저작권법에따른이용자의권리는위의내용에의하여영향을받지않습니다.

Transcription

1 저작자표시 - 비영리 - 변경금지 2.0 대한민국 이용자는아래의조건을따르는경우에한하여자유롭게 이저작물을복제, 배포, 전송, 전시, 공연및방송할수있습니다. 다음과같은조건을따라야합니다 : 저작자표시. 귀하는원저작자를표시하여야합니다. 비영리. 귀하는이저작물을영리목적으로이용할수없습니다. 변경금지. 귀하는이저작물을개작, 변형또는가공할수없습니다. 귀하는, 이저작물의재이용이나배포의경우, 이저작물에적용된이용허락조건을명확하게나타내어야합니다. 저작권자로부터별도의허가를받으면이러한조건들은적용되지않습니다. 저작권법에따른이용자의권리는위의내용에의하여영향을받지않습니다. 이것은이용허락규약 (Legal Code) 을이해하기쉽게요약한것입니다. Disclaimer

2 경영학박사학위논문 Essays on the Loan Loss Provisions in the Banking Industry 은행업의대손상각비에관한연구 2014 년 8 월 서울대학교대학원 경영학과경영학전공 김영준

3

4 Essays on the Loan Loss Provisions in the Banking Industry 지도교수황이석 이논문을경영학박사학위논문으로제출함 2014 년 4 월 서울대학교대학원 경영학과경영학전공 김영준 김영준의경영학박사학위논문을인준함 2014 년 7 월 위원장 곽수근 ( 인 ) 부위원장 황이석 ( 인 ) 위 원 백복현 ( 인 ) 위 원 남혜정 ( 인 ) 위 원 김명인 ( 인 )

5

6 ABSTRACT Essays on the Loan Loss Provisions in the Banking Industry Young Jun Kim Business School Seoul National University This dissertation consists of three related but independent essays on the accounting for loan loss provisions (LLP) in the banking industry. While LLP is the largest accrual item on banks balance sheets, little is known about accounting for LLP. The first and the second essays investigate how the equity market prices LLP of US banks in the post-basel period. The third essay examines how International Financing Reporting Standards (IFRS) affect loan loss provisioning of European banks, and its economic consequences. Below, I briefly elaborate on the three essays. The first essay examines the value relevance of LLP. Prior studies find that banks' discretionary LLP (DLLP) are perceived positively by the market and attribute this to greater LLP signaling to investors the soundness of the bank. However, these studies are based on data from the pre-basel era when LLP increased Tier 1 capital and thus had positive implications. I focus on the post-basel period in which LLP does not affect Tier 1 capital and apply a better specified model to test for the value relevance of loan loss provisions. I find that DLLP is not value relevant in the post-basel period. This result is consistent throughout the recent financial crisis, which is contrary to Ryan s (2011) conjecture that DLLP may be valued positively during economic downturn. In i

7 addition, LLP and non-discretionary LLP (NLLP) are perceived negatively by the market. I also show that findings in the long-window value relevance test still hold in the short-window market reaction test. Overall, in the post-basel period, DLLP provides no value relevant information whereas NLLP conveys value relevant information incremental to earnings. The second essay documents evidence on the mispricing of banks LLP in the equity market. First, I find that equity investors do not correctly price information in LLP: the level of LLP (change in LLP) is strongly (weakly) negatively related to one-year ahead future returns. When LLP is decomposed into non-discretionary LLP (NLLP) and discretionary accruals (DLLP), the level of and change in NLLP appears to be the main driver of return predictability. Second, I show that analysts do not fully impound information in LLP into their one year-ahead earnings forecasts: the level of and change in LLP are negatively related to analyst forecast optimism. Decomposition of LLP suggests that such bias is mainly due to NLLP. In sum, my findings suggest that equity market participants do not fully appreciate the loan-related risk information in LLP. The third essay examines the effect of IFRS on LLP of European banks and their loan origination pro-cyclicality. Using 1,545 bank-year observations in 14 European countries during 1996 to 2009, I find that, contrary to prior studies, there is weak evidence that banks reduce earnings smoothing via LLP in the post-ifrs period. However, there is evidence that banks increase LLP timeliness ii

8 in the post-ifrs period. In particular, the decrease in earnings smoothing is pronounced for low-capitalized banks, whereas the increase in LLP timeliness is more pronounced for high-capitalized banks. My finding raises the possibility that the prior studies are subject to misspecification by omitting LLP timeliness. Next, I find that IFRS adoption does not on average exacerbate the pro-cyclical relationship between LLP and loan growth with the exception of small banks. These findings are robust to various settings. My evidence suggests that loan loss provisioning under IFRS does not threaten the stability of the financial system. Keywords: Analysts forecasts; Loan loss provision; IFRS; Market reaction ; Mispricing; Pro-cyclicality; Signaling; Value relevance Student Number: iii

9 TABLE OF CONTENTS ABSTRACT... i Essay 1 The Value Relevance of Loan Loss Provisions... 1 I. Introduction... 2 II. Literature review and hypotheses development... 6 III. Sample selection and research design III.1 Sample III.2 Research Design IV. Empirical Results IV.1 Value relevance regression IV.2 The effect of the recent financial crisis on pricing DLLP V. Conclusion Appendix References Essay 2 The Mispricing of Loan Loss Provisions I. Introduction II. Literature review and hypothesis development III. Sample selection and research design III.1 Sample iv

10 III.2 Research Design IV. Empirical results IV.1 Stock return test IV.2 Analyst forecast error tests V. Conclusion References Essay 3 The Effect of IFRS on Loan Loss Provision and Loan Origination Pro-cyclicality: Evidence from European Banks I. Introduction II. Related literature and hypotheses development II.1 Loan loss provisioning in Europe II.2 Related Literature II.3 Hypotheses development III. Empirical model III.1 Loan loss provisioning model III.2 Loan origination pro-cyclicality model IV. Sample and descriptive statistics IV.1 Sample selection IV.2 Descriptive statistics V. Empirical results V.1 The impact of IFRS on loan loss provisioning V.2 The impact of IFRS on loan origination pro-cyclicality v

11 V.3 Robustness test VI. Conclusion References Appendix A 국문초록 감사의글 vi

12 LIST OF TABLES Essay 1 The Value Relevance of Loan Loss Provisions... 1 Table 1 Descriptive statistics Table 2 Estimation for the discretionary loan loss provision model Table 3 Regressions of contemporaneous stock returns on level of earnings and various loan loss provisions (level specification) Table 4 Regressions of contemporaneous stock returns on change in earnings and various loan loss provisions (change specification) Table 5 Regressions of contemporaneous stock returns on level of earnings and various loan loss provisions and their changes (level-change specification) Table 6 Regressions of contemporaneous stock returns on level of earnings and various loan loss provisions and their lagged value (level-lag specification) Table 7 Regressions of contemporaneous stock returns on level of earnings and various loan loss provisions and their lagged value by year Table 8 The effect of the recent financial crisis on valuing LLP and its components during (level-lag specification) Table 9 Regressions of stock returns around earnings announcement on level of earnings and various loan loss provisions and their lagged value (level-lag specification) Essay 2 The Mispricing of Loan Loss Provisions Table 1 Descriptive statistics for main variables Table 2 Estimation for the discretionary loan loss provision model vii

13 Table 3 Mean annual returns to various loan loss provisions quintile portfolios and their hedge returns Table 4 Cross-sectional regression test of the explanatory power of loan loss provision with respect to future annual stock returns Table 5 Loan loss provisions and future stock returns, with risk controls based on a three-factor model Table 6 Frequency of profit/loss of annual hedge portfolio returns to various loan loss provisions across year ( ) Table 7 Sell-side analyst forecasts across various loan loss provision portfolios Table 8 Regressions of sell-side analyst forecasts on various loan loss provision portfolios Essay 3 The Effect of IFRS on Loan Loss Provision and Loan Origination Pro-cyclicality: Evidence from European Banks Table 1 Distribution of countries in the sample (Period ) Table 2 Distribution of banks across type and listing status (Period ) Table 3 Descriptive statistics for variables (Period ) Table 4 The comparisons of Gebhardt and Novotny-Farkas (2011) with this Study Table 5 The effect of IFRS Adoption on income smoothing via LLP and LLP timeliness Table 6 The effect of IFRS Adoption on income smoothing via LLP and LLP timeliness by bank quality and bank size Table 7 The effect of IFRS on the association between loan loss provisions and loan growth viii

14 Table 8 The effect of IFRS on the association between loan loss provisions and loan growth by bank quality and bank size Table 9 The effect of IFRS adoption on income smoothing via LLP and LLP timeliness in the pre-crisis period and early crisis period Table 10 The effect of IFRS on the association between loan loss provisions and loan growth in the pre-crisis period and early crisis period Table 11 The effect of IFRS adoption on income smoothing via LLP and LLP timeliness by using the constant sample Table 12 The effect of IFRS on the association between loan loss provisions and loan growth by bank quality and bank size by using constant sample. 158 Table 13 The effect of IFRS adoption on income smoothing via LLP and LLP timeliness across countries Table 14 The effect of IFRS on the association between loan loss provisions and loan growth across countries ix

15 LIST OF FIGURES Essay 1 The Value Relevance of Loan Loss Provisions... 1 Figure 1 Annual regressions coefficients of LLP, NLLP, and DLLP Figure 2 Short-window market reaction to LLP, NLLP, and DLLP during the recent financial crisis (2007:3-2010:4) Essay 2 The Mispricing of Loan Loss Provisions Figure 1 Annual hedge returns to various loan loss provisions ( ) Essay 3 The Effect of IFRS on Loan Loss Provision and Loan Origination Pro-cyclicality: Evidence from European Banks Figure 1 Time Series Behavior of LLP/LOANS during 1996~ Figure 2 Time Series Behavior of LLR/NPL during 1996~ x

16 Essay 1 The Value Relevance of Loan Loss Provisions 1

17 I. Introduction The academic literature has argued that banks are more opaque than nonfinancial firms (Beatty and Liao 2013; Morgan 2002; Flannery et al. 2004; Flannery et al. 2013). Bank opacity is mainly attributable to the opacity of their assets. And among bank assets, loans account for the largest proportion. Accordingly, loan loss provisions (LLP), which is an expense item that is reserved to cover future credit losses, is the largest accrual item in banks accounts and contributes to bank opacity (Beatty and Liao 2013; Gallemore 2013). 1 Thus, it is critical for bank outsiders to understand loan loss provisions when valuing banks. Being an expense item, LLP is bad news in terms of its effect in reducing net loans, assets, and net income. However, academic studies have found LLP to have positive valuation. In particular, discretionary LLP has been considered good news to investors, signaling financial soundness of the bank to outsiders (Beaver 1989; Wahlen 1994; Beaver and Engel 1996). However this signaling hypothesis on the market valuation of LLP was challenged by a later study showing that discretionary LLP is not associated with higher stock returns or higher future earnings (Ahmed et al. 1999). Still, some recent studies find evidence supporting the signaling role of DLLP (Kanagaretnam et al. 2009; Kilic et al. 2013). Thus, the evidence from prior studies is mixed. It is also noteworthy that most prior studies are based on data from the early 1990s. Since then, a major banking regulation, BASEL, has changed the accounting treatment of LLP which may affect the market valuation of LLP. However, 1 Beatty and Liao (2013) shows that the largest accrual, loan loss provisions, accounts for 56% of total accruals in the banking industry and the second largest accrual, the change in other assets and liabilities, accounts for 30% of total accruals. 2

18 no study has investigated whether discretionary LLP is positively perceived by the market based on post-basel data. Given the importance of LLP in valuing banks, it is an interesting empirical question whether and how the market prices discretionary LLP. This study re-examines whether discretionary LLP is positively priced. More broadly, this study investigates the value relevance of LLP and its components: non-discretionary LLP (NLLP) and discretionary LLP (DLLP). In this study, I deal with four limitations in prior studies: two issues regarding the sample period and two issues related to model specification. First, as mentioned above, the sample periods of the prior studies do not fully reflect regulatory change. Specifically, in 1988, the Basel Committee on Bank Supervision (BCBS) changed the accounting treatment of LLP. In the pre-basel period, loan loss provisions (LLP) increased Tier 1 capital but decreased Tier 1 capital in the post-basel period. 2 Prior studies that find positive market valuation of DLLP are likely due to including the pre- Basel period when LLP increased capital. Second, prior studies do not include both economic booms and busts in their sample period. Ryan (2011) points out that DLLP may have positive market valuation only in economic downturn when managers have more incentive to convey the soundness of their banks. Incidentally, prior studies supporting the signaling hypothesis for DLLP study a period of economic bust (Beaver et al. 1989; Wahlen 1994; Beaver and Engel 1996) while a study rejecting the signaling hypothesis for DLLP examines an economic boom period (Ahmed et al. 1999). 2 The Basel Capital Accord (BASEL) is a risk-based capital framework. Banking regulators require banks to maintain minimum capital requirement. If a bank does not meet this requirement, banking regulators take the prompt corrective actions. 3

19 Third, the value relevance regression model used in prior studies is likely misspecified. Most prior studies only include the level of LLP in their model. However, the accounting literature has shown that the level and change variables convey distinct information to the market (Ali and Zarowin 1992; Easton and Harris 1991). Fourth, determinants of LLP used in prior studies do not include recently found determinants to the model estimating normal levels of LLP. Since Beaver and Engel (1996), the most relevant paper to my study, studies have documented that new variables, such as GDP growth, change in the unemployment, and bank asset size, determine non-discretionary LLP (Laeven 2003; Beatty and Liao 2011; Bushman and Williams 2012). I address the aforementioned four drawbacks in prior studies in the following ways. First, I use post-basel period data, after Second, I examine whether the signaling hypothesis is valid during economic downturn by examining the value relevance of LLP during the recent financial crisis. Third, in the value relevance model, I include the change in LLP in addition to the level of LLP. Fourth, in the estimation of normal LLP, I include GDP growth, change in the unemployment, and bank asset size. To test for the value relevance of LLP, I use four different regression specifications. First, I include only the level of LLP in the model to capture the effect of the transitory LLP shock on the market ( level specification ). Second, I include only the change in LLP in the model to capture the effect of the permanent LLP shock on the market ( change specification ). Third, I include both the level of and change in LLP to capture both transitory and permanent effect on the market ( level-change 3 I cannot compare pre- and post-basel periods because Compustat bank does not contain data on nonperforming assets (NPA) prior to 1993 which is required to estimate discretionary loan loss provisions. 4

20 specification ). Fourth, I re-express the level-change specification in terms of only a level and lag variable of LLP to facilitate interpretation of estimation results ( level-lag specification ). To examine whether the signaling hypothesis is valid during the recent financial crisis, I conduct two tests. First, I run the value relevance model each year to see how the market pricing of LLP and its components changes across time. Second, I examine whether the pricing of LLP is different during compared to other years. For robustness test, I conduct the market reaction test to loan loss disclosures around earnings announcement dates. My findings are summarized as follows. First, I find that LLP is negatively associated with contemporaneous stock returns. Second, after decomposing LLP into non-discretionary LLP (NLLP) and discretionary accruals (DLLP), I find that only nondiscretionary LLP (NLLP) is perceived negatively by the market and no effect of DLLP. Third, I show that DLLP is not value relevant even during the financial crisis when it is most likely to serve a signaling role. Lastly, I show that the value-relevance test results still holds in the market reaction test. My study contributes to the literature on capital market pricing of LLP in two ways. First, it is the first to exclusively use post-basel data in the literature on the capital market pricing of LLP. My finding suggests that findings in prior studies about the signaling role of DLLP are likely confined to the pre-basel period. Second, my study is the first to show that DLLP does not serve a signaling role even in economic downturn, contrary to the conjecture by Ryan (2011). My study also contributes to the literature on accounting accruals. Prior studies find that the investors price discretionary 5

21 accruals positively, which is interpreted as managers communicating information about equity value to the market (Chaney et al. 1998; Subramanyam 1996). 4 In contrast, my study shows that the market does not price discretionary loan loss provisions positively. Focusing on a single accrual item, which is less subject to measurement issues in modeling its discretionary component (McNichols and Wilson 1988), my study (i) provides evidence that not all accruals are used to communicate with outsiders; (ii) raises a possibility that the accrual component is not value relevant; and (iii) positive pricing of accruals could be driven by investors earnings fixation (Dechow et al. 2010). The remainder of the paper is organized as follows. Section II includes the literature review and hypotheses development. Section III presents the sample selection and research design. Section IV reports the empirical results. Section V concludes. II. Literature review and hypotheses development My study is in a line of studies on the capital market pricing of loan loss provisions. LLP has two conflicting implications to bank valuation. LLP may have positive implications if bank managers use LLP discretionally to signal to investors that they have the capability of enduring losses. In contrast, LLP may have negative implications because LLP represents the default risk in loans and, as an expense, also decreases reported earnings. 4 Both Chaney et al. (1998) and Subramanyam (1996) use the modified Jones model to measure abnormal accruals. 6

22 Earlier studies find that increase in LLP is priced positively, supporting the signaling hypothesis. These papers examine both the short-window market reaction which captures information content and long-window association which captures value relevance. Most short-window event studies examine bank announcements of increase in LLP for less-developed-country (LDC) debt during March to May 1987 (Grammatikos and Saunders 1990; Musumeci and Sinkey 1990; Elliott et al. 1991; Griffin and Wallach 1991). Large US banks had extended loans to LDCs in the beginning of the 1980s which became troubled loans. These studies document a positive market reaction to 1987 announcements of an increase in LLP for LDC debt and interpret it as banks signaling their ability to manage bad debt. Beaver et al. (1989) is the first study to use long-window associations in their sample period and show a positive relation between market-to-book ratio and loan loss reserves. They interpret this result as managers conveying to investors their ability to withstand a hit to earnings. Using a sample from , Wahlen (1994) finds that the increase in unexpected LLP is related to higher future earnings changes and, more importantly, positively associated with contemporaneous stock returns, supporting the signaling hypothesis. 5,6 Relatedly, Beaver and Engel (1996), which is the most closely related to my study, examine a sample. By decomposing LLP into NLLP and DLLP, they find that NLLP is negatively priced and 5 Wahlen (1994) does not decompose LLP into a non-discretionary component and a discretionary component. He only uses unexpected LLP, which is similarly interpreted as a discretionary LLP. 6 Wahlen (1994) is the only study to employ both the short window and long window analysis. His short window analysis reports that there is a positive reaction to increase in unexpected LLP around earnings announcement date. However, removing outliers of loan loss variable leads to insignificant reaction to the increase in unexpected LLP, which is interpreted that the positive market reaction is attributed to banks with unusually large unexpected LLP. 7

23 DLLP is positively priced, also supporting the signaling hypothesis. Studying a quarterly sample the period, Liu et al. (1997) document a positive association between LLP and contemporaneous stock returns only for banks with low regulatory capital ratios in the fourth quarter. Evidence of negative market implications of LLP emerge from studies using the mid-1990s data. For instance, in a sample from , Ahmed et al. (1999) find that increase in unexpected LLP is not related to higher future earnings changes or higher contemporaneous stock returns. Furthermore, they show that while NLLP is negatively priced, DLLP is not positively priced. Thus, their findings do not support the signaling hypothesis in Wahlen (1994) and Beaver and Engel (1996). Ahmed et al. (1999) attribute this to the difference in sample periods. While they do not specify why the valuation of LLP changes with time, one main difference with the former two studies is that Ahmed et al. (1999) includes post-basel period ( ). 7 In 1988, the accounting treatment of LLP and its impact on bank capital changed with the BASEL capital regulation. This change is likely to have affected the valuation of LLP. In the pre-basel period prior to 1988, LLP increased Tier 1 capital and loan loss allowances were included in Tier 1 capital. In contrary, in the post-basel period, LLP decreases Tier 1 capital by reducing earnings and loan loss allowances are not included in Tier 1 capital. Thus, increase in LLP has some direct positive implications to investors in the pre-basel period but not in the post-basel period. Incidentally, most prior studies on capital market pricing of LLP use data from the pre- 7 The Basel Capital Accord (BASEL) was fully implemented in the U.S. by

24 Basel era. The signaling hypothesis of DLLP supported in studies based on the pre- BASEL period may be due to the accounting treatment for LLP. Ahmed et al. (1999), while finding no evidence to support the signaling role of DLLP, do not relate it to changes in regulation. 8 It is still worthwhile to re-examine the value relevance of LLP, in particular DLLP, because half of the sample period in Ahmed et al. (1999) includes pre- BASEL era and it could somewhat affect the valuation of DLLP. Based on the above arguments, it is likely that DLLP does not play a signaling role of conveying private information to investors in the post-basel period. Thus, the stock market is not likely to price DLLP positively. This hypothesis is presented in null form. H1: The equity market does not price discretionary loan loss provision (DLLP). Ryan (2011) conjectures that the signaling role of DLLP may vary with the change in economic conditions. Specifically, bank managers have more incentive to signal expected improvements during bad economic times while they have little incentive to do so in good economic times. Incidentally, prior studies supporting the signaling hypothesis for DLLP study a period of economic bust (Beaver et al. 1989; Wahlen 1994; Beaver and Engel 1996) while a study rejecting the signaling hypothesis for DLLP examines an economic boom period (Ahmed et al. 1999). Thus, Ryan (2011) attributes the disappearance of positive pricing of LLP in the recent study to including 8 Beatty and Liao (2013) are the first to conjecture, but do not test for, a regulation-based explanations on the disappearing positive valuation implications of DLLP. 9

25 an economic boom period in the sample. He conceives that the recent financial crisis provides interesting research settings to test for capital market pricing of LLP and raises possibilities that positive capital market pricing implications of LLP may revive in the recent financial crisis. 9 On the other hand, there are reasons that the market may perceive DLLP as bad news during the financial crisis. First, since LLP reflects estimated losses in banks loan portfolios, the market could recognize high DLLP as more loan losses to be realized in the near future, especially during the financial crisis. A recent study, Jin et al. (2011) find a significant positive relation between the increase in loan loss reserves and the probability of bank failure during the recent financial crisis ( ). Second, because LLP is not included as core capital under the current BASEL regulation, investors may view high levels of (D)LLP negatively whether it is discretionary or not. 10 Based on the above arguments, it is an empirical question whether positive capital market pricing implications of LLP may revive in the recent financial crisis. The second hypothesis is presented in null form. H2: The equity market does not price discretionary loan loss provision (DLLP) during the recent financial crisis. 9 Ryan (2011) states that there is no study examining the short-window market reaction to loan loss provisions during the recent financial crisis. I conduct the long-window value-relevance test as well as the short-window test in order to test for the second hypothesis. 10 Recent studies, using post-basel data, also show mixed evidence of the signaling hypothesis of DLLP. For example, Kilic et al. (2013) show that DLLP is priced although less so after SFAS 133. Kanagaretnam et al. (2009) document a positive relation between DLLP and stock returns only for clients of Big 4 auditors. 10

26 III. Sample selection and research design III.1 Sample The sample period spans from 1994 to I use post-basel data while most prior studies use pre-basel data. Accounting data are taken from Compustat Bank. I limit the type of banks to commercial banks (SIC code: 6020) to maintain consistency of the business environment. Stock return data is from CRSP. Non-performing assets (NPA) are available since 1993 in Compustat Bank. Analysts forecast data are obtained from IBES. Macroeconomic data, such as GDP growth rate ( GDP) and unemployment rate (UNEMP), are from the Federal Reserve Bank of St. Louis ( I impose a minimal requirement on data to reduce any survivorship bias. I delete bank-year observations missing any variable. Except for oneyear ahead stock returns and macroeconomic data, I winsorize all variables at the top and bottom 1 percent each year to mitigate the influence of extreme observations. These sample selection criteria lead to 5,441 bank-year observations. 11 All variables are deflated by lagged total assets except for one-year ahead stock returns and macroeconomic data. Table 1 provides descriptive statistics for the main variables. LLP has mean and median values of and 0.003, respectively, which suggests that the distribution of LLP is symmetric. NLLP has mean and median values similar to that of LLP which suggests that the distribution of NLLP is similar to that of LLP. DLLP has mean zero, by 11 To use change variables such as LLP, one year data (year 1994) is lost. In addition, one-year ahead forecast error (FE) is constructed using the sample of observations for which analyst forecast data are available. 11

27 construction. The change variables, such as LLP and NLLP, have different distribution characteristics from level variables. Specifically, LLP has mean and median of and 0.000, respectively, which are different from that of LLP. Similarly, the level of earnings (EBTP) has different distribution characteristics from the change in earnings ( EBTP). [TABLE 1 ABOUT HERE] III.2 Research Design To test my hypotheses, I use two types of research design. First, I use a model to decompose LLP into non-discretionary LLP (NLLP) and discretionary LLP (DLLP). Second, I use value relevance models to test whether equity investors positively price DLLP (Hypothesis). Estimation of non-discretionary and discretionary LLP Following prior studies, I estimate the following pooled time-series and cross-sectional regression model (1) and I denote the predicted values as nondiscretionary LLP (NLLP t ) and the residuals as discretionary LLP (DLLP t ). 12 This is analogous to the decomposition of accruals into non-discretionary and discretionary accruals (Jones 1991; Dechow and Dichev 2002). 12 Most LLP studies (Beatty and Liao 2013) use this pooled time-series and cross-sectional regression model. Following studies on accruals, I also run cross-sectional regressions by year and find qualitatively similar results. 12

28 LLP t =α 0 +α 1 NPA t+1 +α 2 NPA t +α 3 LOAN t +α 4 NCO t +α 5 Size+α 6 GDP t + α 7 UNEMP t + u t (1),where LLP t NPA t+1 LOAN t NCO t Size GDP t UNEMP t Loan loss provision (COMPUSTAT pll ) scaled by lagged total assets (COMPUSTAT at ) Change in non-performing assets (COMPUSTAT npa ) scaled by lagged total assets (COMPUSTAT at ) Change in total loans (COMPUSTAT lntal ) scaled by lagged total assets (COMPUSTAT at ) Net charge off (COMPUSTAT nco ) scaled by lagged total assets (COMPUSTAT at ) The natural log of total assets (COMPUSTAT at ) Change in GDP over the year Change in unemployment rates over the year Following Beaver and Engel (1996), I include the next (current) period change in the non-performing assets (NPA), current period loan growth ( Loan t ), and net charge-off (NCO) as determinants of NLLP. Since Beaver and Engel (1996), subsequent studies have added other determinants to the model. Following these studies, I include lagged total assets (Size t-1 ) because banks tend to be regulated based on bank size (Beck and Narayanmoorth, 2013; Bushman and Williams, 2012). I also include the change in GDP over the year ( GDP t ) and the change in unemployment rates over the year ( UNEMP t ) to control for macroeconomic effects on LLP (Beatty and Liao, 2011; Bushman and Williams, 2012). Value-relevance test 13

29 For the value relevance test, I use four different regression specifications with stock return as a dependent variable. 13 First, I include only level variables of earnings and LLP in the regression model, denoted as the level specification (model 1). Second, I use only change variables of earnings and LLP in the model, denoted as the changes specification (model 2). Third, I include both variables of earnings and LLP in the model, denoted as the level-changes specification (model 3). Fourth, I use level variables of earnings and LLP and their lags in the model, denoted as the level-lag specification (model 4). Because model 3 contains two variables (a level variable and change variable) for earnings and each LLP variable, I re-express model (3) in terms of one level variable to facilitate interpretations. Thus, model (4) is equivalent to model (3) (See Appendix for details). To assess the value relevance of components of LLP, I also use a level and change of NLLP and DLLP in place of LLP in each specification. The subscript t indicates year and the subscript i does a specific bank in the below equations. Model (1) R t = α 0 + α 1 EBTP t + α 2 LLP t (α 21 NLLP t + α 22 DLLP t ) + α 5 NPA + ε i,t Model (2) R t = β 0 + β 1 EBTP t + β 2 LLP t (β 21 NLLP t + β 22 DLLP t ) + β 5 NPA + ε i,t Model (3) R t = γ 0 + γ 1EBTP t + γ 2 EBTP t + γ 3LLP t (γ 31NLLP t + γ 32DLLP t )+ γ 4 LLP t (γ 41 NLLP t + γ 42 DLLP t )+ γ 5 NPA + ε i,t Model (4) R t = δ 0 + δ 1 EBTP t + δ 2 EBTP t-1 + δ 3 LLP t (δ 31 NLLP t + δ 32 DLLP t ) +δ 4 LLP t-1 (δ 41 NLLP t-1 +δ 42 DLLP t-1 ) + δ 5 NPA t + δ 6 NPA t-1 + ε i,t 13 Beaver and Engel (1996) use a market value of equity as a dependent variable. However, the use of this variable as a dependent variable in the regression is known to be subject to omitted correlated variables bias (Kothari and Zimmerman 1995). Accordingly, Walden (1994) and Ahmed et al. (1999) use a return as a dependent variable. Thus, I also use a return as a dependent variable. 14

30 ,where R t EBTP t LLP t R t is the 12-month buy-and-hold stock return from the end of the 3rd month of the current fiscal year to the end of the 3rd month of the following year less the contemporaneous buy-and-hold stock returns on the CRSP value-weighted market index. income before taxes and provisions scaled by lagged total assets loan loss provision scaled by lagged total assets NLLP t fitted value from a regression model (1) DLLP t residual value from a regression model (1) EBTP LLP t NLLP t DLLP t NPA t Change in EBTP over the year scaled by lagged total assets Change in LLP over the year scaled by lagged total assets Change in non-discretionary LLP over the year scaled by lagged total assets Change in discretionary LLP scaled by lagged total assets change in non-performing asset over the year scaled by lagged total assets The dependent variable (R t ) is the 12-month stock return minus the CRSP valueweighted market index. The independent variables of interest are a level variable (LLP t ) and a change variable ( LLP). The reason for using both level and change variables is that a level and change of accounting information have different implications for the equity market. Specifically, using the level of earnings implies that shocks to earnings are entirely transitory while using the change in earnings implies that shocks to earnings 15

31 are permanent (Ali and Zarowin 1992; Easton and Harris 1991). Similar to this argument, I assume that the level of LLP captures transitory shocks to LLP and the change in LLP captures permanent shocks to LLP. 14 The level of and change in NLLP and DLLP are interpreted in a similar vein. As control variables, earnings (EBTP) are included because my interest is whether LLP provides incremental information over earnings in the equity market. Thus, I include level and change variables of earnings in each specification. Following Beaver and Engel (1996), I also include change in the non-performing asset (NPA) in the model to control for the possibility that banks do not properly use the non-performing asset (NPA) information in setting LLP. Since I use return as a dependent variable and the non-performing asset (NPA) is a variable from the balance sheet, I use a change variable of NPA (Beaver and Engel 1996). Except for stock returns (AR t ), macroeconomic data such as GDP growth rate ( GDP t ) and unemployment rate ( UNEMP t ) and bank size, all continuous variables are scaled by beginning total assets. I estimate all the regressions using two approaches. The first approach I use is pooled regression with time fixed effects where standard errors are clustered by bank (Petersen 2009). The second approach is a Fama-MacBeth regression (1973). Both approaches address cross-sectional dependence in the residuals. Many recent valuerelevance studies use either approach. I focus on the first approach because my sample comprises only 17 (16) years and the small number of years can lead to insufficient test power in Fama-MacBeth regressions. 14 Soliman (2008) also uses a level-change specification to examine the value relevance of asset turn-over and profit margin. His specification is similar to my model (3) 16

32 Table 2 presents the estimation result of equation (1), which is used to decompose LLP into discretionary and non-discretionary components. The results are similar to those of prior studies. Specifically, net charge-off (NCO) has a strongly negative association with LLP, which is consistent with Beaver and Engel (1996). The coefficient on current period change in non-performing assets ( NPA t ) (0.12) is much greater than that on subsequent change in non-performing assets ( NPA t+1 ) (0.02). GDP growth during the period ( GDP t ) is negatively related to LLP indicating that loan loss provisioning is pro-cyclical (Laeven and Majnoni 2003). Adjusted R 2 is assuring that my model explains much of the variation in LLP. IV. Empirical Results IV.1 Value relevance regression Table 3 presents results on a level specification. Panel A provides estimation results based on the pooled regression. Columns (1) and (2) report the regression of stock returns on the level of earnings and LLP. Columns (3) and (4) report the regression of stock returns on the level of earnings and the two LLP components, NLLP and DLLP. Columns (2) and (4) add NPA to columns (1) and (3) thereby controlling for default risk. Columns (1) and (2) show that the earning level is positively associated with stock returns, confirming the positive value-relevance of earnings documented by prior studies. Moreover, the level of LLP is negatively associated with stock returns, indicating that the equity market views LLP negatively as an expense. Columns (3) and (4) show that the level of NLLP is negatively associated with stock returns. It implies 17

33 that the negative value relevance of LLP is driven by NLLP. More importantly, the level of DLLP is not significantly associated or weakly negatively associated with stock returns, contrary to the signaling hypothesis. Control for NPA does not change the results although the magnitude of the coefficients on LLP and NLLP decreases relative to those in models without NPA. In addition, adjusted R 2 are above 45% which implies that the level model has large explanatory power. [TABLE 3 ABOUT HERE] Panel B presents the Fama-MacBeth regression results. The result in Panel B is consistent with one in Panel A although explanatory power in Panel B is much lower than that in Panel A. Specifically, LLP and NLLP are negatively associated with stock returns and DLLP is not positively related to stock returns with or without controlling for NPA. Again, the signaling hypothesis is not supported. Overall, the level specification in Panels A and B confirm that LLP and NLLP are reduces value and DLLP is value-irrelevant. As mentioned above, the level model only captures transitory shocks, but not permanent shocks, to earnings and LLP. Table 4 reports results using the changes specification, which captures the permanent shocks. Similar to Table 3, Panel A provides estimation results based on the pooled regression model. Columns (1) and (2) report the association between stock returns and LLP. Columns (3) and (4) report the association between stock returns and LLP components, NLLP and DLLP. Columns (2) and (4) add NPA to columns (1) and (3). In each column, the coefficients on change in earnings are positive and significant. Columns (1) and (2) show that the coefficients on change in LLP are negative and 18

34 significant. Columns (3) and (4) report that the coefficients on changes in both NLLP and DLLP are negative and significant. Different from the result of a level specification in Table 3 in which increase in DLLP is not significantly associated or is weakly negatively associated with stock returns, increase in DLLP conveys negative news to the equity market. Adjusted R 2 are as high as the pooled level specifications in Table 3, Panel A. [TABLE 4 ABOUT HERE] Panel B reports Fama-MacBeth regression estimation results. The result in Panel B is similar to that in Panel A. In columns (1) and (2), LLP are negatively associated with stock returns. In columns (3) and (4), NLLP are negatively associated with stock returns. Unlike Panel A in which DLLP is negative and significant, DLLP in Panel B is negative but insignificant. However, note that the statistical inference on the coefficients in Table 4 is based on 16 annual regressions coefficients. Thus, I place more weight on the result in Panel A than Panel B. Taken together, the increases in LLP, NLLP, and DLLP relative to the previous year have negative implications to value and the effect of the change in DLLP on the market is smaller than that of NLLP. Thus, the change in DLLP does not play a signaling role to the market. In Table 3 and Table 4, I separately examine the effect of temporary shocks (level specification) and permanent shocks (change specification) to LLP and earnings on the market. Now, I include both change and level variables of LLP as well as earnings in one regression model to capture both transitory and permanent shocks to LLP and earnings at the same time. Table 5 presents results on a level-change 19

35 specification. Panel A provides estimation results based on the pooled regression model. Similar to Table 3 and 4, columns (1) and (2) are results on the association between stock return and LLP and columns (3) and (4) are results on the relation between stock returns and LLP components, NLLP and DLLP. Column (1) shows that both level and change of LLP are significantly negative. And the coefficient of a level of LLP (-8.14) is of greater (absolute) magnitude compared to that of a change in LLP (-5.84). However, in column (2), after controlling for NPA, while a level of LLP is still significantly negative, the change in LLP becomes insignificant, which implies that transitory shocks to LLP have greater valuation implications compared to permanent shocks to LLP. In column (3), both the level of and change in NLLP are significantly negative and the coefficient on the level of NLLP has greater absolute value than that on the change in NLLP, which is similar to the result for LLP. Note that the level of DLLP is positive and significant at the 5% level. It seems that the market perceives the increase in the level of DLLP as weakly positive conditional on the change in DLLP. Another noticeable feature is that the change in DLLP is significantly negative and it has greater absolute value than the level of LLP and the level of and change in NLLP. This implies that the greater change in DLLP is negative news to the market, in contrast to the level of DLLP. In column (4), once controlling for NPA, the coefficients on the change in NLLP and the level of DLLP disappear, contrary to the signaling hypothesis. However, the level of NLLP and change in DLLP are still negatively significant and the coefficient on the change in DLLP is of greater (absolute) magnitude than that on the level of DLLP and NLLP. Thus, while temporary shocks to NLLP are larger than 20

36 permanent shocks to NLLP, permanent shocks to DLLP are larger than its temporary shocks to DLLP. In addition, both the level and change in earnings (EBTP) have positive implications to value, consistent with prior studies. [TABLE 5 ABOUT HERE] Panel B reports Fama-MacBeth regression results for the level-changes specification. Column (1) shows that both the level of and change in LLP are significantly negative but the coefficient of the change of LLP has greater (absolute) magnitude than that on the level of LLP, which is opposite to column (1) in Panel A. Column (2) documents that the negative significance of the level of LLP disappears but the negative significance of the change in LLP is still present. In column (3), the level of NLLP and the change in DLLP are negatively significant. However, the change in NLLP is not negatively significant and the level of DLLP is positive but insignificant. Column (4) shows that after control for NPA, only the level of NLLP is negatively significant. Note that the Fama-MacBeth regression result provides supplemental information to the pooled regression result. Overall, temporary and permanent shocks to LLP and their components have different impacts on stock returns. Thus, the inclusion of both level and change variables of LLP in the same regression model is necessary for the value relevance study of LLP. More importantly, neither temporary nor permanent shocks to DLLP seem to provide positive news to the equity market, contrary to the signaling hypothesis. Additionally, coefficients on both the level of and change in earnings are significantly positive across columns, consistent with Panel A. 21

37 In Table 5, I examine both temporary and permanent shocks to LLP and their components altogether and interpret each shock, respectively. To facilitate the interpretation of the value relevance of LLP and their components (NLLP, DLLP) information, I use the level-lag specification. 15 As mentioned in section 3, this specification is equivalent to the level-changes specification. Table 6 reports results on the level-lag specification. Panel A provides estimation results based on the pooled regression model. Columns (1) and (2) show that the net effect of LLP information (both LLP t and LLP t ) on stock returns is significantly negative. Columns (3) and (4) document that the net effect of NLLP information (both NLLP t and NLLP t ) is significantly negative. Furthermore, the net effects of DLLP information (both DLLP t and DLLP t ) are negative but insignificant. Considering Panel A of Table 5, it seems that the (weakly) positive effect of transitory shocks to DLLP are cancelled out by (strongly) negative effect of permanent shocks to DLLP. Thus, DLLP information is value-irrelevant. In addition, earnings information (both EBTP t and EBTP t ) is positive and significant. [TABLE 6 ABOUT HERE] Panel B provides estimation results based on the Fama-MacBeth regression. Consistent with Panel A, both coefficients on LLP t and NLLP t are negative and significant and the coefficient on DLLP t is positive but insignificant. In sum, LLP and NLLP information have negative implications to value while DLLP information is 15 I use the term LLP information to describe both the level of LLP and change in LLP. This also applies to the terms NLLP information and DLLP information. 22

38 value-irrelevant. Therefore, I find no compelling evidence that the market positively perceives DLLP information. IV.2 The effect of the recent financial crisis on pricing DLLP In this section, I examine whether the recent financial crisis had an impact on the pricing of DLLP. Ryan (2011) conjectures that the signaling role of DLLP may vary with the change in economic conditions. He points out that, incidentally, while prior studies supporting the signaling hypothesis of DLLP are based on samples concentrated around bad economic conditions, a study rejecting the signaling hypothesis is based on an economic boom period. He conjectures that a signaling role of DLLP may revive in the recent financial crisis. My sample period includes bad economic conditions. The period from 2007 to 2010 is viewed as the recent financial crisis and includes a weak recovery period (Ryan 2011). I investigate whether the signaling role of DLLP is valid during this period. To see how the value relevance of LLP and their components changes with time, I use the level-lag specification and regress contemporaneous stock returns on LLP and its components by year. Table 7 shows the estimation result of annual regressions. Panel A (Panel B) is the regression of stock returns on LLP information (NLLP and DLLP information). Figure 1 summarizes both Panel A and B. The coefficient on the levels of LLP and NLLP are mostly negative, which implies that LLP and NLLP have negative implications to value. On the other hand, DLLP has more positive coefficients than LLP and NLLP. However, there are twice as many negative coefficients on DLLP compared 23

39 to positive coefficients on DLLP. Especially, the coefficients on DLLP during are all negative, which indicates that DLLP has negative valuation implications in the financial crisis. This finding supports the second hypothesis and provides evidence against the conjecture that a positive valuation of DLLP may occur in economic busts. [TABLE 7 ABOUT HERE] [FIGURE 1 ABOUT HERE] By using pooled regressions, I examine the effect of the recent financial crisis on value relevance of LLP and its components. 16 To do so, I construct a dummy variable (Crisis) equal to 1 during and 0 otherwise. I run pooled regressions of stock returns on the interaction term of the crisis dummy and LLP and its components in Table 8. Panel A provides estimation results based on the pooled regression of stock returns on current and lagged values of the level of LLP and the interaction term with Crisis. Column (1) shows that the coefficient on the interaction term (EBTP t *Crisis) is not significant, indicating that the crisis does not have an impact on investors valuation of earnings. Column (2) reports that the coefficient on interaction term (LLP t *Crisis) is negatively significant. It indicates that the crisis have negative impacts of valuing LLP. Column (3) includes the two interaction terms (EBTP t *Crisis and LLP t *Crisis) in the same regression, which yield similar results. [TABLE 8 ABOUT HERE] Panel B provides estimation results based on the pooled regression of stock returns on current and lagged values of the level of NLLP and DLLP, and the interaction 16 The pooled regression does not include year dummy variables because the use of Crisis dummy with year dummy variables leads to multi-collinearity. 24