WPS6948 Policy Research Working Paper 6948 Bank Capital and Systemic Stability Deniz Anginer Asli Demirguc-Kunt The World Bank Development Research Group June 2014 Policy Research Working Paper 6948 Abstract This paper distinguishes among various types of capital The results show that regulatory capital is effective in and examines their effect on system-wide fragility. The reducing systemic risk and that regulatory risk weights analysis finds that higher quality forms of capital reduce are correlated with higher future asset volatility, but this the systemic risk contribution of banks, whereas lower relationship is significantly weaker for larger banks. The quality forms can have a destabilizing impact, particularly paper also finds that increased regulatory risk-weights not during crisis periods. The impact of capital on systemic correlated with future asset volatility increase systemic risk is less pronounced for smaller banks, for banks fragility. Overall, the results are consistent with the located in countries with more generous safety nets, theoretical literature that emphasizes capital as a potential and in countries with institutions that allow for better buffer in absorbing liquidity, information, and economic public and private monitoring of financial institutions. shocks reducing contagious defaults. This paper is a product of the Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at ademirguckunt@ worldbank.org and danginer@worldbank.org The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Bank Capital and Systemic Stability Deniz Anginer, Asli Demirguc-Kunt1 JEL Classifications: G21, L11, L14 Keywords: Systemic risk, bank capital, Basel capital, merton model, distance to default, default risk 1 Anginer, World Bank, danginer@worldbank.org; Demirguc-Kunt, World Bank, ademirguckunt@worldbank.org. We thank Ricardo Meilman for excellent research assistance and Ouarda Merrouche for helpful comments. This paper’s findings, interpretations and conclusions are entirely those of the authors and do not necessarily represent the views of the World Bank, their Executive Directors, or the countries they represent. 1. Introduction One of the important lessons of the 2008 financial crisis was that financial institutions need to be subjected to more effective capital requirements. But while regulatory consensus has been to focus on capital regulation, there has been less agreement among economic theorists on the role of capital. There is continued debate around precisely what kinds of capital requirements are needed and how to structure them. Another important development after the crisis has been a greater emphasis on systemic stability and macroprudential regulation, which requires a focus not on the risk of individual financial institutions, but on an individual bank’s contribution to the risk of the financial system as a whole. In this paper we address both policy issues by studying the empirical relationship between bank capital and systemic risk. Specifically, we distinguish between various definitions of capital and investigate their impact on measures of system-wide fragility, using a bank-level database of over 1,200 publicly traded banks in over 45 countries over the 1998 to 2012 time period. Following Anginer, Demirguc-Kunt and Zhu (2013), we use alternative measures of the correlation in the risk taking behavior of banks to capture each individual institution’s contribution to the risk of the financial system as a whole. Following Demirguc-Kunt, Detragiache and Merrouche (2013), we distinguish amongst various types of capital and examine their effects on these measures of systemic fragility. We find that greater capital reduces system-wide fragility, consistent with the theoretical literature that emphasizes capital as a potential buffer in absorbing liquidity, information and economic shocks. The empirical results suggest that Tier 1 capital, which is of higher quality, has the greatest impact on reducing systemic fragility, while Tier 2 capital – which is less able to absorb losses despite counting as part of capital – has the opposite, destabilizing effect. We find that these results are more pronounced during the crisis years and for larger banks. The link between capital and systemic risk is weaker in countries with more generous safety nets, stronger public monitoring and private information suggesting these can substitute for capital in reducing systemic fragility. We also find that higher regulatory risk-weights are associated with higher 2 future asset volatility, indicating that risk exposure calculations under Basel rules reflect the riskiness of assets at least to some extent, although this relationship is muted for larger banks, which may be able to manipulate these weights. Finally, we find that certain elements of regulatory reforms after the 2008 crisis targeting systemic risk – such as a stronger focus on systemic supervision, conducting system-wide stress tests, and use of countercyclical regulation - have indeed led to a stronger link between capital and systemic risk. Our paper contributes to a large and growing literature on bank capital and risk. Notwithstanding the policy consensus, economic theories are split on the impact of capital on bank risk. An important purpose of stricter capital regulations is to ensure that banks can sustain significant unexpected losses in the values of the assets they hold while still honoring deposit withdrawals and other obligations. Consistent with this argument numerous theories emphasize the role of capital as a buffer to absorb earning shocks (e.g., Repullo, 2004; Von Thadden, 2004). Hence higher capitalization reduces bank risk and increases its survival probability. Another reason why capital requirements are thought to be stabilizing is because they also make bank owners have more “skin in the game,” improving their risk management, and curbing excessive risk-taking incentives due to limited liability and bailout expectations. Consistent with this argument, a number of theories emphasize how higher capitalization improves the borrower screening and risk monitoring functions of banks, thereby reducing bank riskiness (Coval and Thakor, 2005; Holmstrom and Tirole, 1997; Allen, Carletti and Marquez, 2011; Mehran and Thakor, 2011). Another set of theories emphasize the moral hazard angle, and focus on how greater capitalization would lead to the choice of less risky portfolios since risk-shifting incentives would be limited ( Keeley, 1989; Keeley and Furlong, 1990; Calomiris and Kahn, 1991; Rochet, 1992; Freixas and Rochet, 2008; Acharya, Mehran, and Thakor, 2011). However there are also other theories which argue that higher capital regulations may actually reduce bank stability. For example, Koehnan and Santomero (1980) argue higher capital may lead to higher portfolio risk, leading to greater fragility. Besanko and Kanatas (1996) argue reduced moral hazard benefits of higher capital may be offset by the cost of lower effort exerted by insiders whose ownership is diluted at higher levels of capital. Calem and Rob (1999) suggest a U-shaped relationship between bank capital and risk. Initially at low levels of capital, banks choose very risky portfolios to maximize the option value of deposit insurance. But as capitalization increases and future insolvency becomes unlikely, risk-taking incentives are 3 curbed. Finally, at very high levels of capitalization, insolvency becomes so remote that additional capital leads to increased risk-taking because banks want to benefit from the upside. Given the conflicting predictions of theoretical models and policy interest in this issue, there is also a growing empirical literature on the impact of bank capital on risk, again with mixed findings. Estrella, Park, and Peristiani (2000) examine the efficacy of different capital ratios in predicting U.S. bank failures in the early 1990s. They find that simple leverage measures outperform risk-adjusted measures of capital. Using a sample of European banks, Angora et al. (2009) find a positive association between bank risk and bank capital held in excess of capital regulations. Bichsel and Blum (2004) investigate the relationship between the changes in risk and changes in leverage for a panel of Swiss banks and again see a positive correlation between levels of capital and bank risk-taking. For US banks, Berger and Bouwman (2013) examine the impact of capital on individual bank fragility and market shares and find that while capital improves survival probability, higher capital helps medium and large banks primarily during crisis periods. Using an international sample of banks, Demirguc-Kunt, Detragiache and Merrouche (2013) investigate whether bank stock returns react differently to different types of capital ratios, and find that a stronger capital position was associated with stronger performance during the latest crisis, particularly for large banks, and that this relationship was stronger when capital is measured by leverage ratio rather than risk-based ratios, and when higher quality forms of capital are considered. Beltratti and Stulz (2010) also find that large banks with more capital had higher stock returns during the crisis, but these factors did not have a robust impact on bank risk, as measured by the bank’s idiosyncratic volatility and distance-to default. Other papers investigate the usefulness of risk weighted assets concluding that either they do not predict market measures of risk (Das and Sy 2012), or predict bank failure only when the risk of a crisis is very low (Mariathasan and Merrouche, 2012). Acharya, Engle and Pierret (2014) examining the stress tests conducted by U.S. and European regulators find that when capital shortfalls are measured relative to risk-weighted assets, the ranking of financial institutions is very different from stress tests conducted using publicly available information. But, when capital shortfalls are measured relative to total assets, they find similar results. We make a number of contributions to this literature. First, unlike the previous papers that examine the impact of capital on risk, stock market valuation or failure probabilities of individual banks, our paper focuses on systemic risk. Hence, we do not examine solely 4 individual bank risk, but also the co-dependence of those risks, therefore addressing the macro- prudential regulation issues. This is consistent with the growing consensus to adjust capital requirements to better reflect an individual bank’s contribution to the risk of the financial system as a whole, as opposed to absolute level of risk of any individual institution (Brunnermeier, Crockett, Goodhart, Persaud, and Shin 2009, Financial Stability Forum 2009a, 2009b). Second, we are able to examine the relationship between systemic risk and bank capital at the bank level, for different measures of regulatory capital while controlling for a variety of bank characteristics that may affect systemic risk. Third, the cross-country and panel nature of our data set allows us to investigate the impact of institutional and regulatory environment on the capital-systemic risk link, as well as the impact of latest regulatory changes on this relationship. Finally, we examine the extent to which regulatory risk weights reflect future asset volatility, and whether manipulation of risk weights contributes to systemic fragility. The rest of the paper is organized as follows. Section 2 describes the data and the empirical methodology. Section 3 presents the empirical results and discusses the implications. Section 4 concludes. 2. Data and Empirical Methodology 2.1. Sample We obtain bank level financial information from Bankscope. We use stock market information from Compustat Global for international banks and stock market information from CRSP for U.S. banks. The Bankscope database reports detailed balance sheet and income statement information for both public and private banks and covers over 90% of the total banking assets in a given country. The Compustat Global database provides daily stock price information for both active and delisted companies accounting for 98% of the global stock market capitalization. CRSP is the standard source for stock price information of U.S. companies 2.2. Measures of Systemic Fragility As our focus is on systemic stability, instead of looking at the absolute level of risk in individual banks, we examine the contribution of each bank to the risk of the banking system as a whole. We measure systemic risk as a bank’s contribution to the banking system within a given country. 5 Since bank regulation and supervision are conducted at the country level, from a policy perspective, systemic risk measured at the country level (as opposed to at the global level) is more relevant. In addition, Acharya (2011) suggests that banks will have incentives to take on correlated risks if there is an implicit guarantee provided by the state to cover losses stemming from a systemic crisis. Bertay, Demirguc-Kunt, and Huizinga (2012) also suggest that financial safety nets reduce bank internationalization because international banks are unlikely to be bailed out by local governments of the overseas countries where they operate. In the aftermath of the financial crisis of 2007/08, there has been renewed interest quantifying systemic risk inherent in the global banking sector (Acharya et al 2010, Adrian and Brunnermeier 2012, Brownlees and Engle 2012, Huang, Zhou, and Zhu 2009). Instead of relying on a single measure of systemic risk contribution, we use three measures commonly used in the literature. The first measure is the conditional value-at-risk (Covar) measure of Adrian and Brunnermeier (2011). It is the value-at-risk (Var) of the financial system conditional on institutions being under distress. A financial institution’s contribution to systemic risk is the difference between Covar conditional on the institution being under distress and the Covar in the normal state of the institution. Following Adrian and Brunnermeier (2011), we compute a time- series of Covar measures for each bank using quantile regressions and a set of macro state variables. In particular, we run the following quantile regressions: , =∝ + + , , =∝ | + | , + | + |, (1) Above, , is the equity return for bank i in week t. , is the weekly value-weighted return of all financial institutions in a given country. are lagged state variables, and include change in the 3 month t-bill rate (rate), the change in the term spread (term), weekly country stock index return, volatility of the daily country stock index returns over the past 4 weeks. We use weekly stock returns from Compustat Global for international financial firms and weekly stock market information from CRSP for U.S. financial firms. For the aggregate market index, we use the country stock index in which the financial firm is incorporated. 6 Covar variable is computed as the change in the Var of the system when the institution is at the qth percentile (or when the institution is in distress) minus the Var of the system when the institution is at the 50% percentile: ∆ ! = " ! #% ! *'(% | $, − $, + (2) We compute the ∆Covar measure at q=5% for each financial institution in our sample on a rolling three year time periods, in order to accommodate the time varying business conditions (Moore and Zhou 2011). Finally, we invert the Covar variable, so that higher values indicate greater systemic risk. The second measure of systemic risk contribution is the Marginal Expected Shortfall (MES) measure described in Acharya et al (2010). The systemic expected shortfall of an institution describes the capital shortage a financial firm would experience in case of a systemic event. It is based on the notion that a shortage of capital is dangerous for the individual firm, but becomes dangerous for the whole economy if it occurs just when the rest of the banking sector is also undercapitalized. Marginal Expected Shortfall (MES) of a firm is the expected loss an equity investor in a financial firm would experience if the market declined substantially. MES measures the average firm return on days when the market as a whole is in the tail of its loss distribution: , = ,- , . , < 0 (3) Above, , is the financial firm i’s equity return and , is the aggregate market index return. A systemic event is defined as a drop of the market index below a threshold, C, over a given time horizon. The systemic event is thus denoted by < . Acharya, Engle and Richardson (2012) show that MES can be used to set capital limits based on systemic risk contributions. Since the book value of debt will be relatively unchanged while equity values fall by MES, a regulator can require a bank to hold equity to satisfy a prudential capital ratio of k% to make sure that the 5×7 8 9,: systemic risk posed by the bank is zero: ,123 , ≥ . We compute MES using a ; 5<×# =>?@9,: + threshold that corresponds to the index at its lowest 5% level over the previous one year of return 7 data.2 For this computation we use daily stock returns from Compustat for international financial firms and daily stock market information from CRSP for U.S. financial firms. For the aggregate market index, we use the country stock index in which the financial firm is incorporated. We obtain the daily country stock indices data from Compustat Global. The third measure, R-squared, is based on a commonly used measure in the study of convergence of asset prices (Bekaert and Wang 2009, Longin and Solnik 1995, Bekaert and Harvey 2000, and Bekaert, Hodrick and Zhang 2012). R-squared (Rsq) measures the total variation of returns of a given bank explained by returns of all other banks in a given country. Anginer, Demirguc-Kunt and Zhu (2014) use this measure in studying the relationship between competition and systemic risk in the banking sector. R-squared is obtained from regressing returns of an individual bank on average returns of all other banks in a given country. For each bank i, in country j, in year t, we run a time series regression of bank i’s weekly returns on the average return of other banks excluding bank i itself: G 1 ,A, ,B = C ,A, + ,A, F 5,A, ,B + ,A, ,B (4) E 5H ,5I We follow Morck, Yeung, and Yu (2000) and Karolyi, Lee, and Van Dijk (2011) and use the logistic transformation of R-squared from the above regression, which is equal to log(rsqi,j,t) / (1- rsqi,j,t)), to measure systemic risk posed by bank i. R-squared is only computed for banks with at least twenty-six weeks of non-zero volume returns data in a year. In terms of measuring co- dependence, using R-squared has advantages over alternative measures as described in Pukthuanthong and Roll (2009) and Bekaert and Wang (2009). Higher R-squared for a given bank suggests that a bank may be exposed to similar sources of risk as other banks in a given country, and also suggests that there may be channels of inter-dependency between the bank and others in a given country. Both interconnectedness and common exposure to risk makes the banking sector more vulnerable to economic, liquidity and information shocks. 2.3. Capital and Control variables 2 We find similar results using changes in Merton (1970) distance-to-default measure instead of stock market returns in the calculation of MES. 8 Following Demirguc-Kunt, Detragiache and Merrouche (2013) we use a number of alternative definitions of capital. First set of capital ratios uses risk-adjusted assets computed according to Basel rules. We examine three ratios: Tier 1, Tier 2, and Tier 1 plus Tier 2 capital divided by risk-adjusted assets and off-balance sheet exposures (tier 1 capital / rwa, tier 2 / rwa and total capital / rwa). Tier 1 capital is computed as the sum of shareholder funds and perpetual, non- cumulative preference shares. Tier 2 capital is computed as the sum of hybrid capital, subordinated debt, loan loss reserves, and valuation reserves. The second set of ratios uses non- risk weighted assets. As before, we compute ratios with respect to Tier 1, Tier 2, and Tier 1 plus Tier 2 capital (tier 1 capital / ta, tier 2 / ta and total capital / ta respectively). Finally we examine common equity ratio (common / ta) computed as common equity divided by total assets, and tangible common equity ratio (tangible / ta), computed as tangible equity divided by tangible assets. Capital ratios are obtained from Bankscope. In examining the relationship between capital and systemic stability, we control for a number of bank level variables. As with capital ratios, bank level controls come from Bankscope. For each bank, each year, we calculate relative bank size (size), which is the natural logarithm of total assets of a given bank divided by average total assets of other banks in a given country; bank liquidity (liquid assets / ta) which is liquid assets divided by total assets; reliance on deposits for funding (deposits / ta), which is deposits divided by total assets; asset quality (loan loss provisions / ta) which is loan loss provisions divided by total assets; business model (net loans / ta) which is net loans divided by assets. When we examine manipulation of risk weights, we control for earnings (earnings / total loans) which is net income divided by total gross loans. We winsorize all financial ratios at the 1st and 99th levels of their distributions to correct for potential data entry errors and reduce the influence of outliers. As mentioned in the introduction we are interested in the impact of the larger regulatory and institutional framework on the capital and systemic stability relationship. In particular, capital may act as a substitute in reducing systemic risk for poor institutional environments that do not allow for efficient public and private monitoring of banks. We consider three groups of institutional variables. The first set of variables measure the strength of public and private monitoring in each country. The supervisory power is an index measuring supervisory authorities’ power and authority to take specific preventive and corrective actions. The measure ranges from zero to fourteen, with fourteen indicating the highest power of the supervisory 9 authorities. The deposit insurance variable indicates whether a country has explicit deposit insurance (Yes=1/No=0) and whether depositors were fully compensated the last time a bank failed (Yes=1/No=0). The variable ranges from 0 to 2. For both variables, data is obtained from the World Bank regulation and supervision surveys described in Barth, Caprio, and Levine (2008). The second set of institutional variables capture information asymmetry in the lending market. Based on the notion that efficient private monitoring depends on information availability and sharing (Djankov, McLiesh, and Shleifer 2007), we use the depth of credit information sharing (credit info depth) from the World Bank Doing Business Survey as a measure of private monitoring.3 This variable ranges from zero to six, with higher values indicating deeper credit information. From the same data source, we also create two dummy variables indicating whether a public credit registry (public registry) or a private credit bureau (private bureau) operates in a given country. Finally, the third set of variables measure information transparency in the banking sector. The rated variable measures the fraction of ten biggest banks rated by international rating agencies. The audit variable indicates whether an external audit is required of the financial statements of a bank and, if so, by a licensed or certified auditor. The variable ranges from 0 to 2, with a higher value indicating more informative bank account. The disclosure variable indicates whether the income statement includes accrued or unpaid interest or principal on nonperforming loans, whether banks are required to produce consolidated financial statements, and whether bank directors are legally liable if information disclosed is erroneous or misleading. The variable ranges from 0 to 3, with a higher value indicating more informative bank account. Data for these three variables are obtained from the World Bank regulation and supervision surveys (Barth, Caprio, and Levine 2008). Panel A of Table 1 provides summary statistics of all the bank level variables used in the empirical analyses. Panel B lists the number of unique banks and countries that have non- missing regulatory capital ratios and common equity ratios over time. Our sample with regulatory capital covers on average over 750 banks in over 40 countries. The coverage is greater for banks that have non-missing common equity ratios. We have over 1,300 banks in 3 Details on how these variables are constructed are available on World Bank’s Doing Business Survey website at http://www.doingbusiness.org/methodology. 10 over 48 countries. For the full sample, the median risk-adjusted capital asset ratio is 13.1%, which is above the minimum Basel requirement of 8%. The median tangible and common ratios are 8.8% and 7.7% respectively. 3. Empirical results 3.1. Relationship between Capital and Systemic Risk We begin the empirical analyses by examining the baseline relationship between capital and systemic risk. Figure 1 shows the evolution of the regulatory capital ratio and marginal expected shortfall (MES) over the sample period. As expected, we see a significant increase in MES during the financial crisis.4 There is a negative overall correlation between systemic risk and capital ratios. To examine this relationship more formally, we run the following regression: systemic riskijt = β0 + Ω×bank_controlsijt-1 + β1×capitalijt-1 + αj× λt + εij (5) The dependent variable is bank i’s systemic risk (in country j in year t), systemic riskijt, measured using Covar, MES and Rsq described in the previous section. The main explanatory variables of interest are the capital ratios, also described in the previous section. Bank level control variables include relative bank size, bank liquidity, reliance on deposit funding, asset quality and business model. All explanatory variables are lagged by one year. In the regression, we also include country-year fixed effects (αj× λt) to control for all time varying country factors, such as, interest rates, inflation and other macroeconomic variables, differences in levels of economic development and quality of bank regulation and supervision, and differences in accounting and regulatory standards. Time varying fixed effects greatly reduce concerns about possible omitted variables. Table 2 presents the results from the regression specification (5) above. Panel A reports the results using Covar as the measure of systemic risk. Panels B and C report the results using MES and Rsq as measures of systemic risk respectively. The control variables have the expected signs and are consistent with the results in Adrian and Brunnermeir (2011). Bank size is associated with higher systemic risk. Having greater liquid assets and greater reliance on deposit 4 The time-series patterns for the other measures of systemic risk (Rsq and Covar) are similar. 11 funding is associated with lower levels of systemic risk. Asset quality proxied by loan loss provisions is associated with lower systemic risk, as is engagement in more traditional banking activities proxied by net loans over total assets. Across all measures, we find a significant negative relationship between total regulatory capital ratio and systemic risk (reported in column 1). That is, higher levels of capital ratios lead to lower systemic risk. Looking at the components of capital, we find that Tier 1 capital, which is of higher quality, has the greatest impact on reducing systemic fragility (colum 2). Tier 2 capital, on the other hand, has the opposite, destabilizing effect (column 3). This destabilizing effect may be due to components that make up Tier 2 capital. Subordinated debt, hybrid capital, loan loss reserves, and valuation reserves, can provide channels in which information shocks are propagated through the banking system. The results are similar when we consider non-risk weighted assets. Capital ratios using un-weighted total assets are reported in columns 4 to 6. We find that Tier 1 capital reduces systemic risk, while Tier 2 capital has a destabilizing effect. Both tangible capital and common capital are associated with lower levels systemic risk (columns 7 and 8). In terms of economic significance, a one standard deviation increase in Tier 1 regulatory capital decreases MES by 25 basis points which is about 18% of its median value. During crisis years, the impact of Tier 1 regulatory capital goes up to 90 basis points or 65% of the median MES value. 3.2. Controlling for the Leverage Effect Capital can reduce systemic risk by providing a buffer against economic shocks. Shocks would be absorbed by an individual bank and would not be propagated throughout the financial system. Capital acting as a cushion would provide a mechanical relationship between capital and systemic risk. At the extreme case of 100% capital ratios, for instance, there would be no defaults and hence no systemic risk. However, as discussed in the introduction capital can also affect systemic risk through other channels such as asymmetric information after controlling for this leverage effect. In this section, we examine the effect of capital through these other channels by explicitly controlling for the leverage effect. Instead of using equity returns which are affected by the level of leverage, we use asset returns in computing the three systemic risk measures. 12 Although equity values are observed through market prices, we do not observe market value of assets. We use the Merton (1974) structural model to infer market value of assets using equity values, leverage and equity return volatility. We use the approach outlined in Anginer, Demirguc-Kunt and Zhu (2013) to compute these asset values. Specifically, the market equity value of a company is modeled as a call option on the company’s assets: J? = JK LM N;O < − P QM N;OR < + ;1 − LM