* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Accruals, Financial Distress, and Debt Covenants Troy D. Janes
Survey
Document related concepts
Yield spread premium wikipedia , lookup
Financial economics wikipedia , lookup
Securitization wikipedia , lookup
Federal takeover of Fannie Mae and Freddie Mac wikipedia , lookup
Syndicated loan wikipedia , lookup
Stock selection criterion wikipedia , lookup
Debt settlement wikipedia , lookup
Systemic risk wikipedia , lookup
Investment management wikipedia , lookup
Financialization wikipedia , lookup
Debt collection wikipedia , lookup
Global saving glut wikipedia , lookup
First Report on the Public Credit wikipedia , lookup
Debt bondage wikipedia , lookup
Debtors Anonymous wikipedia , lookup
Corporate finance wikipedia , lookup
Transcript
Accruals, Financial Distress, and Debt Covenants Troy D. Janes University of Michigan Business School 701 Tappan Street Ann Arbor, MI 48109 [email protected] (734) 763-3537 This version: January 2003 This paper is based on my dissertation at the University of Michigan Business School. I am grateful to my dissertation committee, Patricia Dechow, Ilia Dichev, Tyler Shumway and Anant Kshirsagar. This paper has also benefited from helpful comments and suggestions from Scott Richardson, Judy Day and student and faculty workshops at the University of Michigan Business School. The author gratefully acknowledges financial support from the University of Michigan Business School, the William A. Paton Scholarship Fund and the American Institute of Certified Public Accountants. Abstract This paper documents that accruals provide information that is useful for predicting financial distress and examines the use of this information by commercial lenders in setting debt covenants. Controlling for the level of earnings, firms with extreme accruals are more likely to experience financial distress than firms with moderate accruals. Tests of the relation between accruals and debt covenant tightness show that, as expected, the debt covenants of borrowing firms with low accruals are set tightly; however, contrary to expectations, the debt covenants of borrowing firms with high accruals are set relatively loosely. Since prior research has shown that lenders possess unique information about borrowers, this result can be interpreted as additional evidence that sophisticated users of accounting information do not fully utilize the information in accruals. However, it is important to note that debt covenants reflect only one aspect use of the use of financial information by lenders, and they may use the information in accruals in other ways. 1. Introduction This paper examines whether commercial lenders incorporate the information about financial distress contained in accruals into debt covenants. It documents that, controlling for earnings, accruals provide incremental information over standard variables used in models for predicting financial distress. It further shows that firms with extreme accruals are more likely to become distressed than firms with moderate accruals. This paper also examines one possible use of the information in accruals by commercial lenders. Results indicate that lenders do not fully consider the relation between accruals and financial distress when setting the initial tightness of debt covenants. As expected, debt covenants are set more tightly for borrowing firms with low accruals, regardless of the level of earnings. However, tests reveal that debt covenants for firms with high accruals are set more loosely than firms with moderate accruals. Because the initial tightness of debt covenants is not consistent with the information in accruals about financial distress, this result adds to prior literature that suggests that sophisticated users of accounting information do not fully utilize the information in accruals. Prior research on the information in accruals has shown that high accruals are associated with declining future performance. Sloan (1996) finds that earnings consisting primarily of accounting accruals are less persistent than earnings predominantly made up of cash flows. His results indicate that the performance of firms with extreme accruals tends to mean-revert more quickly than firms with moderate levels of accruals. This result indicates that firms with high accruals will experience lower earnings performance in the future. 1 Other studies have documented the relation between high accruals and future unfavorable events. Changing auditors is generally regarded as a negative signal about firm performance (Schwartz and Menon, 1985; Johnson and Lys, 1990; Schwartz and Soo, 1995, 1996), and DeFond and Subramanyam (1998) find that firms with high accruals are more likely to change auditors. Dechow, Sloan and Sweeney (1996) find that firms with high accruals are more likely to be subject to SEC enforcement actions for violations of generally accepted accounting principles. High accruals have also been associated with management's attempts to manipulate earnings to avoid problems such as debt covenant violations (Dichev and Skinner, 2002; DeFond and Jiambalvo, 1994). Despite the fact that high accruals have been associated with declining future performance, there is evidence that sophisticated users of accounting information do not fully utilize the information in accruals. Sloan (1996) shows that even though high accruals predict declining performance, stock prices behave as if the market does not understand this information. A study of analyst forecast errors shows that forecast errors are larger for firms with high accruals (Bradshaw, Richardson, and Sloan, 2001), and a study of analyst forecast revisions finds that analysts do not revise their forecasts in anticipation of predictable accrual reversals (Barth and Hutton, 2001). Ahmed, Nainar and Zhou (2001) find that analyst forecasts underweight both accrual and cash flow information, indicating an underutilization of the differing information provided by these measures. Bradshaw, Richardson and Sloan (2001) examine audit opinions and find that future earnings reversals driven by high accruals do not affect auditor opinions.1 1 DeFond and Subramanyam (1998) find that firms that change auditors have higher discretionary accruals than other firms. One interpretation of this finding is that firms that change auditors are “opinion shopping.” If firms with high accruals respond to potential audit opinion qualifications by changing auditors, this result may explain why Bradshaw, Richardson and Sloan (2001) find that high accruals do not affect auditor opinions. That notwithstanding, the fact that these high-accrual firms are able to obtain a clean opinion from a new auditor indicates that the new auditor either does not understand the information in high accruals or is willing to look the other way for the sake of new business. Since DeFond and Subramanyam note that the new auditor is usually smaller than the old auditor (e.g. a change from Big 5 firm to a regional firm), it is possible that the new auditor is less sophisticated than the previous auditor and does not understand the information in accruals. 2 Richardson (2002) finds that short sellers do not appear to actively trade on the information in accruals. In contrast to these studies, Collins, Gong and Hribar (2002) find that institutional investors appear to price accruals more correctly than other investor groups mentioned above. A series of prior studies have examined role of commercial lenders as users and producers of financial information and found that lenders appear to have unique information about borrowers not available to others. A theoretical study by Campbell and Kracaw (1980) suggests that an important role of financial intermediaries (e.g. banks) is the production of information. Empirical studies have found significant market reactions to announcements about bank loans indicating that financial market participants behave as though they believe lenders posses unique information. Best and Zhang (1993) find significant market reactions to the announcement of new loans, and Dahiya, Puri and Saunders (2002) find that negative stock market returns are associated with sales of loans by the originating lender. An objective of this study is to add to prior research by investigating whether commercial lenders, as a sophisticated group of financial information users, use the information in accruals in setting debt covenants. A database of private lending agreements, Dealscan, is used to obtain detailed information on debt covenants. The use of Dealscan has two advantages over prior studies of debt covenants. First, it allows the study of private debt contracts. Prior to the release of Dealscan, there was little publicly available information on private debt contracts. Consequently, most prior studies of debt covenants examine public debt contracts (i.e. bonds). Because of the large number of bondholders involved in a public debt issue, renegotiating a debt contract following a covenant violation can be costly and difficult. Since there are significantly fewer parties involved, the renegotiation of private debt contracts following a covenant violation is easier to carry out. As a result, private 3 debt agreements generally contain a greater number of debt covenants than public debt agreements, and these covenants are set more tightly than those in public debt agreements (Smith & Warner, 1979; Gopalakrishnan and Parkash, 1995). Therefore, covenant levels in private debt agreements are likely to be the product of careful analysis by commercial lenders. The second advantage to using Dealscan results from the detailed information it provides on debt covenants. The database generally provides enough information on the level of debt covenants to allow the calculation of actual debt covenant tightness. Most existing studies on debt covenants use measures such as total debt or the debt-to-equity ratio to proxy for covenant tightness because actual data has not been available (Dichev and Skinner, 2002). Because these proxy measures are noisy, studies using them are difficult to interpret (e.g. Mohrman, 1993; Ball and Foster, 1982). The remainder of this paper proceeds as follows: The next section develops testable hypotheses. Section 3 describes the sample used in testing the relation between accruals and financial distress and presents the results of those tests. Section 4 examines the relation between accruals and debt covenant tightness, and Section 5 concludes and provides suggestions for future work. 2. Hypothesis Development 2.1 Accruals and Financial Distress Prior research has shown that high levels of accruals lead to future declines in performance. However, declining performance does not mean that a firm is financially distressed. In order to examine whether commercial lenders should use accrual 4 information in setting debt covenants, a necessary first step is to show that accruals are useful in predicting financial distress. Prior research has developed several models for predicting financial distress (in particular, bankruptcy). Each of these models uses a similar set of accounting ratios to estimate a firm's risk of bankruptcy. Using discriminant analysis, Altman (1968) develops a bankruptcy prediction model (with a summary statistic known as Altman's Zscore) that includes five accounting ratios: working capital to total assets, retained earnings to total assets, earnings before interest and taxes (EBIT) to total assets, market value of equity to total liabilities, and sales to total assets. Zmijewski's (1984) model includes net income to total assets, total liabilities to total assets, and current assets to current liabilities. Ohlson (1980) developed a model utilizing firm size (log of total assets), total liabilities to total assets, net income to total assets, and working capital (or current liabilities) to total assets. Finally, Shumway (2001) created a hazard model using some of these accounting ratios together with stock market data. His model includes net income to total assets, total liabilities to total assets, relative size (relative to the NYSE/AMEX market), excess returns, and the standard deviation of the firm's stock returns. One common factor in each of these models is a measure of earnings. In each of the models, higher earnings are associated with to a lower risk of bankruptcy. However, in light of Sloan’s (1996) finding that high accruals are associated with lower future earnings, considering the level of earnings alone does not give one a complete picture. All other things being equal, a firm with high earnings and high accruals will experience a greater decline in future earnings than a firm with high earnings and low accruals. It follows that accruals provide information over and above that provided by earnings alone. 5 Although prior research links high accruals and declining earning performance, it does link accruals and financial distress. However, an analysis of the causes of high accruals provides a possible link. High accruals resulting from increases in accounts receivable may indicate that a company is having trouble collecting money owed it. Increases in inventories may indicate that the company’s sales are lagging. Both of these problems suggest that the firm may be experiencing liquidity problems that may lead to financial distress. Or, in a worst case scenario, high accruals may be the result of earnings management intended to artificially inflate earnings (Dichev and Skinner, 2002; DeFond and Jiambalvo, 1994). Likewise, although prior research indicates that low accruals lead to improved earnings performance (Sloan, 1996), low accruals resulting from increases in accounts payable and accrued liabilities may also indicate that the company has an inability to pay its debts. Such liquidity problems may also lead to financial distress. Since it can be argued that both very high and very low accruals may indicate a liquidity problem that may lead to financial distress, the first hypothesis of this paper is as follows: H1: Holding earnings constant, firms with high absolute accruals are more likely to experience financial distress than firms with moderate accruals. 2.2 Accruals and Debt Covenants Early research by Jensen and Meckling (1976) and Smith and Warner (1979) have shown that borrowers have the incentive and the ability to shift wealth from lenders to shareholders. In order to facilitate lending, the lender and borrower write covenants into 6 the debt contract that restrict the actions of the borrower and establish monitoring to ensure that the terms of the debt contract are being met. These covenants take two forms, sometimes referred to as negative and positive covenants. Negative covenants generally prohibit certain activities that result in asset substitution or repayment problems. Examples of negative debt covenants include prohibitions on mergers, limits on additional borrowing, restrictions on dividend payments and excess cash sweeps. Positive covenants require the borrower to take certain actions, such as insuring assets used as collateral or meeting certain benchmarks (usually accounting ratios) that indicate financial health. Common examples of positive debt covenants include minimum or maximum allowable levels of current ratio, leverage ratios, profitability and net worth ratios. Debt covenants are used by commercial lenders as early warning systems to signal impending financial problems among borrowers. When a covenant is violated, lenders have the option to require immediate repayment of the loan. Most of the time, however, after reassessing the borrower's situation, the lender waives the violation and resets the covenant below the current level. If the borrower's performance improves, there is no further problem. If the borrower's performance continues to deteriorate, the covenant is again violated, and the lender once again has the opportunity to evaluate the borrower's performance (Smith, 1993; Chen and Wei, 1993; Gopalakrishnan and Parkash, 1995; Dichev and Skinner, 2002). Although enforcement of debt covenants can vary from situation to situation, there is strong evidence that debt covenants impact firms in several ways. Core and Schrand (1999) find that firms that are close to violating debt covenants experience a greater negative stock price reaction to bad news than do firms that are not close to violating 7 covenants. El-Gazzar (1993) finds a negative stock price reaction to the announcement of new accounting policies that may push firms closer to violating debt covenants. Finally, Beneish and Press (1993) document costs associated with the violation of positive debt covenants, referred to as technical default. Frequently, these violations can be waived or the covenant can be renegotiated, but the borrower incurs costs in doing so, ranging from the actual costs of negotiation (attorney’s fees, etc.) to the addition of new covenants.2 Sweeney (1994) and Dichev and Skinner (2002) find evidence that managers take actions to avoid debt covenant violations, although they are unable to determine whether such actions are cases of earnings management or “real” actions such as accessing equity, selling assets, deferring purchases, etc. Dichev and Skinner also report that debt covenants in private contracts are used in an active monitoring role, with lenders using the covenants as an early warning system to inform them of potential problems with the borrower. Despite the importance of debt covenants in the lending process and the subsequent operation of borrowing firms, there have been few studies on the role of accounting in debt contracts3 (Sloan, 2001). Existing studies that examine characteristics of debt contracts and determinants of debt covenants primarily deal with public debt (i.e. bonds). These studies focus on factors such as the industry in which the borrower operates, the number of lenders involved in syndicating the loan, leverage, profitability, and 2 See Chen and Wei (1993) for a discussion of the determinants of waivers. Recent research on the role of accounting in debt contracts has examined the use of performance pricing, a feature that allows the interest rate charged on a loan to vary based on the borrower’s financial health, as measured by accounting ratios or credit ratings (Asquith, et al, 2001; Beatty, et al, 2001; Doyle, 2002). Beatty, et al (2001) finds that performance pricing and covenants are complements rather than substitutes, particularly when measured over the same variable (e.g. a debt contract that includes performance pricing based on debt-to-EBITDA as well as a covenant requiring the firm to maintain the a minimum level of the same ratio). They conclude that performance pricing addresses improvements in firm health (or credit risk), whereas, debt covenants are used to monitor for declines in firm health. 3 8 probability of bankruptcy (Malitz, 1986; El-Gazaar and Pastena, 1991; Berlin and Mester, 1992). Additionally, Berlin and Mester (1992) show that the restrictiveness of debt covenants is decreasing in the credit worthiness of the borrower.4 Evidence discussed above and in Section 1 shows that debt covenants play a significant role in debt contracting and that lenders have ample opportunity and motivation to use all available information in setting debt covenants. Therefore, one would expect that the initial level of debt covenants would reflect information in accruals about financial distress.5 Stated as Hypothesis 2: H2: The initial tightness of debt covenants is a function of the magnitude of accruals. The results of tests of the hypotheses developed in this section are presented in the next two sections. 3. Relation between Accruals and Financial Distress 3.1 Financial Distress Sample Tests of Hypothesis 1 involve comparisons of the level of accruals of distressed and nondistressed firms. Although a possible research design in cases such as this is to identify a sample of distressed firms and compare that sample to a matched sample of nondistressed firms, such non-random sampling can result in biased parameter and 4 Discussions with commercial lenders confirmed that the findings of prior research are consistent with actual lending practices (Chaika, 2001; Bacevich, 2002). 5 The benchmark contained in the covenant may change over time, generally requiring improving performance by the borrower. However, because factors unrelated to the lender's analysis of the borrower (e.g. economic downturns, etc.) may affect debt covenant tightness during the term of the loan, this study focuses only on the initial tightness of the debt covenant. 9 probability estimates. To avoid such biases, the sample used in this study consists of all firm-years in the Compustat database with sufficient data to compute the variables of interest. Note that the requirement for “complete data” may also introduce bias into the sample, but this bias, in general, does not affect statistical inferences (Zmijewski, 1984). The sample consists of 36,652 firm-year observations from 7,007 firms during the period 1990-2000. 6 367 of these firms (5.2% of the sample) experience financial distress during the sample period. Each firm-year observation consists of data for Year t and the two preceding years. With the exception of firms identified as distressed, all firm-year observations with sufficient data to compute the required variables are included in the sample. As in prior research (Shumway, 1996; Dichev, 1998), CRSP data was used to identify financially distressed firms. There is no generally accepted definition of “financial distress.” A firm that files for bankruptcy is universally considered to be in financial distress, but bankruptcy is the extreme manifestation of financial distress. A firm filing for bankruptcy may have been experiencing financial problems for some time before the filing, but it is difficult to identify when the period of distress began.7 For purposes of this study, I use exchange delisting for performance reasons as an indicator of financial distress. CRSP delisting codes indicate when a firm is delisted from its exchange and for what reason. Reasons for delisting include bankruptcy, insufficient capital, low stock price, failure to make SEC and/or exchange-required filings in a timely manner, etc. One drawback of using delisting to identify financially distressed firms is that delisting is not always timely. Dichev (1998) cites the case of Continental Airlines, which filed for 6 This period was chosen because cash flow data is not available before 1988. The period also corresponds with the availability of debt covenant data used in tests of Hypothesis 2. 7 An additional confounding factor is that fact that bankruptcy filings may be hastened or delayed for strategic reasons. 10 bankruptcy protection in December 1990 but was not delisted until 1992.8 Indeed it is possible that a firm in financial distress may not be delisted at all. However, this limitation of the CRSP data does not weaken the results of the tests in this paper. Failure to identify all firms in financial distress merely means a smaller sample of distressed firms, which lowers the power of the statistical tests and makes it more difficult to find results. The variable of interest is the level of accruals found in the firm’s earnings, computed as earnings before extraordinary items (Compustat Item #18) minus cash flows from operations from the statement of cash flows (Item #308). Additionally, factors identified in previous research as having the ability to predict financial distress are included as control variables (Altman, 1968; Zmijewski, 1984; Ohlson, 1980; Shumway, 2001). Each of these studies identifies somewhat overlapping sets of factors that predict future financial distress (specifically, bankruptcy). Altman’s study is the most well known and often used in practice. Because of this, the factors identified in Altman (1968) are used as control variables for multivariate tests of the relation between accruals and financial distress. The Altman model is: Z = (1.2 x WC) + (1.4 x RE) + (3.3 x EBIT) + (0.6 x MVE) + (0.999 x S) (1) The Altman factors are working capital (WC), retained earnings (RE), earnings before interest and taxes (EARN), market value of equity (MVE), and sales (S). All variables are scaled by total assets except MVE, which is scaled by total liabilities.9 8 Contrasting cases that are high profile at the time of this writing are those of Enron and WorldCom, which were both delisted within weeks of filing for bankruptcy in November 2001 and July 2002, respectively. 9 A firm with a higher Z-score is considered to have a lower probability of bankruptcy. However, recent research has suggested that Altman's coefficients are outdated (Grice and Ingram, 2001). Begley, Ming, and Watts (1996) provide re-estimated coefficients that are more accurate when using recent data. 11 3.2 Tests of the Relation between Accruals and Financial Distress Descriptive statistics for the variables in the financial distress sample are presented in Table 1. For all firms in the sample, as well as for non-distressed and distressed sample firms, this table presents the Altman Z-score factors as well as total accruals (TACC), total liabilities (TL), and current ratio (CR) for the current year, labeled Year 0, and the two preceding years. In the case of the financially distressed firms, Year 0 represents the last annual report issued before becoming distressed (i.e. under the definition used in this study, being delisted). Since each coefficient in the Altman model is positive, the value of each Altman factor is expected to be higher among the non-distressed firms in the sample than among the distressed firms. Table 1 shows that this is the case for all Altman variables except sales (S). Additionally, Table 1 shows that performance of the distressed firms, as measured by the Altman factors, declines rapidly over the three years leading up to becoming distressed, whereas the performance of the non-distressed firm remains relatively stable. Finally, Table 1, Panel B shows that distressed firms typically have lower average accruals than non-distressed firms and more debt on average than non-distressed firms. The current ratios (CR) of distressed and non-distressed firms are not significantly different in Year -2, but the CR for distressed firms is lower than that of non-distressed firms in Years -1 and 0. Table 2 provides additional detail on the sample and univariate evidence about H1. Panel A of Table 2 presents information on quintile portfolios formed on the level of total accruals at Year -2. In the case of financially distressed firms, Year -2 represents the annual report issued approximately three years prior to the firm becoming distressed. Year -2 accruals were used in forming portfolios because Sloan (1996) reports that high 12 accruals generally revert to the mean over three years. Additionally, the average term of the loans in the debt covenant sample is about 40 months, or just over three years.10 Using Year -2 data to form portfolios aligns the time horizon of this analysis with the average time horizon considered by lenders in making lending decisions. H1 states that accruals provide additional information relevant for predicting financial distress above that found in earnings alone. The results shown in Table 2 indicate that accruals do provide additional information regarding the probability of future financial distress. Panel A of Table 2 presents the mean value of earnings (EARN), total accruals (TACC), and current ratio (CR) for quintile portfolios formed on total accruals. For all variables, there is a marked difference between Portfolio 5 (high accruals) and Portfolio 1 (low accruals). However, consistent with the results in Sloan (1996), the mean performance of the high accrual portfolio decreases rapidly in the periods following portfolio formation, while the mean performance of lower accrual portfolios remains more stable. Also consistent with Sloan, mean earnings performance in Portfolio 1 improves over the three-year period. In Year -2, mean EARN in Portfolio 5 is 0.079, compared to a mean EARN of -0.012 in Portfolio 1. By Year 0, however, mean EARN in Portfolio 5 has dropped to 0.028, and mean EARN in Portfolio 1 has risen to 0.008. Rapid reversal of accruals in the high accrual portfolio is also evident, with mean total accruals in Portfolio 5 dropping from 0.096 in Year -2, a number considerably higher than the mean of other portfolios, to -0.045 in Year 0, which is much closer to the other portfolio means in Year 0. To test the hypothesis that, holding earnings constant, firms with higher accruals have a greater risk of financial distress than firms with moderate accruals, portfolios with similar mean earnings are created using a two-pass construction. Following Dechow and Dichev 10 Table 4 provides descriptive data for the debt covenant sample and will be discussed later in the paper. 13 (2001), the sample is sorted into decile portfolios based on the level of earnings. Then, each earnings decile is sorted into quintile portfolios based on the level of total accruals. Then, portfolios are formed by pooling the subportfolios formed in each decile. The final result is five portfolios based on total accruals while controlling for earnings. Portfolio 5 is comprised of the highest quintile of total accruals in each earnings decile; Portfolio 1 is comprised of the lowest quintile of total accruals in each earnings decile, and so on.11 Panel B of Table 2 presents the results of this procedure, with portfolios being formed using Year -2 values of EARN and TACC. The procedure to hold earnings constant was successful, with all portfolios but Portfolio 1 reporting mean EARN between 0.060 and 0.061. The mean EARN of Portfolio 1 is 0.051, which indicates that the sample contains some observations with extremely low earnings.12 Again affirming the findings of Sloan (1996), these results show that, although nearly equal to the mean EARN of other portfolios in Year -2, the mean EARN of Portfolio 5 drops rapidly from 0.061 in Year -2 to 0.015 in Year 0. This figure is well below the mean EARN of the other portfolios, which range from 0.041 to 0.057. It is also interesting to note that the mean EARN of Portfolio 1 remains fairly constant, increasing slightly from 0.051 to 0.052 over the same period. As in Panel A, the mean total accruals of Portfolio 5 drops rapidly after portfolio formation, while the mean total accruals of Portfolio 1 rises over the same period. Because of the presence of outliers in the sample, portfolio medians for the same portfolios are reported in Panel C of Table 2. The median value of EARN in Year -2 is the same (0.081) for all portfolios. It is interesting to note that the median EARN for Portfolios 1 and 2 remains nearly constant over the three years 11 12 I am grateful to Ilia Dichev for suggesting this procedure. Untabulated tests performed on the sample after deleting outliers yield similar results. 14 presented, while the median EARN for Portfolio 1 falls from 0.081 in Year-2 to 0.060 in Year 0, the smallest portfolio mean in Year 0. The relation between total accruals and the occurrence of financial distress is presented in Figure 1, which graphically depicts the occurrence of financial distress in the portfolios discussed above. Figure 1a shows the occurrence of financial distress in portfolios formed on total accruals. Among firms with moderate accruals, the occurrence of financial distress does not vary greatly, ranging from 37 firms (0.5% of the portfolio) in Portfolio 2 to 55 (0.8%) in Portfolio 4. Portfolios 1 and 5 have a much greater occurrence of financial distress with Portfolio 5 (high accruals) containing 88 (1.2%) distressed firms and Portfolio 1 containing 141 (1.9%) distressed firms. Results are similar in portfolios based on total accruals while controlling for earn. Figure 1b presents the occurrence of financial distress in portfolios formed on total accruals, controlling for earnings as done in Table 2, Panel B. Figure 1b shows that financial distress in the moderate accrual portfolios ranges from 55 occurrences (0.76%) in Portfolio 2 to 60 (0.82%) in Portfolio 4. Again, the incidence of financial distress is much higher in the high- and low-accrual portfolios, with Portfolio 1 containing 94 (1.28%) distressed firms, and Portfolio 5 containing 101 (1.41%) distressed firms. These univariate results indicate that firms with the highest and lowest accruals, holding earnings constant, are at greater risk of financial distress than firms with more moderate accruals. Multivariate tests of the relation between accruals and financial distress are presented in Table 3, which reports the results of logistic regressions where the dependent variable is equal to one if the firm is financially distressed following Year 0 and zero otherwise. Model 1 presents the regression of the distress indicator on earnings and the other factors in the Altman model. Model 2 is the same regression including indicator variables for 15 low accruals (LOWACC) and high accruals (HIGHACC). LOWACC equals 1 if the level of accruals for the firm in Year -2 is in the lowest quintile of total accruals, zero otherwise. HIGHACC equals 1 if the level of accruals for the firm is in the highest quintile of accruals, zero otherwise. The results are reported for Year 0, Year -1, and Year -2. As predicted by the Altman model, the coefficient on EARN is negative in both models. The coefficients on WC, RE, and MVE are negative and statistically significant in each regression, indicating that increases in each of these factors decreases the probability of financial distress. The coefficient on S is significant, but in the opposite direction than predicted. Table 2 indicates that firms with high and low accruals have a greater risk of financial distress; therefore, the coefficients on LOWACC and HIGHACC are predicted to be positive. The results indicate that, across all years, LOWACC and HIACC are significant and of the predicted sign. These results indicate that, when considered in addition to earnings, low and high accruals indicate a higher risk of financial distress than more moderate accruals. A comparison of the explanatory power of Model 1 and Model 2 is made by comparing the Akaike Information Criterion (AIC) for each model. A lower AIC indicates more explanatory power (SAS Institute, 1999). For each year, the AIC is lower for Model 2, indicating that accruals provide incremental explanatory power for predicting bankruptcy over earnings alone. In summary, univariate results presented in Table 2 along with the multivariate results in Table 3 indicate that accruals provide more information for the prediction of financial distress than using earnings alone. They also show that the relation between accruals and financial distress is not linear, and firms with the highest and lowest levels of accruals are more likely to experience financial distress than firms with moderate levels of accruals. This finding is not surprising in the case of low accruals, where large negative accruals 16 could be associated with increasing liabilities and related claims on future cash flows. In contrast, high accruals should be associated with increasing cash flows in the future. However, prior research shows that troubled firms sometimes use income-increasing accounting choices to mask their financial condition (Dichev and Skinner, 2002; DeFond and Jiambalvo, 1994). This use of accruals may prevent the firm from taking opportunities to work the problems out, diverting the attention of creditors or shareholders until the firm defaults on a loan, for example (HassabElnaby, 2002). Recent research on bankruptcy emergence supports this idea. Bryan, Tiras and Wheatley (2002) find that bankrupt firms that made income-increasing accounting choices prior to bankruptcy have a lower chance of emerging from bankruptcy. Again, the authors theorize the use of income-increasing accounting choices delays the filing of bankruptcy until the firm’s financial problems are deeper, thus resulting in a lower likelihood that the firm will successfully emerge from bankruptcy. 4 Relation between Accruals and Debt Covenant Tightness 4.1 Debt Covenant Sample Tests of Hypothesis 2 examine whether commercial lenders understand the implications of accruals for financial distress. Data on loans is taken from the Dealscan database provided by LPC Market Access. Dealscan provides a database of over 50,000 loans dating back to 1986. Dealscan consists of loan data gathered from SEC filings, supplemented by research by LPC. The database includes information on the terms of the loan (amount, interest rate, length, etc.) as well as the covenants contained in the debt contract. 17 Loans in the database typically have two or more facilities, or parts. For example, the loan could include a revolving loan and a term loan. Each facility can have slightly different terms, such as different interest rates, but the covenants generally apply to all facilities in the loan. The facility with the longest maturity is assumed to represent the loan and is considered to be the primary part of the loan in this study. If two facilities have equal maturities, the facility with the largest principal amount is selected for inclusion in the sample. The Dealscan database organizes debt covenant information into 12 positive covenants and five negative covenants. Recall from previous discussion that positive covenants generally involve meeting benchmark accounting ratios and negative covenants restrict specific actions. As discussed in Dichev and Skinner (2002), there is a great deal of variation in the definitions of the ratios used in debt covenants. For example, in an examination of Dealscan loans they find over a dozen different ways that the debt-tocash flow ratio is defined in debt contracts. They find similar problems with most other commonly used covenants. Dichev and Skinner (2002) use the current ratio covenant to examine debt covenant violations because they find that it is fairly consistently defined. This allows them to calculate covenant slack using covenant data from Dealscan and data from the borrower’s financials available from Compustat. Since this study also uses Compustat data and Dealscan covenant data together, the current ratio covenant is the primary subject of tests. For inclusion in the sample, data on loan amount, maturity, and current ratio covenant must be available from Dealscan. Observations in the sample must also have sufficient data available on Compustat to calculate the variables used in multivariate tests. The final sample consists of 1,096 loans originating from 1990-1999. 18 4.2 Variable Measurement Typically, accounting-based debt covenants establish a minimum level for the ratio in use. When the lender evaluates the borrower’s financial health, perceived deficiencies will prompt the lender to set the initial level of the covenant more tightly (Chaika, 2001; Bacevich, 2002). Doing so gives the lender more advance warning of deterioration in the borrower’s financial health. A “tight” covenant is one in which the benchmark level in the covenant is close to the actual level of the measure. The difference between the actual measure and the covenant benchmark is referred to as “slack.” I use slack at loan inception as my measure of covenant tightness, with lower slack indicating a tighter covenant. Initial slack is calculated as: SLACK = CR − CRCOV CRCOV (2) where CR = borrower’s current ratio from the annual report immediately preceding the loan, calculated as current assets (Item #4) divided by current liabilities (Item #5) CRCOV = initial current ratio covenant level per Dealscan Control variables include the investment opportunity set of the borrower (IOS), the term (i.e. duration) of the loan (TERM), the indebtedness of the borrower prior to acquiring new debt (DEBT), the size of the borrower as measured by the log of total assets (SIZE), and the amount being borrowed (AMOUNT). Smith and Warner (1978) discuss several opportunities that borrowers have to shift wealth away from the borrower. One of these, 19 referred to as asset substitution, is the ability of the borrower to invest borrowed assets in riskier projects than those approved of by the lender. If a project financed through borrowing is extremely successful, the borrower realizes most of the upside since the payment to the lender is usually fixed by the debt contract. If the project is a failure, the borrower may not be able to repay the loan. Therefore, in making loans, the lender assumes much of the downside risk associated with the assets being loaned. This provides incentive to the borrower to invest in riskier projects than it would if using assets already in place. Since, the IOS is a proxy for the investment opportunities of a firm, a firm with a greater IOS has more opportunity to engage in asset substitution than other firms (Skinner, 1993). It follows that the lender has the incentive to place greater restrictions on borrowers with a greater IOS. Therefore, IOS is included as a control variable, the predicted sign on IOS is negative—higher IOS indicates less slack. The measure used as a proxy for the IOS is that found in Chung and Pruitt (1994): IOS = MVE + BVDEBT + BVPREF TA (3) where MVE = market value of equity (Item #199 x Item #25) BVDEBT = book value of debt (Item #181) BVPREF = book value of preferred stock (Item #130) TA = total assets (Item #6) El-Gazaar and Pastena (1991) and Malitz (1986) show that loans with longer maturities should have more restrictive covenants, but Berlin and Mester (1992) states that longer maturities require looser covenants at inception to account for changes in the firm over time. Therefore, TERM (loan term in months per Dealscan) is included as a determinant 20 of covenant slack, but the predicted sign on TERM is ambiguous. El-Gazaar and Pastena (1991) also show that firms with more debt have tighter covenants, so DEBT (Compustat item #9) is a control variable with a predicted negative sign on the coefficient. Prior research has indicated that the size of the firm is positively correlated with the ability of the firm to repay its debts. Therefore, I include FIRMSIZE (Item #6) as an additional determinant of debt covenant slack, and I predict that the coefficient on FIRMSIZE will be positive. Finally, similar to the finding that firms with more debt have tighter covenants, El-Gazaar and Pastena (1991) find that debt covenant tightness is increasing in the size of the loan. This prompts the inclusion of LOANSIZE (principal amount of the loan per Dealscan) as a control variable with a negative predicted sign. Finally, to separate the effects of low and high accruals, I use indicator variables for low and high accruals (LOWACC and HIGHACC). These indicator variables are computed in the same manner as those in Table 3, with HIGHACC representing borrowers in the highest quintile of total accruals, and LOWACC representing borrowers in the lowest quintile of total accruals. 4.3 Tests of the Relation between Accruals and Debt Covenant Tightness Descriptive data for this sample is found in Table 4. For comparison purposes, the last column of Table 4 presents the mean value of each variable measured over the broader sample used to test Hypothesis 1. Borrowing firms have higher earnings and accruals than the average firm. The IOS for the borrowing firms is lower, suggesting that firms that use commercial loans have lower prospects for growth. Borrowing firms are also smaller and have more debt than the average firm. 21 As with the financial distress sample, the debt covenant sample has been divided into quintile portfolios based on the level of total accruals reported in the annual report preceding the loan closing. Table 5, Panel A presents these results. Similar to the results of tests on the financial distress sample, the firms in Portfolio 5 have the highest earnings (EARN), greatest IOS, which is primarily a function of market value, and highest current ratio (CR). Additionally, firms in Portfolio 5 enjoy the highest slack among the quintile portfolios. The last column in Table 5, Panel A shows that slack increases monotonically from a low of 0.343 in Portfolio 1 to 0.695 in Portfolio 5. This relation is shown graphically in Figure 2a. Panel B of Table 5 presents the results of forming portfolios on total accruals while holding earnings constant, using the same two-pass construction used to form the portfolios in Table 2, Panel B. As shown in the EARN column, the mean earnings of each portfolio are similar, except for Portfolio 1. Mean earnings in Portfolios 2 to 5 range from 0.033 to 0.042, but the mean of Portfolio 1 earnings is 0.003. This result is presented graphically in Figure 2b. Again, the low mean of Portfolio 1 suggests the presence of extremely low earnings in the sample.13 Panel C of Table 5 presents median values of the variables for the same portfolios as in Panel B. The median earnings for each portfolio varies from 0.043 to 0.044. Whether examining mean or median values, the portfolio slack increases monotonically as accruals increase. The results of testing H1 suggest that the relation between the level of accruals and financial distress is non-linear, with high and low accruals leading to greater incidence of financial distress than moderate levels of accruals. However, univariate tests of the debt covenant sample presented in Table 5 indicate that, contrary to expectations, firms with high accruals have looser debt covenants. Table 6 presents the results of multivariate 13 Untabulated tests on a sample with outliers deleted yield similar results. 22 tests of the relation between accruals and covenant slack. Model 1 of Table 6 shows the results of including the HIACC and LOWACC indicator variables in a regression of debt covenant slack on earnings. The coefficient on earnings is 0.567, which is significant at the 1% level, indicating a positive relation between earnings and slack. The coefficient on LOWACC is –0.140, which is statistically significant at the 5% level, which shows that low accruals are associated with lower slack. The coefficient on HIGHACC (0.102) indicates that high accruals are associated with higher levels of slack. However, this coefficient is not significant. The results of this test indicate that, while lenders correctly associate low accruals with greater risk of financial distress, they do not make the same association with high accruals. Model 2 in Table 6 runs the same regression but includes other factors that have been shown to affect debt covenant slack. In Model 2, the coefficient on earnings (0.303) is still positive, but no longer statistically significant. The coefficient on LOWACC (0.136) remains significantly negative, and the coefficient on HIGHACC (0.013) remains statistically insignificant. Of the control variables, only IOS and DEBT have significant coefficients. The coefficient of 0.061 on IOS indicates that a greater investment opportunity set is associated with higher slack. The sign on this variable was predicted to be negative using the reasoning that more investment opportunities were associated with greater opportunities to shift risk to the lender. However, it appears that lenders value the borrower’s opportunities to invest in many projects more than they fear any additional risk the increase in investment opportunity may bring. The coefficient on DEBT is also significant. The coefficient of -0.857 indicates that borrowers with higher ex ante debt levels are subjected to a greater level of monitoring. The above tests show that, in setting debt covenants, commercial lenders fail to fully utilize the information in high accruals for future financial performance. To further 23 examine the effectiveness of the current ratio debt covenant in light of the information in accruals, I examine the number of firms that violate the current ratio debt covenant in the portfolios formed on total accruals. To determine whether a firm violated the current ratio debt covenant, current ratio data was collected from Compustat for the periods following loan inception. Data was collected for all years in the loan term up to and including the year 2000. Firms for which current ratio data was not available for all years were dropped from the sample for this test, leaving a sample of 756 borrowing firms. A borrower whose reported current ratio dropped below the covenant benchmark was considered to be in violation of the loan covenant. The lighter shaded bars in Figure 3 show the results of this analysis. Nearly 45% of the borrowers in Portfolio 1 violated the current ratio debt covenant at some point during the term of the loan. The other portfolios, including Portfolio 5, the high accrual portfolio, had considerably fewer violators, with around 25% of the firms violating the debt covenant in each of the other portfolios. Given the fact that Portfolio 1 borrowers were given considerably less slack than other borrowers, it appears that the number of violators in Portfolio 1 may be due to the lack of debt covenant slack in that portfolio. To address this question, the slack for the borrowers in each portfolio is set equal to the average slack of Portfolio 1, 0.343. The number of violators in each portfolio is then calculated. The results, represented by the darker bars in Figure 3, show that all portfolios now have a similar number of violators. This result suggests that violation of debt covenants is a function of covenant slack; therefore, borrowers in the high accrual portfolio, that receive greater covenant slack despite their higher bankruptcy risk are not monitored enough. Taken together, the results discussed in this section indicate that the debt covenants included by commercial lenders in debt contracts do not reflect an understanding of the information in accruals for future financial distress. 24 4.4 Other Debt Covenant Samples In addition to the current ratio debt covenant sample, analyses have been performed on two additional debt covenants, the debt-to-cash flow covenant and the net worth covenant. The debt-to-cash flow covenant sets a ceiling level for the debt-to-cash flow ratio of the borrower, whereas the net worth covenant sets a floor for the borrowing firm’s net worth (assets minus liabilities). The floor established by net worth covenants typically increases each year by a percentage of the borrower’s net income for that year. Tests similar to those documented in this paper for the current ratio covenant were performed on the debt-to-cash flow and net worth covenant data. For the sake of brevity, the results of these tests are not tabulated. For both samples, the tests did not show a significant relation between accruals and covenant slack. Although these findings provide additional evidence that lenders do not use the information in accruals in setting debt covenants, there are a few alternative explanations for these results. First, the samples were relatively small, so the tests may have lacked power. The debt-to-cash flow sample contained 109 observations, and the net worth sample consisted of 176 observations compared to over 1000 in the current ratio sample. Second, the risk of measurement error was particularly great for the debt-to-cash flow sample. The current ratio is fairly unambiguously defined, and on its face, the debt-tocash flow sample is rather straightforward, too. However, the definitions of debt and cash flow vary from contract to contract. In an examination of Dealscan loans, Dichev and Skinner (2002) find over a dozen different ways that the debt-to-cash flow ratio is defined in debt contracts. Although I was careful to delete any observations with ambiguous definitions from the sample, it is very possible that the measure of the debt-tocash flow ratio constructed using Compustat data did not match up well with the definition the covenant benchmark was based on. Finally, accruals may not be used the 25 same way by lenders in setting various covenants. In my sample data, total accruals and current ratio are significantly, positively correlated, with a Pearson correlation coefficient of 0.22 (p<0.0001). The Pearson correlation coefficient on accruals and the debt-to-cash flow ratio is 0.06 (p=0.554), and the coefficient on accruals and net worth is 0.05 (p=0.488). If lenders understand this and do not consider accruals when setting these two covenants, one would expect to find no relation. 5. Summary and Future Work This paper examines whether a firm’s accounting accruals provide information that is useful in predicting financial distress. It also examines whether commercial lenders use the information in accruals for predicting financial distress as reflected in the initial tightness of debt covenants. Tests of the relation between accruals and financial distress indicate that accruals provide information for the prediction of financial distress above that found in earnings alone. Further, firms with extreme accruals are more likely to experience financial distress. Tests of the relation between accruals and debt covenant tightness show that although borrowing firms with low accruals have tighter debt covenants, borrowers with high accruals do not. Overall, the tests in this paper suggest that accruals provide useful information for predicting future financial distress, but lenders do not incorporate the information in accruals into debt covenants. As a caveat, it is important to note that debt covenants represent only one way in which lenders might use the information in accruals about financial distress. For examples, lenders may respond to the information in accruals by increasing the interest rate on the loan and price-protecting itself. Future work in this area could include an examination of the relation between the information in accruals and interest rates charged on loans. 26 Additional work in this area could incorporate a measure of accrual quality into the analysis of the relation between debt covenant slack and accruals. It may be that the relations reported in this study are not so much due to the level of accruals as they are due to the quality of the accruals. Dechow and Dichev (2001) provide a measure of accrual quality that relates accruals to future realizations of cash flows. This measure may be useful in extending this study. 27 References Ahmed, Anwer S., S.M. Khalid Nainar, and Jian Zhou, 2001, Do analysts forecasts fully reflect the information in accruals?, Syracuse University Working Paper. Altman, Edward I., 1968, Financial ratios, discriminant analysis and the prediction of corporate bankruptcy, The Journal of Finance, 23, 589-609. Asquith, Paul, Anne Beatty, and Joseph Weber, 2001, Performance pricing in debt contracting, Pennsylvania State University Working Paper. Bacevich, Mike, Personal interview, December 12, 2002. Ball, Ray and George Foster, 1982, Corporate financial reporting: A methodological review of empirical research, Journal of Accounting Research, 20, 161-234. Barth, Mary E., and Amy P. Hutton, 2001, Financial analysts and the pricing of accruals, Stanford University Working Paper. Beatty, Anne, Ilia D. Dichev, and Joseph Weber, 2001, The role and characteristics of accounting-based performance pricing in private debt contracts, University of Michigan Working Paper. Begley, Joy, Jin Ming, and Susan Watts, 1996, Bankruptcy classification errors in the 1980s: An empirical analysis of Altman's and Ohlson's models, Review of Accounting Studies, 1, 267-284. Beneish, Messod D. and Eric Press, 1993, Costs of technical violation of accountingbased debt covenants, The Accounting Review, 68, No. 2, 233-257. Berlin, Mitchell and Loretta J. Mester, 1992, Debt covenants and renegotiation, Journal of Financial Intermediation 2, 95-133. Best, Ronald and Hang Zhang, 1993. Alternative information sources and the information content of bank loans, The Journal of Finance, XLVIII, No. 4, 1507-1522. Bradshaw, Mark T., Scott A. Richardson, and Richard G. Sloan, Do analysts and auditors use the information in accruals?, Journal of Accounting Research, 39, 45-74. Bryan, Daniel, Samuel L. Tiras and Clark M. Wheatley, 2002. The association of audit opinion, auditor change and accounting choice with bankruptcy emergence, State University of New York at Buffalo working paper. Campbell, Tim S. and William A. Kracaw, 1980. Information production , market signaling, and the theory of financial intermediation, The Journal of Finance, 35, No. 4, 863-882. 28 Chaika, David, 2001, Personal interview, September 24, 2001. Chen, Kevin C. W., and K.C. John Wei, 1993, Creditors decisions to waive violations of accounting-based debt covenants, The Accounting Review, 68, 218-233. Chung, Kee H. and Stephen W. Pruitt, 1994, A simple approximation of Tobin’s q, Financial Management, 23, 70-74. Collins, Daniel W., Guojin Gong and Paul Hribar, 2002, Investor sophistication and the mispricing of accruals, University of Iowa working paper. Core, John E. and Catherine M. Schrand, 1999, The effect of accounting-based debt covenants on equity valuation, Journal of Accounting and Economics 27, 1-34. Dahiya, Sandeep, Manju Puri and Anthony Saunders, 2002. Bank borrowers and loan sales: New evidence on the uniqueness of bank loans, Forthcoming in Journal of Business. Dechow, Patricia M., and Ilia D. Dichev, 2001, The quality of accruals and earnings: The role of accrual estimation errors, University of Michigan working paper. Dechow, Patricia M, Richard G. Sloan, and Amy P. Sweeny, 1996, Causes and consequences of earnings manipulations: An analysis of firms subject to enforcement actions by the SEC, Contemporary Accounting Research, 13, 1-36. DeFond, Mark L. and James Jiambalvo, 1994, Debt covenant violation and manipulation of accruals, Journal of Accounting and Economics, 17, 145-176. DeFond, Mark L., and K.R. Subrmanyam, 1998, Auditor changes and discretionary accruals, Journal of Accounting & Economics, 25, 35-68. Dichev, Ilia D., 1998, Is the risk of bankruptcy a systematic risk?, The Journal of Finance, LIII, 1131-1147. Dichev, Ilia D. and Douglas J. Skinner, 2002, Large-sample evidence on the debt covenant hypothesis, forthcoming in Journal of Accounting Research. Doyle, Jeffrey T., 2001, Credit risk measurement and pricing in performance pricingbased private debt contracts, University of Michigan Working Paper. El-Gazzar, Samir M., 1993, Stock market effects of the closeness to debt covenants resulting from capitalization of leases, The Accounting Review, 68, No. 2, 258-272. El-Gazzar, Samir M. and Victor Pastena, 1991, Factors affecting the scope and initial tightness of covenant restrictions in private lending agreements, Contemporary Accounting Review 8, No. 1, 132-151. 29 Gopalakrishnan, V., and Mohinder Parkash, 1995, Borrower and lender perceptions of accounting information in corporate lending agreements, Accounting Horizons, 9, 1326. Grice, John Stephen and Robert W. Ingram, 2001, Tests of the generalizability of Altman's bankruptcy prediction model, Journal of Business Research, 54, 53-61. HassabElnaby, Hassan R., 2002. The dynamic relation between earnings management and the violation of debt covenants. Virginia State University working paper. Jensen, Michael C., and William Meckling, 1976, Theory of the firm: Managerial behavior, agency costs and ownership structure, Journal of Financial Economics, 305-360. Johnson, W.B. and T. Lys, 1990, The market for audit services: Evidence from voluntary auditor changes, Journal of Accounting and Economics, 12, 282-308. Malitz, Ileen, 1986, On financial contracting: The determinants of bond covenants, Financial Management 15, No. 2, 18-25. Mohrman, Mary Beth, 1993, Debt contracts and FAS No. 19: A test of the debt covenant hypothesis, The Accounting Review, 68, 273-288. Ohlson, James A., 1980, Financial ratios and the probabilistic prediction of bankruptcy, Journal of Accounting Research, 18, 109-131. Richardson, Scott, 2002, Accruals and short selling. An opportunity foregone?, University of Pennsylvania working paper. SAS Institute, Inc., 1999, SAS OnlineDoc, Version 8. Schwartz, K. and K. Menon, 1985, Auditor switches by failing firms, The Accounting Review, 60, 248-261. Schwartz, K. and B. Soo, 1995, An analysis of Form 8-K disclosures of auditor changes by firms approaching bankruptcy, Auditing: A Journal of Practice and Theory, 14, 125-136. Schwartz, K. and B. Soo, 1996, The association between auditor changes and reporting lags, Contemporary Accounting Research, 13, 353-370. Shumway, Tyler, 1996, The Premium for Default Risk in Stock Returns, Ph.D. dissertation, University of Chicago. Shumway, Tyler, 2001, Forecasting bankruptcy more accurately: A simple hazard model, Journal of Business, 74, 101-124. 30 Skinner, Douglas J., 1993, The investment opportunity set and accounting procedure choice, Journal of Accounting and Economics, 16, 407-445. Sloan, Richard G., 1996, Do stock prices fully reflect information in accruals and cash flows about future earnings?, The Accounting Review, 71, No. 3, 289-315. Sloan, Richard G., 2001, Financial accounting and corporate governance: a discussion, Journal of Accounting and Economics, 32. 335-347. Smith, Clifford W., 1993, A perspective on accounting-based debt covenant violations, The Accounting Review 68, 289-303. Smith, Clifford W. and Jerold B. Warner, 1979, On financial contracting: An analysis of bond covenants, Journal of Financial Economics 7, 117-161. Sweeney, Amy P., 1994, Debt covenant violations and managers’ accounting responses, Journal of Accounting and Economics 17, 281-308. Zmijewski, Mark E., 1984. Methodological issues related to the estimation of financial distress prediction models, Journal of Accounting Research, 22, 59-82. 31 TABLE 1 Comparison of predictors of financial distress in distressed and non-distressed firms 36,652 firm-year observations from the period 1990-2000 Panel A: Altman Z-score factors Year -2 Year -1 Year 0 WC All Non-distressed Distressed t-stat 0.271 0.272 0.234 2.90 *** 0.256 0.257 0.129 9.61 *** 0.242 0.244 0.043 12.90 RE All Non-distressed Distressed t-stat 0.029 0.036 -0.589 12.54 *** 0.008 0.017 -0.962 14.98 *** -0.032 -0.016 -1.565 16.75 *** EARN All Non-distressed Distressed t-stat 0.059 0.060 -0.092 14.04 *** 0.051 0.053 -0.167 17.03 *** 0.044 0.046 -0.208 17.63 *** MVE All Non-distressed Distressed t-stat 4.640 4.645 4.120 1.55 4.195 4.214 2.320 8.23 *** 3.938 3.963 1.475 12.92 *** S All Non-distressed Distressed t-stat 1.235 1.235 1.248 -0.36 1.236 1.235 1.356 -3.19 ** 1.238 1.235 1.505 -6.05 *** *** 32 TABLE 1 (continued) Panel B: Other variables Year -2 Year -1 Year 0 TACC All Non-distressed Distressed t-stat -0.043 -0.042 -0.073 3.52 *** -0.053 -0.051 -0.168 8.46 *** -0.062 -0.060 -0.255 8.34 *** TL All Non-distressed Distressed t-stat 0.491 0.491 0.538 -3.50 *** 0.503 0.501 0.646 -9.58 *** 0.515 0.512 0.772 -13.30 *** CR All Non-distressed Distressed t-stat 2.600 2.602 2.458 1.03 *** 2.388 2.396 1.615 6.40 *** *, **, *** 2.468 2.475 1.794 6.91 Difference between non-distressed and distressed significant at the 10%, 5%, or 1% level, respectively. Sample consists of 36,652 firm-years during the period 1990-2000, including 367 observations from distressed firms. Distressed firms are firms identified by CRSP as having been delisted for performance reasons during the sample period. Year 0 is the firm's annual report immediately preceding delisting. Variables: WC RE EARN MVE S TACC TL CR Working capital divided by total assets Retained earnings divided by total assets Earnings before interest and taxes divided by total assets Market value of equity divided by total liabilities Sales divided by total assets Total accruals divided by total assets Total liabilities divided by total assets Current ratio 33 TABLE 2 Factors predicting financial distress Portfolios formed on total accruals at Year -2 36,652 firm-year observations from the period 1990-2000 Panel A: Variable Means. Portfolios formed on total accruals at Year 2 (1=Lowest) # of Variable Port. Obs. Year -2 Year -1 Year 0 EARN 5 4 3 2 1 7330 7330 7331 7330 7331 0.079 0.080 0.078 0.070 -0.012 0.046 0.067 0.070 0.064 0.009 0.028 0.057 0.065 0.059 0.008 TACC 5 4 3 2 1 7330 7330 7331 7330 7331 0.096 -0.003 -0.039 -0.073 -0.193 -0.018 -0.035 -0.049 -0.063 -0.099 -0.045 -0.047 -0.053 -0.068 -0.097 CR 5 4 3 2 1 7330 7330 7331 7330 7331 3.261 2.914 2.351 2.300 2.177 3.054 2.703 2.220 2.195 2.172 2.900 2.600 2.157 2.170 2.156 34 TABLE 2 (continued) Panel B: Variable Means. Portfolios formed on total accruals controlling for earnings at Year -2 (1=Lowest) # of Variable Port. Obs. Year -2 Year -1 Year 0 EARN 5 4 3 2 1 7330 7330 7331 7330 7331 0.059 0.059 0.059 0.057 0.048 0.028 0.046 0.052 0.058 0.058 0.012 0.037 0.048 0.055 0.051 TACC 5 4 3 2 1 7330 7330 7331 7330 7331 0.098 -0.003 -0.043 -0.085 -0.185 -0.022 -0.038 -0.053 -0.069 -0.091 -0.049 -0.053 -0.058 -0.071 -0.091 CR 5 4 3 2 1 7330 7330 7331 7330 7331 3.319 3.184 2.603 2.301 2.043 3.052 2.914 2.486 2.238 2.059 2.888 2.732 2.433 2.212 2.058 35 TABLE 2 (continued) Panel C: Variable medians. Portfolios formed on total accruals controlling for earnings at Year -2 (1=Lowest) # of Variable Port. Obs. Year -2 Year -1 Year 0 EARN 5 4 3 2 1 7330 7330 7331 7330 7331 0.081 0.081 0.081 0.081 0.081 0.067 0.076 0.079 0.081 0.084 0.060 0.072 0.078 0.081 0.081 TACC 5 4 3 2 1 7330 7330 7331 7330 7331 0.069 -0.005 -0.038 -0.067 -0.130 -0.008 -0.027 -0.041 -0.055 -0.074 -0.022 -0.033 -0.042 -0.054 -0.071 CR 5 4 3 2 1 7330 7330 7331 7330 7331 2.532 2.218 1.881 1.753 1.687 2.413 2.157 1.821 1.733 1.692 2.282 2.056 1.791 1.698 1.681 Sample consists of 36,652 firm-years during the period 1990-2000, including 367 observations from distressed firms. Distressed firms are firms identified by CRSP as having been delisted for performance reasons during the sample period. Year 0 is the annual report immediately preceding delisting. In Panel B, earnings are controlled for by first forming decile portfolios based on EARN, then forming five subportfolios within each earnings portfolio based on TACC. Portfolios in Panel B are formed by grouping together the TACC subportfolios. For example, Portfolio 5 consists consists of the five TACC subportfolios with the highest accruals. Variables: EARN Earnings before interest and taxes divided by total assets TACC Total accruals divided by total assets CR Current ratio 36 TABLE 3 Logistic regressions of occurrence of financial distress on earnings, accruals, cash flows, and control variables 36,652 firm-year observations from 1990-2000 Intercept WC RE Control variables MVE S EARN Pred. Sign - - - - - - Year 0 Model 1 Model 2 -4.790 *** -4.987 *** -1.800 *** -1.806 *** -0.472 *** -0.446 *** -0.228 *** -0.232 *** 0.477 *** 0.448 ** -1.996 *** -1.926 *** 0.492 ** 0.546 *** 853.65 *** 3267.89 873.95 *** 3251.60 Year -1 Model 1 Model 2 -4.680 *** -4.869 *** -1.515 *** -1.594 *** -0.409 *** -0.378 *** -0.106 *** -0.105 *** 0.381 *** 0.342 *** -3.106 *** -2.964 *** 0.566 *** 0.509 *** 620.97 *** 3500.58 642.26 *** 3483.29 Year -2 Model 1 Model 2 -4.737 *** -4.927 *** -0.435 * -0.543 ** -0.380 *** -0.350 *** -0.026 * -0.026 ** 0.243 * 0.197 * -3.059 *** -2.856 *** 0.625 *** 0.489 ** 322.59 *** 3798.95 346.19 *** 3779.36 Period ***, **, * Accruals LOWACC HIGHACC + Likelihood Ratio AIC + Significant at <0.0001, 0.01, 0.05, respectively Above are the results of logistic regressions. The dependent variable is an indicator variable equal to 1 if the firm was financially distressed in Year 0. A firm is considered to be distressed if CRSP indicates that it's stock was delisted for financial reasons. Year 0 is the annual report preceding the delisting. 37 TABLE 3 (continued) Explanatory Variables: EARN LOWACC HIGHACC WC RE MVE S Earnings before interest and taxes divided by total assets Dummy variable equal to one if the firm's total accruals are in the lowest quintile of total accruals (as of Year 2) Dummy variable equal to one if the firm's total accruals are in the highest quintile of total accruals (as of Year 2) Working capital divided by total assets Retained earnings divided by total assets Market value of equity divided by total liabilities Sales divided by total assets 38 TABLE 4 Descriptive statistics on 1,096 loans containing current ratio covenants from the period 1990-1999 Fin. Distress Standard Lower Upper Sample Variable Mean Deviation Quartile Median Quartile Mean EARN 0.030 0.134 0.006 0.043 0.083 0.010 TACC -0.036 0.128 -0.086 -0.031 0.023 -0.064 CR 2.167 1.186 1.379 1.918 2.556 2.462 IOS 1.548 2.014 0.830 1.174 1.743 1.814 DEBT 0.294 0.230 0.111 0.265 0.433 0.241 251.702 525.565 38.880 93.730 250.260 1,658.600 0.389 1.253 0.128 0.242 0.432 n/a 40.720 30.181 18.000 36.000 60.000 n/a CRCOV 1.413 0.447 1.100 1.300 1.500 n/a SLACK 0.558 0.802 0.103 0.357 0.794 n/a FIRMSIZE LOANSIZE TERM Variable Definitions: EARN TACC CR IOS DEBT FIRMSIZE LOANSIZE TERM CRCOV SLACK Net income scaled by total assets. Total accruals at fiscal year-end preceding loan inception, scaled by total assets. Current ratio of borrowing firm. Investment opportunity set of borrowing firm at fiscal year-end preceding loan inception. Total indebtedness of borrowing firm at fiscal year-end preceding loan inception, scaled by total assets. Total assets of borrowing firm at fiscal year-end preceding loan inception. Dollar amount of loan scaled by total assets of borrowing firm. Length of loan term in months. Initial level of current ratio debt covenant. (CR -CRCOV)/CRCOV 39 TABLE 5 Descriptive statistics by portfolio formed on total accruals at loan inception 1,096 loans from 1990-1999 Panel A: Portfolio means: Portfolio formed on total accruals at Year -2 (1=lowest) Portfolio N TACC EARN IOS TERM DEBT FIRMSIZE LOANSIZE CR SLACK 5 219 0.116 0.080 2.062 37.416 0.232 157.926 0.396 2.464 0.695 4 219 0.011 0.049 1.583 37.164 0.266 242.769 0.345 2.393 0.657 3 219 -0.032 0.044 1.470 46.315 0.308 316.264 0.465 2.216 0.541 2 219 -0.074 0.036 1.314 43.290 0.314 320.021 0.367 2.141 0.557 1 221 -0.203 -0.058 1.312 39.683 0.349 222.495 0.376 1.634 0.343 Panel B: Portfolio means: Portfolios formed on total accruals controlling for earnings at Year -2 (1=lowest) Portfolio N TACC EARN IOS TERM DEBT FIRMSIZE LOANSIZE CR SLACK 5 219 0.105 0.038 1.930 36.417 0.258 178.355 0.563 2.469 0.730 4 219 0.009 0.042 1.611 37.868 0.268 240.937 0.343 2.394 0.658 3 219 -0.037 0.035 1.363 44.427 0.314 272.893 0.293 2.164 0.511 2 219 -0.078 0.033 1.424 43.736 0.337 346.166 0.324 2.044 0.469 1 221 -0.178 0.003 1.409 41.081 0.292 219.442 0.425 1.772 0.421 40 TABLE 5 (continued) Panel C: Portfolio medians: Portfolios formed on total accruals controlling for earnings at Year -2 (1=lowest) Portfolio N TACC EARN IOS TERM DEBT FIRMSIZE LOANSIZE CR SLACK 5 219 0.093 0.043 1.095 32.000 0.242 63.955 0.249 2.136 0.485 4 219 0.013 0.043 1.262 35.000 0.258 86.555 0.224 2.060 0.458 3 219 -0.025 0.043 1.117 36.000 0.294 100.290 0.217 1.919 0.361 2 219 -0.062 0.044 1.222 36.000 0.305 154.725 0.224 1.801 0.331 1 221 -0.128 0.043 1.239 36.000 0.258 93.160 0.299 1.636 0.271 Variable Definitions: TACC Total accruals at fiscal year-end preceding loan inception, scaled by total assets EARN Earnings for fiscal year-end preceding loan inception, scaled by total assets IOS Investment opportunity set of borrowing firm at fiscal year-end preceding loan inception TERM Length of loan term in months DEBT Total indebtedness of borrowing firm at fiscal year-end preceding loan inception, scaled by total assets FIRMSIZE Total assets of borrowing firm at fiscal year-end preceding loan inception LOANSIZE Dollar amount of loan scaled by total assets of borrowing firm CR Current ratio at fiscal year-end preceding loan inception SLACK (CR - CRCOV)/CRCOV 41 TABLE 6 Regression of current ratio covenant slack on determinants of debt covenant slack 1,096 loans from the period 1990-1999 Predicted Variable Model 1 Sign Coef. t stat INTERCEPT + 0.604 17.49 EARN + 0.567 LOWACC - HIGHACC - IOS Model 2 Coef. t stat *** 0.867 8.01 2.70 ** 0.303 1.46 -0.140 -1.96 * -0.136 -1.97 * 0.102 1.49 0.013 0.20 - 0.061 4.81 TERM ? -0.0003 -0.04 DEBT - -0.857 -6.42 *** FIRMSIZE + -0.022 -1.07 LOANSIZE - 0.019 1.00 ADJ R2 0.022 *** *** 0.107 ***, **, * Significant at <0.0001, 0.01, 0.05, respectively The dependent variable in each model is the level of initial slack in the current ratio debt covenant. Variable Definitions: SLACK Initial level of slack in the current ratio debt covenant (CR - CRCOV) EARN Earnings for fiscal year-end preceding loan inception, scaled by total assets LOWACC Dummy variable equal to one if the firm's total accruals are in the lowest quintile of total accruals immediately prior to loan inception HIGHACC Dummy variable equal to one if the firm's total accruals are in the highest quintile of total accruals immediately prior to loan inception. Investment opportunity set of borrowing firm at fiscal year-end preceding IOS loan inception TERM Length of loan term in months DEBT Total indebtedness of borrowing firm at fiscal year-end preceding loan inception, scaled by total assets SIZE Total assets of borrowing firm at fiscal year-end preceding loan inception AMOUNT Principal amount of the loan scaled by total assets at fiscal year end preceding loan inception 42 Figure 1a Financially Distressed Firms by Portfolio Formed on Total Accruals 36,652 firm-year observations from 1990-2000 Financially distressed firms (as % of portfolio) 2.5 2 1.5 1 0.5 0 1 2 3 4 5 Portfolios (7,330 or 7,331 obse rvations e ach) forme d on total accruals (1=lowe st) 43 Figure 1b Financially Distressed Firms by Portfolio Formed on Total Accruals after Controlling for Earnings 36,652 firm-year observations from 1990-2000 1.6 Financially distressed firms (as % of portfolio) 1.4 1.2 1 0.8 0.6 0.4 0.2 0 1 2 3 4 5 Portfolios (7,330 or 7,331 obse rvations e ach) forme d on total accruals controlling for e arnings (1 = lowe st) 44 Figure 2a C urre nt Ratio Cove nant Slack by Portfolio Forme d on Total Accruals 1,096 loans from 1990-1999. 0.8 0.7 Current ratio covenant slack 0.6 0.5 0.4 0.3 0.2 0.1 0 1 2 3 4 5 Portfolios forme d on total accruals (1=lowe st) 45 Figure 2b Current Ratio Covenant Slack by Portfolio Formed on Total Accruals Controlling for Earnings 1,096 loans from 1990-1999 0.800 0.700 Current ratio covenant slack 0.600 0.500 0.400 0.300 0.200 0.100 0.000 1 2 3 4 5 Portfolios formed on total accruals controlling for earnings (1=lowest) 46 Figure 3 Current Ratio Covenant Violations by Portfolio Formed on Total Accruals at Loan Inception 756 loans from the period 1990-1999 0.5 0.45 Violators (As % of Portfolio) 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 1 2 3 4 5 Portfolios Formed on Total Accruals at Loan Inception (1=Lowest) Actual Covenant Standardized Covenant 47