Wo r k i n g Pa P e r S e r i e S no 1096 / SePTeMBer 2009 The deTerMinanTS of pOi Bank caPiTal STrucTure EpiSNF by Reint Gropp and Florian Heider WO R K I N G PA P E R S E R I E S N O 10 / S E P T E M B E R 20 THE DETERMINANTS OF BANK CAPITAL STRUCTURE by Reint Gropp and Florian Heider In 2009 all ECB publications feature a motif taken from the €200 banknote This paper can be downloaded without charge from http://www.ecb.europa.eu or from the Social Science Research Network electronic library at http://ssrn.com/abstract_id=1478335 We are grateful to Markus Baltzer for excellent research assistance Earlier drafts of the paper were circulated under the title “What can corporate finance say about banks’ capital structures?” We would like to thank Franklin Allen, Allan Berger, Bruno Biais, Arnoud Boot, Charles Calomiris, Mark Carey, Murray Frank, Itay Goldstein, Vasso Ioannidou, Luc Laeven, Mike Lemmon, Vojislav Maksimovic, Steven Ongena (the editor), Elias Papaioannou, Bruno Parigi, Joshua Rauh, Joao Santos, Christian Schlag, an anonymous referee, participants and discussants at the University of Frankfurt, Maastricht University, EMST Berlin, American University, the IMF, the ECB, the ESSFM in Gerzensee, the Conference “Information in bank asset prices: theory and empirics” in Ghent, the 2007 Tor Vergata Conference on Banking and Finance, the Federal Reserve Board of Governors, the Federal Reserve Bank of San Francisco, the European Winter Finance Conference, the Financial Intermediation Research Society conference and the Banca d’Italia for helpful comments and discussions This paper reflects the authors’ personal opinions and does not necessarily reflect the views of the European Central Bank or the Eurosystem European Business School, Wiesbaden and Centre for European Economic Research (ZEW) Mannheim, Germany; e-mail: reint.gropp@ebs.edu Corresponding author: European Central Bank, Kaiserstrasse 29, D-60311 Frankfurt am Main, Germany; e-mail: florian.heider@ecb.europa.eu © European Central Bank, 2009 Address Kaiserstrasse 29 60311 Frankfurt am Main, Germany Postal address Postfach 16 03 19 60066 Frankfurt am Main, Germany Telephone +49 69 1344 Website http://www.ecb.europa.eu Fax +49 69 1344 6000 All rights reserved Any reproduction, publication and reprint in the form of a different publication, whether printed or produced electronically, in whole or in part, is permitted only with the explicit written authorisation of the ECB or the author(s) The views expressed in this paper not necessarily reflect those of the European Central Bank The statement of purpose for the ECB Working Paper Series is available from the ECB website, http://www.ecb.europa eu/pub/scientific/wps/date/html/index en.html ISSN 1725-2806 (online) CONTENTS Abstract Non-technical summary Introduction Data and descriptive statistics 10 Corporate finance style regressions 13 Decomposing leverage 19 Bank fixed effects and the speed of adjustment 21 Regulation and bank capital structure 23 Discussion and future research 26 Conclusion 29 Figures and tables 30 References 34 Appendices 43 European Central Bank Working Paper Series 49 ECB Working Paper Series No 1096 Septembre 2009 Abstract The paper shows that mispriced deposit insurance and capital regulation were of second order importance in determining the capital structure of large U.S and European banks during 1991 to 2004 Instead, standard cross-sectional determinants of non-financial firms’ leverage carry over to banks, except for banks whose capital ratio is close to the regulatory minimum Consistent with a reduced role of deposit insurance, we document a shift in banks’ liability structure away from deposits towards non-deposit liabilities We find that unobserved timeinvariant bank fixed effects are ultimately the most important determinant of banks’ capital structures and that banks’ leverage converges to bank specific, time invariant targets Key words: bank capital, capital regulation, capital structure, leverage JEL-codes: G32, G21 ECB Working Paper Series No 1096 Septembre 2009 Non-technical summary The objective of this paper is to examine whether capital requirements are a first-order determinant of banks’ capital structure using the cross-section and time-series variation in a sample of large, publicly traded banks spanning 16 countries (the United States and the EU-15) from 1991 until 2004 To answer the question, we borrow extensively from the empirical corporate finance literature that has at length examined the capital structure of non-financial firms The literature on firms’ leverage i) has converged on a number of standard variables that are reliably related to the capital structure of nonfinancial firms and ii) has examined its transitory and permanent components The evidence in this paper documents that the similarities between banks’ and nonfinancial firms’ capital structure may be greater than previously thought Specifically, this paper establishes five novel and interrelated empirical facts First, standard cross-sectional determinants of firms’ capital structures also apply to large, publicly traded banks in the US and Europe, except for banks close to the minimum capital requirement The sign and significance of the effect of most variables on bank capital structure are identical to the estimates found for non-financial firms This is true for both book and market leverage, Tier capital, when controlling for risk and macro factors, for US and EU banks examined separately, as well as when examining a series of cross-sectional regressions over time Second, the high levels of banks’ discretionary capital observed not appear to be explained by buffers that banks hold to insure against falling below the minimum capital requirement Banks that would face a lower cost of raising equity at short notice (profitable, dividend paying banks with high market to book ratios) tend to hold significantly more capital Third, the consistency between non-financial firms and banks does not extend to the components of leverage (deposit and non-deposit liabilities) Over time, banks have financed their balance sheet growth entirely with non-deposit liabilities, which implies that the composition of banks’ total liabilities has shifted away from deposits ECB Working Paper Series No 1096 Septembre 2009 Fourth, unobserved time-invariant bank fixed-effects are important in explaining the variation of banks’ capital structures Banks appear to have stable capital structures at levels that are specific to each individual bank Moreover, in a dynamic framework, banks’ target leverage is time invariant and bank specific Both of these findings for banks mirror those found for non-financial firms Fifth, controlling for banks’ characteristics, we not find a significant effect of deposit insurance on the capital structure of banks This is in contrast to the view that banks increase their leverage in order to maximise the subsidy arising from incorrectly priced deposit insurance Together, the empirical facts established in this paper suggest that capital regulation and buffers may only be of second order importance in determining the capital structure of most banks ECB Working Paper Series No 1096 Septembre 2009 Introduction This paper borrows from the empirical literature on non-financial firms to explain the capital structure of large, publicly traded banks It uncovers empirical regularities that are inconsistent with a first order effect of capital regulation on banks’ capital structure Instead, the paper suggests that there are considerable similarities between banks’ and non-financial firms’ capital structures Subsequent to the departures from Modigliani and Miller (1958)’s irrelevance proposition, there is a long tradition in corporate finance to investigate the capital structure decisions of non-financial firms But what determines banks’ capital structures? The standard textbook answer is that there is no need to investigate banks’ financing decisions, since capital regulation constitutes the overriding departure from the Modigliani and Miller propositions: “Because of the high costs of holding capital […], bank managers often want to hold less bank capital than is required by the regulatory authorities In this case, the amount of bank capital is determined by the bank capital requirements (Mishkin, 2000, p.227).” Taken literally, this suggests that there should be little cross-sectional variation in the leverage ratio of those banks falling under the Basel I regulatory regime, since it prescribes a uniform capital ratio Figure shows the distribution of the ratio of book equity to assets for a sample of the 200 largest publicly traded banks in the United States and 15 EU countries from 1991 to 2004 (we describe our data in more detail below) There is a large variation in banks' capital ratios.1 Figure indicates that bank capital structure deserves further investigation Figure (Distribution of book capital ratios) The objective of this paper is to examine whether capital requirements are indeed a firstorder determinant of banks’ capital structure using the cross-section and time-series variation in our sample of large, publicly traded banks spanning 16 countries (the United States and the EU-15) from 1991 until 2004 To answer the question, we borrow extensively from the empirical corporate finance literature that has at length examined the capital structure of non- The ratio of book equity to book assets is an understatement of the regulatory Tier-1 capital ratio since the latter has risk-weighted assets in the denominator Figure shows that the distribution of regulatory capital exhibits the same shape as for economic capital, but is shifted to the right Banks’ regulatory capital ratios are not uniformly close to the minimum of 4% specified in the Basel Capital Accord (Basel I) ECB Working Paper Series No 1096 Septembre 2009 financial firms.2 The literature on firms’ leverage i) has converged on a number of standard variables that are reliably related to the capital structure of non-financial firms (for example Titman and Wessels, 1988, Harris and Raviv, 1991, Rajan and Zingales, 1995, and Frank and Goyal, 2004) and ii) has examined the transitory and permanent components of leverage (for example Flannery and Rangan, 2006, and Lemmon et al., 2008) The evidence in this paper documents that the similarities between banks’ and nonfinancial firms’ capital structure may be greater than previously thought Specifically, this paper establishes five novel and interrelated empirical facts First, standard cross-sectional determinants of firms’ capital structures also apply to large, publicly traded banks in the US and Europe, except for banks close to the minimum capital requirement The sign and significance of the effect of most variables on bank leverage are identical when compared to the results found in Frank and Goyal (2004) for US firms and Rajan and Zingales (1995) for firms in G-7 countries This is true for both book and market leverage, Tier capital, when controlling for risk and macro factors, for US and EU banks examined separately, as well as when examining a series of cross-sectional regressions over time Second, the high levels of banks’ discretionary capital observed not appear to be explained by buffers that banks hold to insure against falling below the minimum capital requirement Banks that would face a lower cost of raising equity at short notice (profitable, dividend paying banks with high market to book ratios) tend to hold significantly more capital Third, the consistency between non-financial firms and banks does not extend to the components of leverage (deposit and non-deposit liabilities) Over time, banks have financed their balance sheet growth entirely with non-deposit liabilities, which implies that the composition of banks’ total liabilities has shifted away from deposits Fourth, unobserved time-invariant bank fixed-effects are important in explaining the variation of banks’ capital structures Banks appear to have stable capital structures at levels that are specific to each individual bank Moreover, in a dynamic framework, banks’ target leverage is time invariant and bank specific Both of these findings confirm Lemmon et al.’s An early investigation of banks’ capital structures using a corporate finance approach is Marcus (1983) He examines the decline in capital to asset ratios of US banks in the 1970s ECB Working Paper Series No 1096 Septembre 2009 (2008) results on the transitory and permanent components of non-financial firms’ capital structure for banks Fifth, controlling for banks’ characteristics, we not find a significant effect of deposit insurance on the capital structure of banks This is in contrast to the view that banks increase their leverage in order to maximise the subsidy arising from incorrectly priced deposit insurance Together, the empirical facts established in this paper suggest that capital regulation and buffers may only be of second order importance in determining the capital structure of most banks Hence, our paper sheds new light on the debate whether regulation or market forces determine banks’ capital structures Barth et al (2005), Berger et al (2008) and Brewer et al (2008) observe that the levels of bank capital are much higher than the regulatory minimum This could be explained by banks holding capital buffers in excess of the regulatory minimum Raising equity on short notice in order to avoid violating the capital requirement is costly Banks may therefore hold discretionary capital to reduce the probability that they have to incur this cost.3 Alternatively, banks may be optimising their capital structure, possibly much like nonfinancial firms, which would relegate capital requirements to second order importance Flannery (1994), Myers and Rajan (1998), Diamond and Rajan (2000) and Allen et al (2009) develop theories of optimal bank capital structure, in which capital requirements are not necessarily binding Non-binding capital requirements are also explored in the market discipline literature.4 While the literature on bank market discipline is primarily concerned with banks’ risk taking, it also has implications for banks’ capital structures Based on the market view, banks’ capital structures are the outcome of pressures emanating from shareholders, debt holders and depositors (Flannery and Sorescu, 1996, Morgan and Stiroh, 2001, Martinez Peria and Schmuckler, 2001, Calomiris and Wilson, 2004, Ashcraft, 2008, and Flannery and Rangan, 2008) Regulatory intervention may then be non-binding and of secondary importance Berger et al (2008) estimate partial adjustment models for a sample of U.S banks Their main focus is the adjustment speed towards target capital ratios and how this adjustment speed may differ for banks with different characteristics (see also our section 5) Their paper is less concerned with the question of whether capital regulation is indeed a binding constraint for banks See Flannery and Nikolova (2004) and Gropp (2004) for surveys of the literature ECB Working Paper Series No 1096 Septembre 2009 ECB Working Paper Series No 1096 Septembre 2009 37 0.037 -0.072 0.075 0.006*** 0.001 0.027** 0.010 -0.019*** 0.004 0.035 0.004 0.074 0.007*** 0.001 -0.009 0.011 -0.021*** 0.003 -0.027*** se Profits se Log(Size) se Collateral se Dividends se Log(Risk) No Yes 2415 0.76 No No 2415 0.047 0.044 se 0.74 0.79 Yes No 2415 0.044 1.151*** 0.003 0.003 1.260*** 0.002 1.197*** se constant Time Fixed Effects Country Fixed Effects Number of observations R2 0.004 -0.021*** 0.010 -0.003 0.001 0.006*** 0.083 -0.141* 0.034 -0.463*** -0.033*** -0.504*** -0.486*** Market-to-book ratio -0.015*** Market leverage Dependent variable -0.010*** 0.003 -0.008** 0.008 0.036*** 0.001 0.005*** 0.049 -0.089* 0.016 -0.026* 0.46 No No 2415 0.022 0.54 No Yes 2415 0.023 0.820*** 0.001 0.775*** 0.002 -0.016*** 0.003 -0.008** 0.009 0.020** 0.001 0.007*** 0.053 -0.041 0.016 -0.014 Book leverage 0.50 Yes No 2415 0.022 0.760*** 0.002 -0.018*** 0.003 -0.008*** 0.008 0.020** 0.001 0.007*** 0.062 -0.118* 0.015 -0.011 The dependent variable is market leverage (columns to 3) or book leverage (columns to 6) Standard errors are adjusted for clustering at the bank level ***, ** and * denote statistical significance at the 1%, the 5% and the 10% level respectively Lict = β0 + β1MTBict −1 + β2 Profict −1 + β3Ln(Sizeict −1 ) + β4Collict −1 + β5 Divict + β6 Ln( Riskict −1 ) + cc + ct + uict The sample consists of the 200 largest publicly traded banks in the U.S and the EU from the Bankscope database from 1991 to 2004 See Appendix I for the definition of variables The table shows the results from estimating the model below without country and/or time fixed effects: Table VIII: Time and country fixed effects 38 ECB Working Paper Series No 1096 Septembre 2009 0.075 0.537 0.004 0.034*** 0.059 0.118* 0.027 -0.009 -0.011 0.027 -0.009 0.061 -0.099* 0.004 -0.028*** 0.517 -1.924*** 0.066 1.629*** -0.013 0.090 0.120 2408 0.32 0.115 2408 0.35 0.108 2408 0.40 0.103 2408 0.37 se Number of observations R2 1.274*** -0.447*** constant 1.536*** 0.010 -0.005 0.026 -0.005 0.064 -0.100 0.005 -0.028*** 0.584 -2.099*** -0.313*** 0.010 -0.009 0.028 -0.005 0.065 0.125* 0.005 0.035*** 0.553 1.772*** 0.077 Non-deposit liab Deposits (over book value of assets) -0.011 -0.049 0.010 se se se se se se Non-deposit liab Deposits (over market value of assets) -0.137** -0.369*** 0.010 Log(Risk) Dividends Collateral Log(Size) Profits Market-to-book ratio Dependent variable The dependent variable (Liability) is either deposits divided by the market or book value of assets (columns ans4) or minus deposits divided by the market or book value of assets (columns and 3) Standard errors are adjusted for clustering at the bank level ***, ** and * denote statistical significance at the 1%, the 5% and the 10% level respectively Liabict = β0 + β1MTBict −1 + β2 Profict −1 + β3Ln(Sizeict −1 ) + β4Collict −1 + β5 Divict + β6 Ln( Riskict −1 ) + cc + ct + uict The sample consists of the 200 largest publicly traded banks in the U.S and the EU from the Bankscope database from 1991 to 2004 See Appendix I for the definition of variables The table shows the result of estimating: Table IX: Decomposing bank leverage ECB Working Paper Series No 1096 Septembre 2009 39 0.146 2415 0.76 0.88 0.717*** 0.003 -0.016*** 0.007 -0.010 0.013 0.006 0.006 0.013** 0.079 -0.392*** 0.039 -0.118*** se se se se se se se se 2415 0.92 0.92 0.035 0.845*** 0.001 -0.005*** 0.002 0.000 0.007 0.001 0.002 0.003 0.042 -0.244*** 0.006 0.017*** Fixed effects Market leverage Book leverage Number of observations Frac of variance due to bank FE R2 constant Log(Risk) Dividends Collateral Log(Size) Profits Market-to-book ratio Dependent variable Speed of adjustment 0.026 2059 0.027 0.103*** 0.004 -0.004 0.013 0.018 0.020 0.084*** 0.001 0.006*** 0.123 -0.008 0.040 -0.023 0.88 2059 0.019 0.086*** 0.020 0.090*** 0.88 2059 0.70 0.91 0.042 0.417*** 0.045 2059 0.78 0.92 0.041 0.454*** 0.002 -0.003 0.003 0.004 0.012 0.007 0.002 -0.005*** 0.045 -0.096** 0.015 0.032** 0.045 Fixed effects Book leverage 0.124*** 0.450*** 0.468*** Pooled OLS Xict-1 collects the same set of variables as before (market-to-book ratio, profits, size, collateral, dividends and risk) The parameter λ represents the speed of adjustment It measures the fraction of the gap between last period’s leverage and this period’s target that firms close each period The reported coefficients λ and B and their standard errors are obtained using the Delta method since they are non-linear transformations of the originally estimated coefficients (see Flannery and Rangan (2006) for a derivation of the model) For the Pooled OLS estimation (Columns and 4), ci=0 Columns and have Xict-1=0 All standard errors are adjusted for clustering at the bank level ***, ** and * denote statistical significance at the 1%, the 5% and the 10% level respectively Lict = β + λBX ict −1 + (1 − λ ) Lict −1 + ci + uict The dependent variable is either market (Column 1) or book (Column 2) leverage Columns to show the result of estimating a partial adjustment model: Lict = β0 + β1MTBict −1 + β2 Profict −1 + β3 Ln(Sizeict −1 ) + β4Collict −1 + β5 Divict + β6 Ln( Riskict −1 ) + ci + ct + uict The sample consists of the 200 largest publicly traded banks in the U.S and the EU from the Bankscope database from 1991 to 2004 See Appendix I for the definition of variables Columns and show the result of estimating: Table X: Bank fixed effects and the speed of adjustment 40 ECB Working Paper Series No 1096 Septembre 2009 Xict-1 collects the same variables as before (market-to-book ratio, profits, size, collateral, dividends and risk) The parameter λ represents the speed of adjustment It measures the fraction of the gap between last period’s leverage and this period’s target that firms close each period The reported coefficients λ, δ and B and their standard errors are obtained using the Delta method since they are non-linear transformations of the originally estimated coefficients (see Flannery and Rangan (2006) for a derivation of the model) Column has Xict-1=0 All standard errors are adjusted for clustering at the bank level ***, ** and * denote statistical significance at the 1%, the 5% and the 10% level respectively Lict = β + λ (BX ict −1 + δDepct ) + (1 − λ ) Lict −1 + uict The dependent variable is either market (Column 1-4) or book (Column 5-8) leverage Columns 1,2,5 and drop the bank-level variables ((market-to-book ratio, profits, size, collateral, dividends and risk) Dep is given by per capital deposit insurance divided by per capita GDP (Columns 1,3,5 and 7) or divided by average deposits per depositor (Columns 2,4,6 and 8-10) Columns and 10 show the result of estimating a partial adjustment model: Lict = β0 + β1MTBict−1 + β2 Profict−1 + β3 Ln(Sizeict−1 ) + β4Collict−1 + β5 Divict + β6 Ln( Riskict−1 ) + δDepct + cc + ct + uict The sample consists of the 200 largest publicly traded banks in the U.S and the EU from the Bankscope database from 1991 to 2004 See Appendix I for the definition of variables Columns to show the result of estimating: Table XI: Deposit insurance coverage ECB Working Paper Series No 1096 Septembre 2009 41 se se se se se se se se se se Number of observations R2 constant Coverage(Deposits) Coverage(GDP) Log(Risk) Dividends Collateral Log(Size) Profits Market-to-book ratio Speed of adjustment Dependent variable 0.010*** 0.956*** 0.015 2204 0.317 0.315 0.003 0.959*** 0.015 2204 0.006 0.014** 0.002 0.002* -0.021 -0.187*** 0.006*** 0.033*** -0.010*** 0.011 -0.013*** 0.005 0.801 1.247*** 0.049 2204 0.801 1.249*** 0.049 2204 0.001 0.335 0.956*** 0.005 2204 0.002 0.000 0.002 0.337 0.954*** 0.005 2204 0.001 0.596 0.839*** 0.022 2204 0.002 -0.002 0.004 0.004 0.002 0.002 -0.025*** 0.005 -0.025*** 0.003 -0.021*** 0.011 -0.021*** 0.008 0.021* 0.001 0.001 0.020* 0.001 0.005*** 0.090 0.091 0.005*** 0.056 -0.273*** 0.037 0.037 -0.276*** 0.016 -0.466*** 0.596 0.839*** 0.022 2204 0.001 -0.001 0.002 -0.013*** 0.003 -0.010*** 0.008 0.033*** 0.001 0.006*** 0.056 -0.190*** 0.016 -0.021 Book leverage -0.466*** Market leverage Table XI: Deposit insurance coverage (continued) 0.88 0.086*** 0.022 1867 0.002 0.004* 0.090*** 0.023 0.88 0.110*** 0.029 1867 0.002 0.002 0.004 -0.004 0.010 0.008 0.020 0.075*** 0.001 0.006*** 0.129 0.033 0.038 0.130*** 0.029 -0.036 Table XII: Tier capital and banks close to the regulatory minimum The sample consists of the 200 largest publicly traded banks in the U.S and the EU from the Bankscope database from 1991 to 2004 See Appendix I for the definition of variables Column shows the result of estimating: Kict = β0 + β1MTBict −1 + β2 Profict −1 + β3 Ln(Sizeict −1 ) + β4Collict −1 + β5 Divict + β6 Ln( Riskict −1 ) + cc + ct + uict The dependent variable is the regulatory Tier capital ratio Columns and show the result of estimating: Lict = β0 + BXict−1 + CXict−1 * Closeict−1 + cc + ct + uict Xict-1 collects the bank level variables (market-to-book ratio, profits, size, collateral, dividends and risk) The dependent variable is book leverage The interaction dummy Close is equal to if a bank’s Tier capital ratio is below 5% (column 2) or below 6% (column 3) in the previous year Standard errors are adjusted for clustering at the bank level ***, ** and * denote statistical significance at the 1%, the 5% and the 10% level respectively Tier capital ratio Market-to-book ratio -0.019 se 0.020 Dependent variable Book leverage Close is