Financial Distress and Unexpected Cash-Flows

Transcription

1 Financial Distress and Unexpected Cash-Flows J. Douglas Hanna* Graduate School of Business University of Chicago 1101 East 58th Street Chicago, Illinois March 15, 1995 * This paper is based upon my Ph.D. dissertation, completed at Cornell University. I would like to thank my committee members: Professors Thomas Dyckman (chair), Wayne Shaw, Richard Highfield, and George Casella for invaluable guidance during my program. I would also like to thank Professors Peter Wilson, Vic Bernard and Tom Stober for their willingness to answer questions and offer advice. Financial support from the Arthur Andersen Foundation, the Social Sciences and Humanities Research Council of Canada, and the UW/SSHRC grant program during various stages of this work was greatly appreciated. Comments received from the following accounting workshops were appreciated and helped in the evolution of this paper: University of British Columbia, SUNY Buffalo, University of Chicago, Cornell University, University of Michigan, University of Waterloo and the University of Wisconsin at Madison. Specific comments from Jennifer Francis, Thomas Lechner, Charles Lee, Vic Pastena, Tom Schaefer and the doctoral seminar at Florida State University were also appreciated.

2 1. Introduction This paper examines the effect of financial distress on the information contained in the cash-flow and accrual components of earnings, as opposed to that of earnings alone. Association tests (e.g., Bowen, Burgstahler and Daley [1987], Rayburn [1986]) provide mixed evidence about whether knowledge of the the cash-flow and accrual-components of earnings is important for valuing firms securities. Wilson [1987] and Bernard and Stober [1989] used short-window event-study methodologies to examine this relationship. Their results are mixed with respect to whether the relationship between unexpected cash-flow and abnormal stock returns is statistically significant after controlling for the information contained in the earnings announcement. After not finding any significant association, Bernard and Stober hypothesize that the relationship may be contextual with respect to firmspecific or macroeconomic characteristics. The current study examines partitions of the population of firm-quarter observations to determine whether the relationship between unexpected cash-flows and abnormal returns is affected by a firm-specific measure of financial distress. Evidence is provided that the association between unexpected cash-flows and abnormal returns is affected by the firm s level of financial distress. Financial distress is measured using the Ohlson [1980] model of probability of default within one year. The relationship between unexpected cash-flow and abnormal return is stronger for firms that exhibit high levels of financial distress. Wilson [1987] provides evidence of a statistically significant stock-price reaction to the release of firms financial statements that is associated with unexpected cash-flow realizations. Because earnings (the sum of cash-flows and accruals) are known to investors through the usual earnings announcement prior to the release of financial statements, this finding suggests that cash-flow and accrual components of earnings are valued differently. 1

3 Bernard and Stober [1989] examined the robustness of the Wilson [1987] result. Their findings suggest that the relationship between unexpected cash-flow and abnormal returns is not robust across the population of all firms. 1 Bernard and Stober argue that, it is not obvious that cash flows should be preferred to accruals on average (p. 625). In contrast to a hypothesized general preference for cash-flows over accruals, Bernard and Stober [1989] provide two alternative characterizations of the relationship between unexpected cash-flows and abnormal returns. First, the relationship could be affected by highly contextual firm-specific financial situations. Second, macroeconomic factors associated with different time periods could affect the significance of the relationship. Wilson [1987] also suggested that the relationship between unexpected cash-flows and abnormal returns could vary, across different firms and years (p. 320). The current study examines the effect of the level of an announcing firm s financial distress upon the relationship between unexpected cash-flows and abnormal returns. Financial distress is related to cash-flow through the concept of default risk - the risk that a firm will not be able to meet contractual obligations requiring cash payments. The relationship between cash-flows and financial distress has been suggested throughout the accounting literature (e.g., Beaver [1966], Beaver [1968], Beaver, Kettler and Scholes [1970], and Eskew [1979]). Examining a sample of firms that exhibit high levels of financial distress provides an interesting and potentially more powerful test of the relationship between unexpected cash-flows and abnormal returns. 1.1 IMPORTANCE OF STUDYING ACCRUALS/CASH-FLOWS Accounting researchers have found little evidence consistent with the claim that the financial market reacts to the release of disaggregated earnings information in the form of 1 The terms unexpected cash-flows and abnormal returns are operationalized in section three of the current study. 2

4 annual or quarterly financial statements. There is, however, considerable evidence to support the contention that the financial market reacts predictably to publicly-released earnings announcements. Intuitively, it seems unlikely that the information provided at the earnings announcement date (typically only earnings and revenues for the current and past quarters) can be a sufficient statistic for the vast body of information contained in the annual (or quarterly) financial reports. Ou and Penman [1989] find evidence that accounting measures other than earnings are useful for predicting future returns. A simple decomposition of earnings has two components: cash-flows and accruals. If earnings are already known to the market when financial statements are released, then disclosures in financial statements that identify either cash-flows or accruals are identical in their information content. Once earnings and cash-flows are known, accruals can be determined. Conversely, cash-flows can be determined from knowledge of earnings and accruals. For this reason, the terms information content of accruals and information content of cash-flows are equivalent. 2 The current study examines the information content, incremental to that contained in the earnings announcement, of the announcement of the portion of earnings that is made up of either cash-flows or accruals. Claims? by the accounting profession and the popular business press have motivated recent research regarding the information content of cash flows incremental to that contained in earnings alone. 3 Yet, studies examining the stock-price effects of cash-flows information have not demonstrated that cash-flows and current accruals are used differently in the determination of stock valuation. Neill et al. [1991] review the cash-flow literature and conclude that an additional dollar of earnings has the same value, whether that dollar is the result of an accrued amount or a cash receipt. 2 Wilson (1986) examined the relative information content of accruals and cash-flows as opposed to the incremental information content examined in the current study. Most of the research since the publication of Wilson (1987) has concentrated on either confirming or dispelling the claim that the announcement of the proportions of earnings comprised of cash-flows and accruals has information content. 3 Claims to this effect abound. For examples, see FASB 1978; Beauchamp 1987; and Laderman

5 The paper is presented in six sections. Section two reviews the relevant accounting literature. The two-step regression method used in the current study is presented in section three. Sample selection is detailed in section four. Section five uses the two-step regression model to examine whether the relationship between unexpected cash-flows and abnormal returns is affected by the level of financial distress associated with each specific firm. Section six presents the conclusions and suggests areas where further research could help to resolve important questions about the incremental information content of cash-flows. 2. Related Literature Studies of the information content of cash-flow or accrual disclosures are of two types: association tests and event studies (a test of reaction as opposed to association). Recent association studies (e.g., Bowen, Burgstahler and Daley [BBD 1987], Rayburn [1986]) conclude that while cash-flows are associated with stock returns, there is only weak evidence that using cash-flows in addition to earnings improves a model s ability to explain firms abnormal returns in long-window tests. Recent event studies (Bernard and Stober [1989], Wilson [1987]) provide inconclusive evidence regarding whether financial statements are used as a source of cash-flow information by the market. 2.1 ASSOCIATION TESTS BBD [1987] examined the relationship between unexpected cash-flows and abnormal returns using long-window association tests. Their results indicated that cash-flow information has information content useful for explaining annual abnormal returns, incremental to that of earnings and working capital. Cash flows were shown to have incremental information content in their original sample and in their sample after extreme observations were winsorized. However, when extreme observations were removed, the 4

6 results were only weakly consistent with cash flows having information content incremental to that contained in earnings. Rayburn [1986] used a cash-flow proxy that incorporated changes in working capital accounts and annual data to examine the information content of unexpected cash-flows. Cross-sectional regressions of abnormal returns against cash-flow variables yielded results consistent with the existence of an association between abnormal returns and both unexpected cash-flows and unexpected accruals. Rayburn [1986] did not address the question of whether or not cash-flows have information content incremental to that contained in earnings. However, the coefficients on both cash-flows and accruals were similar in both magnitude and significance level. The results of association tests using alternative cash-flow proxies indicate that cashflows are an accounting measure that is consistent with information used by the market in the determination of share prices. The results are inconclusive as to whether investors value cashflows and accruals differently. 2.2 EVENT STUDIES Event studies can provide evidence on whether firms financial statements are used as a source of cash-flow information to the market. Brown and Warner [1985] summarize the methodological issues involved in event studies. The objective is to examine the reaction by the financial market to the release of information (the event). The financial market reaction is measured by some function of share-price changes during a small window of time surrounding the event (the event window). Using event-windows shorter than those used in the association tests, Wilson [1987] and Bernard and Stober [1989] examine the information content of cash-flows. Both Wilson [1987] and Bernard and Stober [1989] examine cash-flow information which becomes available through the release of financial statements after a firm has 5

7 announced its annual earnings. Annual earnings are made up of cash-flows and accruals such that: E = NA + CA + CF (1) where: E is accounting earnings; NA is noncurrent accruals; CA is current accruals; and CF is total cash-flows for the period. This identity suggests three possible partitions of earnings: E = ( NA + CA ) + CF or E = TA + CF (2) where TA is total accruals; E = NA + ( CA + CF ) or E = NA + WCO (3) where WCO is working capital from operations; and E = CA + ( NA + CF ) (4) Prior research has consistently shown that the noncurrent portion of accruals (NA) has little incremental information content (Rayburn [1986], Wilson [1987], and Bernard and Stober [1989]). Noncurrent accruals may not be a surprise to the market either because they can be accurately forecasted or because they are released to the market at an early date. Whatever the reason, the fact that noncurrent accruals have little incremental information content implies that the earnings defined by equation (3) has no surprise information content if noncurrent accruals, NA, are known in advance. Also, if earnings are known and noncurrent accruals can be accurately forecasted, the earnings partition described by equation (4) is equivalent to the equation (2) partition. For these reasons, the current study examines only the partition of earnings given by equation (2). Wilson [1987] related abnormal stock-returns in a nine-day window centered on the date that the annual report arrived at the SEC to cash-flow forecast errors. Observations were taken from two year-end quarters (1981 and 1982). Wilson modeled the market expectation of firms cash-flows in a cross-sectional manner using current earnings and lagged 6

8 accounting measures as independent variables. Unexpected cash-flows were then determined as the difference between actual cash-flows and the market expectation of cash-flows. Abnormal returns were calculated using the standard ordinary least squares market model. In his sample, Wilson found that cash-flow forecast errors were positively correlated with abnormal returns around financial statements release dates. The regression s explanatory power, R 2, was 2.9 percent. 4 These results are consistent with unexpected cash-flows conveying information beyond that contained in the earnings announcement. Bernard and Stober [1989] examined the robustness of the earlier Wilson [1987] result. The authors attempted, where possible, to use Wilson s methods and apply them to a new and larger sample. To facilitate using the larger sample, the authors used a different cash-flow proxy than did Wilson. Wilson s cash-flow proxy was based on data taken from each firm s Statement of Changes in Financial Position (SCFP) and was manually collected. Bernard and Stober chose not to use data from the SCFP and, instead, created a cash-flow measure using Balance Sheet and Income Statement variables available on the Compustat tape for their sample period. The new and larger sample used in Bernard and Stober [1989] included firms quarterly reports from 1977 to In this time frame, they found 170 firms which had the required accounting data for all 32 quarters. 5 Their findings suggested that Wilson s result could not be generalized to their new sample. The coefficient they found for accrual forecast errors was of the same sign as that found by Wilson; however, it was not statistically significant (at the traditional α =.05 level). 4 The thrust of this test was not to explain abnormal returns around the financial statement filing date, so Wilson did not view the low R 2 was not an indication of unsatisfactory results. 5 The requirement that eligible firms had to have complete data for all 32 quarters was imposed so that the authors could examine time-series as well as cross-sectional properties of the relationship between unexpected cash-flows and abnormal returns. 7

9 Whether or not cash-flow announcements are an important source of information to the financial market is a question that remains unresolved after the Wilson [1987] and Bernard and Stober [1989] studies. While association tests suggest that the information contained in cash-flows is impounded in stock prices, these tests cannot ascertain whether the market gets its cash-flow information from the release of financial statements. Given the perceived importance of cash-flows, new and different tests of the information content of cash-flows are appropriate. 2.3 FINANCIAL DISTRESS A relation between a firm s cash-flows and financial distress is suggested in much of the modern accounting literature. Cash-flow is thought to capture the concept of default risk - the risk that a firm will not be able to meet contractual obligations as they come due. In Beaver s [1966] study of firm-failure prediction, the first group of financial ratios examined was cashflow ratios. Other ratios examined in the Beaver study included leverage and liquidity ratios, both affected by a firm s cash-flow. Leverage is a reflection of the amount of debt, in relation to the amount of equity, carried by a firm. Large debt loads are important because they require large, fixed cash payments each period. Liquidity is important as an inventory of cash available for covering deficits during net cash outflow periods. Cash-flow has also been suggested as a measure related to financial distress by Beaver [1968], Beaver, Kettler and Scholes [1970], and Eskew [1979]. Recent studies have examined the issue of financial distress or financial risk with counter-intuitive results. Lang and McNichols [1990] find no evidence that cash-flows are more informative for firms in financial distress. Bond ratings are used to proxy for firms financial condition. Frankel [1992], again using bond ratings to proxy for a firm s level of financial distress, finds only weak evidence of a relationship between unexpected cash-flows and abnormal returns (abnormal returns on both stocks and bonds) for a sample of firms with below investment-grade bond ratings. 8

10 Lang and McNichols [1990] examined earnings quality, measured as the correlation between past income and cash flows; and the level of financial distress, measured as Standard and Poor s rating on firms debt. Their findings were inconsistent with a hypothesized positive correlation between earnings quality and the information content of earnings. Their findings were also inconsistent with a hypothesized negative correlation between financial condition and the information content of cash flows. Specifically, when the authors partition their sample into quintiles based on firms bond ratings, they observe no change across the quintiles in the relation between market-adjusted returns and unexpected cash-flows. In none of the quintiles is the relationship statistically significant. Frankel [1992] examines stock and bond returns around both earnings announcement dates and 10-K filing dates (assumed to be the date that cash-flow information becomes publicly available). For a sample of firms with below investment-grade bond ratings (approximately defined as those bonds with ratings below BBB), Frankel finds only weak correlation between cash-flows and both stock and bond returns in event-study tests. In longwindow association tests, the author finds that stock returns are weakly correlated with cash flows. Both Lang and McNichols [1990] and Frankel [1992] used bond ratings to measure a firm s level of financial distress. The underlying assumption, as provided in Lang and McNichols [1990, p. 5], is that a riskier bond rating indicates that the firm is expected to have more difficulty in meeting its obligations. There may be problems, however, using bond ratings to measure financial distress. One issue is the fineness of the partitioning information. A continuous variable allows the examination of arbitrary partitions of the data. Bond ratings allow only partitions associated with the different rating categories. Another issue is the objective of the measures. The objective in the current study is to explain variation in the return to a firm s equity holders. Bond ratings are understandably oriented towards the return to debt holders. The two returns may be highly correlated, however, they may not be identical if debt is collateralized, etc. 9

11 Ohlson [1980] developed a measure to capture the probability that a firm will experience financial default. Default was defined as the announcement of legal bankruptcy proceedings (filing Chapter 10 or Chapter 11), thus, this definition of financial default has direct cashflow implications. Maximum likelihood estimation of a conditional logit model provided parameter estimates for multiple financial variables in predicting a dichotomous default versus no default future for a firm. Ohlson used 2,163 annual firm-observations for the period to estimate the model. A total of 105 firms in his sample experienced financial default during this period. The accounting variables used by Ohlson are identified in section 3.3. Because filing for bankruptcy is commonly associated with not having enough cash to meet current cash obligations, shareholders of firms with high default probabilities should react strongly to unexpected cash-flows. Examining a sample of firms that exhibit high probabilities of default should strengthen the observed relationship between unexpected cashflows and abnormal returns. Burgstahler, Jiambalvo and Noreen [BJN 1989] use Ohlson s [1980] default probability model to assess effects on a firm s stock price of changes in that firm s probability of default. BJN use a sample of 22,330 annual firm-observations from the period 1976 to They find that unexpected changes in the probability of failure are useful in explaining security returns. The BJN finding provides comfort that the probability of default variable captures information that is used by the market in determining stock prices and that the model estimated by Ohlson in 1980 is still relevant to the market during the period examined in the current study. Altman and Spivak [1983] examined a measure of the probability of default similar to the measure developed by Ohlson. Using a small sample (N=20) of firms with available Standard and Poor s bond ratings and the data necessary to calculate a variant of the Altman [1968] Z- score, the authors found a Spearman correlation of 85.4 percent between bond ratings and the Z-scores. In section 5.2, the probability of default and bond ratings are compared for the 10

12 sample in the current study. The Spearman correlations for the larger sample of firms used in the current study is much lower (approx. 0.68), thus, allowing the possibility that the probability of default and the bond ratings impound different information. 3. Two-Step Regression Method The current study uses a two-step regression method to examine the relation between unexpected cash-flows (or equivalently accruals) and abnormal returns. A first regression forecasts cash-flows for a period. Unexpected cash-flows, defined as the residual from this first regression, capture the new information about cash-flows that becomes available at the financial statement release date. This date is assumed to be the first date that cash-flow information becomes publicly available. The second regression uses the unexpected cashflow proxy as the independent variable in explaining abnormal returns around the financial statement release date. 3.1 PROXY VARIABLES Abnormal returns are calculated around the date that cash-flow information becomes publicly available. The earliest release of cash-flow information is assumed to be the date that the firm files its quarterly interim financial report (10-Q) or, in the case of fourth-quarter filings, its annual report to shareholders (ARS) with the Securities and Exchange Commission (SEC). 6 6 The current study uses a proxy for the actual SEC filing date. The date used is the date that the financial report is received from the SEC by the firm that provides the SEC with microfiche copies of these reports. The two dates were compared for a sample of 325 firm-filings and, consistent with Stice (1988), the proxy date was adjusted forward by one day to more closely align with the actual SEC filing date. Dates were measured in trading-days to avoid problems with weekends and holidays. Wilson (1987, p. 299) found that using the earlier of the ARS or 10-K filing dates weakened the results of the study. 11

13 Abnormal returns were calculated using the standard market model 7 : AR j,t = R j,t - ( αˆ j + βˆ j R m,t ) (5) where: AR j,t is the abnormal return for firm j on day t; R j,t is the actual return for firm j on day t; R m,t is the value-weighted market return on day t; and αˆ j and βˆ j are the estimated market model parameters. Cumulative abnormal returns for quarter, q, around the event date (day 0) were determined by aggregating the daily abnormal returns for firm j over a nine-day event window 8 : 4 CAR j,q = AR j,t (6) t=-4 Before unexpected cash-flow can be determined, cash-flow must be defined. Two approaches are (1) using data from the Statement of Changes in Financial Position (SCFP proxy) and (2) using data from the Balance Sheet and Income Statement (BSIS proxy). Ketz and Largay [1987] document inconsistency in how firms treat events as either operating or nonoperating on the income statement and the Statement of Changes in Financial Position (p. 15). They add further that... [standard-setting bodies ] views governing the 7 To address concerns regarding the robustness of the current paper s analysis with respect to the definition of abnormal returns, market-adjusted returns were also used in all of the analysis in the paper. The results calculated using market-adjusted returns do not change significantly and provide the same conclusions. 8 Wilson used the average abnormal return during the time period which equals the cumulative abnormal return identified in equation 6 divided by the number of days in the event window. Bernard and Stober (1989) and the current paper use cumulative abnormal return for the time period. Wilson s reported coefficients are multiplied by nine to maintain comparability. A longer event-window increases the probability that the stockprice effects of an information announcement are completely captured within the studied time-period. A shorter time window reduces the probability that contaminating information announcements (those other than the announcement of interest) are made during the studied time-period. Given that both the Wilson [1987] finding consistent with a significant stock-price reaction at the announcement date and the Bernard and Stober [1989] finding of no significant reaction were found using nine-day event windows, the current study also examines nine-day event windows. 12

14 composition of funds from operations on the Statement of Changes in Financial Position are less well-developed [than their views regarding the composition on the Income Statement] (p. 15). Because of the less well-developed definition of operations associated with the Statement of Changes in Financial Position, a proxy constructed from Balance Sheet and Income Statement data may reflect cash-flows from operations more consistently across firms than would a proxy constructed from SCFP data. Examples of financial transactions that are often classified differently on the Income Statement and the Statement of Changes in Financial Position include: gains and losses on asset sales, interest expense and revenue, and gains or losses upon the extinguishment of debt. The differing views on the composition of funds from operations on the Income Statement and on the SFCP could also lead to situations where the SCFP proxy systematically captures valuation-relevant aspects of cash-flow that are not reflected in the BSIS proxy. For example, if interest revenues and expenses are perceived as valuationrelevant by the market, and if interest is classified as from operations on the SCFP but is not classified as such on the Income Statement, then the SCFP proxy should reflect abnormal returns more closely than the BSIS proxy. The current study initially uses both SCFP and the BSIS proxies to measure cash-flow from operations. 9 Because the observed relationships between unexpected cash-flows and abnormal returns are uniformly more consistent and significant in tests using the SCFP proxy, partition results are reported for this proxy only A BSIS proxy consistent with that used by Bernard and Stober (1989) was constructed using Balance Sheet and Income Statement data and the following definitions: Working Capital from Operations (WCO) = Operating Income before Depreciation - Interest Expense - Current Income Tax Expense; Current Accruals (CA) = Change in Working Capital - Increases in Cash - Decreases in Short-term Debt; Cash from Operations (CFO) = WCO - CA; Noncurrent Accruals (NCA) = Net Income - WCO. A SCFP proxy consistent with that used by Wilson (1987) was constructed using Statement of Changes data and the following equations: WCO = Funds from Operations; CA = Working Capital Changes Other than Changes in Cash and Cash Equivalents; CFO = WCO - CA; NCA = Net Income - WCO. 10 Pearson and Spearman correlations between the BSIS and SCFP cash-flow proxies were examined (not reported). The quarter-by-quarter correlation coefficients were surprisingly low between the two cash-flow proxies (approximate average 0.25). Similar coefficients were obtained between the respective estimates of unexpected cash-flow. 13

15 Unexpected cash-flow equals the difference between the actual cash-flow for a period and the ex ante market-expectation of the cash-flow outcome. The selection of an expectation model is important, and has been examined by both Bernard and Stober [1989] and Lorek, Schaefer and Willinger [LSW 1992]. Bernard and Stober examined the use of industry-byindustry expectations models. LSW examined firm-specific time-series expectations models. The current study uses a cash-flow expectations model that pools observations both crosssectionally and in time-series to forecast cash-flows. The choice of this model is made for two reasons. First, the LSW [1992] and Bernard and Stober [1989] results are mixed. Bernard and Stober concluded that the industry-specific models they examined did not strengthen the cash-flow relationship significantly. LSW found an improved relationship between unexpected cash-flows and abnormal returns in only some of the fiscal quarters they examine. 11 Second, Wilson [1987] and Bernard and Stober [1989] differ as to the importance of the information content of cash-flows using the same expectations model - the one used in the current study and described below. The chosen expectations model is estimated as a function of current quarter earnings, current quarter revenues and various lagged accounting variables as follows: CF j,q = β 0 + β 1 EARN j,q + β 2 REV j,q + β 3 REV j,q-1 + β 4 (REV j,q-2 + REV j,q-3 ) + β 5 REV j,q-4 + β 6 CF j,q-1 + β 7 (CF j,q-2 + CF j,q-3 ) + β 8 CF j,q-4 + β 9 NA j,q-1 + β 10 (NA j,q-2 + NA j,q-3 ) + β 11 NA j,q-4 where: + β 12 CA j,q-1 +β 13 (CA j,q-2 + CA j,q-3 ) + β 14 CA j,q-4 + β 15 CAPEX j (7) CF is quarterly cash-flows from operations, EARN is earnings, REV is revenues, NA is noncurrent accruals, CA is current accruals, CAPEX is annual capital expenditures for the current fiscal year, the q subscripts represent quarterly time 11 Empirical tests of the relationship between unexpected cash-flow and abnormal returns were reported in earlier versions of LSW [1992]. These results are now discussed in footnote 1 of the current version of their paper. 14

16 periods (e.g., q-1 is the quarter prior to that associated with the information announcement), firms are identified by the j subscripts, and all variables are deflated by end-of-year total assets. The variables are deflated by Total Assets to reduce the effects of size-related heteroscedasticity in their distributions. 12 The forecast cash-flow from this expectation model is used to represent the market expectation of a firm s upcoming cash-flow announcement. 13 Unexpected cash-flows are then the difference between actual cash-flow and the model s forecast. The unexpected cash-flow, UCF j,q, and the abnormal return in the nine-day event window, CAR j,q are then used in a second univariate regression to assess the information content of unexpected cash-flows: CAR j,q = γˆ 1 + γˆ 2 ( UCF j,q ) + ε j (8) where ε j is assumed to be random error independent of UCF j,q. The significance of the linear relationship between unexpected cash-flows and abnormal returns is established by the significance of γˆ 2 in equation (8). Due to the possible limitations of using bond ratings to measure financial distress, the current study uses the probability that a firm will experience financial default within one year, φ j,t, estimated using Ohlson s [1980] model: φ j,t = [ 1 + exp ( -y j,t ) ] -1 (9) 12 The variables Annual Capital Expenditures, CAPEX, and Total Assets, TA, were measured as of the end of the fiscal year that included the fiscal quarter being examined in the observation. Using variables measured contemporaneously with the current-quarter s cash-flow outcome does not allow the expectations model to portray a purely ex ante belief. The expectations model in the current study should underestimate the absolute magnitude of cash-flow surprises because contemporaneously-measured independent variables are used. The contemporaneous variables were used to maintain comparability with Wilson (1987) and Bernard and Stober (1989). 13 To address concerns with the ad hoc nature of the model described in equation 7, the results of the study were repeated using a seasonal random-walk expectation model for cash-flows. The results do not change and are not reported. 15

17 where y j,t = SIZE j,t TLTA j,t WCTA j,t CLCA j,t NITA j,t FUTL j,t INTWO j,t OENEG j,t CHIN j,t and SIZE = ln(total assets / GNP price level), TLTA = (total liabilities / total assets), WCTA = (working capital / total assets), CLCA= (current liabilities / current assets), NITA = (net income / total assets), FUTL = (funds from operations / total liabilities), INTWO = 1 if net income for last two years is negative and 0 otherwise, OENEG = 1 if total liabilities > total assets and 0 otherwise, CHIN = ((net income t - net income t-1 ) / ( net income t + net income t-1 )), subscript j denotes firms and subscript t denotes annual time periods (e.g., t-1 denotes the year before the current year). Since φ j,t is an annual calculation, quarterly observations are assigned the ex ante probability of default determined using the most recent year-end data. The probability of default measure is used to partition the sample of firm-quarter observations. The correlation between f j,t and both the Standard and Poor s (S&P) bond ratings and Stock ratings are discussed in section 5.2 of the current study. 4. Sample Selection The current study examines a sample of firms during the period that begins with the first quarter of 1984 and ends with the third quarter of This time period was chosen due to data availability on the quarterly Compustat II data tape. Compustat began carrying financial information taken from the Statement of Changes in Financial Position with the first fiscal quarter of Sample selection began with data availability on the 48-quarter Compustat tape. The accounting information necessary to constitute a complete observation is documented in Table 1. [ Insert Table 1 about here ] 16

18 As a first step, data were collected for all firm-quarters with available information. Because Compustat did not carry Statement of Changes in Financial Position information before 1984, and because of the requirement for lagged financial information, complete observations for the current study begin with the second quarter of A total of 7,888 firm-quarter observations were found between the second quarter of 1985 and the third quarter of The sample of firm-quarter observations with available Compustat data was matched to a computer-readable tape of Disclosure Inc. financial statement receipt dates to determine the date that financial statements are filed with the SEC. The receipt date serves as a proxy for the date that financial statements become publicly available information. 7,265 firm-quarter observations had an available financial statement receipt date. Observations were matched to the Center for Research on Security Prices (CRSP) tapes. Required returns data included the nine-day period surrounding the financial statement filing date, and the 120-day parameter-estimation period ending one year and 90 trading days before the end of the fiscal quarter whose financial statement release date is being examined. The estimation period was selected to avoid overlap with the periods associated with any of the lagged accounting variables from the cash-flow forecasting model. If returns were missing during the estimation period, earlier days were used. If more than 10 returns were missing, the observation was dropped from the sample. If any returns were missing from the event period (nine-day window), the observation was dropped from the sample. 5,829 observations had the required CRSP data available. To avoid contaminating the nine-day event window with the stock-price changes associated with the earnings announcement, a seven-trading-day interval was required between the earnings announcement and the financial statement release date. Recent studies have shown that the most significant portion of the market reaction to earnings announcements is complete within thirty minutes of the announcement time (e.g., Lee [1992], and Mucklow [1990]). Thus, the seven-day interval between the earnings announcement and 17

19 the release of financial statements is conservative. Requiring the seven-day interval resulted in a sample of 4,420 firm-quarter observations. To maintain comparability with the prior studies, the sample is restricted to firms in SIC codes 1000 to This restricts the sample to a broad definition of manufacturing firms and leaves 3,355 observations in the sample. To use the Ohlson [1980] model, the sample of firms had to be modified. In addition to the data requirements discussed in section four, further accounting information is required to estimate the probability of failure. The additional data were collected from the Compustat Annual tape, except for GNP price-level data which were hand-collected. Collecting the additional required data, identified in equation (9), reduced the sample to 2,770 firm-quarter observations. 4.1 DESCRIPTIVE STATISTICS Summary statistics for the sample of 2,770 firm-quarter observations are presented in Panel A of Table 2. [ Insert Table 2 about here ] Although the sample is subject to survivorship bias, however, the final sample still has a broad diversity of firms with respect to total assets, revenues and earnings. The distribution of each of the variables is skewed to the right; a frequent situation in samples of financial data. To reduce problems associated with the skewness of the variable distributions, each of the variables is deflated by Total Assets. The distribution of firm-quarter observations among the fiscal quarters is presented in Panel B of Table 2. The results of estimating equation 7 are presented in Table 3 for both the SCFP and the BSIS cash-flow proxies. More of the variation in the cash-flow proxy is explained by the forecasting model when using the BSIS proxy. It is important to note, however, that 18

20 goodness-of-fit statistics provide little information regarding whether one of the two proxies reflects the market expectation of cash-flows better. [ Insert Table 3 about here ] Statistics regarding the independent variables and the estimated probabilities of bankruptcy are presented in Table 4. Similar statistics from Burgstahler, Jiambalvo and Noreen [1989] are presented for comparison. [ Insert Table 4 About Here ] The similarity of the current study s parameter estimates to those of BJN [1989] in Table 4 provides some assurance that the BJN result and the statistic, φ j,t, are relevant to the current study s sample of firm-observations. BJN [1989] demonstrated the relationship between changes in the probability of default and abnormal return in long-window tests (12- month). The current study uses narrow-window tests (9-day) to examine the relationship between unexpected cash-flows and abnormal returns, within partitions of the entire sample based upon the probability of default. 5. Financial Distress and the Cash-flow Announcement Table 5 reports the relation between unexpected cash-flows and abnormal returns for the entire population of firm-quarter observations. Consistent with Wilson [1987], the relationship between unexpected cash-flows and abnormal returns is statistically significant. It is somewhat stronger when using the SCFP cash-flow proxy, however, very similar parameter estimates are obtained using either proxy. Before drawing conclusions from the significant coefficient in Table 5, it should be noted that in the sample of 3,355 observations (the sample immediately before applying the data requirements necessary to calculate the φ j,t measure of financial distress), both cash-flow proxies have insignificant coefficient estimates 19

21 (not reported). Also, the ability of both regressions in Table 5 to explain the variability in abnormal return is extremely limited. [ Insert Table 5 About Here ] Throughout the remaining tests, and consistent with the results in Table 5, the SCFP proxy demonstrates a stronger relationship with abnormal returns than does the BSIS proxy. Results reported for the remainder of the current study relate only to the SCFP cash-flow proxy. The mixed signals regarding the information content of cash-flows accurately reflect the current state of the research. The relation between unexpected cash-flows and abnormal returns is not a robust and cannot be generalized across all firms. This observation motivates a search for situations where cash-flow can reasonably be expected to be more important in the valuation of firms securities. The current study examines how a firm s level of financial distress affect the relationship between unexpected cash-flows and abnormal returns. The larger population of firm-quarter observations is partitioned according to this firm-specific characteristic. The partitions are likely to exhibit more internally consistent reactions to cash-flow announcements. 5.1 FIRM-SPECIFIC FINANCIAL DISTRESS Observations were categorized by the magnitude of their estimated probability of failure within one year, φ j,t. Category one includes the large majority of firms (93.1 percent) with estimated probabilities less than or equal to 0.10 (N=2,579). Category two includes firms with probabilities between 0.10 and 0.20 (N=118). Categories three through five include those firms with probabilities in the ranges: 0.20 to 0.30 (N=31), 0.30 to 0.40 (N=13), and greater than 0.40 (N=29), respectively. Partitions of the data were used, as opposed to including the φ j,t measure as an independent variable in an OLS regression, to allow for the possibility that the valuation 20

22 relevant effects of unexpected cash-flow information are nonlinear. That is, cash-flow might not be important in valuing a firm s securities unless the current period s level of financial distress is beyond some threshhold. If a firm is relatively healthy and stable, taking in more cash than is required for continuing working capital requirements, earnings may be a better predictor of future cash-flows and firm value. However, if a firm s financial health is relatively weak, the realizations of each period s cash-flows might be directly related to that firm s ability to continue as a going concern. The ability to continue as a going concern will affect asset values and, therefore, firm value. Table 6 reports the results of a regression of unexpected cash-flow and dummy variables representing intercept changes and slope changes for the five categories of firms relating to the magnitude of probability of failure, φ j,t. Consistent with a nonlinear relationship between the slope coefficient on unexpected cash-flows and the probability of default, the variation in the slope coefficients is observed within a small portion of the larger sample. The coefficient on the first category, where default risk is low, is not significantly different from zero. The coefficients on the next three categories are statistically significant, positive, and increasing in default-risk (statistical significance of the parameter estimates is also increasing in defaultrisk). The fifth category is somewhat anomalous. An investigation of the firms in this category found that many of these firms were in extremely troubled financial states. The extremity of their condition likely required different information dissemination practices than those for healthy firms. 14 The financial statement release date may not have been a timely enough information source for the firms in this category. [ Insert Table 6 About Here ] 14 Some of these firms were already in receivership. A receiver has ready access to information regarding cash and acts on behalf of creditors. In this setting, the creditors are not waiting for periodic announcements of this information to update their beliefs and change their actions. 21

23 The adjusted R 2 from Table 6 (1.23 %) suggests that unexpected cash-flows explain little of the variation in stock prices surrounding the release of financial statements. The large number of firms in category one (for which there is no evidence of any important relationship) may be distorting the true ability of unexpected cash-flows to explain abnormal returns for firms experiencing higher levels of default risk. When the regression is run without the category one firms (not reported), the adjusted R 2 rises to 7.74 percent. Running the regression without both the category one and the category five firms results in an adjusted R 2 of 9.92 percent. The pattern of parameter magnitude and statistical significance remains when either or both of the category one and category five firms are omitted from the regression. Adjusted R 2 statistics of eight to ten percent are comparable to those found throughout much of the literature regarding the relationship between unexpected earnings and abnormal returns. To examine the robustness of the default-risk results, Spearman correlations were calculated by default-risk category. These nonparametric results also suggest a coefficient on unexpected cash-flows that increases in default-risk. The significance levels of the nonparametric correlations are weaker than the parametric tests. The Spearman correlation coefficients and their associated significance levels for categories one through five are: (1) , α 0.26; (2) , α 0.19; (3) , α 0.11; (4) , α 0.22; and (5) , α 0.86, respectively. The pattern in the coefficient magnitudes remains, with the exception of the negative coefficient on the second category. Whether to accept the results of the parametric or the nonparametric tests is unclear. The validity of parametric tests requires that the independent variables be distributed approximately normally. Kolmogorov-D and Shapiro-Wilk statistics suggest rejecting the hypothesis that the data in the default-risk categories are obtained from normal distributions. A trade-off must be made between the robustness of the nonparametric tests to extreme observations versus their inability to use the information contained in the magnitude of the measurements. 22

24 5.2 PROBABILITY OF DEFAULT VERSUS OTHER DISTRESS MEASURES Both Lang and McNichols [1990] and Frankel [1992] use bond ratings to measure a firm s level of financial distress. If bond ratings are primarily a function of a firm s ability to make cash payments, consistent with the probability of default measure, results using the two measures should also be consistent. This section examines different measures of financial distress and the implications of using these measures. The measures examined include the default probability, φ j,t, a firm s S&P bond rating as used in previous studies, a firm s S&P stock rating and a firm s size as measured by the log of total assets. One problem identified earlier with using bond ratings to explain returns to equity holders is that the bond rating is primarily concerned with risks facing the debt holders. The S&P stock rating addresses this issue and is intended as a variable relevant to a firm s equity holders. Size is a variable that has been shown to explain a significant portion of the variability in bond ratings (Kaplan and Urwitz [1979]). Spearman correlations among the measures of financial distress are reported in Table 7. Spearman correlations are calculated because bond ratings and stock ratings are categorical variables with little or no cardinal interpretation. [ Insert Table 7 About Here ] Two observations are noteworthy. First, requiring bond rating data reduces the sample by 58.5 percent, while requiring stock rating data reduces the sample by about 15.6 percent. The sample sizes in the high default-risk partitions are small before requiring bond- or stockrating data. Second, although the correlations are significant at traditional levels and are consistent with intuition with respect to their sign, the magnitude of the correlation coefficients is not as high as might be expected. The highest correlation is approximately 70 percent. These levels leave opportunities for the different measures to capture different information. The combination of the reduced sample size and the lower than expected 23

25 correlation coefficients could explain the different findings in the current versus Lang and McNichols [1990]. Panel B of Table 7 shows that the correlations among the various measures of financial distress are weaker when considering only those observations with a probability of default greater than 10 percent. The sample firms, for which bond ratings are available on the Compustat tape, are partitioned into groups according to their bond rating. The various categories of bond ratings were reordered, to make ordinal sense, from 2 (best) to 24 (worst). Summary statistics on the probability of default for each category are reported in Table 8. [ Insert Table 8 About Here ] The statistics in Table 8 can be used to illustrate one possible reason for the lack of any observed relationship between unexpected cash-flows and abnormal returns in earlier studies. The partitioning of the samples in the earlier studies included much larger proportions of the entire sample than the partitioning used in the current study. The current study finds a relationship exists within the most financially distressed 6.9 percent of the sample. All of the observed relationships between unexpected cash-flows and abnormal returns are observed in a sample the size of the union of bond rating categories 19 through 24. The partitions used in earlier studies did not accurately replicate the partition used in the current study. If the sample were divided into quintiles according to bond rating categories, as in Lang and McNichols [1990], the lowest-rating category would include all bond rating categories from 16 through 24. This would roughly triple the number of observations in this partition (79 obs. vs 230 obs.). If the sample were partitioned according to investment-grade versus junk bonds (below investment grade), as in Frankel [1992], then the partition would include categories 13 through 24 and the number of observations in this partition would roughly quadruple relative to the proportion of the entire sample examined in the current study. 24