Creditor rights and information sharing: the increase in nonbank debt during banking crises

Transcription

1 Creditor rights and information sharing: the increase in nonbank debt during banking crises Abstract We analyze how the protection of creditor rights and information sharing among creditors affect the substitution between bank and nonbank debt during banking crises. Applying a difference-in-differences methodology in a firm-level database across 34 countries, we find that both private and public nonbank credit sources partially substitute bank loans in bank-dependent firms after the onset of the global financial crisis. Strong creditor protection in bankruptcy increases the reduction in bank debt whereas a collateral regime favors the increase in private nonbank debt. Information sharing favors the increase in public debt. The empirical analysis controls for country differences in banking development and competition, shareholder protection, institutional quality and the severity of the crisis, and for alternative model specifications and sample variations. Keywords: banking crises, capital structure, creditor rights, credit registries. JEL classification: G01; G31; G32; O40

2 1. Introduction Literature shows that banking crises cause a reduction in credit supply that damages economic growth, especially in more external finance-dependent firms and in countries with deeper financial development (Kroszner, Laeven, and Klingebiel., 2007; Dell Ariccia, Detragiache, and Rajan, 2008). However, a crucial question for the real effects of a banking crisis is whether firms can compensate for a reduction in credit supply from distressed banks with alternative financing sources. Previous studies basically analyze the role of stock markets (Leary, 2009; Levine, Lin, and Xie, 2016) and trade credit (Nilsen, 2002; Love, Lorenzo, and Sarria-Allende, 2007; García-Appendini and Montoriol-Garriga, 2013; Casey and O Toole, 2014) as alternative corporate financing channels during banking crises. However, less attention has been paid to the substitution between bank and nonbank debt. These studies usually focus on public debt and refer to a specific country. For instance, Leary (2009) and Becker and Ivashina (2014) find evidence of substitution of bank loans with bonds in US listed nonfinancial firms during the global financial crisis. However, Iyer, Lopes, Peydro, and Schoar (2014) suggest there was limited substitution away from bank lending during the recent global crisis for firms in Portugal. Noticeably absent in this recent literature is analysis of the role of private nonbank debt as a potential substitute for bank debt during banking crises. Johnson (1997) and Denis and Mihov (2003) show that private nonbank debt is an economically important financing source even in countries, like the US, with developed financial markets. 1 Denis and Mihov (2003) additionally find that private nonbank debt plays a more important role in the capital structure of firms with low credit quality. This result may suggest an increasing role for private nonbank debt during banking crises because the reduction in the bank credit supply usually coexists with a reduction in the credit quality of borrowers (Kahle and Stulz, 2013). Also noticeably absent in this analysis is the role of creditor protection and information sharing among creditors. Empirical evidence shows the relevance of these two country characteristics for promoting a high bank credit supply and shaping firms capital structure during normal times (Djankov, McLiesh, and 1 Denis and Mihov (2003) show that private nonbank debt accounts for 18.3% of total nonbank debt in US listed firms in the period. Johnson (1997) finds a greater percentage of 53% for nonbank private debt over total debt in a smaller sample of US listed firms in

3 Shleifer, 2007; Haselman and Wachtel, 2010; Vig, 2013). Recently, Levine, Lin, and Xie (2016) have provided evidence that better shareholder protection laws helped equity financing to act as a spare tire during the global financial crisis in financially dependent industries in 36 countries. However, the role of creditor rights and information sharing in shaping the substitution of bank debt with alternative debt financing sources during banking crises remains an unexplored empirical question. We now contribute both to the literature on the real effects of banking crises and the law and finance literature by analyzing how the protection of creditor rights and information sharing among creditors impact on the ability of firms to replace both private and public nonbank debt for bank debt after the onset of the global financial crisis. Our empirical analysis aims to answer the following questions: Do bank-dependent firms substitute bank debt with private and/or public nonbank debt after the onset of a banking crisis? What is the role of a country s creditor protection and information sharing among creditors in this substitution? Which components of creditor rights are most important for the substitution between bank and nonbank debt? Do creditor rights and information sharing have a different role depending on the type of nonbank debt (private versus public) substituting bank financing? The influence of creditor rights and information sharing on the substitution between bank and nonbank debt may operate through different channels. First, strong creditor rights in bankruptcy may promote a greater reduction in bank debt after the onset of the crisis. They allow lenders to seize collateral and force repayment when the crisis increases financial constraints for both banks and debtors. Anticipation by creditors of higher recovery rates in cases of corporate bankruptcy is an argument that is traditionally used to explain a higher supply of bank credit during normal periods in countries with strong protection of creditors (Djankov, McLiesh, and Shleifer, 2007; Haselman and Wachtel, 2010), but it also suggests a greater reduction in bank debt in periods of financial distress in these countries. Consistent with this argument, González (2016) shows a greater reduction in corporate investment after the onset of the global crisis in countries with stronger creditor rights. Moreover, strong protection of creditors in bankruptcy may intensify the higher bankruptcy costs anticipated by shareholders and managers when the crisis increases the probability of bankruptcy for firms. These higher bankruptcy costs provide shareholders and managers with more incentives to seek alternative financing 2

4 sources to bank debt after the onset of the crisis in countries with strong protection of creditor rights. Equity financing, for instance, would reduce bankruptcy costs for shareholders and managers (Denis and Mihov, 2003). Consistent with this prediction, empirical evidence shows that firms in countries with stronger creditor rights have lower asset risk and debt ratios during normal times (Acharya, Amihud, and Litov, 2011; Vig, 2013). Following these arguments, we test whether strong protection of creditors in bankruptcy is associated with a reduction in bank debt ratios after the onset of the global financial crisis. Second, we test whether the same channels that affect the level of private lending in normal times also affect the possibility for firms to replace bank lending with nonbank debt during banking crises. Djankov, McLiesh and Shleifer (2007) show that stronger creditor rights and information sharing in a country promote private lending and increase credit supply. They argue that stronger creditor rights grant more power to creditors in bankruptcy, which makes them more willing to grant credit. Haselman and Wachtel (2010) find that the protection of collateral is even more important than the protection of creditors in bankruptcy for increasing the credit supply and that the presence of a good collateral regime seems to be the prerequisite for an effective bankruptcy regime. Following their arguments, we test whether the protection of collateral, rather than the protection of creditors in bankruptcy, facilitates the substitution of bank debt with nonbank debt after the onset of a banking crisis. We also analyze whether information sharing among creditors favors the substitution of bank with nonbank debt after the onset of the crisis. A banking crisis is associated with new information asymmetries between creditors and debtors when borrowers need to look for new lenders to replace financially constrained bank lenders. Information sharing aims to reduce information asymmetries and may therefore have a significant role in promoting alternative nonbank debt during banking crises. Finally, we test whether creditor rights and information sharing affect the choice between private and public nonbank debt as substitutes for bank debt. The increased probability of financial distress for firms during banking crises may increase the relative attractiveness of private debt over public debt in terms of more flexible renegotiation and liquidation efficiency to reduce the expected costs of financial distress (Chemmanur and 3

5 Fulghieri, 1994; Denis and Mihov, 2003). The stronger the protection of creditor rights, the greater this advantage, so we test whether stronger protection of creditor rights is associated with greater substitution of bank debt with private nonbank debt than with public debt. Moreover, the role of information sharing among creditors in reducing the increased information asymmetries after the onset of the crisis may be more relevant for promoting a greater substitution of public debt for bank debt. Public debt requires lower information asymmetries because arm s-length investors are less effective monitors than private lenders (Diamond, 1984; Fama, 1985). For this reason, public debt is basically issued by high-quality firms with lower information asymmetries, whereas firms with lower quality and/or more information asymmetries use more private debt (Diamond, 1991; Rajan, 1992). As public debt requires lower information asymmetries on borrowers quality, which credit registries aim to reduce, we test whether the role of information sharing during banking crises is associated with facilitating access to public debt more than to private nonbank debt. We use the recent global financial crisis as a natural experiment to identify the exogenous effect of banking crises and country characteristics on corporate debt structure. The global financial crisis simultaneously affected all countries and was less dependent on country characteristics and corporate leverage than any specific banking crisis in a particular country. Moreover, we measure country characteristics at the end of 2006, immediately before the onset of the crisis, to reduce the typical concerns about their potential endogeneity and dependence on banking crises. The global impact of the crisis allows us to assume that different country characteristics before the onset of the crisis are predetermined in our empirical analysis. We use a panel dataset of 6,605 firms from 34 countries and apply a difference-indifferences methodology to compare the ratios of bank and nonbank debt before and after the onset of the crisis. By comparing firms debt structure before and after the onset of the crisis, we can control better for potentially omitted variables (Haselman and Wachtel, 2010; Acharya and Subramanian, 2009; Vig, 2013). Additionally, all the regressions control for the traditional determinants of corporate leverage and potential fixed effects, and include a set of multiple dummy variables to control for any omitted time-variant variables. 4

6 We focus the analysis on bank-dependent firms to better capture the impact of the financial crisis on the substitution between bank and nonbank debt and use firms that are less bank-dependent as a control group. We focus on both private and public nonbank debt, and use the existence of firm ratings for long-term debt before the onset of the crisis as a proxy for access to public debt markets. Firms without a rating do not have access to public debt markets and their changes in nonbank debt are mainly driven by changes in private nonbank debt. Therefore, we use firms without a debt rating to capture the relevance of private nonbank debt as a substitute for bank debt after the onset of the crisis. Differences in changes in nonbank debt ratios in firms with a rating are attributed to the impact of public debt in such firms. We find, on average, that firms which are more bank-dependent substitute nonbank debt for bank debt after the onset of the crisis. We find this substitution effect in firms both with and without a long-term public debt rating. The substitution in firms without a debt rating confirms that bank-dependent firms substitute bank debt with private nonbank debt and not only with public debt. The substitution is only partial because bankdependent firms experience a reduction in total corporate leverage after the onset of the crisis and the increase in nonbank debt ratios does not totally offset the reduction in bank debt ratios. Anyway, firms with long-term public debt ratings suffer a smaller reduction in total debt ratios than firms without a rating and confirm that, on average, access to public debt markets helps to mitigate the effects of a banking crisis on corporate lending. Our analysis confirms the relevance of creditor rights and information sharing to explain differences across countries. Stronger protection of creditor rights and more information sharing among creditors promote greater substitution between bank and nonbank debt in bank-dependent firms after the onset of the crisis. More specifically, we find that the bankruptcy regime is more related to the reduction in bank debt ratios whereas the collateral regime is more important for promoting new private nonbank debt for firms without public debt ratings. Differences in information sharing across countries seem to be more relevant for explaining the change in nonbank debt ratios after the onset of the crisis in firms with a debt rating. The positive influence of both country characteristics for each type of alternative nonbank debt remains after controlling for other country characteristics, such as banking 5

7 development and competition, shareholder protection, overall institutional quality, and the severity of the crisis. We also find that stronger shareholder protection and institutional quality are associated with a lower substitution of bank debt with nonbank debt. This result is consistent with findings by Levine, Lin, and Xie (2016) showing that better shareholder protection promotes greater substitution of bank loans with equity issues during banking crises. Bank market power is associated with a greater increase in nonbank debt ratios in all firms, both with and without a debt rating. The results are robust to alternative specifications and measures of the main variables. As effective protection of creditor rights requires both explicit legal protection and enforcement of the law, we interact the proxies for the legal protection of creditor rights with the rule of law to incorporate both aspects in the proxy for the protection of creditor rights. The results also remain when we define alternative dates for the crisis period, alternative model specifications, and sample variations. Our findings relate to a recent study by Demirgüç-Kunt, Martinez-Peria, and Tressel (2015). They analyze the change in firm leverage and debt maturity after the onset of the global financial crisis in a huge database of listed and non-listed firms from 79 countries. They find, on average, a decline in firm leverage and debt maturity in nonlisted firms and find weaker evidence of a significant change among firms listed on a stock exchange. The reduction in corporate leverage in non-listed firms is greater in countries with less efficient legal systems, weaker information sharing mechanisms, shallower banking systems, and more restrictions on bank entry. However, they only consider the influence of country characteristics on the total leverage of non-listed firms and do not analyze the substitution between bank and nonbank debt, the relevance of private nonbank debt or how country characteristics affect the use of public and private nonbank debt after the onset of the crisis. The rest of the paper is organized as follows. Section 2 describes the data, sample, and variables. Section 3 explains the empirical analysis, and Section 4 presents the results and robustness checks. Finally, Section 5 concludes. 6

8 2. Data and variables 2.1. Data We merge data from several sources. We obtain annual balance-sheet and incomestatement data for publicly-traded firms (in dollars and in real prices) from Compustat Global Vantage Database. Information on bank debt, nonbank debt, and long-term debt ratings is obtained from S&P Capital IQ. This database, unlike Compustat, reports outstanding debt separately, based on whether it is owed to a bank or a nonbank lender. However, it does not provide separate information for private and public nonbank debt. Country-level data on legal protection of creditor rights, information sharing, bank development, bank market power, and institutional quality come from the World Bank Institute s Governance Group, the Heritage Foundation, and the Global Financial Development Database (GFDD) collected by the World Bank. We initially select the 49 countries considered by La Porta, López-de-Silanes, and Shleifer (1998), but then eliminate 10 of them because of lack of data for some firm or country-level variables used in the paper (Ecuador, Mexico, Nigeria, Singapore, South Africa, Sri Lanka, Taiwan, Uruguay, Venezuela, and Zimbabwe). We also eliminate five countries for not having more than 45 observations in all our estimations (Colombia, Egypt, Jordan, Kenia, and Luxembourg). The final number of countries considered is therefore 34, including both developed and developing countries. Finally, since we use lagged values for the firm-level explanatory variables and require data for the pre-crisis and post-crisis sub-periods, we eliminate firms for which we do not have data for more than three consecutive years in each sub-period. We exclude firms with negative book and market equity, negative assets, or negative debt, as of December Further, we retain only those firms that have no missing data for all the variables needed for our baseline empirical specification. We exclude firms whose leverage decisions may reflect special factors: the financial industry (SIC codes ), regulated enterprises (SIC codes ) and not-for profit organizations and governmental enterprises (SICs greater than 8000). We select firms belonging to 20 industrial sectors on a two-digit SIC level. We winsorize all variables at the 5th and 95th 7

9 percentile to lessen the influence of outliers. The final sample contains 6,605 firms from 34 countries. Table 1 reports the final number of observations per country. Following Laeven and Valencia (2012), we consider 2007 as the starting year of the crisis for US and UK, and 2008 for the remaining countries. We use a period of, at least, four years around the onset of the crisis to apply a difference-in-differences approach. Therefore, we compare firms debt structure between the periods and for UK and US, and between the periods and for the remaining countries. A period of several years before and after the onset of the crisis is needed to capture the lagged impact of the crisis on corporate leverage. 2 We check that the results do not change when we define alternative dates for the periods before and after the onset of the crisis. Specifically, we take as the period after the onset of the crisis to better capture a potential delay in the effect of the crisis on the change in firms debt structure, and also follow Demirgüç-Kunt, Martinez-Peria, and Tressel (2015) by considering as the period after the onset of the crisis for all the countries Variables We now describe in detail only the proxies for our main variables: bank dependence, debt structure, creditor rights, and information sharing. Table A1 in the Appendix describes all the variables used in the empirical analysis and their sources. Most of the control variables are self-explanatory and have been used in other cross-country studies. Table A2 in the Appendix reports the overall descriptive statistics and their mean values per country Bank dependence of firms We focus the analysis on bank-dependent firms to better capture the impact of the financial crisis on the substitution between bank and nonbank debt. Following Duchin, Ozbas, and Sensoy (2010) and Kahle and Stulz (2003), we classify firms as bankdependent using data immediately before the onset of the crisis to prevent the 2 Levine, Lin, and Xie (2016) use a period of three years before and after the start year of the crisis to analyze how stock markets substitute bank financing during the global financial crisis. Demirgüç-Kunt, Martinez-Peria, and Tressel (2015) analyze the change in total debt ratios and debt maturity between a period of four years before ( ) and after ( ) the onset of the crisis. 8

10 classification from being endogenous to the crisis. The premise behind this method is the assumption that the supply shock of the crisis is more likely to affect firms that are more financially-dependent on banks before the onset of the crisis. First, we use the ratio of bank debt to total assets in book values at the end of 2006 (Bankdebt06) assuming that a higher ratio indicates greater bank dependence. The results do not change when we define a dummy variable identifying firms with a bank debt ratio above or below the median in the country at the end of Table 1 reports in column (8) the mean of the bank debt ratio in each country at the end of Second, following Kashyap, Lamont, and Stein (1994), Chava and Purnanandam (2011), and Kahle and Stulz (2013), we use the absence of long-term public debt rating at the end of 2006 as a proxy for classifying firms in COMPUSTAT as dependent on private lending. In particular, we define a dummy variable to identify firms with a longterm debt rating (Drating06), which takes the value 1 for firms with a long-term debt rating at the end of 2006, and 0 otherwise. Firms without a rating are more dependent on private lending because they have less access to the bond market and private nonbank debt is the main alternative to bank debt in periods of banking crises. For this reason, we use the absence of a long-term debt rating to proxy for the relevance of private nonbank debt in changes in nonbank debt after the onset of the crisis. Conversely, firms with a debt rating have more access to the bond market and we use differences in the substitution between bank debt and nonbank debt in these firms compared to firms without a rating to proxy for the relevance of public debt. Column (9) in Table 1 shows that the US and Canada have the greatest percentage of firms with a long-term public debt rating, 55.19% and 41.12%, respectively. On the other hand, Pakistan has the minimum percentage and is the only country in our sample where there are no listed firms with a long-term public debt rating. (INSERT TABLE 1 ABOUT HERE) Debt structure We use the ratios of total, bank, and nonbank debt to total assets in book values for each firm in each year to analyze the changes in corporate debt structure after the onset of the global financial crisis. Panel A in Table 1 reports the mean values per country of 9

11 these corporate leverage variables and differences between mean values in the period after and before the onset of the crisis. The average total debt ratio in our sample is 25.24%, with an average bank debt ratio of 14.89% and an average nonbank debt ratio of 10.34%. Columns (3), (5), and (7) show per country the change in each of these three debt ratios after the onset of the crisis. Firms in fifteen countries experienced an average increase in total debt ratios whereas firms in only four countries (Belgium, Hong Kong, India, and Peru) experienced an average reduction in total debt ratios that was statistically significant at conventional levels. Regarding the change in the composition of corporate leverage, fifteen countries experienced on average an increase and eleven a reduction in firms bank debt ratios after the onset of the crisis. Also fifteen countries significantly increased the nonbank debt ratios whereas eight countries reduced firms average nonbank debt ratios. Different average changes are found when we split the sample between firms with high and low bank dependence in Panel B. We classify firms as having high (low) bank dependence if they have a bank debt ratio above (below) the median in the country at the end of We find an average increase in total, bank, and nonbank debt ratios of, respectively, 0.79, 0.64, and 0.15 basis points using data for the whole sample. However, firms with high bank dependence reduced the average bank debt ratio (0.54 basis points) and increased the nonbank debt ratio (0.55 basis points). The total debt ratio did not undergo significant changes as a consequence of this substitution between bank and nonbank debt in firms with high bank debt ratios. A contrary evolution is found in firms with low bank dependence. The bank debt ratio increased and the nonbank debt ratio decreased by a lower amount after the onset of the crisis. The consequence is an increase in the total debt ratio after the onset of the crisis compared to the pre-crisis period in this sub-sample of firms. Figures 1 and 2 confirm a different change in bank and nonbank debt ratios between firms with high and low bank dependence. Figure 1 shows, per country, the mean of bank and nonbank debt ratios before and after the onset of the crisis for firms with high bank dependence. It shows that firms in 22 countries reduced their average bank debt ratio and that firms in 26 countries increased their average nonbank debt ratios after the onset of the crisis. This behavior suggests a substitution of bank debt with nonbank debt in bankdependent firms in most of the countries analyzed in our study. Figure 2 confirms different behavior in firms with bank debt ratios below the median in the country at the 10

12 end of Firms with low bank dependence in 29 countries on average increased their bank debt ratios and only firms in 5 countries reduced their average bank debt ratios after the onset of the crisis. The average reduction in nonbank debt ratios in 19 countries versus the average increase in 15 countries also suggests there is no substitution effect between bank and nonbank debt in this type of firms. However, these descriptive statistics analyze the change in debt ratios after the onset of the crisis without controlling for additional variables affecting corporate leverage. The empirical analysis in Section 3 controls for this aspect. INSERT FIGURE 1 ABOUT HERE INSERT FIGURE 2 ABOUT HERE Creditor rights and information sharing We use several proxies for the legal protection of creditor rights. We begin by using the legal rights index developed by the World Bank to measure a borrower country s overall creditor rights (Creditors06). The strength of the legal rights index measures the degree to which collateral and bankruptcy laws protect the rights of borrowers and lenders. We use values of this index at the end of 2006, immediately before the onset of the crisis, to reduce concerns about its potential endogeneity and dependence on banking crises. Table A2 in the Appendix reports their values per country. In our sample, this index ranges from a minimum value of 2 in Chile to a maximum value of 11 in Hong Kong, New Zealand, and UK. Higher values of this variable indicate that collateral and bankruptcy laws are better designed to protect creditors. Following Haselman and Wachtel (2010), we break down the overall index of creditor rights in two main components: 1) legal rules designed to protect individual creditors claims outside bankruptcy (Collateral06) and 2) the collective enforcement regime established for bankruptcy (Bankruptcy06). The index for the collateral regime is the sum of seven indicators measuring: 1) If a general, rather than specific, description of assets is permitted in collateral agreements; 2) If a general, rather than specific, description of debt is emitted in collateral agreements; 3) If any legal or natural person may grant or take security in the property; 4) If a unified registry that includes charges 11

13 over movable property operates; 5) If secured creditors have priority outside of bankruptcy; 6) If parties may agree on enforcement procedures by contract; 7) If creditors may seize and sell collateral out of court. Higher values of this index indicate higher protection of creditor rights in terms of collateral. As an indicator of the bankruptcy regime, we use the traditional index developed by Djankov, McLiesh, and Shleifer (2007). This index is the sum of four indicators: 1) creditor consent for reorganization; 2) no automatic stay; 3) secured creditors first, and 4) management out. The index ranges from 0 (weak creditor rights) to 4 (strong creditor rights in bankruptcy) and is the most widely used index for the protection of creditor rights in the Law and Finance literature (Haselman and Wachtel, 2010). In our sample, Bankruptcy06 takes a value of 0 in France and Peru. The maximum value of 4 is found in Hong-Kong, New Zealand, and UK. As effective protection of creditor rights not only requires explicit legal protection but also enforcement of the law, we additionally interact the above indicators of the legal protection of creditor rights with a variable capturing countries law enforcement. We use the rule of law measure provided by the World Bank in Kaufman, Kraay, and Mastruzzi (2009) to define these interaction terms. In the sample, this proxy for the legal protection of creditor rights (Creditors06-RoL) ranges from a maximum value of in UK to a minimum of in Pakistan. We use three indicators from the World Bank to proxy for information sharing among creditors at the end of Our main indicator is the depth of credit information (Depth information06). It theoretically ranges from 0 to 8, with higher values indicating the availability of more credit information. We check that the results remain when we use the coverage of public registries (Cov-Public registry06) and private bureaus (Cov- Private bureau06). These two indicators measure the number of individuals and firms listed, respectively, in a public credit registry and in a private bureau with information on repayment history, unpaid debts, or credit outstanding from the past five years, scaled by the country s adult population. Houston, Lin, Lin, and Ma (2010) used these three indicators to analyze the impact of information sharing on bank risk-taking and economic growth. 12

14 Firm and country-level control variables We use the traditional determinants of firms capital structure indicated by Rajan and Zingales (1995) as firm-level control variables. These include profitability (Profitability), which proxies for higher tax benefits and higher internal cash available for investment funding; the ratio of market value to book value of assets (QTobin), which proxies for better investment opportunities and thus lower potential agency costs; firm size (Size), which proxies for lower bankruptcy costs and information asymmetries; and asset tangibility (Tangibility), which proxies for higher collateral value and bankruptcy recovery rates. We use a wide set of country control variables to rule out the possibility that effects attributed to the protection of creditor rights and/or information sharing among creditors are not caused by alternative country characteristics at the onset of the crisis. We specifically control for countries banking development (Bank development06), measured as the ratio of private credit by deposit money banks to GDP, because previous studies find a greater negative bank lending channel effect for banking crises in countries with more developed banking systems (Rajan and Zingales, 1998; Kroszner, Laeven, and Klingebiel, 2007). We use the Lerner index (Lerner06) as a proxy for bank market power because there is extensive evidence suggesting that bank competition affects the impact of banking crises on debtors (Ongena, Smith, and Michalsen, 2003; Carvalho, Ferreira, and Matos 2015). We use the Kaufman, Kraay, and Mastruzzi (2009) KKZ index as a proxy for the overall quality of institutions in each country (KKZ06). We additionally include an index for information disclosure (Disclosure06) and an index of shareholders protection (Antidirector06). We need to control for both overall institutional quality and shareholder protection laws because they boost stock market development by reducing the ability of corporate insiders to expropriate resources from minority shareholders (La Porta, López-de-Silanes, and Shleifer 1997, 1998). Levine, Lin, Xie (2016) have recently shown that stock markets act as a spare tire during banking crises, providing an alternative corporate financing channel. For these reasons, we expect better institutional quality and shareholder protection laws to be associated with a greater reduction in corporate leverage after the onset of the banking crisis. 13

15 All country-level variables are measured at the end of 2006 to make them exogenous to the financial crisis. Regulators may change some rules affecting investor protection and the institutional characteristics of each country. Surviving banks buying failed banks during the crisis might also affect bank competition. However, the simultaneity of the global crisis increasing firms financial constraints across countries makes it possible to assume that country characteristics are predetermined immediately before the onset of the crisis. 3. Empirical analysis We employ a difference-in-differences methodology in which we compare bank and nonbank debt ratios before and after the onset of the crisis. The identification strategy uses the global financial crisis as a shock to credit and exploits exogenous variation in the degree to which firms are bank-dependent. Following previous studies of banking crises, we assume that firms that depend more on banks before the onset of the crisis would be more affected by a reduction in the credit supply as a consequence of problems in banks, so would be more likely to substitute bank debt with nonbank debt (Chava and Purnanandam, 2011; Kahle and Stulz, 2013). Therefore, less bank-dependent firms act as the control group. In the regressions, we explicitly control for traditional firm-level variables explaining capital structure, use a firm fixed-effects model to capture unobserved heterogeneity in the firm-specific (i.e., time-invariant) bank and nonbank debt, and control for additional unobserved heterogeneity including industry and country timevariant characteristics. The first baseline specification is: Debtit=α0 + α1 Crisiskt + α2 Crisiskt*Bankdep06i + α3 Crisiskt*Drating06i + α4 Firm controlsit-1 +λjt + θkt+ γi + μit [1] where subscripts i, j, k, and t indicate, respectively, firm, industry, country, and year. Debt is the particular ratio of total, bank, or nonbank debt. Crisis is a dummy variable that takes value 1 after the onset of the crisis. Otherwise, Crisis takes value 0. 14

16 Bankdep06 is the ratio of bank debt to total assets at the end of 2006 and is our main proxy for firms bank dependence. Drating06 is a dummy variable identifying firms with a long-term public debt rating at the end of Firm controls refers to the set of firmspecific control variables, which include firm profitability, growth opportunities, size, and asset tangibility. We lag all firm control variables by one year in order to avoid simultaneity with corporate leverage variables. We use alternative specifications of dummy variables to control for potentially omitted country, industry, and firm effects, both time invariant and time variant. Initially, all regressions include three specific effects: industry-year (λjt), country-year (θkt), and firm-specific (γi) effects. The three sets of specific effects should control for most shocks affecting capital structure, and their inclusion reduces concerns about the potential omission of control variables. 3 The industry-year specific effect controls for worldwide industry shocks. The country-year specific effects control for aggregate country-specific shocks and changes in institutional and regulatory country characteristics. This approach has the advantage that it is not necessary for additional country and industry time-varying variables to be included in the regression on their own. As the crisis simultaneously occurs in most of the countries, country-year dummy variables could absorb the effect of the crisis dummy variable. For this reason, we check that the results do not change when country-year dummy variables are excluded from the regressions. Finally, firm-specific effects control for omitted firm, industry, and country-level variables that are timeinvariant. 4 Standard errors are heteroskedasticity-consistent and clustered at country level to capture correlations of different firms affected by the same country-level characteristics. Initially, we only include the crisis dummy variable and the interaction Crisis*Bankdep06. Under this initial specification, α1 captures the change in corporate leverage after the onset of the crisis in firms that are less bank-dependent and α2 captures the different impact of the crisis on the particular ratio of corporate leverage in firms that are more bank-dependent. Bank-dependent firms are more negatively affected by a bank 3 Mclean, Zhang, and Zhao (2012) apply a similar structure of dummy variables to analyze the impact of investor protection on investment, finance, and growth in regressions using firm-level data from 44 countries. 4 For instance, corporate governance variables, such as ownership structure and the characteristics of the board of directors, are omitted but as they are usually stable over time, their effect on investment behavior is controlled for the firm-fixed effects included in the regressions. 15

17 supply shock and can look for alternative financing to bank loans after the onset of the crisis. For this reason, we expect a negative value of α2 when the dependent variable is the bank debt ratio and a positive coefficient when the dependent variable is the nonbank debt ratio. When the interaction Crisis*Drating06 is included, α2 captures the different impact of the crisis on the particular debt ratio in firms that are more bank-dependent and do not have a long-term debt rating. As such firms have less access to public debt markets, we use the change in nonbank debt ratios in firms without a long-term debt rating as a proxy for the change in private nonbank debt. α3 now captures the different impact of the crisis on firms with debt ratings and we associate their potential different behavior with their access to public debt markets and changes in public nonbank debt. The firm-fixed effects subsume the level effect of Bankdebt06 and Drating06 because they are measured only once per firm (at the end of 2006). We extend model [1] to analyze how differences in the protection of creditor rights and information sharing across countries shape the impact of the financial crisis on the substitution between bank and nonbank debt in bank-dependent firms. We estimate the following model in the sub-sample of bank-dependent firms: Debtit=β0 + β1 Crisiskt + β2 Crisiskt*Creditors06k+ β3 Crisiskt*Registry06k + β4 Crisiskt*Bankdep06i + β5 Crisiskt*Drating06i + β6 Crisiskt*Country controls06k + β7 Firm controlsit-1 + λjt + θkt+ γi + μit [2] Creditors06k and Registry06k are, respectively, the proxies for the legal protection of creditor rights and information sharing in country k at the end of β2 and β3 capture how the impact of the crisis on bank and nonbank debt ratios changes when, respectively, the protection of creditor rights is stronger or when there is more information sharing among creditors. The fixed-effects model subsumes the level effect of creditor rights and credit registries (because they are measured once per country), and other country-industry fixed effects affecting corporate leverage as long as they are time-invariant. The interaction Crisiskt*Country controls06k controls for other country variables, apart from creditor rights and information sharing, potentially affecting the substitution between 16

18 bank and nonbank debt after the onset of the global financial crisis. Country controls06k includes bank development, bank market power, overall quality of institutions, information disclosure, and the legal protection of shareholders. As with Creditors06k and Registry06k, all the country control variables are measured at the end of Other variables are the same as in model [1]. 4. Results 4.1. Substitution between bank and nonbank debt during the global financial crisis Table 2 reports the results of model [1] using total, bank, and nonbank debt ratios as dependent variables. The non-significant coefficients of the crisis dummy variable in columns (1), (4), and (7) indicate that there are no significant changes in the ratios of total, bank, and nonbank debt after the onset of the crisis when we focus on the overall sample without considering firms bank dependence and after controlling for traditional determinants of corporate leverage. Different conclusions emerge when we focus on firms bank dependence by including Crisis*Bankdep06 in the regressions. The coefficient for Crisis*Bankdep06 is negative in column (5), where the dependent variable is the bank debt ratio, and positive in column (8), where the dependent variable is the nonbank debt ratio. These coefficients indicate that firms that were more bank-dependent increased their nonbank debt ratios, replacing bank debt, after the onset of the crisis. However, the substitution of bank debt with nonbank debt in bank-dependent firms is only partial because the negative and significant coefficient for Crisis*Bankdep06 in column (2) suggests an average reduction in total debt ratios after the onset of the crisis in more bank-dependent firms. The positive and significant coefficient for Crisis*Drating06 in column (9) indicates that firms with a long-term debt rating were able to increase their nonbank debt ratios more through access to public debt markets than bank-dependent firms without a rating. The coefficient for Crisis*Bankdep06 continues to be positive and significant in column (9), after including Crisis*Drating06, and suggests the relevance of private nonbank debt as a substitute for bank debt. Firms that are more bank-dependent and do not have a long-term debt rating are able to partially substitute bank lending for nonbank debt after the onset of the crisis. The positive and significant coefficient for Crisis in 17

19 column (5) and its non-significant coefficient in column (8) indicate different behavior for firms that are less bank-dependent. Such firms even increased their bank debt ratios after the onset of the crisis but their nonbank debt ratios did not change. In economic terms, the results in column (8) imply that the increase in nonbank debt ratios after the onset of the crisis in firms at the 75 th percentile of bank dependence is 8.16% greater than in firms at the 25 th percentile of bank dependence. Focusing on firms without rating, the coefficients in column (9) indicate that firms without a long- term debt rating at the 75 th percentile of bank dependence increase their nonbank debt ratio after the onset of the crisis by 9.6% more than firms without a debt rating at the 25 th percentile of bank dependence. This latter increase indicates the average relevance of private nonbank debt as a substitute for bank debt after the onset of the crisis. The coefficients of the firm-level control variables are as expected. The significant negative coefficients of Profitability in all the estimations are consistent with higher possibilities of retained earnings reducing the use of debt, following traditional arguments of the pecking order theory. This result is consistent with previous findings by Rajan and Zingales (1995), and González and González (2008) in international samples of firms. The coefficients of QTobin are significant and positive when the dependent variable is the total debt ratio. Otherwise, the coefficients of QTobin are not statistically significant. The positive coefficients of QTobin are again consistent with the arguments of the pecking order theory because higher growth opportunities increase financing needs and the relevance of information asymmetries in the firm. These characteristics lead to a higher preference for debt over equity according to the pecking order theory. The positive coefficients of Size in all the estimations are consistent with firm size being an inverse proxy for the probability of bankruptcy. This result is consistent with findings by Rajan and Zingales (1995) and Flannery and Rangan (2006), among others. The coefficients of Tangibility are significant and positive. These coefficients suggest that firms with a higher proportion of tangible assets have greater corporate leverage and a higher proportion of both bank and nonbank debt. This result is consistent with tangible assets increasing the value of firm assets as collateral to facilitate long-term debt and with previous findings by, among others, Rajan and Zingales (1995), Flannery and Rangan (2006), and González and González (2008). 18

20 INSERT TABLE 2 ABOUT HERE Table 3 shows the robustness of the results to alternative specification models, definitions of the crisis period, and sample variations. To save space, we only report results for bank and nonbank debt ratios in the specification capturing differences in bankdependent firms and in firms with long-term debt ratings. The results do not change in columns (1) and (7) when we exclude country-year dummy variables to avoid any potential correlation with dummy crisis variables. Nor do they change in columns (2) and (8) when standard errors are not clustered at the country level, or when we use alternative definitions for the period after the onset of the crisis. In particular, we use the period in columns (3) and (9) to better capture a lagged change in firms capital structure after the onset of the crisis, and we use 2008 as the starting year of the crisis for all the countries in columns (4) and (10). Finally, the results indicating a substitution between bank and private nonbank debt do not change in columns (5) and (11) when UK and US firms are excluded from the regressions, and in columns (6) and (12) when Japanese firms are also excluded. INSERT TABLE 3 ABOUT HERE We additionally check whether the results remain unchanged when we split the sample in two sub-samples around the median of the bank debt ratio in the particular country at the end of Panel A in Table 4 reports the results of model [1] for firms with a bank debt ratio below the median (Bankdebt06LOW) and Panel B for firms with a bank debt ratio above the median in the particular country at the end of 2006 (Bankdebt06HIGH). The results are different in each sub-sample. The significant positive coefficient for Crisis and the non-significant coefficients for Crisis* Bankdep06 and Crisis*Drating06 in columns (1)-(3) of Panel A indicate an increase in bank debt ratios after the onset of the crisis in firms that are less bank-dependent, both with and without a debt rating. Moreover, we do not find significant changes in nonbank debt ratios during the crisis in less bank-dependent firms because none of the coefficients of Crisis, Crisis* Bankdep06, and Crisis*Drating06 are statistically significant at conventional levels in columns (4)- 19

21 (6) of Panel A. However, the results in Panel B for firms that are more bank-dependent confirm the partial substitution between bank and nonbank debt in this sub-sample of firms. The negative and significant coefficient for Crisis in column (1) of Panel B indicates that firms that are more bank-dependent reduced their bank debt ratios after the onset of the global banking crisis. The negative coefficients of Crisis* Bankdep06 in columns (2) and (3) even suggest that the reduction in bank debt ratios is higher, the greater the bank dependence in this subsample of firms. The positive and significant coefficient for Crisis* Bankdep06 in columns (5) and (6) indicates that firms that are more bank-dependent increase their nonbank debt ratios after the onset of the crisis, the greater their bank dependence before the onset of the crisis. These results are again consistent with a substitution effect between bank debt and nonbank debt in firms that are more bank-dependent. Finally, the positive coefficient for Crisis*Drating06 in column (6) indicates that firms with high bank dependence and a long-term debt rating substitute bank debt with nonbank debt to a greater extent, consistent with their greater access to public debt markets. INSERT TABLE 4 ABOUT HERE 4.2. The role of creditor rights and information sharing We now analyze how the country s legal protection of creditor rights and the degree of information sharing among creditors shape the substitution between bank and nonbank debt in firms with high bank dependence. We focus on firms with high bank dependence (firms with a bank debt ratio above the median in the country at the end of 2006) as previous tables indicate that these are the only firms that substituted bank with nonbank debt after the onset of the crisis. Table 5 reports the results of model [2] using alternative proxies for countries creditor protection. We use both the indicator of the overall legal protection of creditor rights (Creditors06) and its two components (Collateral06 and Bankruptcy06). We additionally interact the above variables with the country s law enforcement to better capture the effective protection of creditor rights (Creditors06-RoL, Collateral06-RoL and Bankruptcy06-RoL). We use the depth of credit information (Depth information06) as the main proxy for information sharing among creditors. The coefficients for 20