Bank Earnings and Regulatory Capital Management using Available for Sale Securities

Transcription

1 Bank Earnings and Regulatory Capital Management using Available for Sale Securities Mary E. Barth * Graduate School of Business Stanford University Javier Gomez-Biscarri Departamento de Economía y Empresa Universitat Pompeu Fabra and Barcelona Graduate School of Economics Ron Kasznik Graduate School of Business Stanford University Germán López-Espinosa Departamento de Empresa Universidad de Navarra June 2014 * Corresponding author: mbarth@stanford.edu. We appreciate the helpful comments and suggestions from Min Ji Lee, Doron Nissim, Stephen Penman, and seminar participants at the American Accounting Association annual meeting, especially John McInnis the discussant; Barcelona Accounting Seminar Series; European Accounting Association Annual Congress; Spanish Finance Association annual conference, especially Ricardo Gimeno the discussant; Universidad de Navarra; University of Neuchatel; Universitat Pompeu Fabra; Universidad de Valladolid; and the VII International Research Symposium for Accounting Academics (Madrid), especially Pedro J. García the discussant. We also appreciate the data collection efforts of Min Ji Lee. Mary E. Barth and Ron Kasznik acknowledge funding from the Center for Global Business and the Economy at Stanford Graduate School of Business; Germán López-Espinosa (Javier Gomez-Biscarri) acknowledges financial support from the Ministerio de Ciencia e Innovación project ECO (ECO ) and from the Gobierno de Navarra Jerónimo de Ayanz program (Barcelona GSE). Electronic copy available at:

2 Bank Earnings and Regulatory Capital Management using Available for Sale Securities ABSTRACT We address banks use of available-for-sale (AFS) securities to manage earnings and regulatory capital. Although prior research investigates banks use of realized securities gains and losses to smooth earnings and regulatory capital, results are mixed. Creation of AFS securities and enhanced disclosures permit more powerful tests and new insights. We find banks realize gains and losses on AFS securities to smooth earnings and regulatory capital, and banks with more accumulated unrealized gains and losses do so to a greater extent. Banks with negative earnings realize losses to take a big bath, unless they have accumulated unrealized gains that offset the negative earnings. If so, they smooth earnings. Our inferences apply to non-listed and listed banks, which suggests the incentives do not derive solely from public capital market pressures. Our findings reveal the discretion afforded by historical cost-based accounting for AFS securities gains and losses enables banks to manage earnings and regulatory capital. Keywords: Banks, earnings management, available-for-sale securities, realized gains and losses, regulatory capital. JEL Classification: M41, G12. Electronic copy available at:

3 Bank Earnings and Regulatory Capital Management using Available for Sale Securities 1. INTRODUCTION The question we address is whether, and in what way, banks use available-for-sale (AFS) securities to manage earnings and regulatory capital. 1 AFS securities can be an earnings and regulatory capital management tool because although these securities are measured at fair value in the statement of financial position, their unrealized gains and losses are recognized in other comprehensive income. The gains and losses are recognized in earnings and, largely, in regulatory capital only when they are realized. We find consistent evidence that banks both publicly listed and non-listed use AFS securities to smooth earnings and regulatory capital, engage in big bath earnings management, and avoid negative earnings. Taken together, our findings reveal that the accounting for AFS securities does not eliminate banks incentive to use these securities as an earnings and regulatory capital management tool, and that bank managers seem to view earnings, and not comprehensive income, as the target for earnings management. Although prior studies address the question of whether banks use realized securities gains and losses to smooth earnings and regulatory capital, the studies focus on investment securities, not AFS securities, and report mixed results. We revisit this question in light of changes in accounting and disclosure that enable us to conduct more powerful tests and provide new insights. Key to our research design is Accounting Standards Codification (ASC) Topic 320 (formerly Statement of Financial Accounting Standards No. 115, FASB 1993). ASC Topic 320 requires all entities, including banks, to classify securities as trading, held-to-maturity (HTM), or available for sale. Trading securities are measured at fair value, and changes in their fair value are included in earnings and regulatory capital. As a result, there is no incentive to realize 1 We use the term earnings and net income interchangeably. Electronic copy available at:

4 trading securities gains and losses to manage earnings or regulatory capital. Although HTM securities are measured at amortized cost and, thus, these securities could be used to manage earnings and regulatory capital, selling HTM securities risks violating the tainting provisions in ASC Topic 320. AFS is the largest category of banks securities. AFS securities are measured at fair value in the statement of financial position, with changes in fair value, i.e., unrealized gains and losses, recognized in other comprehensive income. These gains and losses are only recognized in earnings when they are realized. This historical cost-based accounting treatment for AFS securities gains and losses provides the opportunity for banks to engage in earnings and regulatory capital management by selling these securities and realizing selected gains and losses. However, if bank managers view comprehensive income as the earnings management target, we will not observe evidence of this selective selling behavior to manage earnings. Gains and losses on AFS securities also generally only affect regulatory capital when they are realized. Thus, banks decisions regarding when to sell AFS securities and realize any unrealized gain or loss also affects regulatory capital. Banks disclosure of accumulated unrealized gains and unrealized losses on AFS securities permits us to incorporate into our research design the extent to which a bank has accumulated unrealized gains or losses and, thus, the opportunity to realize gains and losses to smooth earnings and regulatory capital or to take a big bath. Our research design also benefits from publicly available quarterly bank regulatory reports that include detailed information about a bank s regulatory capital in addition to its securities. Prior research needed to estimate regulatory capital, which introduced estimation error into the empirical tests. In addition, the reports cover all banks, publicly listed and non-listed, whereas prior research generally only 2

5 could include the largest listed banks. Our sample comprises 136,879 bank-quarter observations for 6,300 banks, with 15,204 (121,675) observations for 728 (5,862) listed (non-listed) banks from 1996 to We base our inferences on estimations of the relation between realized gains and losses on AFS securities and earnings and regulatory capital, each calculated before the effect of the realized gains and losses, and control variables identified in prior research. We begin by focusing on smoothing of earnings and regulatory capital. Consistent with smoothing, we find a significant negative relation between realized AFS securities gains and losses and pre-gains-andlosses earnings and regulatory capital. We also find that the relation is significantly more negative when the bank has more accumulated unrealized gains and losses at the beginning of the quarter. This finding provides new evidence that banks with more opportunity to smooth earnings and regulatory capital by realizing heretofore unrealized gains and losses on AFS securities do so to a greater extent. We next focus on testing whether banks with negative earnings take a big bath, which prior research does not consider. We test for big bath earnings management by permitting the relation between realized gains and losses on AFS securities and earnings before the realized gains and losses to differ for banks with positive and negative pre-gains-and-losses earnings. We find evidence of earnings smoothing by banks with positive earnings and evidence of taking a big bath by banks with negative earnings. In addition, we find that banks with negative earnings take big baths to a greater extent when they have more accumulated unrealized losses and engage in earnings smoothing to a greater extent when they have more accumulated unrealized gains. Estimating the relations separately for publicly listed and non-listed banks reveals that the findings relating to smoothing earnings and regulatory capital and taking a big bath apply to both 3

6 types of banks. We also find evidence that non-listed banks with more accumulated unrealized gains and losses engage to a greater extent in regulatory capital smoothing; we do not find this evidence for listed banks. These findings contrast with those in prior research that pre-dates ASC Topic 320, uses considerably smaller samples, and finds weaker evidence of earnings management for non-listed banks. Our findings reveal that incentives to manage earnings and regulatory capital do not derive solely from public capital market pressures. Regarding loss avoidance, we find evidence that listed banks realize just enough AFS securities gains to offset negative earnings before the realized gains and losses; non-listed banks do not realize quite enough gains to offset their pre-gains-and-losses negative earnings. We also find that banks with accumulated unrealized gains on AFS securities that are not large enough to convert negative earnings before realized AFS securities gains and losses into positive earnings take a big bath, and banks for which the sum of earnings before realized gains and losses on AFS securities and accumulated unrealized gains is positive engage in earnings smoothing. Additional findings support our inferences. First, our inferences are robust to considering simultaneously loan loss provisions, securitization gains, and changes in equity financing as alternative earnings and regulatory capital management tools. Second, using the median bank s earnings before taxes (regulatory capital) as the earnings (regulatory capital) target reveals consistent evidence of earnings (regulatory capital) management. Third, consistent with our expectations, we find no evidence that banks use fair value gains and losses on trading securities to smooth earnings or regulatory capital. The remainder of the paper proceeds as follows. Section 2 discusses related research and explains the basis for our empirical predictions. Section 3 develops the research design. Section 4

7 4 describes the data and Section 5 presents the results. Section 6 offers a summary and concluding remarks. 2. INSTITUTIONAL BACKGROUND AND RELATED RESEARCH 2.1 Available-for-sale securities The accounting for investment securities in the US is specified in FASB Accounting Standards Codification (ASC) Topic Prior to ASC Topic 320, banks classified securities as trading or investment. Trading securities, which typically represented a small fraction of bank assets (Barth 1994), were debt securities the bank intended to trade actively, and were measured at fair value with changes in fair value recognized in earnings. Investment securities, which represented most bank securities, were securities the bank had the ability and intent to hold to maturity, and were measured at amortized cost. Although the fair value of investment securities was disclosed, there was no accounting recognition either on the statement of financial position or the statement of income based on those fair values. Only realized investment securities gains and losses were recognized in earnings and equity. 3 ASC Topic 320 requires all entities, including banks, to categorize securities into one of three categories. Trading securities are securities that the bank holds for the purpose of selling them in the near term; this category was largely unchanged by ASC Topic 320. Held-to-maturity (HTM) securities are debt securities that the bank has the positive intent and ability to hold to maturity. Although this description is similar to that for investment securities, the classification criteria in ASC Topic 320 are more stringent. Most notably, ASC Topic 320 introduced tainting rules, under which sales of HTM securities could taint the bank s assertion that it had a positive intent to hold the securities to maturity, thereby precluding the bank from using the 2 ASC Topic 320 formerly was Statement of Financial Accounting Standards (SFAS) No. 115 (FASB, 1993). Throughout we refer to ASC Topic 320, rather than SFAS Securities held for sale were measured at the lower of cost or market. Held-for-sale securities comprised a small fraction of banks securities. 5

8 HTM category. The third category is available-for-sale (AFS). AFS securities are securities other than those categorized as trading and held to maturity. Shortly after ASC Topic 320 became effective the FASB offered a brief reclassification period during which banks could reclassify securities from HTM, i.e., securities previously classified as investment securities, to AFS (FASB 1995). Most banks did so. 4 As a result, AFS replaced investment securities as the largest category of banks securities. Unlike investment securities, AFS securities are measured at fair value. However, changes in AFS securities fair value, i.e., unrealized gains and losses, are recognized in other comprehensive income. These gains and losses are only recognized in earnings when the security is sold or otherwise disposed of, or when impairment is deemed other-than-temporary. That unrealized gains and losses on AFS securities do not affect earnings admits the possibility that bank managers selectively sell or dispose of these securities to manage earnings. However, if bank managers view comprehensive income as the earnings management target, we will not observe evidence of this selective selling behavior. Accumulated unrealized gains and losses on AFS securities also are excluded from Tier 1 regulatory capital and, thus, only affect regulatory capital when they are realized. 5 Thus, banks decisions regarding when to sell AFS securities and realize any unrealized gains or losses determines when the gains or losses are included in regulatory capital as well as earnings. 4 Untabulated statistics based on our sample banks reveal that during this brief reclassification period for approximately 80% (69%) of publicly listed (non-listed) banks the increase in AFS securities, as a percentage of total securities, was within 1% of the decrease in HTM securities. 5 45% of unrealized gains and losses on AFS equity securities are included in Tier II capital. However, untabulated statistics based on our sample of banks reveal that the percentage of equity AFS securities the banks hold is small; the median is zero and the 75 th percentile is < 1%. Thus, we proceed as if AFS securities are debt securities. The existence of equity AFS securities weakens the ability to realize gains and losses on such securities to manage regulatory capital, thereby biasing against us finding evidence consistent with banks using AFS securities to manage regulatory capital. Section considers this possibility. 6

9 Our tests focus on realized gains and losses on AFS securities as an earnings and regulatory management tool. Trading securities are measured at fair value, with changes in fair value, i.e., unrealized gains and losses, recognized in earnings and regulatory capital. Thus, there is no incentive to realize trading securities gains and losses to manage earnings or regulatory capital. Because HTM securities are measured at amortized cost, realizing gains and losses on HTM securities affects earnings. However, selling HTM securities risks violating the tainting rules, which means that banks are unlikely to use realized gains and losses on HTM securities as an earnings or regulatory capital management tool. Using realized gains and losses on AFS securities is a potentially attractive way to manage earnings and regulatory capital because it is likely less costly than managing accruals or engaging in other types of real earnings management. 6 Although sales of securities involve transactions costs, such sales are not as subject to ex post scrutiny such as from auditors as accounting accruals because there is a bona fide transaction. Such scrutiny is costly and, coupled with accounting rules, limits the extent to which banks can manage accruals. Although the extent to which a bank can realize AFS securities gains and losses to manage earnings and regulatory capital is limited by the extent to which the bank has unrealized gains and losses, it avoids the other costs associated with accruals-based earnings management. In addition, unlike other real earnings management tools, such as engaging in asset securitizations, banks can essentially eliminate any real effects of the sales of securities, which are typically easily obtainable financial instruments, by repurchasing the securities shortly even momentarily after their sale. Thus, a bank can realize AFS securities gains and losses without affecting its 6 Real earnings management involves entering into transactions that are reflected in financial reporting (Goel and Thakor 2003). Graham, Harvey, and Rajgopal (2005) reports results of a survey of chief financial officers that indicate real earnings management is more prevalent than accrual-based earnings management. 7

10 operations or risk profile, which makes securities sales less costly than other real transactions as earnings and regulatory capital management tools. Opportunistically realizing gains and losses on AFS securities also has the potential to affect earnings and regulatory capital to an economically meaningful extent. Nissim and Penman (2007) and Laux and Leuz (2010) report that banks classify 95% of their non-trading securities as AFS, which represents 16% of total assets. Although trading securities represent more than 12% of total assets for the largest bank holding companies, trading securities represent almost 0% of total assets for the majority of banks Related research on earnings management in banks The question of how entities use accounting discretion to manage earnings and other performance metrics is of longstanding interest in the accounting literature. The literature on earnings management is extensive, and offers several incentives for firms to manage earnings, such as smoothing earnings (e.g., Healy and Wahlen 1999; Dechow, Myers, and Shakespeare 2010b), managing earnings relative to a target (e.g., Burgstahler and Dichev 1997; Degeorge, Patel, and Zeckhauser, 1999), and recognizing large losses when a loss is unavoidable to increase the likelihood of positive earnings in the future, i.e., take a big bath (see, e.g., Dechow, Ge, and Schrand 2010a for a review). All of these incentives apply to banks. In addition, banks are subject to regulatory capital requirements and, thus, have an incentive to manage regulatory capital. As a result, banks could manage earnings to accomplish that objective. Beatty and Liao (2013, particularly section 5) provides a recent review of the large empirical literature on the banking industry, including that addressing earnings management. 7 Untabulated statistics reveal that, on average, our sample banks classify 86% of their non-trading securities as AFS, which represents 17% of total assets. Although trading securities represent 15% of total assets for the ten largest bank holding companies, they represent less than 0.06% of total assets for the other banks. 8

11 Because the loan loss provision is a large bank accrual and requires considerable judgement to determine its amount, many studies investigate whether banks use the provision to smooth earnings. Some studies find evidence that they do (Ma 1988; Greenawalt and Sinkey 1988; Wahlen 1994; Beatty, Chamberlain, and Magliolo 1995; Collins, Shackelford, and Wahlen 1995; Bhat 1996; Kim and Kross 1998; Kanagaretnam, Lobo, and Mathieu 2003; Pérez, Salas- Fumás, and Saurina 2008). However, other studies do not (Wetmore and Brick 1994; Ahmed, Takeda, and Thomas 1999). Wall and Koch (2000) provides a summary of this literature. Kilic, Lobo, Ranasinghe, and Sivaramakrishnan (2013) finds that adoption of SFAS 133, which reduces banks ability to use derivatives to smooth earnings, is associated with an increase in the use of the loan loss provision to smooth earnings. The evidence relating to banks using the loan loss provision to manage regulatory capital is more consistent (Moyer 1990; Beatty et al. 1995; Ahmed et al. 1999; Pérez et al. 2008). Although we focus on realized AFS securities gains and losses, we also consider whether the simultaneous use of the loan loss provision as an earnings and regulatory capital management tool affects our inferences. DeFond (2010) observes that research on real earnings management is scant. However, some studies investigate real earnings management by banks and focus on the strategic timing of asset sales, usually to recognize gains. A few studies find a significant association between earnings management incentives and asset securitizations (Karaoglu 2005; Dechow and Shakespeare 2009; Dechow et al. 2010b). 8 Although these studies do not consider the sale of AFS securities, we consider whether the simultaneous use of asset securitizations as an earnings and regulatory capital management tool affects our inferences. 8 Song and Linsmeier (2010) investigates the use of interest rate swaps to manage net interest income for a sample of 546 bank-year observations for the largest bank holding companies from 1995 to However, our interest is in earnings and regulatory capital, not net interest income. 9

12 Several studies consider banks use of realized securities gains and losses to manage earnings and regulatory capital and report mixed evidence. Regardless, all of these studies predate ASC Topic 320 and, thus, the existence of AFS securities. As section 2.1 explains, at the time of these studies, investment securities were measured at amortized cost. AFS securities, the focus of our study, are measured at fair value with changes in fair value recognized in other comprehensive income until realized, when they are recycled into earnings. Moyer (1990) predicts that banks opportunistically realize securities gains and losses to increase regulatory capital. Based on a sample of 845 bank-year observations relating to 142 banks from 1981 to 1986, Moyer (1990) finds evidence consistent with banks using realized securities gains and losses to increase regulatory capital, but only for banks with regulatory capital below the minimum. 9 Based on a sample basically the same as that in Moyer (1990), Scholes, Wilson, and Wolfson (1990 SWW) finds a significant negative relation between regulatory capital and realized securities gains and losses, which indicates that banks manage regulatory capital through the realization of these gains and losses. SWW also finds a significant negative relation between realized securities gains and losses and the loan loss provision. Although SWW interprets this finding as consistent with banks using realized securities gains and losses to smooth earnings by offsetting the earnings effect of the loan loss provision, SWW does not provide evidence directly supporting this interpretation. Collins et al. (1995 CSW) investigates the impact of banks changing levels of capital and earnings on several capital-raising decisions, including realizing securities gains and losses. Based on a sample of 1,760 bank-year observations for 160 banks from 1971 to 1991, CSW 9 This finding is consistent with our regulatory capital smoothing prediction because we predict that banks with lower regulatory capital realize more AFS securities gains. However, very few of our bank-quarter observations have regulatory capital below the required minimum. See footnote 24. Thus, the incentive to avoid violating minimum regulatory capital requirements largely is not present in our sample. 10

13 estimates bank-specific regressions and reports inconclusive findings regarding whether banks use realized securities gains and losses to smooth earnings and regulatory capital. CSW finds that realized securities gains and losses are not consistently negatively related to earnings across estimation specifications, and are not significantly negatively related to regulatory capital after Beatty et al. (1995 BCM) investigates how banks manage regulatory capital and earnings using realized miscellaneous gains and losses, which is the sum of gains and losses on investment securities and all other gains, including those on physical assets. A key innovation of BCM is the simultaneous estimation of a system of equations to consider miscellaneous gains and losses, in addition to the loan loss provision, equity issues, and other items, as tools to manage earnings and regulatory capital. BCM s sample comprises 638 bank-year observations for 148 banks from 1985 to Even though BCM finds that miscellaneous gains and losses are negatively related to earnings and regulatory capital, the relations are not consistently significant. 10,11 Other studies investigate whether the use of realized securities gains and losses to manage earnings differs for publicly listed and non-listed banks. Beatty and Harris (1999) compares realized securities gains and losses for 297 publicly listed and 553 non-listed bank-year observations from 1991 and 1992, and finds that although both groups of banks use realized securities gains and losses to smooth earnings, the relation is significantly stronger for listed banks. Using a sample of 4,762 (4,940) yearly observations for 707 publicly listed (1,160 non- 10 For example, BCM tables 6, 7, and 8 reveal a significant negative relation between earnings and miscellaneous gains and losses, which is consistent with using the gains and losses to smooth earnings. However, BCM section 5.3 reports that the relation is not significant after 1987 if the estimating equation includes year fixed effects. 11 Barth, Beaver, and Wolfson (1990) and Ahmed and Takeda (1995) find that the relation between returns and realized securities gains and losses is weaker for banks with lower regulatory capital and earnings, which is consistent with the banks having incentives to manage earnings using realized securities gains and losses. 11

14 listed) bank holding companies from 1988 to 1998, Beatty, Ke, and Petroni (2002) finds evidence that publicly listed banks are more likely to use realized securities gains and losses to transform small decreases in earnings before the gains to small increases in earnings. Taken together, prior research does not provide consistent evidence that banks use realized securities gains and losses to smooth earnings and regulatory capital. We re-address this question in light of the changes that have taken place since the time of this prior research that should enhance our ability to conduct more powerful tests. 12 In particular, the availability of quarterly bank regulatory reports for all publicly listed and non-listed banks enables us to test our predictions on a considerably larger sample with greater cross-sectional variation. Whereas samples in prior studies generally comprise fewer than 2,000 bank-year observations relating to 200 of the largest banks, our sample comprises 136,879 bank-quarter observations relating to 6,300 banks. In addition, the reports include the amount of a bank s regulatory capital. Prior research needed to estimate regulatory capital based on its general definition, which likely introduced estimation error. Moreover, we provide insights that prior research could not provide. First, we use the disclosure of accumulated unrealized gains and unrealized losses on AFS securities to take into account the extent to which a bank has the opportunity to realize gains and losses on AFS securities, not just the incentive, and to distinguish accumulated unrealized gains and unrealized losses, thereby permitting us to test for big bath earnings management as well as earnings smoothing. Second, we exploit the fact that quarterly regulatory reports are available for all banks publicly listed and non-listed, and regardless of size not just large bank holding 12 Lifschutz (2002) is more recent, but is limited in scope. For a sample of 88 bank holding companies from , the study finds that the motivation for bank managers to engage in gains trading is negatively related to the level of earnings before taxes and securities net gains. 12

15 companies. These reports enable us to conduct powerful tests of banks using AFS securities to manage earnings and regulatory capital for both publicly listed and non-listed banks. Third, because unrealized gains and losses on AFS securities are recognized in comprehensive income but not earnings, we provide evidence on whether bank managers view earnings or comprehensive income as the earnings management target. This, in turn, indicates whether the change in accounting for these securities removed the incentive for banks to use these securities as an earnings management tool. Some studies find that other comprehensive income is value relevant, particularly the unrealized securities gains and losses component (e.g., Dhaliwal, Subramanyam, and Trezevant 1999; Biddle and Choi 2006; Chambers, Linsmeier, Shakespeare, and Sougiannis 2007; Bamber, Jiang, Petroni, and Wang 2010). Except for Dhaliwal et al. (1999), the sample periods in these studies as largely is ours are after the issuance of ASC Topic 220 (formerly SFAS 130 FASB 1997), which led to more prominent disclosure of the components of other comprehensive income. Hirst and Hopkins (1998) shows that more prominent presentation of other comprehensive income decreases the tendency for firms to manage earnings. As a result, the findings of prior research relating to using realized securities gains and losses to manage earnings may no longer hold. 3. RESEARCH DESIGN 3.1 Realized gains and losses on AFS securities, earnings, and regulatory capital Our first two predictions are that banks use realized gains and losses on AFS securities to smooth earnings and regulatory capital, and the extent to which they do so depends on their opportunity. To test these predictions, we estimate equations (1a) and (1b). 13 RGL NI RegCap it 1 it 2 it UGL Liquid Int Sec VIX Unemp LIBOR 1 it 1 2 it 3 it 4 it 5 t 6 t 7 t it (1a) 13 For ease of exposition, we use the same notation for coefficients and error terms in each equation. 13

16 RGL NI RegCap NI UGL RegCap UGL it 1 it 2 it 3 it it 1 4 it it 1 UGL Liquid Int Sec VIX Unemp LIBOR 1 it 1 2 it 3 it 4 it 5 t 6 t 7 t it (1b) RGL is realized gains and losses on AFS securities. NI is net income before taxes and RGL. RGL and NI are scaled by beginning of quarter total assets. RegCap is the bank s end of quarter regulatory capital ratio, which is allowable Tier 1 plus allowable Tier 2 regulatory capital before RGL and after taxes scaled by risk-weighted assets. i and t denote bank and quarter. Equations (1a) and (1b), and all specifications that follow, include bank fixed effects. In all specifications, we cluster standard errors by bank and quarter when constructing t-statistics (Petersen 2009; Gow, Ormazabal, and Taylor 2010). To the extent that banks use AFS securities to smooth earnings and regulatory capital, we predict the coefficients on NI and RegCap, 1 and 2, are negative in equation (1a). Equation (1b) includes two interaction variables, NI UGL t 1 and RegCap UGL t 1, to test whether smoothing is more pronounced for banks with more accumulated unrealized gains and losses on AFS securities at the beginning of the quarter, i.e., when banks have more opportunity to engage in smoothing. To the extent this is the case, we predict 3 and 4 are negative. To the extent that banks realize gains and losses on AFS securities that were not unrealized at the beginning of the quarter to smooth earnings and regulatory capital, we also predict the coefficients on NI and RegCap, 1 and 2, are negative in equation (1b). Equations (1a) and (1b) both include four variables as controls for bank characteristics likely associated with realized gains and losses on AFS securities. UGL t 1 is accumulated unrealized gains and losses on AFS securities; the more accumulated unrealized gains and losses a bank has at the beginning of the quarter, the more likely it will realize them during the quarter (Scholes et al. 1990; Beatty and Harris 1999). Thus, we expect the coefficient on UGL t 1, 1, is 14

17 positive. Liquid is the amount of liquid assets (Beatty et al. 1995), Int is total interest income, and Sec is the total amount of securities; all three variables are scaled by beginning of quarter total assets. Because the extent of securities gains and losses likely depends on economic conditions, equations (1a) and (1b) also include controls for macroeconomic conditions. The presence of these variables, which vary by quarter, eliminates the need to include quarter fixed effects. They are VIX, the implied volatility of options on the S&P 500 Index; Unemp, the oneyear-ahead consensus forecast of the US unemployment rate; and LIBOR, the difference between the London Interbank Offer Rate and overnight indexed swap rates. Because RGL reflects securities gains and losses, we have no expectations for the coefficients on the control variables other than UGL t 1. Our next predictions are that some banks use AFS securities to take a big bath and the extent to which they do so depends on their opportunity. To test these predictions, we expand equation (1a) to test whether banks with positive (negative) NI realize more losses (gains) from AFS securities, which is consistent with earnings smoothing, or whether banks with negative NI realize more losses, which is consistent with taking a big bath. We also expand equation (1b) to test whether these relations are more pronounced when banks have more accumulated unrealized gains or unrealized losses on AFS securities at the beginning of quarter. We estimate equations (2a) and (2b): RGL Dpos negni ps o NI RegCap UL UG it 0 it 1 it 2 it 3 it 0 it 1 1 it 1 Liquid Int Sec VIX Unemp LIBOR 2 it 3 it 4 it 5 t 6 t 7 t it (2a) RGL Dpos negni posni RegCap negni UL it 0 it 1 it 2 it 3 it 4 it it 1 negni UG posni UL posni UG RegCap UGL 5 it it 1 6 it it 1 7 it it 1 8 it it 1 UL UG Liquid Int Sec VIX Unemp LIBOR 0 it 1 1 it 1 2 it 3 it 4 it 5 t 6 t 7 t it (2b) 15

18 where Dpos is an indicator variable that equals one if NI is greater than zero, and zero otherwise. posni (negni) is NI if NI is greater (less) than zero, and zero otherwise. UG (UL) is accumulated unrealized gains (unrealized losses) on AFS securities. All other variables are as previously defined. We do not partition the regulatory capital variable because no banks have negative regulatory capital and we are unaware of any incentives for a bank to engage in big bath regulatory capital management. In both equations, if banks use AFS securities to smooth earnings, as in equations (1a) and (1b), we predict the coefficients on negni and posni, 1 and 2, are negative. If banks with negative NI take a big bath, we predict the coefficient on negni, 1, is positive. In equation (2b), if banks with negative (positive) NI are more likely to smooth earnings when they have more accumulated unrealized gains (losses) on AFS securities, we predict the coefficient on negni UG t 1 (posni UL t 1 ), β 5 (β 6 ), is negative (positive). If banks with negative NI are more likely to take a big bath when they have more accumulated unrealized losses on AFS securities, we predict the coefficient on negni UL t 1, β 4, is negative; otherwise we have no prediction for β 4. We have no prediction for the coefficient on posni UG t 1, β 7 ; we estimate it for completion. All other predictions are as in equations (1a) and (1b). Accordingly, we predict the coefficients on UL t 1 and UG t 1, 0 and 1, are positive. 3.2 Simultaneous consideration of loan loss provisions, securitizations, and equity financing Prior research discussed in section 2.2 finds evidence that banks manage earnings and regulatory capital using the loan loss provision and manage earnings using securitization gains. Changes in equity financing also can be used as a direct means to manage regulatory capital. Thus, following Beatty et al. (1995), we next test whether the simultaneous consideration of 16

19 these earnings and regulatory capital management tools affects our inferences regarding AFS securities. We use two-stage least squares regression to estimate systems of equations based on equations (1a) and (1b). 14 For ease of exposition, we present the system based on equation (1b); the system based on equation (1a) is the same, except that it omits the interaction variables. 15 RGL LLP EqFin SG NI RegCap NI UGL ^ ^ ^ it 1 it 2 it 3 it 4 it 5 it 6 it it 1 ^ 14 7RegCapit UGLit 1 8UGL it 1 9Secit k 10 kcommoncontrols it it LLP RGL EqFin SG NI RegCap ^ ^ it 1 it 2 it 3 it 4 it 5 it NPL LLP Sec CommonControls 5 13 it 7 it 1 8 it k 9 k it it SG RGL LLP EqFin NI RegCap ^ ^ it 1 it 2 it 3 it 4 it 5 it 10 SG CommonControls 6 it 1 k 7 k it it EqFin RGL LLP SG RegCap ^ it 1 it 2 it 3 it 4 it EqFin Lev CommonControls 11 5 it 1 6 it 1 k 7 k it it (3a) (3b) (3c) (3d) LLP is the loan loss provision and SG is securitization gains. EqFin is change in equity other than that associated with current quarter comprehensive income, which we calculate as the change in total equity during the quarter minus total comprehensive income. NI^ (RegCap^) is net income (regulatory capital) before RGL, LLP, and SG, and CommonControls comprises Liquid, Int, VIX, Unemp, and LIBOR. Our predictions for the coefficients in equation (3a) are the same as for equation (1b). For equations (3b) and (3c), if banks use the loan loss provision and securitization gains to 14 To estimate the system, in the first stage, we estimate versions of equation (3a) through (3d) that include as explanatory variables all of the explanatory variables in the system, except for RGL, LLP, EqFin, and SG. In the second stage, we estimate equations (3a) through (3d) replacing RGL, LLP, EqFin, and SG as explanatory variables with their predicted amounts from the first stage. 15 We also estimate, but do not tabulate statistics from, systems of equations similar to equations (3a) through (3d), but analogous to equations (2a) and (2b). Section 5.4 reports the results. 17

20 smooth earnings and regulatory capital, we predict the coefficients on NI^ and RegCap^, 4 and 5, are negative. 16 For equation (3d), to the extent that banks use equity financing to manage regulatory capital, we predict the coefficient on RegCap^, 4, is positive. We have no predictions for the coefficients on the alternative earnings and regulatory capital management tools, 1, 2, and 3, because we have no predictions for which earnings management tools banks use or in which order. To identify the system, each equation includes variables omitted from the others. Equation (3a) includes UGL t 1 because we expect accumulated unrecognized gains and losses on AFS securities at the beginning of the quarter to be associated with realized gains and losses on the securities during the quarter. Equation (3b) includes LLP t 1 and ΔNPL, the change in nonperforming loans and leases scaled by beginning of quarter total assets, because we expect the bank s prior quarter loan loss provision and the change in nonperforming loans during the quarter to be associated with the loan loss provision (Beatty et al. 1995). Equations (3a) and (3b) include Sec because it captures the bank s asset composition; banks with higher investment in securities tend to have relatively lower investment in loans. Equation (3c) (equation (3d)) includes SG t 1 (ΔEqFin t 1 and Lev t 1 ) because we expect the bank s prior quarter securitization gains (equity financing and leverage) to be associated with the current quarter s securitization gains (equity financing). 4. SAMPLE, DATA, AND DESCRIPTIVE STATISTICS We base our sample on commercial banks and bank holding companies in the US Federal Reserve Bank of Chicago Bank Regulatory Database, which contains quarterly accounting information from forms filed by regulated depository financial institutions. Our sample 16 We code LLP as a positive number. Thus, higher LLP corresponds to larger expense. 18

21 comprises all top holder banks with data required for our tests from 1996 to We exclude observations from Q through Q to avoid the potential for the financial crisis to affect our inferences. 18 The sample begins in 1996 when regulatory capital ratios became publicly available. 19 We obtain our macroeconomic variables from the Federal Reserve. The resulting sample comprises 136,879 bank-quarter observations for 6,300 banks, with 15,204 (121,675) observations for 728 (5,862) publicly listed (non-listed) banks. 20 Realized gains and losses on AFS securities, RGL, is reported in line 6b of Schedule HI (RI) Income Statement for bank holding companies (other commercial banks). Accumulated unrealized gains and losses on AFS securities, UGL, is the difference between total fair value and amortized cost for these securities as reported in Schedule HC-B (RC-B) Securities for bank holding companies (other commercial banks). Accumulated unrealized gains and losses are reported separately for twenty-one categories of securities. We use the totals across categories to construct UGL. We use the by-category information to construct UG (UL), accumulated unrealized gains (losses) on AFS securities. Specifically, if the difference between fair value and amortized cost for a particular category of securities is positive (negative), we include that difference in UG (UL). Table 1 presents distributional statistics for the variables we use in our estimating equations. Table 1 reveals that, on average, banks realize gains on AFS securities (mean RGL = 0.01), although the median is zero. On average, net income before realized gains and losses on 17 All commercial banks file regulatory reports. Bank holding companies control, directly or indirectly, one or more commercial banks. A bank holding company can control another bank holding company, which in turn controls one or more other commercial banks. The entity at the top of the control chain is called the top holder, which could be a single commercial bank. 18 Nonetheless, untabulated findings from estimating equation (1a) including these observations reveal inferences identical to those revealed by the tabulated findings is the first year ASC Topic 320 required the classification of securities into trading, available for sale, and held to maturity. Untabulated findings reveal that beginning the sample period in 1994 does not affect our inferences. 20 The sum of 728 and 5,862 exceeds 6,300 because 290 banks changed status during the sample period. 19

22 AFS securities, NI, also is positive (mean = 0.27; median = 0.32). The mean (median) regulatory capital ratio before realized gains and losses on AFS securities, RegCap, is 19.20% (15.86%). Table 1 also reveals that, on average, banks have positive accumulated unrealized gains and losses on AFS securities (mean UGL = 0.08), but this is the net of unrealized gains (mean UG = 0.26) and unrealized losses (mean UL = 0.18). Untabulated statistics reveal that 57% (42%) of the observations have positive (negative) UGL and that 81% (70%) of the observations have positive (negative) UG (UL). Table 2 presents Pearson (Spearman) correlations above (below) the diagonal. Consistent with our predictions, the table reveals that realized gains and losses on AFS securities, RGL, is significantly negatively related to both net income and regulatory capital before RGL, i.e., NI and RegCap (Pearson corr. = 0.08 and 0.01; Spearman corr. = 0.07 and 0.06). Although almost all of the correlations are significantly different from zero, most are less than The exceptions are the correlations between unrealized gains and losses on AFS securities, UGL, and its components unrealized gains, UG, and unrealized losses, UL, and between UG and UL, which is to be expected. The only other exceptions are the correlations between RegCap and securities, Sec. Regardless, we rely on multivariate regressions to test our predictions. 5. FINDINGS 5.1 Realized gains and losses on AFS securities, earnings, and regulatory capital Table 3 presents regression summary statistics from estimating equations (1a) and (1b). Consistent with predictions, both sets of findings reveal that realized gains and losses on AFS securities, RGL, is significantly negatively related to net income before the gains and losses, NI (t-stats. = 3.32 and 3.40) and regulatory capital before the gains and losses, RegCap (t-stats. = 20

23 8.06 and 8.12). These findings are consistent with banks using AFS securities to smooth earnings and regulatory capital. Also consistent with predictions, table 3 reveals that the coefficients on the interaction variables in equation (1b), NI UGL t 1 and RegCap UGL t 1, are significantly negative (t-stats. = 4.99 and 3.80). These findings indicate that banks smooth earnings and regulatory capital using AFS securities to a greater extent when they have more accumulated unrealized gains and losses at the beginning of the quarter. The findings support the inferences we draw from equation (1a) by linking accumulated unrealized gains and losses at the beginning of the quarter to earnings and regulatory capital smoothing during the quarter. In addition, the findings provide evidence that bank managers perceive earnings, and not comprehensive income, as an earnings management target. Thus, recognizing unrealized gains and losses on AFS securities in other comprehensive income did not eliminate banks incentive to realize the gains and losses to manage earnings. Regarding the control variables, as expected, the coefficients on unrealized gains and losses on AFS securities at the beginning of the quarter, UGL t 1, are significantly positive in both equations (t-stats. = 6.95 and 6.23), which indicates that banks realize more gains and losses during the quarter when they have more accumulated unrealized gains and losses at the beginning of the quarter. Several of the coefficients on the other control variables are significantly different from zero. Table 4 presents summary statistics from estimating equations (2a) and (2b), which permit the coefficient on NI to differ depending on whether NI is positive or negative. 121,234 (15,534) observations for 6,120 (3,341) banks have positive (negative) NI. Table 4 reveals that the coefficients on Dpos and posni are significantly negative (t-stats. range from 3.44 to 21

24 5.55). These findings indicate that banks with positive earnings and with more positive earnings realize fewer gains on AFS securities, or more losses, which is consistent with earnings smoothing. Table 4 also reveals evidence consistent with banks with negative earnings taking a big bath, not smoothing earnings. The coefficients on negni are significantly positive (t-stats. = 4.96 and 3.36), which indicates that banks with more negative earnings realize more losses on AFS securities, or fewer gains. Equation (2b) reveals that the coefficient on negni UL t 1 is significantly negative (t-stat. = 4.60), which indicates that banks with more negative earnings realize greater net losses when they have more accumulated unrealized losses at the beginning of the quarter. This finding reveals that the big bath finding from equation (2a) largely is attributable to negative earnings banks with larger accumulated unrealized losses. As predicted, the coefficient on negni UG t 1 in equation (2b) is significantly negative (t-stat. = 3.09), which indicates that banks with more negative earnings realize greater net gains on AFS securities when they have more accumulated unrealized gains at the beginning of the quarter. This is consistent with earnings smoothing by banks with negative earnings when they have unrealized gains to realize. However, equation (2b) reveals no evidence that banks with more positive earnings and more accumulated unrealized losses at the beginning of the quarter smooth earnings; the coefficient on posni UL t 1 is not significantly different from zero (t-stat. = 0.77). Together, these findings reveal that the earnings smoothing finding in table 3 largely is attributable to negative earnings banks with larger accumulated unrealized gains, not to banks with positive earnings and larger accumulated unrealized losses. Regarding the control variables, the coefficients on UL t 1 and UG t 1 are both significantly positive (t-stats. range from 3.72 to 12.37), which indicates that banks with more accumulated 22

25 unrealized gains (losses) on AFS securities at the beginning of the quarter realize more gains or fewer losses (more losses or fewer gains) during the quarter. Inferences relating to all other coefficients in table 4 are the same as those relating to the corresponding coefficients in table Listed versus non-listed banks Because our tests do not require equity market variables, we are able to test our predictions on a large sample of publicly listed and non-listed banks. However, to the extent the incentive to manage earnings derives from public capital market pressures, we expect non-listed banks to engage in less earnings management (Beatty and Harris 1999; Beatty et al. 2002). To test whether the inferences we draw from tables 3 and 4 apply to publicly listed and non-listed banks, we estimate equations (1a) through (2b) separately for each subsample. Table 5 presents the findings. Panel A presents regression summary statistics relating to equations (1a) and (1b). Panel A reveals inferences for both categories of banks consistent with those revealed by table 3. The only exceptions are for listed banks in equation (1b); there is no evidence that listed banks smooth regulatory capital more when they have more accumulated unrealized gains and losses on the securities at the beginning of the quarter (t-stat. = 0.01 on RegCap UGL t 1 ) and, although the coefficient on NI UGL t 1 is significantly negative (t-stat. = 5.64), the coefficient on NI is not (t-stat. = 1.22). Panel B of table 5 presents regression summary statistics relating to equations (2a) and (2b). Panel B reveals inferences for both categories of banks that are the same as those revealed by table 3, with two exceptions relating to listed banks and equation (2b). First, the coefficient on posni UL t 1 is significantly negative in panel B, whereas it is insignificantly different from zero in table 4 and in equation (2b) for non-listed banks (t-stat. = 2.82 versus 0.77 and 0.38). This finding indicates that listed banks with more positive net income and more accumulated 23