0,9 0,8 0,7 0,6 0,5 0,4 0,3 Total Liabilities/Total Assets Total Liabilities minus Deposits/Total Assets 0,2 0, After the recent financial crisis, theoretical and empirical research on banks capital structure and regulation is growing (see Thakor (2014) for the review of existing research). In the next section I review the articles which model capital structure by connecting banks and non-financial companies. These articles provide a theoretical background for our study. 3. Theoretical Background Diamond and Rajan (2000) explain the optimal capital structure of banks, by modelling the interactions between banks liquidity-creation and credit-creation functions. The authors model optimal capital structure using the interaction between depositors, equity (debt) holders and borrowers of a bank. They show that trade-offs between liquidity-creation, credit-creation and bank stability determines the optimal capital structure. Diamond and Rajan (2000) argue that banks capital structure also determines the nature of banks customers because different customers rely to a different extent on liquidity and credit. The model of Diamond and Rajan (2000) derives 6

7 the implications of a tough capital structure on intermediary s behavior towards borrowers and effect of minimum capital requirements on the banks, its lenders and its borrowers. Sundaresan and Wang (2014) analytically solve for the liability structure of banks by connecting banks and non-financial companies. Another paper which relates capital structure decisions of banks their borrowers is Gornall and Strebulaev (2015). They argue that tax benefits from debt originate only at the banks level and banks and companies leverages act as strategic substitutes and strategic complements. Strategic complementarity effect arises because banks pass tax benefits from debt to their borrowers. Strategic substitution effect arises because banks pass distress costs to their borrowers. We reckon that leverages of borrowers and lenders may indeed be related. Banks earn their margins on the difference between the interest expenses from deposits and interest incomes from loans. The more deposits banks have on their balance sheet, the higher the leverage. Higher leverage implies excess liquidity and allows borrowers receiving more debt because banks need to invest their cash. Thus, high leverage of banks allows their borrowers to receive more debt and increase the leverage. The aim of our paper is to test empirically if the leverages of borrowers and lenders are related. Using the information on the debt contracts, we identify the relationship between lenders and borrowers. As banks have multiple borrowers as well as borrowers have multiple lenders, we compute the weighted average of leverages and construct the average borrower for each bank and average lender for each borrower. Use of average borrowers and average lenders allows us to remove any potential effect from relationship lending. Relationship lending in our context is a lending with long history of cooperation between a borrower and a lender. Relationship lending can result in superior contract conditions for a particular borrower (for example lower interest rate) because of personal relationship between the bank s and borrower s managers. Rather than analyzing the effect of one particular lender to one particular borrower, our approach disentangles the average effect of lenders leverage on borrowers leverage (and vice versa). Next section describes the computation of weighted averages of lenders and weighted averages of borrowers. 4. How do we relate Borrowers and Lenders To identify the relationship between borrowers and lenders we use debt contracts, in particular syndicated loan contract. Use of syndicated loan contracts perfectly suits for connecting borrowers 7

8 and lenders due to the following reasons: 1) total amount of debt contract allows determining the extent to which a borrower depends on a lender; 2) repayment schedule of debt contract allows to track changes in borrower-lender relationship over time; 3) syndicated loan contracts allow relating a borrower to multiple lenders and a lender to multiple borrowers, which most realistically models the borrowers-lenders relationship. We get the information on syndicated loans contracts from DealScan, the database on loan facilities issued by financial institutions in 161 countries over a 30-year period. Most of the loan facilities in the DealScan originate from the USA, but data for European and Asia-Pacific countries is also available. Data allows us to study joint capital structure decisions between borrowers and lenders across different countries and over time Hypotheses tested in the study According to Gornall and Strebulaev (2015), the effect of lenders leverage on companies leverage is non-linear, depending on the level of lenders leverage. For moderately high levels of lenders leverage, borrowers receive more tax benefits and borrow more from their lenders (strategic complementarity effect). For very high levels of lenders leverage, companies stop borrowing from their lenders because the borrowing costs in the case of bankruptcy are too high (strategic substitution effect). For very low levels of lenders leverage (but high enough to transfer tax benefits), the probability of bankruptcy and hence borrowing costs are low and companies borrow more from a lender. Leaving the analysis of non-linearity for further stages of research, we start our analysis from identifying if linear relationship between the lenders and borrowers leverage exists. In other words, we assume that the trade-off theory holds, tax benefits are important for the company s financial decisions, tax benefits originate only at the bank s level and bank transfer tax benefits to borrowers by issuing loans 3. The first hypothesis we test in our study is the following: 3 According to Gornall and Strebulaev (2015), the debt benefits originate only at the bank level because of fundamental asymmetry between final users of financing ( downstream borrowers) and the intermediaries which pass financing along ( upstream borrowers). 8

9 Hypothesis 1: The relationship between a borrower s and its lenders leverages is positive because debt benefits originate at the lenders level. The higher the leverage of lenders, the more tax benefits lenders transfer to borrowers and the higher the leverage of borrowers. Using the second hypothesis, we test if the borrowers leverage has any effect on lender s leverage. Hypothesis 2: The relationship between lender s and its borrowers leverages is negative. The seniority of bank s debt explains this negative relationship. Seniority implies that in the case of bankruptcy, corporate borrowers paid their debt to the bank before paying to other creditors. If the company s leverage decreases, the leverage of its bank increases correspondingly because the larger fraction of company s debt becomes senior. For a bank, decrease in leverage of its borrowers means that portfolio of the bank loans became less risky, but due to the seniority of bank debt and diversification of loan portfolio, bank prefers to have high level of leverage to be able to earn the margins on its activities. To test the hypothesis described in two previous paragraphs, we compute the weighted average of borrowers leverage for each bank and weighted average of lenders leverage for each borrower. Next section describes the methodology we use to calculate the weighted averages Borrower-Lenders Relationship To test relationship between lenders and borrowers capital structures, one needs to know how much a lender lent to a company at specific point of time. Usually, the information on banks loans is confidential, but DealScan database provides the information on the syndicated loan transaction of large corporate and middle market commercial loans filed with the Securities and Exchange Commission or obtained through other reliable public sources. We consider each loan facility in DealScan as a debt contract because the loan facility provides necessary attributes of a debt contracts: total amount, maturity, repayment schedule, name of a borrower, lenders and the amount each lender allocates within a particular syndicated loan (allocation). We only consider the facilities if all mentioned attributes are available. DealScan does not provide company-specific data for lenders and borrowers. To compute leverage ratios for borrowers and lenders, we link DealScan with Compustat North America, and S&P Capital IQ databases. 9

10 To relate a borrower to lenders, we use the amount of outstanding debt: total amount of a loan minus repayment installment 4. Repayment of a loan can begin at the year of the loan issue or with a time lag. We construct a matrix D of outstanding debt of company k at time t. Each element in the debt matrix looks as follows: 0 DD iiiiiiii = LL iiiiiiii iiii tt < tt pp, (1) LL iiiiiiii pp iiiiiiii iiii tt tt pp where tt pp is the start date of the loan s repayment, LL iiiiiiii is the amount of loan i borrower k received from lender j at time t; pp iiiiiiii is the payment installment repaid at a period t. We use the debt matrix D to construct the weight which relates a borrower and lenders. We compute the weight of lender j in company s k debt at time t as follows: ww jjjjjj = ii DD iiiiiiii ss iiiiiiii, ii jj DD iiiiiiii ss iiiiiiii (2) where DD iiiiii is the amount of outstanding debt of company k, ss iiiiiiii is lender s j allocation of debt to a borrower k within a syndicate i at time t. ss iiiiiiii is based on the BankAllocation variable from DealScan, we compute ss iiiiiiii as follows: ss iiiiiiii = "BankAllocation" 100 (3) Using the weight computed in formula (2), we compute the weighted average of lenders leverages for each borrower k at time t as follows: YY kkkk JJ = jj=11 ww jjjjjj YY jjjj, (4) where YY jjjj is the leverage of lender j at time t. Section 6.2 provides the details on how we define the leverage of lender j. To test the Hypothesis 1 (lenders leverage has a positive effect on the leverage of their borrowers) we estimate the following equation: BB ZZ kkkk = ββ 00 + ββ 11 YY kkkk 11 + ββ 22 XX kkkk 11, (5) 4 We distinguish between different repayments frequencies available at DealScan: quarterly, semi-annualy, annualy, monthly, daily, weekly, bi-annualy, tri-annualy or by final bullet payment. 10

11 where ZZ kkkk is the leverage of borrower k at time t computed as the ratio of total debt to total assets, YY kkkk 11 BB as described above and XX kkkk 11 is the matrix of control variables on the borrower s level. BB Section 5 describes how we define ZZ kkkk and what we include in XX kkkk Lender-Borrowers Relationship Similar to the previous section, we use debt contracts to identify the relationship between a lender and its borrowers. But in this section we interpret the debt matrix D as the matrix of outstanding loans issued by lender j to borrower k. We weight the borrowers leverage ZZ kkkk by the amount of loan each lender allocated to a borrower relative to the total amount of loans issued by a lender to all borrowers at time t. We compute the weighted borrowers leverage ZZ jjjj as follows: ZZ jjjj = KK kk=1 ww jjjjjj ZZ kkkk, (6) where ww jjjjjj and ZZ kkkk as described in the previous section. To illustrate the difference in computation of Y kt and Z jt, let us consider how the elements of matrix D look after the multiplication with lenders allocation (D ikt s ijkt ). For simplicity, we consider only three lenders j=1, 2, 3 and two borrowers k=1,2. We assume each borrower has only one loan facility at time t. The leftmost column of Table 1lists the indices for borrowers and at the top row lists the indices for lenders. Each cell in the table shows the outstanding debt of each borrower to each lender if we consider the borrowers (columns); and the amount of loan issued by each lender to each borrower if we consider the lenders (rows). To compute Y kt we sum the weights for each lender w kj, i.e. we sum the columns of the lenders-borrowers matrix and to compute Z jt we sum the rows of the lenders-borrowers matrix. To test the second hypothesis of this study (the relationship between lender s and its borrowers leverages is negative), we estimate the following model: YY jjjj = αα 00 + αα 11 ZZ kkkk 1 LL Section 5 describes how we define YY jjjj and what we include in XX jjjj 11. LL + αα 22 XX jjjj 1 (7) 11

12 Table 1 Computation of weighted average of lenders and borrowers leverage (an illustration) Lenders Borrowers k1 ww 1111 = DD 1111ss 1111 j1 j2 j3 Weighted Lender s Leverage 33 jj=11 DD kkkk ss kkkk DD 1111 = 00 ww 1111 = DD 1111ss jj=11 DD kkkk ss kkkk 33 YY 11 = ww kkkk jj=11 YY jj k2 ww 2211 = DD 2222ss jj=11 DD kkkk ss kkkk DD 2222 = 00 DD 2222 = YY 22 = ww kkkk YY jj jj=11 Weighted Borrower s Leverage 33 ZZ 11 = ww kkkk jj=11 ZZ kk 22 ZZ 11 = ww kkkk jj=11 ZZ kk 5. Econometric models To test for borrower-lenders relationship we construct a panel of borrower-year observations and estimate fixed effect panel data regressions of the borrower s leverage on the weighted average of lenders leverage and a number of control variables. The model we estimate looks as follows: ZZ kkkk = ββ 0 + ββ kk + ββ 1 YY kkkk 1 BB + ββ 2 XX kkkk 11 BB + uu kkkk (8), where ZZ kkkk is the leverage of a borrower k at time t, ββ 0 is constant, ββ kk is borrower s fixed effect, BB YY kkkk 1 is weighted average of lenders leverages as described in section 4.2 and XX kkkk 11 is a matrix BB of borrower-specific control variables which we describe in the next paragraph, uu kkkk is a borrower-specific error term. We use first lags of independent variables to address the potential endogeneity problem arising from simultaneity bias. To understand what we mean by simultaneity 12

13 bias, consider company s leverage as a dependent variable and company s profitability as a regressor. If we observe both the dependent variable and the regressor at the same time t, we cannot distinguish if a company has low leverage because of high profitability or the company is more profitable because of low leverage. By lagging the regressors we measure the effect of their realized values on dependent variable at t-1. We define the dependent variable ZZ kkkk in two different ways: - book leverage = book value of debt (long term debt plus debt in current liabilities) divided by total assets; - market leverage = book value of debt divided by market value of assets (market value of equity plus book value of debt). In our definition of leverage we follow numerous literature on companies capital structure (see for example Korajczyk and Levy (2003) ). We use three different measures of lenders leverage: debt to-book assets, total liabilities- to- assets and deposits- to- assets. BB The control variables in matrix XX kkkk 11 are the borrower-specific determinants of capital structure according previous studies (Fama and French (2002), Jandik and Makhija (2001), Korajczyk and Levy (2003), Leary and Roberts (2014)). We summarize control variables, their definitions and expected signs in Table 2. To test for lender-borrowers relationship we construct a panel of lender-year observations and estimate fixed effect panel data regressions of the lender s leverage on the weighted average of its borrowers leverage and a number of control variables. We estimate the following model: YY jjjj = αα 00 + αα jj + αα 11 ZZ kkkk 1 LL + αα 22 XX jjjj 1 LL + uu jjjj (9) Where YY jjjj is leverage of a lender j at time t, αα 0 is constant, αα jj is lender s fixed effect, ZZ kkkk 1 is LL weighted average of borrowers leverages as described in previous section, XX kkkk 1 is a matrix of LL lender-specific control variables described in the next paragraph and uu jjjj is a lender-specific error term. Similar to borrower-lenders case we use the first lag of dependent variables to address the endogeneity problem arising from simultaneity bias. We use three different measures of YY jjjj : debt to assets, total liabilities to assets and deposits to assets. Some control variables and their expected signs such as profitability, size, risk, investment opportunities are the same as in borrower-lenders regression described in Table 2. However, the 13

14 definition of collateral for the lenders case is slightly different. Due to the essence of their activity, lenders (which are mostly banks in our case) do not usually own a lot of buildings, land or machinery, but they can use securities and cash as collateral for the short-term borrowings. We define collateral available for the banks in the following way: Bank collateral= (mortgage backed securities + investment securities + net property plant and equipment + cash) /total assets. As a measure of risk, we use the ratio of non-performing loans to total assets. We expect to find a negative relationship between the non-performing loans and banks leverage. Creditors can consider high ratio of non-performing loans as signal of banks distress and will be reluctant to lend to such bank. Finally, we control for the total amount of loans on banks balance sheet. By controlling for a total amount of loans, we control for a demand for credit from the bank s side. If a bank has large amount of loans, it needs to receive more debt to be able to finance the loans. We expect to have a positive relationship between the amount of loans and banks leverage. 14

15 Table 2 Control variables for borrower-lenders regression: proxies, expected sign and rationale for predictions Determinant of Proxies used in our study Capital Structure Profitability Opertaing income before deprectiation, tax and interest expenses Investment Market Value of opportunities Company/Book Value of Company Collateral Net Property Plant and Equipment/Total Assets Size The natural logarithm of sales Risk Volatility of earnings computed as a standard deviation of earning for the past five years Expected Rationale for expected sign sign +/- More profitable firms have more book leverage (trade-off model); controlling for investment opportunities firms with more profitable assets have less market leverage (pecking order model) -/+ Controlling for profitability, firms with larger investments have lower book and market leverage(trade-off model)/given the profitability firms with more investments have more book leverage (pecking order model) + More collateral allows firms issue more debt and increase leverage. + Expected costs of financial distress are likely to be lower for large (arguably older and more stable) companies (Weiss (1990) ) and hence larger firms can issue more debt. - Higher volatility of earnings can signal unstable environment and debt providers can be reluctant to issue debt. 15

16 6. Data and Results To relate borrowers to lenders, we use DealScan, a database which provides historical information on terms and conditions of syndicated loans in the global commercial market. DealScan provides the information on amount, maturity, payment schedule and participants of each loan, but it lacks the data on financial statements of the companies. To include in the analysis financial statements information, we link DealScan with Compustat North America and S&P Capital IQ. Most of the information for the borrowers we download from Compustat North America using the matching provided by Chava and Roberts (2008). We do hand-matching of lenders with S&P Capital IQ because this database allows finding the information easily even if the company is renamed or merged. We match lenders by their name, country and state (for the United States), SIC code and parent s company name. The sample period is from 1995 to 2014 because most of the information in DealScan is available for this period. To construct weighted average of borrowers and lenders leverage we use the data on 3195 lenders and 2478 borrowers. The sample with non-missing data for all variables consists of around 1000 borrowers 5 with the average of 4.5 observations per borrower and around 1200 lenders with an average of 6.4 observations per lender Sample of Borrowers Sample of borrowers consists of non-financial companies, identified as borrowers in syndicated loans. Left panel of Figure 2 illustrates that majority of our sample constitute North American companies (66.41 % of the sample); Asian and European companies comprise 20.96% and 9.45% respectively. Around 66% of North American companies are the companies from the USA; majority of Asian companies are from Taiwan (5.68 %) and Hong Kong (5.12%). European companies are mostly from the countries - members of European Union. Appendix 2 lists the frequencies of observations for different countries in our sample. Our study is the first one which analyzes companies from different regions in one sample. Data from different regions allows to investigate the differences in capital structure in general rather than differences in capital structure within a particular region or a country. To account for heterogeneity of companies from different 5 The number of borrowers and lenders differs if the dependent variable is book or market leverage. 16

17 countries, we control for time-invariant firm-specific characteristics using the fixed effect panel regression. Figure 2. Distribution of the borrowers and lenders over the geographical regions Borrowers Figure 2 Lenders 20.96% 32.31% 41.28% 9.453% 66.41% 24.85% Africa Europe Middle East Asia Latin America North America Africa Europe Middle East Asia Latin America North America The industries distribution of borrowers in our sample is diverse. The sample includes 58 industries as measured by standard industry classification (SIC) with two-digits codes. Figure 3 presents the histogram of industries distribution of borrowers. As the histogram illustrates, none of the industries dominates the sample considerably: the percentage of most frequently observed industry (SIC 48 Communications ) is around 11 %. The second most frequent industry is SIC 36 ( Electronic and other Electrical Equipment and Components, except Computer Equipment ) 7.2 % and the third most frequent industry is SIC 73 ( Business services ) 5.6%. As similar factors affect the financial policies of the companies within an industry endogeneity, we exclude financial companies from the sample of borrowers to avoid potential endogeneity. Table 3 presents descriptive statistics for the sample of borrowers. To mitigate the influence of extreme observations, we Winsorize all variables at the 1st and 99th percentiles. The upper part of Table 3 shows the descriptive statistics for dependent variables and borrower-specific control variables; the lower part of the table shows the statistics for lenders-specific regressors. Similar to previous studies (see for example Jandik and Makhija (2001), Frank and Goyal (2009)), non-financial companies in our sample finance with debt 40 % of book value and 45.8% of market value of assets. Average profitability (EBITDA-to-Assets), market-to-book, tangibility and 17

18 size are similar to the previous studies. These similarities indicate that our sample is an unbiased selection from a population. Similar to Jandik and Makhija (2001), we measure risk as standard deviation of the percentage change in companies operating income for the past five years, including the year of interest. Some authors (Frank and Goyal (2009)) measure risk as a variance of stock returns, but we prefer to use the standard deviation in operating income because more data is available for the later measure. The lower part of Table 3 presents the descriptive statistics for weighted averages of lenders leverages (Section 4.2 explains computation of weighting).in contrast to borrowers, of lenders have large proportion of liabilities on their balance sheets: lenders finance 70 % of their assets with liabilities (deposits and non-deposit liabilities). On average, the proportion of deposits to total assets is 46.5 %, while the proportion of debt to total assets (lenders leverage) is only around 19%. Appendix 2 presents the correlation matrix for all the variables. Figure 3 Histogram of industries distribution of borrower. Industries (SIC2) percent

19 Table 3 Descriptive statistics for the sample of borrowers The table presents number of observations, means, standard deviations (Std.Dev.), minimums (Min) and maximums (Max) for the borrower-lenders regression. Sample of borrowers consists of non-financial companies, identified as borrowers in the syndicated loans; sample of lenders consists of financial companies identifies as lenders in DealScan. Size is measured as natural logarithm of sales. All sales are converted to U.S. dollars by the exchange rate as of the end of the corresponding year. Appendix 1 provides the definition of all variables. All variables are Winsorized at the 1st and 99th percentiles. The upper part of this table shows dependent variables and borrower-specific control variables; the lower part of the table shows lenders-specific regressors. Mean Std.Dev. Mmin Max Borrower-Specific Factors Book Leverage Market Leverage EBITDA-to-Assets Size Market-to-Book Tangibility Risk Lender-Specific Factors Lenders Leverage Lenders Liabilities-to- Assets Lenders Deposits-to-Assets Observations

20 6.2. Sample of Lenders Sample of lenders consists of the financial companies identified as lenders in syndicated loans by DealScan. We cover the period from because DealScan provides most of the information for this period. Our unbalanced panel of lender-time observations includes around 1200 lenders (depending on which dependent variable we use) with the average of 6.4 observations per lender. Majority of lenders are banks: Commercial Banks (SIC 602) constitute 61 % of the sample, Foreign Banking and Branches and Agencies of Foreign Banks (SIC 608) constitute 17% and Business Credit Institutions constitute around 6%. The rest of the sample is distributed among 23 different financial industries. Right panel of Figure 2 illustartes geographical distribution of lenders in our sample, majority of the sample are North American companies (41.28%) and Asian companies (32.31%). Most of the North American lenders are the U.S. lenders; most of Asian lenders are from Japan (8.17%) and Taiwan (7%). European countries are mostly the members of European Union: Germany (7.23%), France (5.35%), Italy (3.36%). Appendix 2 lists the frequencies of observations for different countries in our sample.table 4 presents summary statistics for lender-borrowers regression. To mitigate the influence of extreme observations, we Winsorize all variables at the 1st and 99th percentiles. The sample consists of around 1200 financial companies identified as lenders in a syndicated loan contracts. Similar to the 80 years average for Amercian banks (Figure 1), total liabilities to assets in our sample has very high average (0.937) and low standard deviation (0.038). Deposits-to-Assets and Non-Deposit Liabilities have the averages of and correspondingly. These averages illustrate that 56.8% of banks assets are funded from deposits and this financing structure is similar to the values presented at Gropp and Heider (2010). The ratios of deposits and non-deposists liabilities are more volatile (with a standard deviation almost 6 times higher) than total liabilities-to-assets. Figure 4shows that most of the variation in the banks leverages comes from the changes in structure between deposits and non-deposits financing. Banks total liabilities-to-assets are stable over time (average varies only between the values of ). The difference between the market and book leverages is striking because market value of assets (measured as market capitalization plus total debt) is on average lower than book value of 20

21 assets. Consider for example one of the lenders in our sample - Allied Irish Banks, p.l.c. (ISE:AIB). Its market capitalization by the end of 2009 was only million Euro, in contrast to total assets of million Euro and total debt of million Euro. Due to the lower market value of assets, the ratio of debt to total assets and deposits to total assets higher than market leverage, but total libilities to market value of assets measure as 1- (total equity/market value of assets) is lower. Weighted averages of the borrowers leverage (0.346) has a typical value for a leverage of non-financial companies (see for example Jandik and Makhija (2001)). Similarities of summary statistics of our sample to the statistics from previous studies suggest that our sample represents the unbiased selection from the population. Correlation matrix of all the variables is at the Appendix 3. Table 4 Summary statistics for lender-borrowers regression The table presents number of observations, means, standard deviations (Std.Dev.), minimums (Min) and maximums (Max) for the lender-borrowers regression. All variables are Winsorized at the 1% level. The sample includes financial companies identified as lenders by DealScan and non-financial companies identified as borrowers by DealScan. The sample consists of American, Asian and European companies for the period Appendix 1contains definitions of all the variables. The upper part of the table shows dependent variables and lender-specific control variables; the lower part of the table shows borrowers-specific regressors. Mean Std.Dev. Min Max Lender-Specific Factors Debt-to-Book Assets Total Liabilities-to-Assets Deposits-to-Assets Total Liabilities-to-Market Value Debt-to-Market Value Deposits-to-Market Value Collateral Loans-to-Assets EBITDA-to-Assets Size Non-Performing Loans Market-to-Book Borrower-Specific Factors Borrowers Book Leverage Borrowers Market Leverage N

22 Figure 4. Structure of Lenders Liabilities Non-Deposit Liabilities Deposits-to-Assets year year Liabilities-to-Assets year 6.3. Effect of the Lenders Leverage on the Leverage of their Borrowers Previous research on banks leverage use total liabilities to assets as main dependent variable (see for example Gropp and Heider (2010)). We use lenders leverage (total debt to total assets) as our main regressor because it suits best for testing the hypotheses of our study due to the following two reasons. Firstly, lenders leverage defined as total debt to total assets is consistent with the definition of borrowers leverage (total debt to total assets). Secondly, if lenders and borrowers leverages are related through tax benefits of debt, the use of total liabilities is not appropriate as they also include non-interest bearing liabilities and deposits. Non-interest bearing liabilities do not provide tax-benefits; and deposits are different from debt of non-financial companies because they can run and they are insured by the government. We argue that banks do not choose deposits solely because of tax benefits, but rather deposits reflect the traditional activities of the banks. If the tax benefits indeed originate only at the banks level (Gornall and Strebulaev (2015)), the banks transmit the tax benefits from their debt to the borrowers debt, rather than deposits or non-deposits liabilities. 22

23 Panel A of Table 5 presents results of estimation of equation (8) with the dependent variable equal to total debt to total assets. To simplify interpretation, the coefficients in Table 5 are standardized by the standard deviation of a corresponding variable. Columns (1)-(5) present estimation results for different specifications. The specification in the first column has only one regressor (lenders book leverage) and the fifth column presents the model with all control variables as described in Table 2. The coefficient on lenders book leverage is positive and significant in all the specifications. Keeping the effect of size, tangibility, market-to-book, profitability and risk fixed, one standard deviation increase in lenders book leverage corresponds to 6.2 % increase in borrowers leverage. This result accepts the first hypothesis of our study about the positive effect of lenders leverage on the leverage of their borrowers. Among the control variables, tangibility has the largest effect on leverage of non-financial companies. We interpret tangibility (ratio of net property plant and equipment to book assets) as collateralizable assets, which increase the ability of a company to issue debt. One standard deviation increase in borrowers collateral, ceteris paribus, corresponds to 16.8% increase of their leverage. Our results of positive effect of tanginbility are similar to Frank and Goyal (2009) and Leary and Roberts (2014). Similar to their studies, we find negative and significant effect of profitability on leverage. Kayhan and Titman (2007) explain negative relation of leverage and profitability by firms passively accumulating profits. The effect of size and growth (market-to-book) is ambiguous. Panel B of the Table 4 presents results of estimation of equation (8) with the dependent variable equal to total debt to market value of assets. Market value of assets is the sum of market capitalization book value of debt. In contrast to Panel A, the coefficient on lenders leverage becomes statistically and economically insignificant. Insignificant coefficient of lenders leverage in regression with market leverage of borrowers is not surprising. Theoretical model of (Gornall and Strebulaev (2015)) describes the leverage measured by debt to total assets, but theoretical predictions about market leverage is missing. 23

24 Table 5 The sample of borrowers consists of nonfinancial companies identified as borrowers in syndicated loans by DealScan. Lenders are financial companies, mostly banks.. All independent variables except of Risk are lagged by one year. Risk is a standard deviatiof a percentage change in the operating income for the past five years and already reflects the information on the past activities of a company. Estimated coefficients are scaled by the corresponding variable s standard deviation. In Panel A, the dependent variable is debt scaled by total assets and in Panel B, dependent variable is total debt scaled by market value of assets. Appendix 1 provides detailed definitions of variables. Standard errors, robust to heteroscedasticity and within-borrower dependence, are in parentheses. Statistical significance at the 1% and 5% and 10% is denoted by *,** and *** respectively. The table shows results of estimating the following equation, with Y kt 1 as lenders average debt to assets and Z kt as borrowers debt to book assets (Panel A) and borrowers debt to market value of assets (Panel B): BB BB ZZ kkkk = ββ 0 + ββ kk + ββ 1 YY kkkk 1 + ββ 2 XX kkkk 1 + uu kkkk Panel A: Book Leverage (1) (2) (3) (4) (5) Lenders Leverage ** * * * * (0.068) (0.070) (0.070) (0.073) (0.068) Size (0.008) (0.008) (0.008) (0.008) Tangibility *** *** ** (0.057) (0.060) (0.069) Market-to-Book (0.005) (0.005) EBITDA-to-Assets *** (0.070) Risk (0.001) Observations R Adjusted R Panel B: Market Leverage (1) (2) (3) (4) (5) Lenders Leverage (0.099) (0.097) (0.097) (0.083) (0.081) Size 0.121* 0.140* *** (0.012) (0.012) (0.010) (0.011) Tangibility 0.183*** 0.169*** 0.177*** (0.089) (0.081) (0.086) Market-to-Book *** *** (0.012) (0.019) EBITDA-to-Assets *** (0.115) Risk (0.002) Observations R Adjusted R Constant Yes Yes Yes Yes Yes Borrower Fixed Effect Yes Yes Yes Yes Yes 24

25 Table 6 illustrates the results of estimation of equation (8) with two other measures of lenders leverage: total liabilities to assets and deposits to assets. As expected the coefficients on both measures of leverage are statistically and economically insignificant. This implies that lenders do not transfer tax-benefits to borrowers through leverage as measured by liabilities to assets and deposits to assets. The explanation for this insignificance is that total liabilities also include non-interest bearing liabilities, which do not provide tax benefits. We also doubt if deposits can provide tax benefits of debt because deposits reflect traditional banking operations, rather than banks chose deposits solely because of tax benefits of debt. Table 6 The sample of borrowers consist of nonfinancial companies identified as borrowers in syndicated loans by DealScan. Lenders are financial companies, mostly banks. All independent variables except of the Risk are lagged by one year. As Risk is a standard deviation of percentage change in the operating income for the past five years and already reflects the information on the past activities of a company. Estimated coefficients are scaled by the corresponding variable s standard deviation. In columns (1)-(2), the dependent variable is debt scaled by total assets and in columns (3)-(4), the dependent variable is total debt scaled of market value of assets. Appendix 1 provides detailed definitions of variables. Standard errors, robust to heteroscedasticity and within-borrower dependence, are in parentheses. Statistical significance at the 1% and 5% and 10% is denoted by *,** and *** respectively. The table shows results of estimating the following equation, with Y kt 1 as lenders average liabilities to assets or deposits to assets: ZZ kkkk = ββ 0 + ββ kk + ββ 1 YY kkkk 1 BB BB + ββ 2 XX kkkk 1 + uu kkkk (1) (2) (3) (4) Book Leverage Book Leverage Market Leverage Market Leverage Lenders Liabilities-to- Assets (0.035) (0.033) Lenders Deposits-to-Assets (0.045) (0.049) Size *** *** (0.008) (0.009) (0.011) (0.011) Tangibility ** ** *** *** (0.068) (0.068) (0.083) (0.084) Market-to-Book *** *** (0.007) (0.007) (0.017) (0.017) EBITDA-to-Assets *** *** *** *** (0.075) (0.077) (0.108) (0.109) Risk (0.001) (0.001) (0.002) (0.002) Observations R Adjusted R Constant Yes Yes Yes Yes Borrower Fixed Effect Yes Yes Yes Yes To test if banks pass to their borrowers cost of distress as described in the first paragraph of Section 4.1, we distinguish between the groups of distressed banks and distressed borrowers. We present 25