Earnings and Cash Flows in Debt Evaluation by Private Debt Holders. N. Bugra Ozel

Transcription

1 Earnings and Cash Flows in Debt Evaluation by Private Debt Holders N. Bugra Ozel Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy under the Executive Committee of the Graduate School of Arts and Sciences COLUMBIA UNIVERSITY 2010

2 UMI Number: All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion. Dissertation Publishing UMI Copyright 2011 by ProQuest LLC. All rights reserved. This edition of the work is protected against unauthorized copying under Title 17, United States Code. ProQuest LLC 789 East Eisenhower Parkway P.O. Box 1346 Ann Arbor, Ml

3 2010 N. Bugra Ozel All Rights Reserved

4 ABSTRACT Earnings and Cash Flows in Debt Evaluation by Private Debt Holders N. Bugra Ozel This study investigates whether private debt holders focus more on earnings or cash flows of their borrowers in debt evaluation. I utilize estimates of credit losses and realizable value of loans as reported by commercial banks in regulatory filings to explore how private debt holders react to information about borrowers' operating performance. I find that changes in estimates of credit losses are significantly associated with measures of borrowers' current and future operating performance, especially with operating income growth. I also find that when assessing borrowers' performance, private debt holders consider some accruals (e.g., working capital accruals) more informative than others (e.g., depreciation). Furthermore, I show that the estimates of credit losses provide incremental information about borrowers' future short term profitability growth over several measures of equity investors' expectations (valuation ratios, stock returns, past growth rates, and analyst forecasts). While this is consistent with the widely held belief that private debt holders have superior information about their borrowers, it also suggests that other investors may be ignoring some useful information provided by private debt holders.

5 TABLE OF CONTENTS TABLE OF CONTENTS LIST OF FIGURES AND TABLES ACKNOWLEDGEMENTS DEDICATION i iii iv v 1. INTRODUCTION 1 2. RELATED STUDIES AND BACKGROUND Review of Literature Regulations on Loan Loss Reserves Empirical Predictions DATA AND SUMMARY STATISTICS Sample Formation and Variable Definitions Summary Statistics EMPIRICAL FINDINGS Earnings vs. Cash Flows Components of Loan Loss Reserves High Credit Risk Firms vs. Low Credit Risk Firms Effect of Financial Condition of Banks Predictive Ability of Loan Loss Reserves Robustness Checks Alternative Methods for Calculations Alternative Performance Measures Macroeconomic Variables 41 i

6 4.7. Discussion of Findings CONCLUSION 46 REFERENCES 49 FIGURES 55 TABLES 61 ii

7 LIST OF FIGURES AND TABLES FIGURES: Figure 1: Operating Performance and ALLR over Time 56 Figure 2: Operating Performance and ALLR 57 Figure 3: Distribution of Leverage for Bank and Finance Company Borrowers 58 Figure 4: The Effect of Figure 5: Univariate Regressions of Future Operating Performance on E/P Ratio and ALLR 60 TABLES: Table 1: Summary Statistics 62 Table 2: Correlation Matrix 64 Table 3: Regressions of ALLR on Earnings and Cash Flow Growth Variables 65 Table 4: Components of ALLR 67 Table 5: Firm Portfolios Based on Leverage 68 Table 6: Effect of Financial Condition of Lender 69 Table 7: Future Earnings Growth, Proxies of Public Information and ALLR 70 Table 8: Regressions of Change in Nonperforming Commercial Loans on Earnings and Cash Flow Growth Variables 72 Table 9: Regressions of ALLR on Earnings/Cash Flow Growth Variables and Macroeconomic Variables 73 iii

8 ACKNOWLEDGEMENTS I am indebted to Divya Anantharaman, Bjorn Jorgensen, Nahum Melumad, Doron Nissim, Stephen Penman and Gil Sadka for their valuable comments and support. I also thank Sudhakar Balachandran, Shira Cohen, Emre Karaoglu, Jon Kerr, Urooj Khan, Hanna Lee, Part ha Mohanram, Julian Yeo and Yuan Zhang as well as workshop participants at Baruch College, Carnegie Mellon University, Columbia University, INSEAD, London Business School, New York University, The U.S. Securities and Exchange Commission, and University of California at Los Angeles for their helpful comments and suggestions. All errors are my own. iv

9 DEDICATION I dedicate this dissertation to my family. Particularly, to my parents, Ahmet and Suzan Ozel for the encouragement and support they give all the time and to my brother and my best friend Cagri, whose guidance and advice has always been invaluable to me. I must also thank my uncle and my grandmother, who have never left my side. I also dedicate this dissertation to my friends. Especially to Benan Zeki Orbay,Martin Dumav and Suat Teker for encouraging me to start the doctoral program. A special feeling of gratitude to Doron Nissim who has been very influential in the development of my research abilities and very supportive of my work. Finally, I would like to give my heartfelt thanks to Bjorn Jorgensen, Nahum Melumad and Gil Sadka who have generously spared their time to help me whenever I needed it. v

10 1 EARNINGS AND CASH FLOWS IN DEBT EVALUATION BY PRIVATE DEBT HOLDERS 1. INTRODUCTION Accounting researchers have extensively studied how equity holders use earnings and cash flow information in evaluating the future prospects of firms. Complementing these studies, a growing body of literature focuses on the use of earnings and cash flows in debt markets. 1 This literature provides evidence that secondary market participants react to accounting-based performance measures and this reaction is different from that of equity holders (e.g., Easton, Monahan, and Vasvari 2009). Much less work, however, has been done to understand how private debt holders use earnings and cash flow information of borrowers in debt evaluation. In fact, private debt holders have an instrumental role in corporate finance, and differ from other investors in several dimensions. In this study, I investigate whether private debt holders focus more on earnings or cash flows of the borrowers when estimating potential credit losses. Merton (1974) asserts that the value of corporate debt instruments depends on the risk free rate, the probability of default, and various debt provisions and restrictions (e.g., maturity date, coupon rate, and call terms). Of those pricing components, the probability of default involves the most significant estimation, based primarily on the issuer's financial reports. Therefore, like stockholders, debt holders monitor the performance of the firms they invest in, and reassess their investments in the presence of new financial information. However, the risk-return trade-off for debt holders is different from that for 1 Recently, several studies focused on the use of accounting information in debt markets and produced interesting findings on different fronts (e.g., Dichev and Skinner 2002, Khurana and Raman 2003, Asquith, Beatty and Weber 2005, Ball, Bushman and Vasvari 2008, DeFond and Zhang 2008, Wittenberg-Moerman 2008, Bushman, Smith and Wittenberg-Moerman 2009). Still, the number of studies in this area is relatively small.

11 2 stockholders. In addition, compared to equity investors, short-term creditors are less concerned about the long-term prospects of the firms. These distinctions suggest that the same information may have different implications for debt and equity instruments. Prior research on equity investors has documented that equity values react to information about firm's performance (e.g., Ball and Brown 1968) and that cash flows are more value relevant than accruals. Still, compared to cash flows, earnings have a stronger association with stock price (e.g., Dechow 1994). Studies have also found that cash flows are more persistent than accruals (e.g., Sloan 1996), but current earnings can predict future cash flows better than current cash flows does (e.g., Dechow, Kothari and Watts 1998). Since debt represents claims on the same assets and profits as equity, properties and differential implications of cash flows and accruals should also be relevant for debt holders. Prior studies on the use of accounting information in debt evaluation have focused primarily on bondholders, perhaps because corporate bond prices inform on the probability of default (e.g., Bhojraj and Swaminathan 2009, Easton, Monahan and Vasvari 2009). Nevertheless, it is also intriguing to study the use of accounting measures by private debt holders for at least three reasons. First, private debt is the predominant source of debt financing. For example, the total amount of syndicated loans issued in 2007 was approximately $1,680 billion, compared to $1,127 billion of corporate bond issues. 2 Second, private debt holders are presumed to have access to inside information (e.g., Fama 1985), and provide more efficient monitoring than equity holders and public debt holders (e.g., Diamond 1984, 1991). Private debt holders' reaction to financial 2 Loan Pricing Corporation and SIFMA.

12 3 information may therefore differ from that of the other investors. For instance, if private debt holders obtain more detailed and timely information regarding borrowers' operating performance, or if they evaluate borrowers' operating performance more accurately through the efficient processing of available information, they may anticipate impairment charges that surprise other investors. Third, private debt is less dependent on borrower characteristics, such as size or reputation, than is public debt (e.g., James 1987, Diamond 1991, Carey, Post, and Sharpe 1998). These same characteristics are also correlated with the relative magnitude of accruals and cash flows, which implies that these items may have different implications for private and public debt. Small companies may be more dependent on operating cash flows, so that excessive working capital accruals for small firms may be more suspicious. Empirically, extreme accruals are more common among young and small firms (e.g., Lev and Nissim 2006, Mashruwala, Rajgopal, and Shevlin 2006). 3 For public debt, price is a timely measure of the investors' perception of the firm's performance. In contrast, for private debt no such metric is readily available. Still, it is possible to measure private debt holders' reaction to borrowers' cash flow and accrual information using debt holders' financial disclosures. Specifically, regulators require most private lenders to disclose comprehensive information about lending activities, including direct and indirect measures of the changes in the riskiness and value of debt. The most notable set of such measures is the loan quality information 4 provided 3 In addition to these differences, unlike public debt, private debt is not subject to the filing requirements under sections 13 and 15(d) of the Securities Exchange Act of 1934, and is often more senior to public debt in the event of default. 4 Examples of loan quality measures include - but are not limited to - loan loss reserves (estimates of uncollectible amounts of loans), delinquencies (the loans where borrower is

13 by commercial banks, the largest single group of commercial and industrial debt 4 financers in the economy. 5 I employ the disclosures on loan loss reserves a key loan quality measure that represents the bank's best estimate of the losses inherent in its loan portfolio and total loans to study the reaction of private lenders to new financial information of borrowers. I use economy-wide data to conduct the tests. The setting can be viewed as an economy composed of a representative private lender (i.e., commercial bank) and a representative publicly traded firm. The firm borrows from the lender and the lender monitors the firm's financial condition in order to determine whether the firm will be able to repay the loan. In its financial statements, the lender presents the total amount of loan given to the firm and the amount reserved for the uncollectible portion of the loan. These two items reveal the lender's evaluation of the ability of a firm to service its debt at a given point in time. Within this framework, I examine the association between the lender's evaluation and the firms' current and future operating performance. While an economy-wide setting provides several advantages 6, it is prudent to acknowledge that findings based on the aggregates may not apply to each individual bank within the economy and that a firm-level study could provide valuable insights, as well. However, commercial banks are not required to present a firm-level breakdown of loans or loan behind/late in payment), and charge offs (value of loans that are confirmed uncollectible and written off from the books). 5 Carey, Post, and Sharpe (1998) report that banks' share in total C&l loans stayed above 85% historically. 6 Such as the inclusion of public and private banks/corporations, averaging out the effects of bank/firm specific activities, and identifying the results for a representative bank

14 quality measures and, to the best of my knowledge, no other source provides detailed information on credit evaluation at the individual firm level. 5 I find that private debt holders' estimates of credit losses are significantly negatively correlated with borrowers' operating performance. Moreover, when evaluating credit losses, private debt holders focus primarily on borrowers' operating earnings, not on operating cash flows. In particular, both the current and future growth rates in operating earnings are more strongly correlated with changes in loan loss reserves than are cash flow growth rates. This supports the argument that debt holders consider accruals informative about the ability of borrowers to service their debt. The result is especially significant because the loan losses are recognized when the lender has doubts about the viability of the borrower, and going concern is an underlying assumption of the accrual concept. I also find that private debt holders do not weigh all accruals equally. Among the accounting performance measures tested, growth in operating earnings before depreciation and amortization has the strongest association with changes in loan loss reserves. This is consistent with private debt holders regarding certain accruals (e.g., working capital) to be more instructive about the borrowers' repayment ability than other accruals (e.g., depreciation). When the sample is split into groups based on leverage, an almost monotonic relation prevails between leverage and the relevance of operating earnings growth for changes in loan loss reserves. In other words, the earnings growth of high leverage firms is more strongly correlated with changes in loan loss reserves than is the earnings growth of low leverage firms. While this result is unsurprising per se, the disturbance of this

15 6 monotonic relation in the highest leverage group supplements prior research on borrower characteristics. In particular, it is consistent with the evidence in the finance literature (e.g., Carey, Post, and Sharpe 1998, Denis and Mihov 2003) that firms in the highest risk categories tend to borrow from non-bank lenders. Of the two components of credit losses, confirmed losses (i.e. net charge-offs) are mainly associated with the current operating performance, especially cash flows, of borrowers. On the other hand, unconfirmed losses (i.e. provisions for loan losses after controlling for net charge-offs), are associated with both current and future operating income of borrowers. This suggests that lenders rely on information about current and expected future earnings for unconfirmed losses while they focus mainly on borrowers' current cash flows for confirmed losses. Finally, I show that changes in loan loss reserves are informative about the future operating performance of the borrowers even after controlling for measures of public information including typical valuation ratios, stock returns, analyst forecasts and past operating performance. The association weakens as future earnings growth is measured over longer periods. Specifically, in horizons exceeding one year, loan loss reserves cannot predict future earnings growth whereas the predictive power of earnings to price (E/P) ratio increases. 7 The findings suggest that private debt holders provide some valuable information about short term future earnings of borrowers. However, given that loan quality measures are publicly disclosed, it is unclear why this information is not incorporated into the measures of public information. 7 This is intuitive given that regulatory guidance on loan loss reserves suggests that these reserves should be adequate to cover one year's losses.

16 7 This study contributes to the literature in several ways. First, using a unique setting, it explores the use of earnings and cash flow information by private debt holders in debt evaluation. Accordingly, it complements the studies that examine the use of these accounting performance measures by equity holders (e.g., Dechow 1994), and by public debt holders (e.g., Easton, Monahan and Vasvari 2009). Second, this study provides evidence relating to the debate over the accounting performance measure that debt holders care more about in debt evaluation. 8 Some academics (e.g., Bhojraj and Swaminathan 2009) implicitly assume that debt holders are more interested in cash flows than earnings because a firm's ability to repay debt depends on the amount of cash it generates. Some professionals share the same perspective. For example Standard and Poor's 2008 Corporate Ratings Criteria notes: "cash flow adequacy is typically the single most critical aspect of credit rating analysis", 9 On the other hand, several academics (e.g., Watts 2003, Ball, Robin, and Sadka 2008) argue that timely measures are important for debt holders and that debt holders should focus on earnings as it is timelier than cash flows. I find evidence consistent this latter view. Moreover, my findings support the claims that certain accruals are more valuable than others in debt evaluation. While this may provide a justification for the use of modified GAAP measures for earnings and cash flows in studies relating to debt holders (e.g., James 1995, Carey, Post, and Sharpe 1998, Beaver, McNichols, and Rhie 2005), to the 8 It is important to emphasize that the arguments in the literature, as well as in the current study, relate to the relative usefulness of earnings and cash flows. In other words, the discussion is not about whether debt holders use only earnings or only cash flows but is about which of these measures explain debt holders' reactions better. Clearly, debt holders do not need to choose a single measure and would optimally utilize both earnings and cash flows. 9 For another example see the Fitch Research's special report named "Cash is King in Bond Analysis".

17 extent that those modified measures include (exclude) accruals that debt holders use, such studies may over (under) state the role of actual earnings or cash flows. 8 Finally, this study adds to the evidence that private debt holders possess superior information about the borrowers. Theoretical studies in finance often argue that due to reasons such as relationship banking, better processing of information and economies of scale, compared to public investors, private debt holders have better information about future prospects of the borrowers (e.g., Sharpe 1990, Diamond 1991). The existing empirical studies provide indirect support to these claims (e.g., James 1987, Bharath, Sunder and Sunder 2008). While the present paper corroborates the findings of prior studies using a more direct test, it also demonstrates that public investors may be neglecting some useful information that private debt holders disclose. The remainder of this paper is organized as follows. Section 2 provides background information and states the empirical predictions. Section 3 describes the data and provides summary statistics. Section 4 presents the empirical findings. Section 5 summarizes and concludes the paper. 2. RELATED STUDIES AND BACKGROUND This section has three subsections. Section 2.1 summarizes relevant literature. Section 2.2 provides an overview of regulations on loan loss reserving. The empirical predictions of the study are explained in section 2.3.

18 Review of Literature In spite of their instrumental role in corporate finance, private debt holders have received limited attention in accounting literature. 10 The extant empirical research primarily focuses on the use of accounting information in debt contracting (see Armstrong, Guay and Weber (2009) for a detailed review of contracting literature). 11 In one of the earliest works in this field, Leftwich (1983) finds that the accounting information used in private debt contracts is systematically different from GAAP measures. Later studies document that conservative, persistent and relatively nondiscretionary accounting-based measures are important in debt contracting (e.g., Beatty, Weber and Yu 2008, Zhang 2008, Li 2008). They also document that certain characteristics of borrowers (e.g., accounting quality) and the form in which accounting information is included in the contract are important determinants of contracting costs (e.g., Asquith, Beatty, and Weber 2005, Bharath, Sunder and Sunder 2008). Studies on debt contracting provide somewhat different findings on the use of earnings and cash flows in contract terms. Dichev and Skinner (2002) show that the most widely used covenant in debt contracts is debt to cash flow ratio. However, Dichev and Skinner underline that this ratio does not have a standard definition and metrics used in the numerator and denominator vary among contracts. Li (2008) also demonstrates that debt to cash flow covenant is included in the majority of debt contracts, and argues that 10 The number of studies in this field is on the rise since the introduction of Dealscan database. Dichev and Skinner (2002) were among the first to investigate debt contracts using this new data. 11 Some studies also examine the monitoring role of private debt holders (e.g., Beatty, Liao and Weber 2007), and the factors that determine private debt holders' loan related decisions such as loan granting decisions (Libby 1979), and waiver decisions (Chen and Wei 1993).

19 banks view debt to cash flow ratio as a particularly valuable signal of credit quality. On the other hand, Asquith, Beatty and Weber (2005) report that performance pricing provisions often include earnings-based measures. 12 Prior research has typically considered debt provisions and restrictions as tools for reducing agency problems and renegotiation costs rather than as direct measures of risk or value. For example Gorton and Kahn (2000) provide a theoretical model to demonstrate that initial terms of loans are not set to price credit risk, but rather are set to balance bargaining power in later renegotiations. 13 Assessment of risk and value of the lending takes place in periodic evaluations, which are internal events to the lender. In fact, periodic evaluations are aimed at identifying the borrowers for whom lender's credit-granting decision may prove erroneous. Accordingly, measures used in contracting and evaluation processes may differ. For example, prior studies have found that debt holders often exclude temporary and/or highly discretionary items from debt covenants. Nevertheless, items such as unusual gains/losses are not necessarily irrelevant when evaluating borrowers' repayment ability since those can alter the repayment potential of the debt. The present study's primary focus is to identify how private lenders measure the performance of the borrower when assessing value and risk of their lending. To 12 Beatty, Dichev and Weber (2002) provide findings consistent with performance pricing complementing debt covenants. On the other hand Zhang (2008) states that it remains unclear whether performance pricing substitutes or complements debt covenants. 13 Waivers of debt covenants can be interpreted as empirical evidences in support of this view. If covenants are used as direct tools for debt evaluation, then debt holders would be reluctant to waive the violations. In contrast, Dichev and Skinner (2002) report that 30% of the firms in their sample have covenant violations and that these violations are frequently waived as most of those firms are not in financial distress. They conclude that debt covenants are used as screening devices. Moreover, Beneish and Press (1995) provide some evidence that firms may default on debt without violating covenants beforehand, even when covenants are not "too slack".

20 11 effectively address this issue, I test the association of both reported and modified earnings/cash flow measures with private debt holders' evaluation of potential credit losses. The study of use of earnings and cash flows in credit loss evaluation process relates to diverse strands of literature. One line of research investigates the role of financial information in failure prediction. Beaver (1966), the study which pioneered modern studies in this area, finds that cash flow to debt ratio is an early indicator of firm failure. 14 Altaian (1968) is first to introduce multiple discriminant analysis method into this field and he finds that Beaver's results can be improved by using a larger set of financial information. Altman's study includes earnings measures but not cash flow measures due to lack of consistent data. Later studies vary in their use of earnings and cash flow information (e.g., Deakin 1972, Norton and Smith 1979, Altman 1977, Ohlson 1980). 15 The wide variety of ratios used in this literature led to the question of whether earnings or cash flows is more relevant in failure prediction. Largay and Stickney (1980) examine the bankruptcy of W.T. Grant Company and demonstrate that cash flow from operations provided a stronger and earlier signal of the bankruptcy than earnings. Casey and Bartzcak (1984) provide a careful examination of the usefulness of cash flow measures and compare them with the predictive ability of accrual based measures. 14 Due to unavailability of cash flow data, early studies often use fund flows, ebida, ebitda and similar measures as proxies for cash flows. 15 Related to failure prediction research, prior studies investigate the types of information credit rating agencies rely on to determine the corporate bond ratings. Similar to research on failure prediction, these studies also differ in their use of earnings and cash flows (e.g., Ang and Patel 1975, Kaplan and Urwitz 1979, Gentry, Newbold and Whitford 1988).

21 Although the results indicate that cash flow from operations predict bankrupt companies better than accrual based model, based on overall (bankrupt and non-bankrupt firms) predictive accuracy, authors conclude that cash flow measures are not useful in failure prediction. 16 Several other researchers studied the question and reached different conclusions about the usefulness of earnings and cash flows in failure prediction (e.g., Gombola et al. 1987, Aziz and Lawson 1989, Mossman et al. 1998). More recently, Jones and Hensher (2004) investigate a sample of Australian firms and show that cash flow measures are significant in predicting financial distress. Extant studies in bankruptcy prediction literature explore the relation between the default risk and different measures, including accounting variables. However, a better understanding of how debt holders use those accounting variables in evaluating borrowers' default risk may provide new directions for this literature. The present paper's contribution to the literature is twofold. First, this study takes a positive, rather than a normative, perspective and analyzes private debt holders' approach rather than the optimal approach in determining credit losses. Second, it provides a convenient framework for studying private debt holders' reaction to new financial information. The current study also relates to the literature that focuses on equity holders' and public debt holders' use of earnings and cash flow information. Following the seminal 16 Casey and Bartzcak (1985) reiterate this conclusion. While consistent with Casey and Bartzcak (1985), Gentry, Newbold and Whitford (1985a, b) also show that three cash flow components (dividends, investments and receivables) provide additional predictive power. White, Sondhi and Fried (2002) discuss two weaknesses of these studies. First, due to costs of bankruptcy, correct prediction of bankrupt firms is more important than correct prediction of nonbankrupt firms. Second, cash dividends, which appear to be a significant predictor of bankruptcy in Gentry, Newbold and Whitford (1985 a) may be capturing the cash generation ability of the firms. It is also important to note that several other researchers (e.g., Gombola and Ketz 1983, Zmijewski 1984, Shumway 2001, Campbell, Hilscher and Szilagyi 2008) criticize the bankruptcy prediction studies for problems in methodologies, models, and sample constructions.

22 13 paper of Ball and Brown (1968), there has been intensive research on equity holders' use of accounting information. Several studies examine the relative usefulness of earnings and cash flows in equity valuation (e.g., Rayburn 1986, Wilson 1987, Livnat and Zarowin 1990, Ohlson 1990, Dechow 1994, Sloan 1996, Lev and Nissim 2004, Penman and Yehuda 2009). This literature documents that accruals are less value relevant than cash flows, but earnings explain stock returns better than cash flows. Additionally, findings in Dechow, Kothari and Watts (1998) and Barth, Cram and Nelson (2001) indicate that current earnings predict future cash flows better than do current cash flows. The number of studies that investigate bondholders' use of earnings and cash flows is relatively small. Early papers in finance literature (e.g., Fisher 1959) document the relationship between earnings variability and risk premium on bonds. Later studies provide models for pricing bonds (e.g., Brigham 1966, Van Home 1974) which essentially relate bond values to accounting earnings. More recent researches, in both accounting and finance, focus on detailed analysis of bondholders' reaction to earnings (e.g., Davis, Boatsman and Baskin 1978, Datta and Dhillon 1993, Plummer and Tse 1999, Khurana and Raman 2003). These studies generally find that bond prices react to earnings related information and as the financial condition of a firm deteriorates, earnings becomes more relevant for bondholders. To date, the only study that provides a large sample comparison between the use of earnings and cash flows by public debt holders is Easton, Monahan and Vasvari (2009). Their study finds that both accruals and cash flow from operations are correlated with bond returns, especially when firm reports a loss. An emerging line of research also focuses on the secondary syndicated loan market (e.g., Wittenberg-Moerman 2008, 2009, Bushman, Smith, Wittenberg-Moerman

23 2009). The loans traded in this market have similarities to both bonds and privately held 14 debt. 17 There are very few studies that examine the use of operating performance measures in secondary loan market. In a recent study, Allen, Guo and Weintrop (2009) find loan prices react to earnings, and this reaction takes place around the time syndicate members receive monthly covenant reports. Unlike studies on equity holders and secondary debt market participants, research on private debt holders is yet to examine how this group of stakeholders uses earnings and cash flow information in evaluation of their investments. Nonetheless the present study and more generally studies on private debt holders' use of accounting information, can contribute to literature in at least two respects. First, further research can explain how demands for and uses of accounting information differ between various groups of stakeholders. Second, such studies can aid in building a more complete picture of the direct and indirect effects different stakeholders have on each other. A third strand of relevant research is the theoretical studies in finance literature that model banks as possessors of superior information about borrowers. Those studies often argue that due to reasons such as relationship banking, better processing of information and economies of scale, banks have better information about future prospects of the borrowers (e.g., Sharpe 1990, Diamond 1991). Some empirical evidence is consistent with this assumption. For example, James (1987) finds that the stock market responds more positively to announcement of bank loans compared to announcement of other sources of debt. He asserts that banks provide a special service, such as better 17 Secondary trading in loan market is relatively thin but it has been growing fast since 1990s. According to Gadanecz (2004), in 2003 the trading volume in secondary market was equal to nine percent of outstanding syndicated loans. Traded loans differ from non-traded loans in various dimensions, such as seniority, security and maturity (see Wittenberg-Moerman 2008).

24 15 monitoring, that is not provided by other lenders. Similarly, Bharath, Sunder and Sunder (2008) find that firms with poor accounting quality prefer bank loans probably because banks reduce adverse selection costs for borrowers. The current paper adds to the growing body of evidence on informational superiority of banks. It offers a direct test to compare expectations of public investors and private debt holders with regard to future profitability. Similar to previous empirical studies however, it does not distinguish between the sources of informational differences. Finally, the paper provides a new dimension to the research conducted at the economy-wide level. Recently several studies (e.g., Kothari, Lewellen, and Warner 2006, Sadka 2007, Anilowski, Feng and Skinner 2007, Hirshleifer, Hou, and Teoh 2007, Ball, Sadka and Sadka 2009, Jorgensen, Li and Sadka 2008, Sadka and Sadka 2009) investigate various aspects of economy-wide level earnings-stock price relation and provide interesting results. The present study extends the focus of this literature to private debt holders Regulations on Loan Loss Reserves In the US, commercial banks are regulated by several banking regulators including The Office of Comptroller of the Currency (OCC), The Federal Reserve, and The Federal Deposit Insurance Corporation (FDIC) in addition to the FASB, IRS and for publicly traded banks the SEC. The estimation of loan quality measures have long been a concern for these institutions. Over the last three decades the regulators have spent considerable effort to have banks report as accurate information about loan quality as

25 16 possible. Regulations often place a special emphasis on loan loss reserves (a.k.a. allowance for loan losses) because in contrast to many other measures, estimation of this measure includes a significant amount of forward looking information and therefore, is subjective. 18 Moreover regulators use this metric as an input for measuring the financial health of the banks. Loan loss reserves is a contra asset account on a bank's balance sheet. It is established and maintained by charges against bank's operating income. Banks often make additions to this reserve when (i) it becomes likely that a loan or group of loans will be in part or wholly uncollectible, (ii) an unanticipated charge off occurs for which the bank did not set aside reserves or (iii) the amount of loans increase. When a loss is confirmed (e.g., borrower goes bankrupt), the amount is written off from total loans and from loan loss reserves. Total book value of a bank's loans less the reserves for loan losses demonstrate the bank's best estimate of the net realizable value of the loan portfolio as of the financial statement date. Few banks provided allowance for loan losses prior to 1947; the year the Commissioner of Internal Revenue differentiated treatment of loan loss reserves from bad debt reserves of firms in other industries. 19 Tax considerations had been the major determinant of loan loss reserves until to mid-1970s. In 1975, FASB issued FAS 5: "Accounting for Contingencies", which became the highest hierarchical element in 18 One main concern of FASB is to measure net income for the period. Therefore in FASB's view loan loss reserves are not aimed at identifying expected future losses, but rather at identifying potential losses that will occur due to the events that took place in the current period. In contrast, economists, such as FRB officials, argue that expected loss approach is more meaningful. 19 In 1948 only 38% of the banks were holding loan loss reserves. This figure reached 61% in 1963 and 94% by 1975 (Walter 1991).

26 17 GAAP relating to loan loss reserves. FAS 5 stated that loss should be recognized when probable and can be reasonably estimated at the date of financial statement. This core definition of loss recognition has not changed since then. Until early 1980s both due to taxation issues and regulator influence, reserving was often based on arbitrary percentages of total loans or prior years' actual loan losses. The increase in bank failures during 1980's led regulators to encourage bank managers to use their own judgment in determining the adequacy of loan loss reserves. Accordingly, measures are taken to curb the opportunistic use of this discretion. For example in 1986 the Securities and Exchange Commission issued FRR 28 "Loan Loss Allowance Methodology and Documentation Issues". This statement had the authority of FAS 5 for the publicly traded banks and was consistent with FAS 5. It underlined that all registrants are required to provide adequate documentation of procedures followed in reviewing loan portfolios and in determining amounts of allowances for loan losses. Also in 1986, the IRS issued the Tax Reform Act which eliminated the rule that had allowed tax deductions on additions to loan loss reserves in excess of actual charge offs. In 1989, bank capital adequacy regulations were substantially changed and loan loss reserves were removed from Tier I (Primary) Capital calculations. Also as a part of total capital, loan loss reserves were no longer allowed to account for more than 1.25% of risk weighted assets. While the regulation did not affect the measurement of potential loan losses, it affected the incentives of banks. For example Ahmed, Takeda and Thomas (1999) document that prior to 1990, banks were using loan loss reserves for capital management purposes; however, after 1990, the strength of this practice diminished significantly.

27 18 In 1994, FASB issued FAS 114: "Accounting by Creditors for Impairment of a Loan". This standard did not change the fundamental recognition criteria provided by FAS 5, but it required certain methods of measurement for the loans that are individually considered for impairment. 20 In particular, it provided more specific guidance on measurement and disclosure for (i) loans that are identified for evaluation and individually deemed to be impaired and (ii) loans that are restructured in a troubled debt restructuring involving modification of the terms. Today, banks are still required to estimate their loan loss reserves in accordance with FAS 5 and FAS 114. To ensure the appropriate application of these standards and to assist banks on important issues about loan loss reserving, regulators have issued a number of guiding documents since Some noteworthy examples include the 1999 Joint Interagency Letter to Financial Institutions, EITF D-80, 2001 Policy Statement on ALLL Methodologies, and 2006 Interagency Policy Statements on the ALLL. In spite of several regulations and guidance regarding loan loss reserves, its estimation still involves substantial judgment. 21 The estimation process starts with categorizing loans based on their segments and risk levels. For loans that are suspect to losses, banks follow certain methods depending on properties of the loan. Generally, for loans evaluated under FAS 5 regulators encourage banks to start from historical loss rates and make adjustments for external factors (such as credit quality, current trends and changes in industrial, geographical and political environment) and for internal factors 20 FAS 118 amended FAS 114 in It is primarily about recognition of interest income on impaired loans. 21 In her testimony before the US House of Representatives on July 16, 1999 Donna Tanoue, who was chairman of FDIC then, mentioned : "Establishing appropriate levels for loan loss reserves is an art, not a science"

28 (such as changes in experience and ability of lending management staff, current nature and volume of loan portfolio, changes in quality of loan review system). 22 For determining loan loss reserves for loans evaluated under FAS 114, the standard requires the use of one of three valuation methods: (i) present value of expected future cash flows, (ii) fair value of collateral or (iii) observable market price of loan. Regulations are silent about defining a fixed period of losses that the loan loss reserves should cover. However, since losses should be incurred at the time of recognition, it is likely that reserves are mainly created for the losses that will be confirmed in the short or medium horizon. Borrowers' current and expected future operating performances are considered important signals in determining loan quality. 23 OCC Comptroller's Handbook mentions that loan review systems must not only respond to obvious problems such as delinquency, but also recognize more subtle warnings such as deterioration in a borrower's financial statements. It cites reduced profitability and future prospects of borrowers among the reasons for recognizing loan losses. In SAB No. 102, SEC also suggests that expected future profitability and cash flows of borrowers should be taken into consideration when measuring loan quality. Disclosures by commercial banks are aligned with these requirements. For example in their financial reports many banks cite financial Regulators stress the importance of providing detailed documentation of all these processes. Federal Reserve BHC Supervision Manual notes: "Examples of underlying supporting evidence could include but not limited to relevant articles from newspapers, other publications that describe economic events affecting particular geographic area, economic reports and data and notes from discussions with borrowers." 23 While regulators require loan losses to be estimated based on current events, not expected future events, they imply that the expected impact of current events on future performance must be incorporated into the estimates.

29 condition/performance of borrowers among the important factors that are used in determining loan loss reserves Empirical Predictions Credit losses occur when a borrower no longer has the ability to repay a loan. Due to accounting conservatism, private debt holders recognize estimates of credit losses before the loss is confirmed. The estimates depend on debt holder's assessment of borrower's debt repayment ability, which in turn depends on borrower's operating performance. Operating performance measures have different implications about borrower's debt repayment ability. Cash flows are necessary to service debt. However, earnings are timelier than cash flows. Anecdotal evidence suggests that private debt holders are more interested in cash flow information than equity holders do. For example SFAS 95, "Statement of Cash Flows", states that the majority of respondents to the exposure draft were commercial lenders and that those respondents claimed: "... creditors are more exposed to fluctuations in net cash flow from operating activities than to fluctuations in net income...". On the other hand, some accounting practitioners, academics and regulators assert that earnings is a more meaningful figure for debt holders. For instance FASB's Concept Statement 1 notes: "/Investors, creditors and others] 's interest in an enterprise's future cash flows... leads primarily to an interest in information about its earnings [comprehensive income] rather than information directly about its cash flows".

30 21 Easton, Monahan and Vasvari (2009) demonstrate that bond holders react to both accruals and cash flows. However, as discussed in previous sections of this paper, there are several reasons why reaction of private debt holders may differ from that of secondary market participants. Moreover, in the setting of this study, accruals may not be relevant since recognition of loan losses essentially indicates that the probability of bankruptcy for the borrower is increasing and accruals are based on going concern assumption. In light of arguments supporting both sides, which performance measure (earnings or cash flows) should receive more weight in debt evaluation is an empirical question. Therefore, while I expect estimates of potential loan losses to have a negative correlation with borrowers' operating performance, I do not make a prediction about the relative strengths of the association with earnings and cash flows. GAAP performance measures, like earnings and cash flow from operations, are comprehensive measures and are aimed to satisfy information needs of various stakeholders. Each component of these measures has different implications about firms' conditions and debt holders may consider certain items more relevant than others when evaluating borrowers' debt service ability. Some borrowers fall in this view and provide customized measures in their financial statements. For example Neiman Marcus's 10-Q for the third quarter of 2008 reads: "We present the non-gaap financial measures EBITDA and Adjusted EBITDA because we...believe the presentation of these measures will enhance investors' ability to... evaluate our ability to service our debt". Another example is available in Radio Shack Corp.'s 10-K for 2007: "We believe free cash flow is a relevant indicator of our ability to repay maturing debt, change dividend payments or

31 fund other uses of capital that management believes will enhance shareholder value" 24 Moreover, several studies relating to debt holders (e.g., James 1995, Carey, Post and Sharpe 1998, Beaver, McNichols and Rhie 2005) implicitly assume that some components of performance measures are less relevant for debt holders. Accordingly those studies use modified GAAP measures rather than reported numbers for earnings or for cash flow from operations. While prior studies show that private debt holders modify GAAP numbers for contracting purposes (e.g., Lefhvich 1983, El-Gazzar and Pastena 1990, Li 2008), it is unknown whether such modifications take place in estimation of credit losses. These arguments provide the basis for the second prediction of this study. I predict that when determining potential credit losses, private debt holders do not value every component of the GAAP performance measures equally. The third prediction of this study is based on two observations. First, estimates of credit losses incorporate forward looking information about borrowers' profitability. Thus, these estimates should have predictive power on future performance. Second, a widely held belief holds that private debt holders possess superior information about the future prospects of the borrowers. In particular, several studies in finance and accounting argue that private lenders have informational advantage over other investors (e.g., Fama 1985, James 1987, Diamond 1991). However, the proponents of efficient markets would probably disagree. For example, Black (1975) states that: "[In an efficient market] if the bank's loan officers have the ability to obtain and identify information on companies that is not already identified, they would be better off as security analysts". 24 It is important to note that both the names and the calculations of such non-gaap measures vary among firms, see examples in Mulford and Comiskey (2002) and Mulford and Maloney (2004).

32 The current setting allows for a natural test of the question: "Do private debt holders have more information about the future profitability of the borrowers than equity holders do?" Clearly, it is infeasible to measure the whole information set of either private debt holders or equity investors. However, it is possible to compare the predictive ability of private debt holders' estimates (i.e., loan loss reserves) with that of equity holders' estimates (e.g., valuation ratios, growth forecasts etc.). If public investors utilize the information provided by private debt holders, then the information content of loan loss reserves should be subsumed by the estimates of future operating performance based on public information. Thus, I predict that after controlling for proxies of public information, loan loss reserves cannot predict future operating performance. 3. DATA AND SUMMARY STATISTICS 3.1. Sample Formation and Variable Definitions The sample for the empirical tests includes firms with available accounting data on the Compustat quarterly file from 1990 to Following firms are excluded from the sample: (i) firms that have fiscal quarter ends in months other than March, June, September, or December (to match the timing of loan data and firm data) (ii) financial firms (to avoid any potential mechanical relation between loan quality measures and performance measures), and (iii) firms that are not incorporated in the US (foreign firms are likely to borrow from local banks rather than the US banks). 25 Prior to 1989, the necessary cash flow information is unavailable for a representative number of firms and loan loss reserves were extremely volatile in late 1980s due to less developed country crisis.

33 I use four performance measures in the main tests due to their relevance: (i) operating income (01), (ii) operating income before depreciation (OIBDP), (iii) cash flow from operations (CFO), and (iv) cash flow from operations before interest and tax payments (CFOBIT). I test several other measures and discuss their results in the robustness checks section. I measure operating performance as growth in these variables. I calculate growth using two different methods. Under the first method, growth is equal to the sum of economy-wide income (cash flows) for last four quarters minus the same figure for prior year's same quarter, divided by the latter. This method allows testing results using rolling four quarter growth rates. The second method is more suitable for investigating the association in the short horizon. This method is based on seasonally differenced quarterly measure, defined as economy-wide income (cash flows) in the current quarter minus the same figure for four quarters prior, divided by the latter. Hence to be included in the sample, firms must have data for all relevant quarters. As discussed in section 2.2, both current performance and expected future performance are cited among the important factors in determination of loan loss reserves. Accordingly, both current and future growth rates are included in the tests. Data on loan quality measures, except for that on provisions for loan losses, are available in Federal Reserve Economic Database (FRED ). 26 Loan loss reserves information as reported in the database is equal to total loan loss reserves by all commercial banks divided by total gross loans. Similarly, percentage of nonperforming commercial loans equals total nonperforming commercial loans divided by total commercial loans. I calculate the change in these ratios as the difference between the 26 FRED contains quarterly information from Report of Condition and Income. These filings are mandatory for every national, state member and insured non-member bank.

34 value of the ratio at time t and that at time t-4. Net charge-offs are equal to total net charge offs during the quarter divided by total gross loans. Finally, I calculate provisions for loan losses as total provisions for loan losses divided by total gross loans, using data from call reports. Since net charge-offs and provisions for loan losses are flow variables, for each of them current and past three quarters' values are summed up to make the variable consistent with the loan loss reserves measure. Proxies for public information used in the study include financial ratios (earnings to price (E/P), book to market (B/M) and dividend yield (DvYld)), stock returns and analyst forecasts. E/P ratio (DvYld) is calculated as rolling four quarter trailing aggregate operating income before depreciation/amortization (dividends) divided by current aggregate market value. B/M ratio is equal to current aggregate book value divided by current aggregate market value. Only firms with available data for both numerator and denominator are included in the calculations. I calculate earnings growth forecasts using I/B/E/S database. For consistency, I restrict this data to nonfinancial, US firms that have fiscal quarter ends at March, June, September, or December. Earnings growth forecast for a period is equal to the sum of current quarter consensus earnings forecasts minus actual aggregate earnings at quarter t-4, divided by the latter. Because there is not enough time series data on analysts' operating earnings forecasts, I use net income growth forecasts. For consensus earnings forecasts, I use the last consensus forecast issued at least 30 days before the report date. I manually eliminated one influential firm quarter observation, which is likely to be a data error, from I/B/E/S sample. Inclusion of this observation is favorable for the results of the paper. Finally, the returns used in the tests are CRSP value weighted quarterly returns for AMEX, NYSE and NASDAQ stocks.

35 3.2. Summary Statistics In Table 1, I report the summary statistics for the data. Quarterly OIBDP (01) growth is 7.3% (7.65%) for the sample period. This compares to 7.8% reported in Kothari, Lewellen and Warner (2006) for period. 27 On average, growth rate in cash flows is higher than the growth rate in earnings in the sample period. Correlation between concurrent values of OIBDP and CFOBIT is 64% and correlation between current OIBDP and future CFOBIT is 56% (untabulated). Figure 1 graphically illustrates the behavior of growth rates over time. High profitability growth during mid-1990s is followed by negative growth rates in early 2000s. The average number of firms used in quarterly earnings and cash flow from operations growth calculations are similar and above 4300 firms per quarter. Data availability constrains the sample size for cash flow from operations before interest and tax payments. To ensure the robustness of the results to differences in number of firms, I replicated all tests first replacing missing observations for interest and tax payments with zero, second using the sample of firms that has available data for all relevant variables. Conclusions remain similar. The average change in loan loss reserves is negative. This might be seen as an indication of reduction in the credit risk during the sample period although the post-2007 events arguably suggest otherwise. Figure 1 demonstrates that while the change in loan loss reserves is negative in most of the quarters, in early 1990's, and early 2000's loan loss reserves grew. There is a noticeable negative association between the growth rates in operating performance measures and the change in loan loss reserves. This is intuitive 27 The measure of earnings in that study is earnings before extraordinary items.

36 since banks would recognize loan losses when the borrower's performance falls short of necessary level to repay debt. Table 2 shows that the correlation is negative for all operating performance measures and stronger for earnings growth measures. The same result also holds for the current growth rates (untabulated). The average provisions for loan losses and net charge offs are very similar for the sample period. This is not surprising since in the long run provisions should be equal to actual losses. Figure 2 shows the distribution of operating performance on changes in loan loss reserves. I use three valuation ratios in the study (E/P, DvYld and B/M). E/P ratio has a mean of approximately 13%. This is lower than the 20% reported in Lewellen (2004) for Presumably the boom in 1990's and the stock market bubble contribute to this difference. E/P ratio is negatively correlated with one-quarter-ahead growth rates, consistent with E/P ratio reflecting future profitability growth. The mean B/M ratio is 34%. Similar to E/P ratio, B/M and DvYld are negatively correlated with one-periodahead operating performance whereas analysts' growth forecasts are positively correlated with it. Quarterly average return on the CRSP value weighted index is 3.1%. 4. EMPIRICAL FINDINGS 4.1. Earnings vs. Cash Flows The empirical analysis begins with testing the relation between estimates of credit losses and borrowers' operating performance measures. I use two different models to test the relation:

37 28 AOP. ^ AOP, ALLR t = a+ Pi + P2 + OP t 4 0P t-3 (Quarterly Differencing) IL 3 OP - 2: OP OP - 2.OP ALLR t = a + Pi * _. ^ ' + P2 f S l L «^ + * 5^_-OP A-JOP (Four Qtr. Rolling) / LLR \ / LLR \ = (Total Loans j t " (TOUI Lo, P s) t, 4 where v 't t_4 and OP is aggregate earnings or cash flows measure. I present results of this regression analysis in Table 3. Results for the univariate regression analysis are similar to those for the multivariate analysis, and thus, are not reported. Panel A contains the results for the first model (quarterly differencing) and panel B contains the results for the second model (four quarter rolling window). To account for the autocorrelation in the errors, I report Newey-West (1987) standard 28 errors. Table 3 provides three main results. First, in both Panel A and Panel B, all coefficients on growth rates are negative and, all except one is statistically significant. This suggests that in the US, commercial banks' estimates of credit losses are negatively associated with borrowing firms' operating performance measures. Second, earnings growth measures explain the variation in changes in estimates of credit losses better than the cash flow growth measures. In Panel A, the regression model with OIBDP growth rates has an adjusted R-squared of 66% compared to 35% provided by the model with CFOBIT growth rates. Vuong test indicates that unadjusted R-squared for the two models are significantly different at 1% level (Z-statistic 3.39). Results in Panel B are similar to 28 The maximum number of lags is set to 4 in Panel A and to 8 in Panel B.

38 those in Panel A. This suggests that commercial banks' estimates of credit losses relate more to the earnings of borrowers than to their cash flow. Third, exclusion of certain items, such as depreciation or cash payments for taxes and interest from performance measures increases the explanatory power of the models. Specifically, both coefficients in the regression model with OIBDP growth rates are statistically significant whereas coefficient on future OI value is statistically insignificant. According to Vuong test, OIBDP model also has significantly higher explanatory power compared to OI model in both panels (p value < 0.01). This suggests that information about depreciation is not as valuable as working capital accruals in estimation of credit losses. Similarly CFOBIT model has higher explanatory power compared to CFO model, indicating that interest and tax payments are not as relevant as other operating cash flows in estimation of credit losses. In fact, CFO model has the lowest explanatory power among all four models. Table 3 also demonstrates that in earnings-based models, current earnings growth rates have larger coefficients than future earnings growth rates. In contrast, in cash-flowbased models, future cash flow growth rates have larger coefficients than those of current cash flow growth rates. This finding is perhaps a result of conservative accounting and the correlation between current earnings and future cash flows. To test if cash flow measures have any incremental explanatory power over earnings measures, the last two models in both panels include both OIBDP and cash flow growth rates. 29 In those 29 In the four quarter rolling window tests, calculations of current and future growth rates have four overlapping periods, therefore it is prudent to note that collinearity is not a significant concern. The correlations between current and future growth rates are less than 40%. Variance inflation factor is less than 1.15 for all measures and condition number is less than 3.5 for all measures except for CFO model, which has a condition number of 6. These values are considerably lower than the traditional upper threshold of 10. Moreover conclusions using univariate regressions remain the same. Similarly, in regressions that include both earnings and cash flow variables collinearity statistics do not indicate a problem.

39 30 columns, OIBDP growth rates remain negative and significant while cash flow growth rates have coefficients indistinguishable from zero. This suggests that OIBDP subsumes most of the relevant information contained in cash flow rates. In summary, results in Table 3 lead to three conclusions. First, both earningsbased and cash-flow-based operating performance measures of the borrowers are negatively correlated with the estimates of credit losses. Second, operating income measures explain changes in estimates of potential credit losses better than operating cash flow measures do. Operating cash flow measures have explanatory power but their information content is mostly subsumed by operating income measures. Third, not all items are equally useful in explaining the variation in estimates of credit losses. Performance measures that exclude certain items (e.g., depreciation/amortization) explain variation in estimates of credit losses better than the measures which include those items Components of Loan Loss Reserves Loan loss reserves increase with provision for loan losses and decrease with loan charge-offs. Charge-offs are mainly related with confirmed losses while provisions are made for both confirmed and unconfirmed losses. Accordingly it is useful to examine how these two components that change loan loss reserves are affected by the operating performance of the borrowers. Table 4 reports the results of this analysis. In Panel A, confirmed losses (i.e. net charge-offs) are regressed on current and future operating performance measures of borrowers. First column shows that current OIBDP measure is significant in explaining confirmed losses while future OIBDP is not.

40 31 In second column, both current and future CFOBIT measures are significant. Additionally, the coefficient on current CFOBIT is higher than that on future CFOBIT, which contrasts with the coefficients on CFOBIT on Table 3. CFOBIT model also has higher explanatory power compared to OIBDP model however Vuong test fails to reject the equality of r-squareds. When both OIBDP and CFOBIT are included in the regression, OIBDP measures become insignificant whereas CFOBIT measures remain significant. Results are similar for four quarter rolling window tests, except for the combined test in which current CFOBIT measure also becomes insignificant. In Panel B, unconfirmed losses are examined. Since provisions for loan losses are created for both unconfirmed losses and unexpected confirmed losses, in this regression net charge-offs are included in the models as a control. The first column in Panel B suggests that current and future OIBDP measures have similar coefficients and both are related with the unconfirmed losses. In the second column, current CFOBIT measure is insignificant but future CFOBIT measure is significant, indicating that future cash flows are correlated with unconfirmed losses. When OIBDP and CFOBIT measures are included in the same model, coefficients on both OIBDP measures remain significant while current CFOBIT measure becomes positive and future CFOBIT is insignificant. Four quarter rolling window tests confirm the same findings. Overall, table 4 suggests that confirmed loss component of changes in loan loss reserves is mainly related with the cash flows of the borrowers whereas unconfirmed loss component is related with operating income. Moreover, it is the current operating performance of the borrowers that matters more for calculation of the confirmed losses while for unconfirmed losses both current and future operating performance is important.

41 High Credit Risk Firms vs. Low Credit Risk Firms In this section, I extend the analysis in previous section by focusing on firm portfolios based on credit risk. In particular, firms with high credit risk are more likely to default on their debt payments. Therefore loan loss reserves should be more strongly correlated with the operating performance of high credit risk firms rather than with that of low credit risk firms. To test this prediction and to refine the findings reported in Table 3, I conduct a second analysis. In this test, for every quarter, I classify firms into different portfolios based on leverage. 30 Next, I calculate OIBDP growth for each portfolio. Then, similar to the analysis in Table 3, I run a regression of aggregate changes in loan loss reserves on each portfolio's current and future earnings growth. Table 5 presents the results of this second analysis. In Panels A and B I split the sample into two and four leverage portfolios, respectively. In Panel A, both quarterly differenced, and four quarter rolling window analysis shows that earnings growth for high-leverage firms is more significantly correlated with loan loss reserves than that of low-leverage firms. Coefficients on current earnings growth for high-leverage firms are about 50% larger than those on low-leverage firms. Coefficients on future earnings growth for low-leverage firms are insignificant, whereas for high-leverage firms these coefficients are significant. R-squared of the regression for high-leverage group is almost twice that of low-leverage group. Vuong test suggests that 30 Following Carey, Post and Sharpe (1998), I define leverage as book value of debt divided by the sum of itself and book value of equity. I exclude firms with negative/missing debt or equity values from the sample. Results are robust when leverage is defined as total liabilities to total assets ratio.

42 the difference is statistically significant at 1%. These findings suggest that the results in table 3 are mainly driven by more leveraged firms. In Panel B, I further split the sample into four leverage portfolios. This more detailed examination reveals an interesting pattern among the portfolios. In particular, the explanatory power of the earnings growth rates increases monotonically with leverage except for the highest leverage portfolio. While it may seem counter intuitive, this disturbance supports the results of the previous studies in finance literature which have found that firms in highest risk groups tend to borrow from non-bank lenders. Carey, Post and Sharpe (1998) report that high leverage firms borrow from finance companies rather than from banks. 31 Figure 3 is taken from their study and demonstrates the borrowing patterns across different leverage groups. The figure indicates that the distribution of loans exhibits a symmetric bell shape for banks, in contrast to the markedly skewed distribution towards the high leverage firms for finance companies. Thus, the disturbance of the monotonic relation between earnings growth and loan loss reserves in the last leverage group is not unexpected and in fact, adds to the evidence that firms in the highest risk groups tend to borrow from sources other than banks Effect of Financial Condition of Banks Prior research has found some evidence regarding income smoothing and capital management via loan loss measures (e.g., Wahlen, 1994, Ahmed, Takeda and Thomas 1999). To the extent that these activities are idiosyncratic, the expected effect on the 31 Similarly Denis and Mihov (2003) reports that firms with low credit quality are more likely to borrow from non-bank private lenders.

43 current results would be zero. However, Liu and Ryan (2006) assert that these activities may have a systematic component. In particular, authors conclude that in the boom periods, specifically during 1990s, banks managed loan loss provisions and charge offs upwards to smooth their earnings. 32 To test whether financial condition of banks has any impact on the current findings, I test whether results differ when on average banks have lower capital ratios or ROAs. I calculate aggregate ROA for banks as the sum of earnings before provision for loan losses for all banks in a given quarter divided by the sum of one quarter lagged total assets for the same banks using data from call reports. For capital ratio, since call reports do not contain this data until 2001, I use data from Compustat Bank database which contains data starting Because the number of banks change dramatically from 1993 to 1994, sample starts from 1994 for capital ratio tests. Capital ratio in a given period is equal to median capital ratio of banks with available data in that period. I assign each quarter into one of two categories for each variable using the median value of the variable for the whole time series (i.e. high capital ratio vs low capital ratio, high ROA vs low ROA). The results of the tests are reported in table 6. Capital ratio is negatively associated with the changes in loan loss reserves, indicating that in periods with lower capital ratio loan loss reserves tend to increase. In contrast, ROA has a positive relation with changes in loan loss reserves. When high 32 Loan loss reserves increase with loan loss provisions and decrease with charge offs. Accordingly, findings of Liu and Ryan (2006) may contribute to the current results only if banks inflate charge offs more than they do the loan loss provisions. Otherwise, if the upward management of provisions is stronger than that of charge offs, then the loan loss reserves will increase in boom times which would bias against the findings of the current study.

44 capital ratio dummy (takes value 1 if the capital ratio in the quarter is higher than median 35 capital ratio of the whole sample period, 0 otherwise) and its interaction term with current OIBDP is included in the regression along with current OIBDP, dummy variable remains negative and significant whereas interaction term is positive and significant. This suggests that, loan loss reserves become less sensitive to operating performance of borrowers when capital ratios are relative high. When high ROA dummy (takes value 1 if the capital ratio in the quarter is higher than median capital ratio of the whole sample period, 0 otherwise) and its interaction term with current OIBDP is included in the regression along with current OIBDP, both the dummy variable and the interaction term becomes insignificant. With one exception, these results are similar when OIBDP is measured using four quarter rolling window. The ROA dummy variable becomes positive and significant when four quarter rolling window is used Predictive Ability of Loan Loss Reserves Tables 3 reveals that changes in loan loss reserves are significantly associated with future operating performance of the borrowers. Accordingly, in this section I analyze whether loan loss reserves provide any information about future operating performance that is not captured by proxies of public information. I use following models and report their results in Table 7: AQl t +t _ = «f fcallrt + fcplt + c (One-quarter-ahead growth) = cr+ PiALLR, + p 2 PI { + e (One-year-ahead growth)

45 36 where ALLR t = LLR Total Loans t LLR.Total Loans, 01 is the operating earnings before depreciation, and PI is the proxy for public information. In Panel A, the dependent variable is one-quarter-ahead operating income before depreciation growth. Proxies for expected earnings growth rates include both stock market based variables (current quarter's stock market returns (VW Return), analysts' earnings growth forecasts (Growth Forecast)) and credit market based variables (changes in 10 year t-bond -1 year t-bill yield spread (ATERM), changes in Moody's baa -10 year t-bond yield spread (ASPREAD)) as well as average of past three quarters' aggregate earnings growth rate (Average Past Growth). 33 Standard errors are calculated using Newey-West (1987) method with four lags. The first column in Panel A indicates that ALLR has a decent ability to predict one-quarter-ahead aggregate OIBDP growth. A one percent increase in loan loss reserves as a percentage of total loans indicates an approximately forty percent decrease in future profits. In univariate regressions, coefficients on all proxies are significant. As expected, coefficients are positive for market returns, analysts' forecasts, and average earnings growth rate. ATERM and ASPREAD have negative coefficients. 34 Conclusions remain similar when alternative measures such as future stock returns (quarter t+1), six month stock returns (from t-1 to t+1), Moody's baa bond yields, Moody's baaaaa yield spread, and/or four or five quarter average earnings growth rates are used. 34 These findings complement Kothari, Lewellen and Warner (2006) who find a negative relation between term structure and concurrent earnings growth. They also find a negative relation between market returns and concurrent earnings growth. In the current sample period the correlation between market returns and concurrent earnings growth is positive. However further investigation reveals that this positive relation is driven by period, and when this period is excluded, the correlation becomes negative.

46 When ALLR is included into the models, market returns and ATERM become insignificant. This suggests that ALLR already incorporates the information about onequarter-ahead operating performance incorporated into these two variables. An intuitive result emerges when ALLR and average earnings growth rate variable are included in the same model. The significance of the average earnings growth variable disappears and the goodness of fit improves only slightly compared to the regression reported in the first column. This probably reflects the regulators' requirement that historical data should be incorporated into the estimation of credit losses. ALLR also provides additional explanatory power over growth forecasts and ASPREAD. While coefficients on analysts' earnings growth forecasts and ASPREAD remain significant, the information content of ALLR is not fully captured by either of these variables. 35 Final column shows the results when ALLR is included in the model together with all proxies. Panel B reports the results for one-year-ahead earnings growth using four quarter rolling window. Figure 4 shows the behavior of one-year-ahead OIBDP growth and aggregate E/P ratio over time. Clearly, the relation between E/P ratio and earnings growth is significantly distorted between period. Several events, such as standard changes, and the burst of dot-com bubble, might contribute to this finding. 36 Accordingly, The use of net income growth estimates as analysts' growth forecast, instead of operating income growth estimates, may add noise to this measure. However, even when actual earnings growth rates are calculated using net income measures as reported in l/b/e/s database, the change in loan loss reserves remains negative and statistically significant. Thus, the result is not attributable to the difference between earnings growth measures. In fact, the use of l/b/e/s earnings growth significantly biases against finding results for loan loss reserves, because as results suggest and as discussed in the robustness checks not all accruals are equally valuable to debt holders. 36 Prior studies, (e.g., Jorgensen, Li and Sadka 2008) also show that this period distorts earnings-return relation, but less so when earnings is measured using operating income. In the current study, it is likely that the shorter length of sample period and the use of rolling window exacerbate the problem.

47 results for valuation ratios are sensitive to different treatments of this period. Particularly results for E/P, B/M and DvYld become weaker and often insignificant when this period is included. On the other hand when the period is excluded, results for ALLR are weaker and certain econometric problems arise report results for both treatments in Panel B. In the univariate regressions in Panel B, the coefficient on ALLR increases to -15 (to -31 when is included). This change is in the expected direction since loan loss reserves, by definition, aim to measure potential credit losses in the short term. When period is interpolated, E/P and DvYld have negative and significant coefficients. In the multivariate regressions, results are similar. The coefficient on ALLR becomes slightly smaller but remains significant when used with valuation ratios. Overall results in Panel A and Panel B demonstrate that ALLR have significant explanatory power on OIBDP growth even after controlling for several proxies of public information. Figure 5 shows the coefficients and the R-squareds from the univariate regressions of future OIBDP growth on ALLR (E/P ratio) for different horizons. In all of the regressions period is excluded and replaced by linearly interpolated data. Earnings growth over a 2 year horizon is calculated as the geometric mean of earnings growth in years t+1 and t+2. Panel A demonstrates that the predictive power of ALLR erodes over time. On the contrary, the predictive power of E/P ratio increases. In the oneyear-ahead earnings regression, predictive power of ALLR and E/P ratio becomes similar, and in the two-years ahead earnings growth regression E/P ratio has higher explanatory 37 Exclusion of this period causes econometric problems. Specifically, if the gap in the time series is ignored, methods like Newey-West (1987) correction can taint the results and direction of the effect is unclear. To alleviate these problems, linear interpolation method is used to replace the missing data.

48 39 power than ALLR. Panel B shows that coefficient on ALLR increases from -21 in one quarter regression to -7 in 2 year regression and becomes insignificant. In contrast, coefficient on E/P ratio becomes more significant in longer investment horizons. These results are consistent with ALLR performing well in short term and E/P ratio working better in longer term Robustness Checks Alternative Methods for Calculations Several tests are conducted to test the robustness of the results to alternative measurement methods. First, I test different measures of earnings growth. Following prior studies (e.g., Kothari, Lewellen and Warner 2006) I scale earnings changes by (i) lagged book value and (ii) lagged market value instead of lagged earnings. Results (untabulated) remain similar for both quarterly differencing and rolling four quarter window tests. Second, I conduct tests to check whether firms in Compustat database are representative of the economy. Jorgensen, Konchitchki, Ozel and Sadka (2009) find that in the US, the correlation between the ROA for publicly traded firms and that for privately held firms has been high (69%). Thus, publicly traded firms are likely to be representative of the economy. Since data source in Jorgensen et al. (2009) do not provide data for pre-2000 period, I employ an alternative data source, Global Insight Database, to check whether results are robust to inclusion of private firms. This database provides data on aggregate earnings and cash flows for all the corporations, including

49 40 private firms, in the United States. 38 When I repeat the tests using this data, results are weaker but earnings measure still dominates cash flow measure. Thus, the main conclusion remains robust. Third, I test the robustness of loan loss reserves measure. Banks report loan loss jq reserves for their loan portfolio but not separately for commercial loans. To check if the noise affects the results, I replicate the tests using non-performing commercial loans. Loan loss reserve for a loan and the nonperforming status are highly correlated, but loan loss reserves involve a more significant subjective component. Loans are often classified as nonperforming when they are regarded as nonaccrual or when they are past due for 90 days or more. I measure non-performing commercial loans as total non-performing commercial loans divided by total commercial loans at the end of quarter. This measure is also available in FRED. The results, reported in Table 8, are similar to those in Table 3. In Table 8, earnings growth variables are mostly significant and the models including those variables have higher explanatory power than models that include cash flow growth variables. OIBDP growth rates again outperform OI and cash flow growth rates. Finally, to ensure that results are not specific to a certain measurement method, I retest the first two predictions by measuring the variables in ratio form rather than in growth form. In particular, loan loss reserves is measured as total loan loss reserves divided by total loans and operating performance is measured as four quarter rolling 38 The drawback of the data is that since Bureau of Economic Analysis (BEA) collects this data for governmental use, it is subject to certain adjustments. Therefore, it is only a noisy proxy for GAAP measures. Moreover, BEA does not provide a breakdown of different components (e.g., accruals) of the measures. 39 In the sample period, the size of outstanding commercial loans has been on average 1.7 times that of consumer loans.

50 performance measure (i.e., earnings or cash flows) divided by beginning of period total 41 aggregate assets. 40 Again, the performance measure calculated using operating income before depreciation outperforms the other three measures Alternative Performance Measures The main tables present results for four performance measures. I test four additional performance measures. These are revenues, net income, free cash flows and total cash flows 41 The results (untabulated) indicate that while revenue and net income growth rates are significant and negatively related with changes in loan loss reserves, their coefficients and explanatory powers are weak compared to OIBDP and OI growth rates. Moreover, neither total cash flows nor free cash flows have a significant negative relation with loan loss reserves. Overall, these results provide additional evidence that accruals are relevant in determining loan loss reserves and different earnings components are not equally weighted in estimation of credit losses Macroeconomic Variables In determination of loan loss reserves, macroeconomic changes are also taken into consideration. To test whether results are driven by some macroeconomic events, I use following measures of macroeconomic environment: GDP growth rate, industrial 40 For this test, because of data trend, a time variable is included into the models. Also collinearity becomes a problem therefore conclusion is based on Vuong statistic for explanatory powers of only current (not both current and future) values of performance measures. 41 Aggregate net income and free cash flows are negative in some years. Therefore, to compare these alternative measures with the four measures used in the main tests, I use lagged book value in the denominator for consistency. I calculate free cash flows as cash flow from operations plus cash flow from investments.

51 42 production growth, change in University of Michigan consumer sentiment index, change in 10 year T-bond rate, and Case-Shiller Housing Price Index growth rate. First four are common measures used in the literature for capturing the changes in macroeconomic environment, and the last measure is included to capture the effects of changes in real estate prices. Table 9 reports the results of the models. 42 Changes are measured as changes from time t-4 to t and growth rates are calculated using data at time t-4 and t. Results suggest that the correlation between earnings measures and loan loss reserves cannot be attributed to macroeconomic changes. The other macroeconomic variables that are tested but not tabulated include: changes in Moody's Aaa-Baa yield spread, changes in 1 year t-bill rates, changes in term structure, quarterly returns on S&P 500/CRSP value weighted portfolio, changes in University of Michigan consumer sentiment index, returns on MIT/CRE Commercial Property Index, growth rates of personal disposable income, and growth rates of commercial and industrial loans. The results are robust to addition of these variables. I also examine whether economic downturns affect the relation between operating performance and ALLR. To test this relation, I include a dummy variable for recession quarters (as defined by NBER) and its interaction term with operating performance measures to the models in Table 3. Results (untabulated) suggest that findings are not driven solely by recession periods. Results using macroeconomic variables should be interpreted with caution. To the extent macroeconomic factors and loan loss reserves are co-determined with corporate performance, addition of these factors will capture information content of corporate performance measures. The effect may not be symmetric for each of the performance measures.

52 Discussion of Findings The findings in table 3 indicate that private debt holders focus primarily on operating earnings of borrowers when evaluating credit losses. In particular, among various measures of performance, growth in operating earnings before depreciation and amortization has the highest association with the changes in loan loss reserves. While it is outside the scope of the current study to explain why debt holders prefer earnings-based measures over cash flow-based measures, it is pertinent to elaborate on this question. The straightforward interpretation of the result is that certain, but not all, accruals provide additional explanatory power about borrower's debt service capacity in the short run. For example, Stumpp (2000) argues that in a deep recession a company can curtail CAPEX to pay principal and interest. Accordingly it is legitimate to exclude depreciation from performance measures when evaluating firms that are in unfavorable financial conditions. An alternative explanation of the finding is that debt holders use earnings-based measures to approximate cash flows perhaps because available cash flow information is not adequate for debt evaluation purposes. This might be the case if, for example, debt holders are interested in details of operating activities which are often reported only in accrual basis (e.g., segment disclosures). Assuming that collection of the same information on cash basis is infeasible or too costly, debt holders may prefer to base their analysis on income statement information for a consistent and detailed assessment. The following lines from SFAS 95 also suggest that debt holders demand more information about cash flows than what regulations currently require: "Both commercial lenders and equity analysts who responded to the Exposure Draft asked that more detail on cash flows from operating activities be required. Some said that degree of detail is more

53 important than manner of presentation The Board understands, however, that determining operating cash payments in more detail than the minimum specified in paragraph 27 might involve significant incremental costs over those already required to apply the indirect method... " As a third interpretation, debt holders may not be using information in the cash flow statement optimally. For example, a number of banking professionals contend that in commercial banking, ascendant definitions of cash flows have not necessarily been the ones that are the most relevant for the analysis but rather those that are popular among investment community (e.g., Strischek 2001). The second main finding is that change in loan loss reserves outperforms proxies of public information in predicting aggregate operating income growth in the short run. The result is present only in the short term presumably because of the nature of loan loss reserves. In particular, although standards do not define a fixed time frame of loss coverage, regulatory guidance states that coverage of one year's losses is often adequate. This guidance probably is an important determinant of banks' loan loss reserving policies. An interesting question is: "Why do not proxies of public information fully incorporate the information content of loan loss reserves?" To the extent that proxies reflect public investors' expectations about future performance, Table 7 suggests that public investors can improve their estimates of future performance by using information content of loan loss reserves. At least, two arguments can be made against this claim. The first argument is based on the anonymity of loan loss reserves. In particular, because banks are not required to reveal the list of the clients for whom specific reserves are

54 created for, equity investors are often unable to track the loan loss reserve information back to the borrower. This obscurity may hinder the use of loan quality information in equity valuation. 43 Although this argument may hold at the firm level, it is not clear why it is applicable at the aggregate level. In other words, in the absence of firm specific information, equity holders may still utilize this information by constructing portfolios that are representative of the economy. Thus, as long as equity investors use information efficiently, the anonymity of loan loss reserves should not lead to these findings. A second argument can be made based on the inherent differences between private debt holders' estimates and equity holders' estimates. Specifically, equity holders are often concerned about the long term prospects of the firms whereas for private debt holders shorter term profits are relatively more important. Thus, the estimates of private debt holders can explain short term performance better than valuation ratios. This explanation is consistent with the finding that the explanatory power of loan loss reserves erodes as earnings growth is measured over longer periods. However, the explanation has other shortcomings; for example, it does not explain why loan loss reserves remain significant after controlling for analysts' estimates. In rare instances investors may make inferences from loan quality disclosures. For example, First Union and Bank of America were known to hold two very large pieces of a syndicated loan to Sunbeam Corp. in In November 14, 2000 when First Union Corp announced that it will raise charge offs due to a large problematic loan but declined to name the borrower, several analysts claimed that the borrower is Sunbeam. At the time, Sunbeam's stock price was already reflecting the significant problems the firm has. But on the day of First Union's announcement Sunbeam's stocks plummeted approximately 8%. Sunbeam's stock price dropped another 9% the next day when Bank of America announced that it too will increase (more than double) the charge offs due to a loan to an "unnamed borrower". Sunbeam went bankrupt in February 2001.

55 5. CONCLUSION In this study, I investigate whether private debt holders focus more on earnings or cash flows of the borrowers when estimating potential credit losses. I find that private debt holders consider earnings-based measures more relevant than cash flow-based measures in determination of potential credit losses. While it is mostly the current performance that is important in explaining confirmed losses, both current and future performances are important in explaining unconfirmed losses. I also find that private debt holders do not consider all accruals equally informative about borrowers' debt repayment ability. Moreover my findings suggest that, compared to public stakeholders, private debt holders have superior information about borrowers' future operating performance. This study contributes to the research on debt markets in two key respects. First, while extant studies mainly focus on secondary debt market participants, this paper explores the debt evaluation process by private debt holders. Given the significance of the role private debt holders play in debt markets and the differences that distinguish private debt holders from other stakeholders, a better understanding of how private debt holders use accounting information is necessary. Second, the present paper employs a novel framework for studying private debt holders' reactions to new financial information about borrowers. Such a framework may prove useful for other studies on debt markets. My findings provide empirical support to two arguments that are of interest to both academics and practitioners. First, they are aligned with the view that due to certain characteristics of earnings, such as timeliness, debt holders would rather focus on earnings than cash flows. However, this does not necessarily imply that debt holders

56 47 focus on earnings for those characteristics alone. Alternative explanations are possible; for example, if borrowers are unable to provide detailed cash flow information for certain transactions (e.g., because GAAP does not require this information and hence it is not collected) then private lenders may be using earnings-based measures as proxies of cash flows. Another explanation that some banking professionals offer is that commercial bankers may not be using cash flow information optimally in the analysis of credits. Second, my findings are consistent with the arguments in finance and accounting literatures that private debt holders have an informational advantage over other investors who rely only on public information. This finding is perhaps even more interesting because public disclosures of private debt holders (i.e., loan loss reserves) are used as the measure of private debt holders' estimates of future performance. This indicates that equity holders may be neglecting this source of information. It is important to stress that findings are based on economy-wide level data and as such reflect the behavior of private debt holders on average. These findings may not apply to every individual private debt holder within the economy. Moreover, the paper studies the relative usefulness of earnings and cash flows and the findings cannot be interpreted as earnings being the only accounting performance measure used in the estimation of credit losses. Debt holders can and optimally will utilize information from both income statement and cash flow statement in debt evaluation. The results raise additional issues for future research. Of particular interest is the comparative ability of earnings-based measures and cash flow-based measures in measuring default risk. To date, studies on bankruptcy prediction have produced conflicting results. More conclusive research is necessary to understand why private debt

57 48 holders favor earnings over cash flows in debt evaluation. Additional studies might focus on establishing whether debt financing induces incentives for earnings management, or on exploring whether the source of financing is an important determinant in accrual quality. Also, future studies are necessary to explain why equity investors appear to neglect the information content of loan loss reserves regarding future operating performance.

58 49 REFERENCES Ahmed, A. S., Takeda, C., & Thomas, S. (1999). Bank loan loss provisions: a reexamination of capital management, earnings management and signaling effects. Journal of Accounting and Economics, 28(1), Allen, L., Guo, H., & Weintrop, J. (2008). The information content of quarterly earnings in syndicated bank loan prices. Asia-Pacific Journal of Accounting and Economics, 15 (1), Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance, 23 (4), Altman, E. I. (1977). ZETA(TM) analysis : A new model to identify bankruptcy risk of corporations. Journal of Banking and Finance, 1 (1), Ang, J. S., & Patel, K. A. (1975). Bond rating methods: comparison and validation. Journal of Finance, 30 (2), Anilowski, C., Feng, M., & Skinner, D. J. (2007). Does earnings guidance affect market returns? The nature and information content of aggregate earnings guidance. Journal of Accounting and Economics, 44 (1-2), Armstrong, C. S., Guay, W. R., & Weber, J. P. (2009). The role of information and financial reporting. Working Paper. Asquith, P., Beatty, A., & Weber, J. P. (2005). Performance pricing in bank debt contracts. Journal of Accounting and Economics, 40 (1-3), Aziz, A., & Lawson, G. H. (1989). Cash flow reporting and financial distress models: Testing of hypotheses. Financial Management, 18 (1), Ball, R., & Brown, P. (1968). An empirical evaluation of accounting income numbers. Journal of Accounting Research, 6 (2), Ball, R., Robin, A., & Sadka, G. (2008). Is financial reporting shaped by equity markets or by debt markets? An international study of timeliness and conservatism. The Review of Accounting Studies, 13 (2-3), Ball, R., Sadka, G., & Sadka, R. (2009). Aggregate earnings and asset prices. Journal of Accounting Research, 47 (5), Ball, R., Bushman, R. M., & Vasvari, F. P. (2008). The debt-contracting value of accounting information and loan syndicate structure. Journal of Accounting Research, 46 (2), Barth, M. E., Cram, D. P., & Nelson, K. K. (2001). Accruals and the prediction of future cash flows. The Accounting Review, 76 (1), Beatty, A., Dichev, I. D., & Weber, J. P. (2002). The role and characteristics of accounting-based performance pricing in private debt contracts. Working Paper. Beatty, A., Weber, J. P., & Yu, J. J. (2008). Conservatism and debt. Journal of Accounting and Economics, 45 (2-3),

59 Beatty, A., Liao, S., & Weber, J. P. (2008). Evidence on the determinants and economic consequences of delegated monitoring. Working Paper. Beaver, W. H. (1966). Financial ratios as predictors of failure. Journal of Accounting Research, 4, Beaver, W. H., McNichols, M. F., & Rhie, J. W. (2005). Have financial statements become less informative? Evidence from the ability of financial ratios to predict bankruptcy. Review of Accounting Studies, 10 (1), Beneish, M. D., & Press, E. (1995). Interrelation among events of default. Contemporary Accounting Research, 12 (1), Bharath, S., Sunder, J., & Sunder, S. (2008). Accounting quality and debt contracting. The Accounting Review, 83(1), Bhojraj, S., & Swaminathan, B. (2009). How does the corporate bond market value capital investments and accruals? Review of Accounting Studies, 14 (1), Black, F. (1975). Bank funds management in an efficient market. Journal of Financial Economics, 2 (4), Brigham, E. F. (1966). An analysis of convertible debentures: Theory and some empirical evidence. Journal of Finance, 21 (1), Bushman, R. M., Smith, A. J., & Wittenberg-Moerman, R. (2008). Price discovery and dissemination of private information by loan syndicate participants. Working Paper. Campbell, J. Y., Hilscher, J., & Szilagyi, J. (2008). In search of distress risk. Journal of Finance, 63 (6), Carey, M., Post, M., & Sharpe, S. A. (1998). Does corporate lending by banks and finance companies differ? Evidence on specialization in private debt contracting. The Journal of Finance, 53 (3), Casey, C. J., & Bartczak, N. J. (1984). Cash flow-it's not the bottom line. Harvard Business Review, 62 (4), Casey, C. J., & Bartczak, N. J. (1985). Using operating cash flow data to predict financial distress: some extensions. Journal of Accounting Research, 23 (1), Chen, K. C., & Wei, K. C. (1993). Creditors' decisions to waive violations of accounting-based debt covenants. The Accounting Review, 68 (2), Datta, S., & Dhillon, U. S. (1993). Bond and stock market response to unexpected earnings announcements. Journal of Financial and Quantitative Analysis, 28 (4), Davis, D. W., Boatsman, J. R., & Baskin, E. F. (1978). On generalizing stock market research to a broader class of markets. The Accounting Review, 53 (1), Deakin, E. B. (1972). A discriminant analysis of predictors of business failure. Journal of Accounting Research, 10 (1), Dechow, P. M. (1994). Accounting earnings and cash flows as measures of firm performance: The role of accounting accrual. Journal of Accounting and Economics, 18 {1),

60 Dechow, P. M., Kothari, S. P., & Watts, R. L. (1998). The relation between earnings and cash flows. Journal of Accounting and Economics, 25 (2), DeFond, M. L., & Zhang, J. (2008). The information content of earnings surprises in the corporate bond market. Working Paper. Denis, D. J., & Mihov, V. T. (2003). The choice among bank debt, non-bank private debt, and public debt: evidence from new corporate borrowings. Journal of Financial Economics, 70 (1), Diamond, D. W. (1984). Financial intermediation and delegated monitoring. The Review of Economic Studies, 51 (3), Diamond, D. W. (1991). Monitoring and reputation: The choice between bank loans and directly placed debt. Journal of Political Economy, 99 (4), Dichev, I. D., & Skinner, D. J. (2002). Large-sample evidence on the debt covenant hypothesis. Journal of Accounting Research, 40 (4), Easton, P. D., Monahan, S. J., & Vasvari, F. P. (2009). Initial evidence on the role of accounting earnings in the bond market. Journal of Accounting Research, 47 (3), El-Gazzar, S., & Pastena, V. (1990). Negotiated accounting rules in private financial contracts. Journal of Accounting and Economics, 12 (4), Fama, E. F. (1985). What's different about banks? Journal of Monetary Economics, 15 {1), Federal Reserve Bank. (2002). Bank Holding Companies Supervision Manual. Fisher, L. (1959). Determinants of risk premium on corporate bonds. Journal of Political Economy, 67(3), Fitch Research. (1994). Cash is King in Bond Analysis. Special Report. Gadanecz, B. (2004). The syndicated loan market: structure, development, and implications. BIS Quarterly Review,December, Gentry, J. A., Newbold, P., & Whitford, D. T. (1985a). Classifying bankrupt firms with funds flow components. Journal of Accounting Research, 23(1), Gentry, J. A., Newbold, P., & Whitford, D. T. (1985b). Predicting bankruptcy: If cash flow's not the bottom Line, what is? Financial Analysts Journal, 41 (5), Gentry, J. A., Newbold, P., & Whitford, D. T. (1988). Predicting industrial bond ratings with a probit model and funds flow components. Financial Review, 23 (3), Gombola, M. J., & Ketz, J. E. (1983). A note on cash flow and classification patterns of financial ratios. The Accounting Review, 58 (1), Gombola, M. J., Haskins, M. E., Ketz, J. E., & Williams, D. D. (1987). Cash Flow in Bankruptcy Prediction. Financial Management, 16 (4),

61 Gorton, G., & Kahn, J. (2000). The design of bank loan contracts. Review of Financial Studies, 13 (2), Hirshleifer, D. A., Hou, K., & Teoh, S. H. (2008). Accruals,cash flows and aggregate stock returns. Journal of Financial Economics, 91 (3), James, C. (1987). Some evidence on the uniqueness of bank loans. Journal of Financial Economics, 19 (2), James, C. (1995). When do banks take equity in debt restructurings? The Review of Financial Studies, 8(4), Jones, S., & Hensher, D. A. (2004). Predicting firm financial distress: a mixed logit model. Accounting Review, 79 (4), Jorgensen, B. N., Konchitchki, Y., Ozel, B. N., & Sadka, G. (2009). The legal environment and the differential performance of publicly-traded and privately-held firms. Working Paper. Jorgensen, B. N., Li, J., & Sadka, G. (2008). Are accounting standards diversifiable? Evidence of the aggregate effects of standards. Working Paper. Kaplan, R. S., & Urwitz, G. (1979). Statistical models of bond ratings: A methodological inquiry. Journal of Business, 52 (2), Khurana, I. K., & Raman, K. K. (2003). Are fundamentals priced in the bond market? Contemporary Accounting Research, 20 (3), Kothari, S. P., Lewellen, J. W., & Warner, J. B. (2006). Stock returns, aggregate earnings surprises and behavioral finance. Journal of Financial Economics, 79 (3), Largay, J. A., & Stickney, C. P. (1980). Cash flows, ratio analysis and the W.T. Grant Company bankruptcy. Financial Analysts Journal, 36 (4), Leftwich, R. (1983). Accounting information in private markets: Evidence from private lending agreements. The Accounting Review, 58 (1), Lev, B., & Nissim, D. (2004). Taxable income, future earnings, and equity values. The Accounting Review, 79(4), Lev, B., & Nissim, D. (2006). The persistence of the accruals anomaly. Contemporary Accounting Research, 23 (1), Lewellen, J. (2004). Predicting returns with financial ratios. Journal of Financial Economics, 74 (2), Li, N. (2008). Negotiated measurement rules in debt contracts. Working Paper. Libby, R. (1979). The impact of uncertainty reporting on the loan decision. Journal of Accounting Research, 17, Liu, C., & Ryan, S. G. (2006). Income smoothing over the business cycle: Changes in banks' coordinated management of provisions for loan losses and loan charge-offs from the pre-1990 bust to the 1990s boom. The Accounting Review, 81 (2),

62 Livnat, J., & Zarowin, P. (1990). The incremental information content of cash-flow components. Journal of Accounting and Economics, 13 (1), Mashruwala, C., Rajgopal, S., & Shevlin, T. J. (2006). Why is the accrual anomaly not arbitraged away? The role of idiosyncratic risk and transaction costs. Journal of Accounting and Economics, 42(1-2), Merton, R. C. (1974). On the pricing of corporate debt: The risk structure of interest rates. Journal of Finance, 29 (2), Mossman, C. E., Bell, G. G., Swartz, L. M., & Turtle, H. (1998). An empirical comparison of bankruptcy models. Financial Review, 33 (2), Mulford, C. W., & Comiskey, E. E. (2002). The financial numbers game: detecting creative accounting practices. John Wiley & Sons, Inc. Mulford, C. W., & Maloney, D. K. (2004). Corporate reporting practices for free cash flow. %20free%20cash%20flow%20computation%20survey.pdf. Newey, W. K., & West, K. D. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55 (3), Norton, C. L., & Smith, R. E. (1979). A comparison of general price level and historical cost financial statements in the prediction of bankruptcy. The Accounting Review, 54 (1), Office of Comptroller of Currency. (1998). Allowance for Loan and Lease Losses. Comptroller's Handbook. Ohlson, J. A. (1980). Financial ratios and the probabilistic prediction of bankruptcy. Journal of Accounting Research, 18 (1) Ohlson, J. A. (1990). A synthesis of security valuation theory and the role of dividends, cash flows, and earnings. Contemporary Accounting Research, 6 (2), Penman, S. H., & Yehuda, N. (2009). The pricing of earnings and cash flows and an affirmation of accrual accounting. Review of Accounting Studies, 14 (4), Plummer, C. E., & Tse, S. Y. (1999). The effect of limited liability on the informativeness of earnings: Evidence from the stock and bond markets. Contemporary Accounting Research, 16 (3), Rayburn, J. (1986). The association of operating cash flow and accruals with security returns. Journal of Accounting Research, 24, Sadka, G. (2007). Understanding stock price volatility: The role of earnings. Journal of Accounting Research, 45 (1), Sadka, G., & Sadka, R. (2009). Predictability and the earnings-returns relation. Journal of Financial Economics, 94 (1), Sharpe, S. A. (1990). Asymmetric information, bank lending and implicit contracts: A stylized model of customer relationships. Journal of Finance, 45 (4),

63 Shumway, T. (2001). Forecasting bankruptcy more accurately: A simple hazard model. Journal of Business, 74(1), Sloan, R. G. (1996). Do stock prices fully reflect information in accruals and cash flows about future earnings? The Accounting Review, 71 (3), Standard and Poor's. (2008). Corporate Ratings Criteria. Strischek, D. (2001). E-B-l-T-D-A it doesn't spell "Cash Flow". RMA Journal, 84 (3), Stumpp, P. M. (2000). Putting EBITDA in perspective : Ten critical failings of EBITDA as the principal determinant of cash flow. Moody's Investors Service. Van Home, J. C. (1974). Financial management and policy. Englewood Cliffs, N.J.: Prentice Hall. Wahlen, J. M. (1994). The nature of information in commercial bank loan loss disclosures. The Accounting Review, 69 (3), Walter, J. R. (1991). Loan loss reserves. Economic Review, Watts, R. L. (2003). Conservatism in accounting part I: Explanations and implications. Accounting Horizons, 77(3), White, G. I., Sondhi, A. C., & Fried, D. (2002). The Analysis and Use of Financial Statements. John Wiley & Sons, Inc. Wilson, P. G. (1987). The incremental information content of the accrual and funds components of earnings after controlling for earnings. The Accounting Review, 62 (2), Wittenberg-Moerman, R. (2008). The role of information asymmetry and financial reporting quality in debt trading: Evidence from the secondary loan market. Journal of Accounting and Economics, 46 (2-3), Wittenberg-Moerman, R. (2009). The impact of information asymmetry on debt pricing and maturity. Working Paper. Zhang, J. (2008). The contracting benefits of accounting conservatism to lenders and borrowers. Journal of Accounting and Economics, 45 (1), Zmijewski, M. E. (1984). Methodological issues related to the estimation of financial distress prediction models. Journal of Accounting Research, 22,

64 FIGURES

65 56 Figure 1: Operating Performance and ALLR over Time Panel A: Growth in Operating Income before Dep. and ALLR vv ALLR. <_ Growth in Operating Inc. Bef. Dep. (Quarterly, Current) Panel B: Growth in Cash Flow from Operations before Interest and Taxes and ALLR ir% Irt/V* 1\ * I.IV t. A * A/ \ i»v i 1 r! J r / if h iv #»\i «i"» i % % ALLR Growth in CFO Bef. interest & Tax {Quarterly, Current) For illustration purposes ALLR values are multiplied by 100.

66 57 Figure 2: Operating Performance and ALLR -0.2 pf» 0.2 «p.3 I % t Growth in CFO Growth in CFOBIT # 0.2 ALLR V I _ -J # % -0.2 Growth in OIBDP

67 58 Figure 3: Distribution of Leverage for Bank and Finance Company Borrowers 0.3 O Bank Borrowers Finance Company Borrowers Leverage This figure compares the distributions of leverage for bank borrowers and finance company borrowers. Leverage is measured as the book value of debt divided by the sum of itself and book equity. The figure is taken from Carey, Post and Sharpe (1998) study. Loan data for their study is taken from Loan Pricing Corporation, SEC filings and leverage data is taken from Compustat. White (Black) bars indicate the loans given to that group of leveraged companies as a percentage of total loans issued by banks (finance corporations).

68 Figure 4: The Effect of

69 60 Figure 5: Univariate Regressions of Future Operating Performance on E/P Ratio and ALLR Panel A: Adjusted R-Squareds Adj. R-Squareds Fro^ Univariate Regressions 30.00% 25.00% 20.00% 15.00% * E/P "ALLR 10.00% 5.00% 0.00% 1/4 1 2 Horizon of Future Earnings Growth (Years) Panel B: Coefficients (t-statistics) Coefficients (t-statistics) From Univariate Q.OO Regressions Horizon of Future Earnings Growth (Years) This figure compares the R-squareds (Panel A) and coefficients (Panel B) from univariate regressions of growth in future OIBDP on ALLR and E/P ratio. Growth is calculated for three different horizons: one quarter ahead, one year ahead and two years ahead. Two years ahead growth is equal to geometric mean of growth in year t and year t+1. Rolling four quarter window is used for calculating annual growth rates period is replaced with interpolated data in all regressions. Coefficients on change in LLR are divided by 100. Adjusted R-Squareds are from OLS regressions. Standard errors (shown in parenthesis in Panel B) are calculated using Newey- West (1987) method.

70 TABLES 61

71 Table 1: Summary Statistics Earnings Growth Cash Flow Growth Loan Quality Ratios Public Information Returns ^ Average No of Firms Mean St Dev Min Max Operating Inc. Before Depreciation (Q) Operating Income (Q) Operating Inc. Before Depreciation (Roll) Operating Income (Roll) Cash Flow From Ops. Bef. Interest and Taxes (Q) Cash Flow From Ops. (Q) Cash Flow From Ops. Bef. Interest and Taxes (Roll) Cash Flow From Ops. (Roll) A Loan Loss Reserves/G. Loans Net Charge offs/g.loans Provision For Loan Losses/G. Loans A Nonperforming Loans/G. Loans (Commercial) E/P Ratio B/MRatio Dividend Yield Earnings Growth Forecasts Value Weighted Returns (Q) CT> N3

72 The table reports the average, standard deviation and the extreme values of aggregate earnings growth, aggregate cash flow growth, change in aggregate loan quality ratios, proxies for public information and market returns. All variables are in percentages, except N, number of observations and average number of firms per year. Aggregate earnings is measured using two alternative metrics: i) Operating Income Before Depreciation (OIBDP), ii) Operating income (OI). Aggregate cash flows is also measured using two alternative metrics: i) Cash Flow From Operations (CFO) and Cash Flow From Operations Before Interest and Tax Payments (CFOBIT). Growth in the earnings variable is calculated using two methods: i) seasonal differencing (i.e., growth in aggregate earnings is equal to aggregate earnings for quarter t minus aggregate earnings for quarter t-4, divided by the latter) ii) four quarter rolling window (i.e., growth in aggregate earnings at time t is equal to sum of aggregate earnings for last four quarters (from t-3 to t) minus the same value calculated at the quarter t-4, divided by the latter). Growth in cash flow from operations is calculated similarly. Aggregate E/P (Dividend Yield) is calculated as sum of four quarter-rolling aggregate operating income before depreciation (dividends) divided by total market value of firms at the end of the given period. Aggregate B/M ratio is calculated as aggregate book value of the firms at the end of quarter t divided by the aggregate market value of the firms at the end of quarter t. Accounting data is taken from Compustat North America database. Change in loan loss reserves/total gross loans and non-performing commercial loans/total commercial loans are measured as change in ratio from quarter t-4 to quarter t. Provisions for loan losses/total loans and net charge offs/total loans are measured as the sum of current and past three quarters' values of these ratios. Loan quality data, except provisions for loan losses, are available in Federal Reserve Economic Database (FRED). Data for provisions for loan losses are extracted from call reports and aggregated. Earnings growth forecasts are calculated using analyst forecast data from I/B/E/S database. Earnings growth forecast for a period is equal to aggregate consensus earnings forecasts for quarter t minus aggregate actual earnings in quarter t-4 divided by the latter. For consensus earnings forecasts, the last consensus forecast issued at least 30 days before the report date is taken. Returns are CRSP quarterly value weighted returns for AMEX, NYSE, and NASDAQ stocks. Sample consists of firms with March, June, September or December fiscal quarter ends. Financial firms (sic>5999 & sic<7000) and non US firms are excluded from the earnings, cash flows, earnings forecast and valuation ratio calculations. In calculation of growth variables only firms with available data for all relevant quarters are used, where applicable.

73 Table 2: Correlation Matrix ALLR FOIBDP FOI FCFOBIT FCFO (Q) (Q) (Q) (Q) E/P B/M DvYld ALLR FOIBDP(Q) FOI(Q) FCFOBIT(Q) FCFO(Q) E/P B/M DvYld Growth Forecast V.W. Returns (Q) Growth Forecast V.W. Returns (Q) The table reports correlations among the change in aggregate loan loss reserves, aggregate one-quarter-ahead earnings growth, aggregate onequarter-ahead cash flow growth, market returns and proxies for public information. Aggregate earnings is measured using two alternative metrics: i) Operating Income Before Depreciation (OIBDP), ii) Operating income (OI). Aggregate cash flows is also measured using two alternative metrics: i) Cash Flow From Operations (CFO) and Cash Flow From Operations Before Interest and Tax payments (CFOBIT). One-quarter-ahead growth rate in the earnings variable is calculated using quarterly seasonal differencing (i.e., growth in aggregate earnings is equal to aggregate earnings in quarter t minus aggregate earnings in quarter t-4, divided by the latter). Growth in cash flow from operations is calculated similarly. Aggregate E/P (Dividend Yield) is calculated as sum of 4 period-rolling window aggregate operating income before depreciation (dividends) divided by total market value of firms at the end of the given period. Aggregate B/M ratio is calculated as aggregate book value of the firms at the end of quarter t divided by the aggregate market value of the firms at the end of quarter t. Accounting data is taken from Compustat North America database. Change in loan quality ratios is measured as change in ratio from quarter t-4 to quarter t. Loan quality data is available in Federal Reserve Economic Database (FRED). Earnings growth forecasts are calculated using analyst forecast data from I/B/E/S database. Earnings growth forecast for a period is equal to aggregate consensus earnings forecasts for quarter t minus aggregate actual earnings in quarter t-4 divided by the latter. For consensus earnings forecasts, the last consensus forecast issued at least 30 days before the report date is taken. Returns are CRSP quarterly value weighted returns for AMEX, NYSE, and NASDAQ stocks. Sample consists of firms with March, June, September or December fiscal quarter ends. Financial firms (sic>5999 & sic<7000) and non US firms are excluded from the earnings, cash flows, earnings forecast and valuation ratio calculations. In calculation of growth variables only firms with available data for all relevant quarters are used, where applicable. CT)