Accounting Discretion, Loan Loss Provisioning, and Discipline of Banks Risk-Taking

Transcription

1 Accounting Discretion, Loan Loss Provisioning, and Discipline of Banks Risk-Taking Robert M. Bushman Kenan-Flagler Business School University of North Carolina-Chapel Hill Christopher D. Williams Ross School of Business University of Michigan December 2009 We thank Ryan Ball, Dan Amiram, Dave Larcker, Ed Maydew and workshop participants at Chinese University of Hong Kong, Columbia University, London Business School, MIT, Peking University, University of Missouri, University of Texas at Dallas, Washington University in St. Louis, Yale University, and the 2007 Duke/UNC Fall Camp, for helpful comments. We thank Kenan-Flagler Business School, University of North Carolina at Chapel Hill, for financial support.

2 Accounting Discretion, Loan Loss Provisioning, and Discipline of Banks Risk-Taking Abstract This paper empirically delineates economic consequences associated with differences in accounting discretion permitted to banks under existing regulatory regimes. Exploiting crosscountry variation in loan provisioning practices, we generate country-level measures of observable discretion allowed to banks within a given country. We examine implications of discretion for both the informational properties of loan provisions and for bank transparency. First, we investigate the extent to which banks in countries allowing higher discretion use this enhanced flexibility to infuse loan provisioning practices with a more forward looking orientation relative to banks in lower discretion countries. Next, we investigate whether discretion impedes the ability of regulators and outside investors to monitor and discipline bank risk taking. We have three main findings: (1) There is no evidence that banks in high discretion countries impound more forward looking information in loan provisions relative to banks in low discretion countries; (2) Sensitivity of changes in bank leverage to changes in asset volatility is lower in high discretion regimes relative to low discretion regimes; and (3) Banks in high discretion regimes exhibit more risk-shifting relative to banks with less discretion. Our results are consistent with discretion degrading transparency of banks and weakening discipline exerted over bank risk taking. Keywords: Smoothing, Loan Loss Provisions, Discretion, Risk, Banks JEL Classifications: E58, G21, G32, M41

3 1. Introduction The recent financial crisis has energized politicians and regulators to scrutinize financial accounting standards as never before, creating significant pressure for change. For example, recent high profile proposals by Financial Stability Forum (2009) and U.S. Treasury (2009) call for standards setters to re-evaluate the incurred loss model underlying current loan loss provisioning requirements and consider a range of alternative approaches. A premise of these proposals is that loan loss accounting should adopt a more forward looking orientation that allows for recognition of future expected loan losses earlier in the credit cycle, which in turn would dampen pro-cyclical forces in periods of financial crisis. 1 A key aspect of these alternatives is that, relative to the incurred loss model, they would generally increase the scope for judgment and discretion in determining loan loss provisions. However, as has long been recognized (e.g., Watts and Zimmerman (1986)), accounting discretion is a double-edged sword. On the one hand, increased discretion can facilitate incorporation of more information about future expected losses into loan provisioning decisions, but on the other hand it increases potential for opportunistic accounting behavior by bank managers, which may degrade the transparency of banks and lead to negative consequences. The main objective of this paper is to empirically delineate significant economic consequences associated with observable differences in discretion permitted to banks under existing regulatory regimes. Exploiting significant cross-country variation in observed loan provisioning practices, we generate country-level measures of the discretion allowed to banks within a given country, where discretion is estimated relative to an incurred loss model. Using 1 Financial Stability Forum (2009) defines pro-cyclicality as the dynamic interaction between the financial and the real sectors of the economy that amplify business cycle fluctuations and cause or exacerbate financial instability. See also Dugan (2009). 1

4 these country-level measures of observed discretion, we perform three fundamental analyses geared towards isolating implications of discretion for both the information properties of loan provisions and for bank transparency. First, we investigate the extent to which banks in countries allowing higher discretion use this enhanced flexibility to infuse loan provisioning practices with a more forward looking orientation relative to banks in lower discretion countries. Next, we investigate the possibility that discretion imposes costs on the banking system by impeding the ability of regulators and outside investors to monitor and discipline bank risk taking. We capture the extent of discipline over bank risk taking using two measures. First, we study the relation between discretion and the sensitivity of changes in bank capital to changes in the riskiness of a bank s assets. Secondly, we examine the relation between discretion and the observed riskshifting behavior of banks. 2 A fundamental role of financial reporting is to provide credible and relevant firm-specific information about the financial performance and condition of businesses. Such information is of central importance for governance of firms and to investors in their resource allocation decisions, as well as to regulators charged with prudential oversight of financial institutions. Loan loss provisioning is a key accounting choice that can significantly influence the information properties of banks financial reports with respect to reflecting changes in the fundamental risk attributes of the underlying loan portfolios. 3 Current accounting procedures under both U.S. GAAP and IFRS utilize an incurred loss framework where a provision for loan losses is 2 As discussed in more detail later in the paper, risk shifting refers to the phenomenon where banks equity holders benefit themselves at the expense of deposit insurers by increasing the risk of asset portfolios without adequately increasing bank capital simultaneously. 3 A number of papers examine the value relevance of bank loan loss provisions. See for example Beaver et al. (1989), Wahlen (1994), Liu and Ryan (1995), Liu et al. (1997) and Kanagaretnam, Krishnan and Lobo (2009), among others. 2

5 recognized only after loss impairment events have already occurred prior to the financial reporting date that are likely to result in non-payment of loans in the future. The focus is on reflecting losses expected to result from events during a given period, while limiting consideration of expected effects of future events. An alternative to the incurred loss model posits that regulatory capital should operate as a buffer against unexpected losses (i.e., large, infrequent occurrences), while loan loss reserves should deal with expected losses ((Wall and Koch (2000), Basel Committee on Banking Supervision (1991)). This perspective underlies calls for loan loss accounting to fully incorporate all future expected losses regardless of whether a loss impairment event has occurred. Such proposals often introduce elements of earnings smoothing. 4 In current deliberations, the FASB proposes adopting fair value accounting for loans (FASB (2009), while the IASB favors an expected loss approach where expectations of future losses over the life of a loan are incorporated ex ante into effective interest rates (IASB (2009). While differing in the details, such proposed regimes generally allow greater discretion to incorporate a broader range of available credit information and create an expanded role for managerial judgment in assessing future expected losses. But as noted earlier, there are important trade-offs associated with increased discretion, and any benefits must be weighed against the costs of increased opportunism by managers. To investigate the economic consequences associated with discretion, we use a large sample of 4 For example, Borio et al. (2001), posits that credit risk builds up in periods of growth and materializes in downturns, and so loan loss reserves should be built up in good times that can be drawn down in bad times. Laeven and Majnoni (2003) also consider benefits of earnings smoothing in which loan loss provisions account for differences between incurred losses today and expected future loan losses. Borio et al. (2001) and Laeven and Majnoni (2003) recognize that such schemes create possibility for opportunistic accounting by bank managers. See Benston and Wall (2005) for discussions of other accounting alternatives. 3

6 banks from 23 countries to estimate two measures of observed discretion allowed to banks within a given country under the existing regulatory regime. The first is a smoothing measure defined as the coefficient on earnings in a regression of loan loss provisions on a vector of nondiscretionary provisioning determinates and earnings. A higher coefficient is posited to reflect more discretion to deviate from the incurred loss model and smooth via the loan loss provision. The second measure is the incremental explanatory power achieved by adding earnings to a regression of loan loss provisions on a vector of non-discretionary determinants, where a higher incremental explanatory power is posited to capture higher levels of discretion. Using these measures of discretion, we first ask if banks use discretion to more fully reflect future expected loan losses in current loss provisions. To address this, we test whether the relation between current loan loss provisions and future changes in non-performing loans increases with discretion. We find no evidence that banks in high discretion countries employ loan provisioning practices with a more forward looking orientation relative to banks in low discretion countries. Failing to find evidence that more discretion is associated with more forward looking provisioning, we next investigate the possibility that discretion imposes costs on the banking system by impeding outside discipline over bank risk taking activities. In our first approach to examining discipline over risk taking activity, we estimate the impact of increased discretion on the relation between changes in the volatility of bank assets and changes in bank leverage. This analysis posits that outside discipline imposed by regulators and investors pressures banks to decrease leverage (increase capital) in response to increases in risk, and that more intense outside discipline is associated with a higher sensitivity of changes in leverage to changes in risk. We find that, indicative of discretion dampening disciplinary 4

7 pressure, the sensitivity of changes in bank leverage to changes in asset volatility is lower in high discretion regimes relative to low discretion regimes. In our second approach to risk discipline, we investigate the relation between discretion and bank risk-shifting behavior. Explicit and implicit deposit guarantees create incentives for banks to shift risk onto the deposit guarantee agency by increasing the risk of assets without simultaneously increasing capital adequately. Countering such incentives, deposit insurers and uninsured creditors have incentives to monitor and discipline bank risk taking behavior. Exploiting Merton s (1977) characterization of deposit insurance as a put option, and methodology developed by Duan et al. (1992) and Hovakimian and Kane (2000), we provide evidence that banks in high discretion regimes exhibit more risk-shifting relative to banks in low discretion countries. Our results are consistent with discretion being used by bank managers to degrade the transparency of banks and thereby weaken the ability of regulators and outside investors to monitor and discipline bank risk taking. Our paper makes several fundamental contributions to the literature. First, we provide new evidence on the impact of discretion on information properties of loan provisions and discipline exerted over bank risk taking activities. This evidence is important in light of the current push to fundamentally change the accounting for loan provisioning. To the best of our knowledge, no previous research has investigated connections between the discretionary use of loan loss provisioning and banks risk-taking behavior. There also exists a significant literature examining the use of discretionary loan loss provisioning to smooth earnings (Moyer, 1990; Beatty el al., 1995; Collins et al., 1995; Ahmed et al, 1999; Laeven and Majonni, 2003; Liu and Ryan, 2006). Our paper adds to this literature by documenting significant consequences of smoothing behavior by exploiting cross-country variation in loan loss provisioning practices. 5

8 Our paper also complements and extends the literature on the role of market discipline in the regulation of banks (Flannery (1998) and Rochet (2005)) and the growing academic literature focused on understanding what works best in bank regulation and supervision. Market discipline seeks to harness market forces in the service of bank regulation and is embedded in the Basel II Capital Accord. 5 We contribute to this literature by demonstrating that the extent of market discipline over risk taking varies inversely with the extent of discretionary loan provisioning. That is, while price changes reflect shifts in banks perceived value that can trigger outside scrutiny of banks activities, it is plausible that credible, disaggregated information in accounting reports aids regulators and others in understanding sources of changes in price. This relates to the distinction between monitoring and influence raised by Bliss and Flannery (2001). We conjecture that textured accounting information is an important ingrdient for converting the monitoring role of prices into disciplinary influence. Our results suggest that in regimes with significant discretion over loan loss provisions, the transparency of banks financial reports is compromised, weakening market discipline. Finally our paper complements several recent papers that examine various facets of banks accounting discretion. Huizinga and Laeven (2009) examine accounting discretion by U.S. banks during the time frame, documenting that banks used discretion to overstate the value of distressed assets, and that banks with large exposures to mortgage-backed securities provisioned less for bad loans. Also, Vyas (2009), using a novel measure of financial reporting transparency, shows that for U.S. financial institutions during the ongoing financial crisis, exposure to risky assets is reflected in stock prices on a timelier basis for transparent firms 5 The Basel II Capital Accord is based around three complementary pillars Pillar 3 recognizes that market discipline has the potential to reinforce minimum capital standards (Pillar 1) and the supervisory review process (Pillar 2), and so promote safety and soundness in banks and financial systems. 6

9 The rest of the paper is organized as follows. Section 2 puts our paper in context relative to the extant research on bank accounting and transparency, and also on the role of market discipline as complementary aspect of bank regulation. Section 3 presents the main empirical analysis on the relations between country-level provisioning regimes and the discipline of bank risk-taking. Section 4 concludes. 2. Related Literature In section 2.1 we discuss the relation of our research to existing literature on loan loss provisioning and earnings smoothing. In section 2.2 we relate our paper to the literature on role of market discipline in the regulation of banks and the literature focused on understanding what works best in bank regulation and supervision. Finally, section 2.3 describes how we draw from the extant literature in order to carefully control for other aspects of bank regulatory regime as well as institutional features of the country in general. 2.1 Loan Loss Provisioning and Smoothing There exists a significant literature examining the use of discretionary loan loss provisioning to manage earnings, and in particular via earnings smoothing. Using data on U.S. banks, Greenwald and Sinkey (1988), Collins et al. (1995), Liu and Ryan (2006), and Fonseca and Gonzalez (2008) document earnings smoothing via loan loss provisions, while Beatty et al. (1995) and Ahmed et al. (1999) do not. Also, Collins et al. (1995), Beaver and Engel (1996), and Ahmed et al. (1999) document that discretionary loan-loss provisions are negatively related to capital, while Beatty et al. (1995) find the opposite result. We extend and complement this research by documenting significant consequences of discretion by exploiting cross-country variation in loan loss provisioning practices. 7

10 A long standing debate in the literature concerns whether earnings smoothing increases the information content of earnings by revealing innate fundamentals or whether it obscures fundamentals and reduces information in earnings. While some argue that income smoothing reveals information (e.g., Arya et al., 2003; Chaney and Lewis, 1995; Tucker and Zarowin, 2006; Trueman and Titman, 1988; Sankar and Subramanyam, 2001; Demski, 1998), others argue that income smoothing distorts information (e.g., Barth et al., 2007a, 2007b; Francis et al., 2004; Lang et al., 2003; Leuz et al., 2003). While the existence of earnings smoothing has been documented, it has proven difficult to empirically distinguish whether smoothing enhances or obscures information. Our design offers a unique setting in which to address this issue by investigating whether earnings smoothing at banks is associated with more forward looking provisioning or whether it imposes costs on the banking system by impeding outside discipline over bank risk taking activities. 2.2 Market Discipline as a Bank Regulatory Tool The premise that financial accounting information can play a fundamental role in the prudential oversight of banks is consistent with the Basel II Capital Accord which posits a central role for informational transparency in bank regulation in facilitating market discipline. 6 While Basel Pillar 3 envisions a range of disclosures that may or may not be part of the formal financial accounting rules of a given country (BCBS (2001), financial accounting systems form the foundation of the firm-specific information set available to interested parties outside the firm 6 The Basel II Capital Accord is based around three pillars. Pillar 3 recognizes that market discipline has the potential to reinforce minimum capital standards (Pillar 1) and the supervisory review process (Pillar 2). The Basel Committee on Banking Supervision (BCBS) (2001) notes Market discipline imposes strong incentives on banks to conduct their business in a safe, sound and efficient manner, including an incentive to maintain a strong capital base as a cushion against potential future losses arising from risk exposures. 8

11 and are a logical starting point for investigating properties of information important for addressing moral hazard problems at banks. It is also plausible that the quality of banks financial accounting information is correlated with the quality of bank disclosures that fall outside of a country s financial accounting rules. Bank transparency can complement prudential supervision by supporting the financial analysis of banks by private investors which, by impounding such information into prices, can supplement the information already possessed by supervisors (Rochet (2005), Hovakimian and Kane (2000), Kane (2004), and Flannery and Thakor (2006)). Transparency may also enhance ex-ante discipline on bank risk taking activities as managers will anticipate that informed investors in uninsured liabilities will be more likely to discern increased risk-taking and respond quickly to greater risks by demanding higher yields on their investments. 7 Barth, Caprio, and Levine (2004) examine regulations on capital adequacy, deposit insurance system design features, bank supervisory power, regulations fostering information disclosure and private sector monitoring of banks, and government ownership of banks, among other factors. Their main results suggest that policies relying on regulatory features that foster accurate information disclosure, empower private-sector oversight of banks, and foster incentives for private agents to exert corporate control work best to promote bank development, performance and stability. Further, Beck, Demirgüç-Kunt, and Levine (2006) find that a supervisory strategy that empowers private monitoring of banks by forcing banks to disclose 7 Better disclosure may not beneficially impact bank risk taking behavior. For example, Blum (2006) demonstrates that benefits of market discipline via subordinated debt contracts depends on the ability of banks to credibly commit to a given risk level. Cordella and Yeyati (1997) show that public disclosure can serve to reduce bank risk taking, but only to the extent that the bank controls the risk of its portfolio. Plantin, Sapra and Shin (2007) and Allen and Carletti (2008) focus on potential negative effects of mark-to-market accounting on bank soundness. 9

12 accurate information to the private sector, tends to lower the degree to which corruption of bank officials is an obstacle to firms raising external finance. Demirgüç-Kunt, Detragiache, and Tressel (2006) study whether compliance with the Basel Core Principles for Effective Banking Supervision improves bank soundness. They document that countries which require banks to regularly and accurately report their financial data to regulators and market participants have sounder banks (measured with Moody s financial strength ratings). They note that these findings highlight the importance of transparency in making supervisory processes effective and strengthening market discipline. Tadesse (2006), using a range of survey-based metrics find that banking crises are less likely in countries with greater regulated disclosure and transparency. Finally, Nier and Baumann (2006) look at the role of bank transparency in providing incentives for banks to limit their risk. Where we focus on risk shifting behavior, Nier and Baumann (2006) look at the extent to which higher levels of transparency enhance market discipline and provide more incentives for banks to limit their risk of default by holding larger capital buffer. Nier and Baumann s primary measure of transparency is a bank level index of disclosure constructed by counting the number of individual disclosures available from BankScope. Our paper complements and extends this literature by demonstrating that the extent of market discipline may depend not only on the existence of informed market prices, but also on the availability of more textured information provided in banks financial reports that combines with price triggers to facilitate disciplinary actions by regulators and investors. Our focus on publicly traded banks implies that all banks in our sample have prices available, and so differences in market discipline across banks must depend on more than just the existence of 10

13 market prices. 8 We conjecture that while prices transmit aggregated information that can trigger outside scrutiny of the bank s activities, supplementing this aggregated information with more textured information amplifies the intensity of disciplinary responses. Stock price changes reflect shifts in a bank s perceived value, where credible, disaggregated information in the bank s accounting reports can aid regulators and others in understanding the source of changes in firm value. This conjecture is related to the distinction between monitoring and influence raised by Bliss and Flannery (2001), where we are contending that textured financial information is a key element for converting the monitoring role of prices into disciplinary actions by outsiders. Our results suggest that in regimes where discretion over loan loss provisions allows banks to smooth earnings, bank transparency is compromised, thus weakening the market-disciplining role of prices. It is also important to distinguish our analysis from Laeven and Levine (2009) who focus on conflicts between bank managers and owners over risk, and document that bank risk is generally higher in banks that have large owners with substantial cash flow rights. While we do not have data to control for intricate aspects of bank governance, it also not our objective to focus on the level of risk that banks take. Instead, we explore whether the sensitivity of bank leverage (capital) to changes in risk varies with accounting discretion. 2.3 Controlling for other aspects of bank regulatory regime To isolate the economic consequences of discretion in loan loss provisioning, it is crucial to control for other key aspects of countries bank regulatory regimes as well as other country- 8 As discussed further in section 3.6, we explicitly control for the possibility that market efficiency may vary across countries. 11

14 level institutions. Fueled by recent availability of rich cross-country data on bank regulations and supervisory practices, the literature discussed in section 2.2 above provides evidence on which of the many different bank regulations and supervisory practices employed around the world work best, if at all, to promote banking-sector development, performance, and stability. Barth, Caprio and Levine (2006) design and implement a survey funded by the World Bank to collect information on extensive array bank regulations and supervisory practices for many countries. We rely on Barth, Caprio and Levine (2006) as our source of bank regulatory variables, and supplement this with variables from other sources. All variables and their sources are described in detail in Appendix A of the paper. As discussed above, the Basel II Capital Accord is based around three pillars: (1) Capital adequacy standards; (2) The supervisory review process; and (3) Market discipline. We follow Barth, Caprio, and Levine (2006) in controlling for these elements as follows: Regulations on capital adequacy: CapIndex which is an index constructed by Barth, Caprio and Levine (2006) to measure the stringency of the capital requirements in each country. Supervisory power: Official is a measure of the official supervisory power that bank regulators have over the operations of the bank. This measure is taken from Barth, Caprio and Levine (2006) and represents an index constructed from answers to individual questions contained in the survey of bank regulatory practices. Private-sector monitoring of banks: We include a control variable, Private, developed by Barth, Caprio, and Levine (2006) which captures the extent to which bank regulations in a country foster accurate information disclosure, empower private-sector oversight of banks, and create incentives for private agents to exert corporate control over banks. 12

15 Properties of the general contracting environment: Judicial is an assessment of the efficiency and integrity of the country s legal system. In robustness analyses, we include a wide range of additional control variables. These include Disclosure, which is a measure of the general disclosure requirements of a country s securities regulations; Rights, a measure of the investor protection rights present in the country; StateBank, to control for the extent of state ownership of banks; Liquidity, which measures the share turnover of a country s equity market, a common measure of market development; and MrktCap, which represents the total market capitalization of a country s stock market as a percentage of GDP, again a common measure of market development. 3. Empirical Analysis This section is organized as follows. 3.1 Sample selection criteria 3.2 Estimating Country-level Discretion in Loan Loss Provisioning Practices 3.3 Discretion and Forwarding-Looking Provisions 3.4 Discretion and Risk-Taking Behavior Sensitivity of Leverage to Changes in Risk 3.5 Discretion and Risk-Taking Behavior Risk shifting 3.6 Robustness 3.1 Sample selection criteria The sample period of our study spans All bank financial statement data is taken from Bankscope and all market data is from Datastream. Country-level variables derive from five different sources; the World Development Indicators Database, Barth et al. (2006), Demirgüç-Kunt et al. (2005), La Porta et al. (1998), and La Porta et al. (2002). Detailed information concerning variable construction and data sources are included in Appendix A. To be included in the sample a bank is required to have all necessary bank-level data spanning a period of at least three years. We also require that the bank have more than $5 billion in total 13

16 assets. For a country to be included, we require all country-level data to be available. In all analyses the data is trimmed at the 1 and 99 percentiles. These general requirements yield a sample of 14,062 potential bank-year observations across 23 different countries. Table 1 Panel A provides descriptive statistics on the sample. 3.2 Estimating Country-level Discretion in Loan Loss Provisioning Practices As discussed earlier, the first step in our analysis is to empirically derive estimates of the amount of discretion allowed to banks in a given country, where discretion is measured relative to an incurred loss model. We generate two different measures of discretion. The empirical specification to derive these measures requires that we control for the basic determinants of loan loss provisions under an incurred loss model. Our basic specification is consistent with extant banking research in accounting and economics that has previously examined discretionary loan loss provisioning. The main contribution of our paper is not in the estimation of discretion, but rather in our investigation of the economic consequences of discretion. Our first measure of discretion in a country employs the following model which is estimated using OLS: LLP itj 0 1 Ebllp itj 2 NPL itj 3 NCO itj 4 LLR it 1, j 5 CAP it 1, j 6 Loans it 1, j 7 Size it 1, j 8 RLG itj 9 % GDP tj (1) LLP itj is the loan loss provision scaled by lagged total loanss for bank i, in country j, at time t. To control for aspects of the incurred loss model we include both the contemporaneous change in non-performing loans ( ) and net charge-offs (NCO), both scale3d by lagged total loans, to capture observed changes in portfolio performance and ultimate collectability. We also include 14

17 current period real loan growth (RLG). This variable is included to control for any impact on loan provisioning related to statistical provisioning applied to an increasing base of homogeneous loans (Liu and Ryan (2006)), and also to control for increased riskiness of loans. Existing literature suggests that loan growth represents an important driver of the riskiness of banks (e.g., Foos et al. (2009), and so we include it to make sure earnings is not simply picking up risk. 9 The percentage change in GDP per capita (% ) is also included to control for macroeconomic events that may trigger a need to provision. In addition to pure incurred loss proxies, we also include the level of the loan loss reserve scaled by lagged assets (LLR) and equity capital to total assets (CAP) at the beginning of the period. Finally, we include the percent of the bank s asset in the loan portfolio (Loans) and the size of the bank (Size) to control for size and asset mix effects. Our main variable of interest in (1) is earnings before taxes and loan loss provisions (Ebllp). Under the incurred loss model earnings should not explain contemporaneous provisioning behavior after controlling for incurred loss proxies and other determinates. Our first measure of discretion utilizes the estimated coefficient on Ebllp from (1). Prior research (e.g., Moyer, 1990; Beatty et al., 1995; Collins et al., 1995; Ahmed et al., 1999; Leaven and Majonni, 2003; Liu and Ryan, 2006) interpret the coefficient on Ebllp as earnings smoothing via the loan loss provision. We utilize the same interpretation, and presume that such smoothing behavior represents an important manifestation of discretion allowed to banks under a country s regulatory regime. 9 In untabulated results, we also try alternative measures of risk as controls including the volatility of equity and the volatility of assets extracted from an option pricing framework with no qualitative differences in results. 15

18 We begin by first estimating equation (1) using a pooled regression across all banks in all countries (table 1, panel B). Our objective here is to provide evidence on the general smoothing behavior of banks in our sample countries. Table 1 reports descriptive statistics of the sample in Panel A, while Panel B reports the coefficients from the pooled estimation of (1). Consistent with prior cross-country research by Laeven and Majonni (2003), we find a positive coefficient on Ebllp indicating that on average banks around the world smooth earnings via the loan loss provision, and a negative coefficient on the percentage change in GDP, %, consistent with banks provisioning less when the business cycle is upward trending. Also, we see a large, positive coefficient on the contemporaneous change in non-performing loans,. To carry out our investigation of the economic consequences of discretion, we need a measure of discretion at the individual country level. We thus estimate (1) by country where the coefficient on Ebllp is used as our first country-level measure of managerial discretion, termed Smoothing. Table 2 reports the descriptive statistics on the country specific estimates of Smoothing. The mean coefficient estimate is with and standard deviation of Peru has the highest coefficient (0.7414), whereas Singapore has the lowest ( ). These negative values raise an interpretation issue, as a large negative could be interpreted as high discretion, where we are in essence interpreting these negatives as the lowest discretion. To deal with this, we also use the following measure of discretion which does not rely on the coefficient on earnings. Our second measure is the incremental R 2 attributed to Ebllp in (1). Our first measure relies on the sign of the coefficient on Ebllp (i.e., more positive coefficient, more smoothing), while our second measure is a general measure of how important earnings are in explaining 16

19 variation in provisioning behavior. To obtain the incremental explanatory power of Ebllp we estimate (1) then estimate (2): LLP itj 0 1 NPL itj 2 NCO itj 3 LLR it 1, j 4 CAP it 1, j 5 Loans it 1, j 6 Size it 1, j 7 RLG itj 8 % GDP tj (2) We then subtract the R 2 in (2) from the R 2 in (1) and call the resulting measure LLP Opacity, with higher values interpreted as more discretion in LLP relative to an incurred loss model. Table 1, Panel C reports the results of (1) and (2) and the difference for a pooled regression including year, bank-type and country fixed effects. Table 1, Panel C reports that the R 2 for (1) is and the R 2 from (2) is , a difference (LLP Opacity) equal to To test whether the difference in R 2 is significant we use a Vuong test. Results in Panel C show that the Vuong z-statistic is with a p-value < The results in Table 1 Panel C provide evidence that on average Ebllp explains a significant portion of LLP, with the inclusion of Ebllp increases the R 2 17%. We next create a country-level LLP Opacity measure and report the country-level results in Table 2. The mean LLP Opacity is with a standard deviation of Similar to Smoothing Peru has the high LLP Opacity (0.4556), but South Africa has the lowest (0.0000). Table 2 Panel B reports both the Spearman and Pearson correlations between our measures of discretion and measures of countries bank regulatory regimes and other countrylevel institutions. We see that the two measures of provisioning discretion are highly correlated (0.85- Pearson). It is interesting to note the lack of correlation between discretion and measures of bank supervisory power (Official), regulations on capital adequacy (CapIndex) and judicial efficiency of the legal system (Judicial). However, both Smoothing and LLP Opacity are negatively correlated with Private. Recall Private captures the extent to which bank regulations 17

20 in a country foster accurate information disclosure, empower private-sector oversight of banks, and create incentives for private agents to exert corporate control over banks. These univariate correlations indicate that the regulatory features captured by Private are associated with less discretion by banks in loan loss provisioning. It is interesting to note that Barth et al. (2004) show that Private is associated with better bank development, performance and stability. 3.3 Discretion and Forwarding-Looking Provisions As discussed in the introduction, Financial Stability Forum (2009), U.S. Treasury (2009) and others favor increasing the scope for managerial judgment and discretion in determining loan loss provisions to achieve a more forward looking orientation that allows for recognition of future expected loan losses earlier in the credit cycle. While we cannot directly investigate the properties of the specific regimes proposed as such regimes have not yet been widely implemented, we can empirically examine the extent to which discretion allowed under existing regulatory regimes is actually used to infuse a forward looking orientation to provisioning. In this spirit, we test whether the relation between current period loan loss provisions and future changes in non-performing loans increases as a function of discretion. 10 We employ the following OLS specification for this purpose: LLP itj 0 1 LLPREGIME j * NPL it 1, j 2 NPL tt 1, j 3 NPL itj 4 NPL it 1, j 5 NPL it 2, j 6 LLPREGIME j 7 Ebllp itj 8 CAP t 1 itj (3) 10 Liao and Beatty (2009) also exploit the relation between provisions and future changes in non-performing loans to capture the forward looking orientation of provisioning practices. 18

21 where LLP,, Ebllp and CAP are as defined above. LLP REGIME is either Smoothing or LLP Opacity depending on the specification. If discretion is utilized by bank managers to generate more forward-looking provisions, then we would expect 0. We also control for other country-level institutional features including Offical, CapIndex, Private and Judicial as both main effects and interactions along with bank-type and year fixed effects. Table 3 columns I and II report results using Smoothing as the measure of discretion, and columns III and IV report the results using LLP Opacity. Columns I and II show that without controlling for other country-level features the coefficient on is less than zero, but when we control for other features of the regimes the coefficient is insignificantly different from zero. This suggests that Smoothing regimes do not promote forward-looking provisioning! Columns III and IV further confirm this result using LLP Opacity. Summarizing this analysis, we find that discretion over loan loss provision practices does not appear to promote forward-looking provisioning. Also, while our two measures of discretion (Smoothing and LLP Opacity) are positively correlated (table 2, panel B), columns I and III suggest that there are differences in the measures, and so in what follows we will consider both measures. Failing to find evidence that more discretion is associated with more forward looking provisioning, we next investigate the possibility that discretion imposes costs on the banking system by impeding outside discipline over bank risk taking activities. 3.4 Discretion and Risk-Taking Behavior Sensitivity of Leverage to Changes in Risk As discussed in section 2.2, financial accounting information can play a fundamental role in the prudential oversight of banks by facilitating market discipline. Loan loss provisioning is a key accounting choice that can significantly influence the information properties of banks 19

22 financial reports with respect to reflecting changes in the fundamental risk attributes of the underlying loan portfolios. In this section, we take the first of two approaches to investigating the impact of discretion on the discipline of bank risk taking behavior. This first approach examines whether the sensitivity of changes in bank leverage (or capital) to changes in risk is impacted by discretion in loan provisioning. The idea that capital should be an increasing function of risk is a basic tenet of prudential bank regulation and is reflected in the risk-weighted capital requirements laid out in the Basel II Accord. To empirically operationalize this construct, we follow Duan et al. (1992) and model equilibrium relations between increases in risk and changes in leverage by positing an equilibrium relation that specifies leverage as a linear function of asset risk. Using this specification, we estimate the sensitivity of leverage to changes in the risk of the underlying assets, and investigate whether this sensitivity varies with discretion. Specifically we estimate the following using OLS: Δ Δ (4) where following Duan et al. (1992), D is the face value of debt, V is the market value of bank assets, Δ is the change in the volatility of bank assets, and LLP REGIME is either Smoothing or LLP Opacity depending on the specification. To estimate of V and we focus on publicly traded banks and exploit the concept that a firm s equity can be characterized as a call option on the firm s assets, where the strike price is the face value of debt. Using measures of face value of the reported total liabilities (D), the observed market value of equity, and the estimated standard deviation of stock returns, we obtain values for V and (see Appendix B for further details). 20

23 Disciplinary pressure should generally result in 0 in (4), where the bank decreases leverage (increases capital) in response to an increase in risk. To investigate whether discretion impedes disciplinary responses, we interact Smoothing and LLP Opacity with Δ. A result that 0 implies that discretion over provisioning behavior is associated with leverage being less sensitive to changes in risk. In estimating (4) we also include other country and bank level controls as both main effects and interactions. Table 4 presents the results of the estimation of (4), where for brevity we do not report the main effects (available upon request). In column I, we document that 0, consistent with Duan et al. (1992) and the basic intuition that risk discipline should lead banks to decrease leverage in response to an increase in risk. In table 4, column II we find that the coefficient on the Smoothing interaction term is 3.18 (p-value < 0.01), consistent with discretion via Smoothing dampening (making less negative) the sensitivity of leverage to changes in risk.. The same is found for LLP Opacity in table 4, column III. These findings suggest that when banks are given discretion over provisioning practices, there is less disciplinary pressure response in leverage when risk changes. 3.5 Discretion and Risk-Taking Behavior Risk-Shifting Deposit insurance provides an explicit or implicit guarantee that in the event of default by the bank, depositors will receive some proportion of the face value of the deposits. 11 Merton (1977) characterizes deposit insurance as put option written by the deposit insurer to the equity holders of the bank. The value of this put option in essence represents the fair insurance 11 Under many deposit insurance schemes the full face value (or any fraction of the value) of the debt may not be explicitly guaranteed. 21

24 premium to the bank for deposit insurance. Our strategy is to exploit the economics of this put option framework to generate an empirical specification with which to estimate risk-shifting behavior by banks. Merton (1977) derives a theoretical pricing model for the deposit insurance put option that is a non-linear function of the volatility of the bank s assets and the bank s leverage. The existence of this put option creates incentives for banks to shift risk onto the guarantee agency by increasing the risk of assets without simultaneously increasing capital adequately: Risk-shifting occurs when banks manage to increase the risk-adjusted cost of deposit insurance that deposit insurance agencies are unable to pass onto individual banks. Let IPP represent the value of the put option per dollar of deposits, the volatility of the market value of the bank s assets, and the leverage of the bank (defined as the face value of debt divided by the market value of the assets). 12 Following Ronn and Verma (1986) and Duan et al. (1992), consider first the following linear approximation for the value of the deposit put option:.. (5) Note that and in (5) represent the partial derivatives of IPP with respect to volatility and leverage, respectively. However, deposit insurers and uninsured creditors will generally attempt to impose discipline on the bank s risk taking. As a result, banks are not generally unconstrained in their choice of risk, leverage pairs. To incorporate this disciplinary force into (5), we revisit the equilibrium relation discussed above in section 3.4 that specifies leverage as a linear function of asset volatility (i.e., equation (4)). That is, consider the relation: 12 Merton (1977) derives the comparative static results that 0 and 0. 22

25 , (6) where, 0 in (6) represents the natural equilibrium relation between risk and leverage (capital). We embed the disciplinary force from (6) in (5) by substituting the right hand side of (6) into (5) for D/V and simplify, yielding, (7) where. The coefficient captures the net effect of the struggle between risk-shifting banks and outside disciplining forces. The first term in (4),, captures the bank s incentive to increase risk, while the second term,, is generally negative and captures the offsetting impact of the disciplinary response to increased risk via. This second term is generally negative as is expected to be negative and 0. The overall interpretation is that 0 is consistent with observed risk-shifting as the disciplining effect does not completely neutralize incentives to increase risk. The economic intuition behind the interpretation of as observed risk-shifting is that if banks find risk-shifting behavior beneficial (i.e. profit maximizing) then they would manage their overall risk levels in such a way that would increase the actuarially fair value of their insurance. If banks do not find risk-shifting behavior to be advantageous, they would have no incentive to manage their risk in such a way because any increases in the risk profile would be borne by the equity holders of the bank. 23

26 Following Duan et al. (1992), Hovakimian and Kane (2000), and Hovakimian, Kane and Laeven (2003), we estimate variants of (7) in changes to examine the relation between discretion and risk shifting. In particular, we estimate the following: IPP 0 1 v 2 v * LLPREGIME 3 LLPREGIME (8) where IPP is the change in the fair deposit insurance premium (see Appendix A for calculation) and other variables are defined above. The coefficient captures the impact of discretion on risk shifting, where >0 would be consistent with higher discretion increasing risk shifting. In Table 5, panel B, column I reports the estimation of risk shifting forces in general, documenting that 0. That is on average, banks risk shifting incentives dominate the disciplinary pressure imposed on them by regulators and investors. We next examine whether Smoothing and LLP Opacity exacerbate observed risk shifting. Columns II and III of Table 5 Panel B, document a coefficient of (p-value < 0.01) for the interaction with Smoothing, and a coefficient of (p-value < 0.01) for the interaction with LLP Opacity. Taken together the results show that in regimes where banks are given more discretion over the loan loss provision there is more observed risk shifting. Incentives to risk shift should generally increase with declining performance, of the bank (e.g., Eisdorfer (2008), Loktionov (2009)), implying that the effects documented in Table 5 should be more pronounced in poor performing banks. To test this conjecture we first rank our banks by return on equity (ROE) into quintiles and then take the top quintile as good performers and the bottom as poor performers. We predict that we should observe risk shifting in low ROE firms; we then interact country-level discretionary provision measure and expect that the opacity magnify the effect in the low ROE firms. Table 6 panels A and B report the results. 24

27 Focusing first on Panel A, the first two columns shows that for poor performers, the coefficient on the change in risk is (p-value < 0.01), indicative of poor performers having strong incentives to risk shift. In contrast, for the high ROE group, the coefficient on the change in risk is insignificantly different from zero. Turning now to impact of discretion, and focusing our discussion on the Smoothing results in panel A (LLP Opacity provides similar results in panel B), we find for the low ROE group the coefficient on the Smoothing interaction is positive (0.4553) and significant (p-value < 0.01), whereas for the high ROE group the coefficient is insignificant (0.0960, p-value > 0.10). These results suggest that in time of poor performance banks have the incentive to risk shift and moreover, discretion over loan provisioning exacerbates the effect. Overall, the results of sections 3.4 and 3.5 are consistent with discretion resulting in lower bank transparency which in turn weakens the ability of regulators and outside investors to monitor and discipline bank risk taking 3.6 Robustness We first address basic issues of correlated omitted variables. We rerun our risk-shifting specification from tables 5, but now include a wide range of additional variables that might affect risk shifting. All variables are defined in detail in Appendix A. First we include the well known measure of the general securities market disclosure in a country from LaPorta et al. (2006) (Disclosure). We also control for whether the bank is state owned (StateBank) as such banks may have differing incentives. To control for cross-country differences in shareholder rights, following LaPorta et al. (1998) we include a proxy for shareholder rights (Rights). Because the risk shifting analyses relies on market prices, we control for development of the equity markets in a country by including the country s market liquidity (Liquidity) (LaPorta et al., 2006), and 25