Financial Distress and the Earnings-Sensitivity-Difference Measure of Conservatism

Authors


  • Earlier versions of this paper have benefited from comments of an anonymous reviewer and of participants at seminars at the University of Essex, the University of Cambridge, the National Taiwan University, the European Accounting Association Congress (Lisbon) and the British Accounting Association Conference (Dundee).

Audrey Wen-Hsin Hsu is an Assistant professor in the Department of Accounting, National Taiwan University; John O'Hanlon (j.ohanlon@lancaster.ac.uk) is Professor of Accounting and Ken Peasnell Distinguished Professor of Accounting in the Lancaster University Management School.

Abstract

Following Basu (1997), the difference between the sensitivity of accounting earnings to negative equity return (proxy for bad news) and its sensitivity to positive equity return (proxy for good news) is interpreted as an indicator of conditional accounting conservatism. However, there is concern that the earnings-sensitivity difference (ESD) may be affected by factors other than conditional conservatism, and that this may impair its reliability as an indicator of conditional conservatism. Motivated by such concerns and by recognition that financial distress could contribute to an ESD through a conditional-conservatism route and/or through a non-conditional-conservatism route, we examine the association between financial distress and the ESD for U.S. non-financial firms. By decomposing the association into an element arising from accruals, which can reflect conditional conservatism, and an element arising from cash flow from operating activities (CFO), which cannot directly reflect conditional conservatism, we seek evidence as to whether such association arises through a conditional-conservatism route or through a non-conditional-conservatism route. We find that positive association between financial distress and the ESD arises predominantly through the accruals component of earnings rather than the CFO component, consistent with it arising primarily because of a higher degree of conditional conservatism in relatively financially distressed firms. The inference that there is a positive association between financial distress and conditional conservatism is supported by other non-equity-return-based measures of conditional conservatism. The evidence in this paper suggests that the effect of financial distress does not significantly impair the reliability of the ESD as an indicator of conditional conservatism.

Basu (1997) defines accounting conservatism as a tendency on the part of accountants ‘to require a higher degree of verification for recognizing good news than bad news in financial statements’, resulting in accounting earnings being more timely in its recognition of bad news than in its recognition of good news. This concept of conservatism is sometimes termed conditional conservatism. Following Basu, a piecewise-regression-based measure of the excess of the sensitivity of accounting earnings to contemporaneous negative equity return over its sensitivity to contemporaneous positive equity return is widely interpreted as an indicator of asymmetric timeliness in the accounting recognition of bad news and good news and, therefore, of conditional conservatism. Many studies have reported positive association between the Basu earnings sensitivity difference (ESD) and likely contracting-related sources of demand for conditional conservatism. The ESD has been found to be associated with measures of debt-contracting-related demand for accounting conservatism (Ball et al., 2008; Beatty et al., 2008; Zhang, 2008), measures of governance-related demand for accounting conservatism (Ball et al., 2000; Beekes et al., 2004; Ahmed and Duellman, 2007; LaFond and Roychowdhury, 2008; LaFond and Watts, 2008; García Lara, García Osma and Penalva, 2009), the expertise of auditors (Krishnan, 2005), the litigation environment (Givoly and Hayn, 2000; Bushman and Piotroski, 2006; Lobo and Zhou, 2006) and a change in rate regulation (Sivakumar and Waymire, 2003). The ESD has also been found to be associated with the conservatism of previous periods (Pae et al., 2005; Roychowdhury and Watts, 2007) and differences in earnings constructs with respect to predicted timeliness (Pope and Walker, 1999), consistent with a conditional-conservatism interpretation of the ESD. However, some studies have questioned the conditional-conservatism interpretation of the ESD on the grounds that the ESD can be affected by factors not directly related to conditional conservatism. Such studies include Dietrich et al. (2007) and Patatoukas and Thomas (2009), which suggest that the ESD can arise from econometric phenomena associated with the partitioning of data with respect to the sign of equity return, and Givoly et al. (2007), which suggests that the ESD is affected by the nature and clustering of economic events and by firms' disclosure policies.

In light of concerns that have been expressed about the conditional-conservatism interpretation of the ESD and the possibility that the effect of financial distress could give rise to an additional source of concern regarding this interpretation, we examine in this paper the association between measures of financial distress and the ESD for U.S. non-financial firms. We recognize that association between the ESD and financial distress could arise through a conditional-conservatism route if relatively financially distressed firms exhibit a relatively high degree of conditional accounting conservatism. We also recognize that it could arise through a non-conditional-conservatism route because of possible financial-distress-related non-linearity in the response of equity prices and returns to news reflected in equity earnings, or because firms' cash-flow response to bad news and good news is related to financial distress. By decomposing the association between the ESD and measures of financial distress into an element arising from accruals, which can reflect conditional conservatism, and an element arising from cash flow from operating activities (CFO), which cannot directly reflect conditional conservatism, we seek evidence as to whether such association arises through a conditional-conservatism route and/or through a non-conditional-conservatism route. We also provide additional evidence on the association between financial distress and conditional conservatism based on non-equity-return-based measures of conditional conservatism.

We find a positive association between our financial-distress measures and the ESD. We find that this association arises predominantly through the accruals component of earnings rather than through the CFO component, consistent with the association arising primarily because relatively financially distressed firms exhibit a relatively high degree of conditional conservatism. The inference that there is a positive association between financial distress and conditional accounting conservatism is supported by evidence from the non-equity-return-based measures of conditional conservatism, and is robust to a number of methodological variations. Overall, the evidence in this paper suggests that the effect of financial distress does not significantly impair the reliability of the ESD as an indicator of conditional accounting conservatism.

POSSIBLE SOURCES OF ASSOCIATION BETWEEN FINANCIAL DISTRESS AND THE EARNINGS-SENSITIVITY DIFFERENCE (ESD)

We recognize that financial distress could be associated with the ESD because of association between financial distress and conditional conservatism and/or because of financial-distress-related non-linearity in the earnings-return relationship that is not driven by conditional conservatism. In this section, we describe conditional-conservatism and non-conditional-conservatism routes by which financial distress might be associated with the ESD, and outline relevant evidence from prior studies.

Possible Association Between Financial Distress and the ESD Through Conditional Accounting Conservatism

It is likely that debt-contracting is a principal source of demand for conservative accounting. Conservative accounting tends to activate lenders' rights under financial-reporting-based debt covenants sooner rather than later in response to adverse events, thereby allowing lenders to impose more timely constraints on the ability of managers and shareholders to reduce the value of lenders' claims. Also, because of links between earnings and dividend payments, conservative accounting may limit the potential adverse effects of conflicts between debtholders and shareholders with regard to distributions. Empirical evidence of debt-related demand for conservative accounting can be found in a number of studies. Ball et al. (2008) predict a positive association between the size of countries' debt markets relative to their GNP and the degree of conditional accounting conservatism observed in those countries. Their prediction is based on the expectation that demand for accounting to produce timely violation of debt covenants in response to adverse events is relatively high where debt markets are relatively important. Their empirical results confirm their prediction. Zhang (2008) predicts that relatively conservative accounting increases the probability that covenant violations will be triggered in the wake of negative price shocks, and that part of the ex ante benefit of conservative accounting by borrowers accrues to the borrowers in the form of reduced cost of debt. Using a number of measures of conservatism, including the Basu (1997) ESD, she reports evidence in support of these predictions. The study by Ahmed et al. (2002) is motivated by the potential for large dividends to cause wealth transfers from debtholders to shareholders because of increases in default risk, and the potential for accounting conservatism in conjunction with balance-sheet-ratio-based covenants to act as an indirect constraint on such wealth transfers. That study uses accounting-conservatism measures other than the ESD and a number of proxies for the likelihood of debtholder–shareholder conflict to examine the association between potential debtholder–shareholder conflict with regard to dividend policy and accounting conservatism. It reports that firms that face relatively severe debtholder–shareholder conflict over dividend policy tend to be more conservative in their accounting. Motivated by the Guay and Verrecchia (2006) argument that lenders and borrowers can write contracts that make financial-reporting conservatism unnecessary for debt-contracting purposes, Beatty et al. (2008) report evidence of a debt-contracting-related demand for financial-reporting conservatism even where debt-contract provisions are based on conservatively adjusted GAAP. Evidence of the likely benefit of conditional conservatism for financially distressed firms can be found in the study by García Lara, García Osma and Neophytou (2009). Using measures of conservatism other than the ESD, that study reports that U.K. firms that eventually failed had a history of relatively low conditional conservatism in years preceding failure.

From the foregoing, there is evidence of demand for conservative accounting to reduce the cost of debtholder–shareholder conflicts by forcing the timely activation of debtholder rights under debt covenants in the wake of adverse events and by preventing dividend payments that might result in wealth transfers from debtholders to shareholders. Pressure for conservative accounting from this source is likely to be associated with the probability of default. It is therefore possible that an association between financial distress and the ESD could arise because relatively financially distressed firms adopt relatively conservative financial-reporting practices.

Possible Association Between Financial Distress and the ESD Other Than Through Conditional Accounting Conservatism

Shareholders' termination options can cause wealth changes associated with the equity earnings of poorly performing firms to be borne in part by stakeholders other than shareholders and, related to this, can cause losses to be transitory. This can cause the equity return of poorly performing firms to be relatively insensitive to news reflected in equity earnings. This raises the possibility that association between financial distress and the ESD arises because financial distress affects the properties of equity return as an indicator of news rather than because it induces a greater degree of conditional conservatism in accounting.

A number of studies, both empirical and analytical, have provided evidence that non-linearity in the relationship between equity price or equity return and equity earnings can be attributed to factors related to termination options. Burgstahler and Dichev (1997) report a convex relationship between the market value of equity (vertical axis) and earnings (horizontal axis), and attribute the flattening of the market value-earnings relationship to the availability of adaptation options in respect of unprofitable assets. Hayn (1995) reports that the slope of the relationship between equity returns and earnings is substantially lower for loss cases than for profit cases. This is attributed to the existence of liquidation options that are more likely to be exercised in the presence of losses. Plummer and Tse (1999) report that the sensitivity of equity returns (bond returns) to changes in earnings decreases (increases) monotonically from the highest-credit-rated firms to the lowest-credit-rated firms. This suggests that, as the probability of liquidation increases, the bondholders bear an increasing proportion of the value changes reflected in equity earnings. Subramanyam and Wild (1996) report that the response of abnormal equity return to changes in earnings is relatively low for relatively financially distressed firms. This is attributed to the lower informativeness of earnings where the going-concern status may be in question. Dhaliwal and Reynolds (1994) report that the slope coefficient from a reverse regression of unexpected earnings on abnormal equity return is positively associated with default risk as inferred from bond ratings, consistent with abnormal equity returns of relatively high-default-risk firms being relatively unresponsive to earnings. Fischer and Verrecchia (1997) demonstrate analytically that the limited liability of shareholders gives rise to convexity in the relationship between equity price (vertical axis) and earnings (horizontal axis). Valuable insights regarding the impact of default risk on the return-earnings relationship are provided by Beaver and Ryan (2009). They demonstrate that, if realized returns on debt become increasingly sensitive to realized returns on net operating assets as the latter become smaller and as the probability of exercise of the shareholders' put option on the net operating assets increases, realized equity returns are a convex function of realized returns on net operating assets. They also observe that, prior to the introduction of SFAS 159, The Fair Value Option for Financial Assets and Financial Liabilities (FASB, 2007), U.S.-GAAP net income reflected return on net operating assets but could not reflect the share of that return borne by the debtholders when debt is risky. This leads to the prediction that risky debt induces a convex relationship between equity returns (vertical axis) and net income (horizontal axis). The connection between default risk and convexity in equity prices is also highlighted by Easton et al. (2008).

The existence of financial-distress-related convexity in the relationship between equity price or equity price changes (vertical axis) and news reflected in earnings (horizontal axis) could give rise to financial-distress-related concavity in the relationship between earnings (vertical axis) and equity returns (horizontal axis) not directly due to conditional conservatism.

Another possible source of non-conditional-conservatism-related association between financial distress and the ESD is an effect referred to by Ball et al. (2009). They observe that the greater sensitivity of CFO to bad news than to good news that has been reported in a number of studies could have arisen because firms' corrective actions in response to adverse events resulted in operating cash outflows that occurred relatively immediately in response to those adverse events. If such an effect were associated with financial distress, it could also give rise to a non-conditional-conservatism association between financial distress and the ESD.

Prior Evidence Relevant to the Source of Association Between Financial Distress and the ESD

The seminal paper by Basu (1997) briefly considers the possibility that the effect of termination options documented by Hayn (1995), rather than conditional conservatism, might be the source of the ESD. He reports that, although the termination-option theory predicts some of his results, it does not predict all of them. Among other things, he reports a higher sensitivity difference for earnings than for cash flows, and notes that the termination-option theory does not predict this result whereas the conditional-conservatism theory does.

Using U.S. data for periods from 1999 to 2006, Wang et al. (2009) seek evidence as to whether the ESD is affected by financial-distress-related distortion of the return measure. They do so by comparing (a) the association between a measure of financial distress and the ESD where the ESD is based on equity return and equity earnings with (b) the corresponding association where the ESD is based on estimates of entity return and on pre-interest earnings. A rationale for this comparison is that an entity-return measure that accurately impounded all periodic changes in the value of the entity, including changes in the value of creditors' claims that arise because value changes reflected in equity earnings are borne partly by creditors, would not be subject to the potential financial-distress-related distortions referred to above. To do this comparison properly, however, it is necessary to construct accurate series of liability-value-inclusive periodic entity returns, which is very difficult to do for a large sample of firms. Wang et al. (2009) report that there is an association between financial distress and the normal equity-return-based ESD but no association between financial distress and the ESDs that are based on their estimates of entity return, consistent with the association between financial distress and the equity-return-based ESD arising primarily from a non-conservatism route.

Summary

From the foregoing there is reason to believe that financial distress could be associated with the ESD, and that such association could arise either through a conditional-conservatism route and/or through a non-conditional-conservatism route. Evidence on the role of financial distress with respect to the ESD appears mixed. In the remainder of this paper, we seek further evidence, based on a decomposition of the ESD into an element due to accruals and an element due to CFO, as to whether association between financial distress and the ESD arises through a conditional-conservatism route and/or through a non-conditional-conservatism route.

RESEARCH DESIGN

Measurement of Association Between Financial Distress and the Earnings-Sensitivity Difference (ESD)

Our principal source of evidence on whether association between financial distress and the ESD arises through a conditional-conservatism route or through a non-conditional-conservatism route is a decomposition of the association into an element due to the accruals component of earnings and an element due to the CFO component. The rationale for this is as follows. As outlined in the previous section, we envisage that financial distress could contribute to the ESD through a number of routes. First, financial distress could contribute to the ESD because it affects the degree of conditional conservatism with which firms account for their activities, for example, by inducing relatively financially distressed firms to recognize asset impairment on a relatively timely basis. Any contribution by this route must operate through accruals and not through CFO.1 Second, financial distress could contribute to the ESD because it causes equity return to be relatively unresponsive to news reflected in the equity earnings of relatively poorly performing firms. As noted by Basu (1997), there is no reason to predict that such an effect on the ESD would be concentrated in one or other of the accrual and CFO elements of the ESD. Third, financial distress could affect the ESD because it influences the phenomenon envisaged by Ball et al. (2009), whereby firms' CFO response to good news and bad news is asymmetric. Any contribution by this route must operate through CFO. In summary: any association directly due to conditional accounting conservatism must be reflected in the accruals component of earnings and not in the CFO component; any association due to equity return being relatively unresponsive to news reflected in the equity earnings of relatively poorly performing firms could be reflected in the accruals component and the CFO component; and any effect due to asymmetric CFO response to good news and bad news must operate through the CFO component. On the basis of this, we conclude that evidence that association between financial distress and the ESD arises predominantly from accruals would suggest that the association arises primarily through a conditional-conservatism route, and that evidence that it derives predominantly from CFO or in equal measure from accruals and CFO would suggest that it arises to a significant degree through a non-conservatism route.

In common with many previous studies, the earnings construct for which we measure the ESD is earnings before extraordinary items (EBEI). We also measure the ESD with a control for the beginning-of-period book-to-market ratio. We do so in light of evidence in Pae et al. (2005) and Roychowdhury and Watts (2007) that conditional conservatism as measured by the ESD is relatively high when the cumulative conservatism of earlier periods is relatively low, as reflected in a relatively high book-to-market ratio. We measure the association between financial distress and the ESD for EBEI by including a financial-distress interaction term in the ESD-measurement regression models. We investigate whether any association between financial distress and the ESD arises though a conditional-conservatism route or through a non-conditional-conservatism route by estimating regression models of the same forms for the accruals component of EBEI and the CFO component of EBEI. In each of our regression models, we use pooled cross-section and time-series data. Following Petersen (2008) and Gow et al. (2010), t- statistics for all of our regression models are based on standard errors clustered by firm and by year.2 As a sensitivity test, we also estimate each regression model separately for each of the seventeen years covered by the study without allowing for clustering, with test statistics being based on the seventeen-year averages of the regression coefficients. Because the explanatory variables in each of the ESD-measurement regression models described below are the same for EBEI, accruals and CFO and because EBEI is defined to be the sum of accruals and CFO, a regression coefficient for EBEI from any of these models must equal the sum of the corresponding regression coefficients for accruals and CFO. This means that the coefficients measuring the ESD and the effect of financial distress on the ESD are each equal to the sum of the corresponding coefficients for accruals and CFO.

In order to aid comparability with other studies, we first estimate models that measure the ESD without taking account of financial distress. Model (1) is the standard Basu (1997) regression model, and model (2) is the same model with a control for the beginning-of-period book-to-market ratio:

image(1)
image(2)

where Xit is either EBEI, CFO or accruals, where accruals are defined as EBEI less CFO, for firm i for the accounting period ended at balance-sheet date t, scaled by beginning-of-period market value; Rit is the equity return for firm i for the accounting period ended at balance-sheet date t; DRit is a dummy variable that takes the value of one where Rit is negative, and 0 otherwise; BMi,t−1 is the annually computed percentile rank of the book-to-market ratio for firm i as at the beginning of the accounting period ended at balance-sheet date t; the β terms are regression coefficients; and the ε terms are error terms. Where earnings is the dependent variable, the coefficient β1,4 in (1) is the Basu (1997) ESD. A significant positive value for this coefficient for earnings is conventionally interpreted as evidence that earnings is more timely in its recognition of bad news than in its recognition of good news, consistent with conditional conservatism. In (2), the coefficient β2,8 reflects the effect of the beginning-of-period book-to-market ratio on the sensitivity difference and β2,4 measures the sensitivity difference after controlling for the book-to-market ratio.

Models (3) and (4) address the key object of interest in this study. Through interaction terms for financial distress, they measure the effect of financial distress on the sensitivity differences for EBEI, accruals and CFO:

image(3)
image(4)

where FDi,t−1 is the annually computed percentile rank of a financial-distress measure, for which further details are given below, as at the beginning of the accounting period ended at time t, and other notation is as previously defined. In (3), the principal focus of interest is β3,8, which reflects the effect of financial distress on the sensitivity difference. A significant positive value of β3,8 for EBEI would indicate a positive association between financial distress and the ESD. A significant positive value of β3,8 for CFO would indicate that association between financial distress and the ESD arises through CFO, consistent with such association arising at least in part through a non-conditional-conservatism route. A significant positive value of β3,8 for accruals would indicate that association between financial distress and the ESD arises through accruals. In the absence of a corresponding significant coefficient for CFO, this would be consistent with the association arising primarily because relatively financially distressed firms exhibit a relatively high degree of conditional conservatism. In model (4), which includes a control for the beginning-of-period book-to-market ratio, the principal focus of interest is β4,8. This reflects for each of the dependent variables the effect of financial distress on the sensitivity difference in the presence of a control for the book-to-market ratio.

Measurement of Association Between Financial Distress and Non-Equity-Return-Based Measures of Conditional Conservatism

For additional evidence on whether financial distress is associated with conditional conservatism, we use two measures that do not use equity returns. The first of these measures, due to Basu (1997), reflects the difference between the next-period reversal of negative earnings changes and the next-period reversal of positive earnings changes. The rationale for this measure is that the recognition of a large proportion of a time t−1 economic event in the earnings of time t−1 will cause, other things equal, a substantial reversal at time t in the earnings change that occurred at time t−1. Under conditional conservatism, the accounting recognition of bad news, giving rise to negative earnings changes, is more timely than the accounting recognition of good news, giving rise to positive earnings changes, and it is therefore expected that the next-period reversal of negative earnings changes will be greater than the next-period reversal of positive earnings changes. The earnings-change-reversal measure of conditional conservatism is given by the following regression model:

image(5)

where ΔEBEIit is the change in EBEI of firm i from the accounting period ended at balance-sheet date t−1 to the period ended at balance-sheet date t, scaled by total assets at balance-sheet date t−1; DΔEBEIi,t−1 is a dummy variable that takes the value of one where ΔEBEIi,t−1 is negative and 0 otherwise; and other notation is as previously defined. β5,4 indicates whether earnings-change reversal is greater for negative earnings changes than for positive earnings changes: β5,4 < 0 implies more timely recognition of bad news than of good news, consistent with conditional accounting conservatism. The addition of financial-distress interaction terms in model (6) enables us to observe the effect of financial distress on conditional conservatism as measured by reference to earnings-change reversal:

image(6)

where notation is as previously defined. β6,8 indicates whether conditional conservatism measured by reference to earnings-change reversal is associated with financial distress: β6,8 < 0 implies that it is more pronounced for relatively financially distressed firms. If such a financial distress effect is attributable to accounting conservatism, then this effect should be observable in the accruals element of earnings and not in the cash-flow element. As supplementary evidence with respect to results from model (6), we therefore estimate regression models (7) and (8). These models are of the same form as model (6), but with change in accruals and change in CFO, respectively, in place of change in earnings. The models are as follows:

image(7)
image(8)

where ΔACCitCFOit) is the change in accruals (CFO) of firm i from the accounting period ended at balance-sheet date t−1 to the period ended at balance-sheet date t, scaled by total assets at t−1; DΔACCi,t−1 (DΔCFOi,t−1) is a dummy variable that takes the value of one where ΔACCi,t−1CFOi,t−1) is negative and 0 otherwise; and other notation is as previously defined.

The second of our non-equity-return-based measures follows Ball and Shivakumar (2005). It comprises the difference between the sensitivity of accruals to negative CFO and the sensitivity of accruals to positive CFO. The rationale for this measure is as follows. As indicated by Dechow (1994) and Subramanyam (1996), there is a mechanical negative association between accruals and CFO. In bad-news periods as proxied by negative cash flows, this negative association is likely to be reduced because economic losses are recognized on a timely basis through negative accruals whereas economic gains are recognized only when realized and are accounted for on a cash basis. The accruals-CFO-based measure of conditional conservatism is given by the following regression model:

image(9)

where ACCit is accruals for firm i for the accounting period ended at balance-sheet date t, scaled by beginning-of-period total assets; CFOit is CFO for firm i for the accounting period ended at balance-sheet date t, scaled by beginning-of-period total assets; DCFOit is a dummy variable that takes the value of one where CFOit is negative and 0 otherwise; and other notation is as previously defined. β9,4 indicates whether the response of accruals to contemporaneous CFO is less negative for negative CFO than for positive CFO: β9,4 > 0 implies more timely recognition of bad news than of good news, consistent with conditional accounting conservatism. The addition of financial-distress interaction terms in model (10) enables us to observe the effect of financial distress on conditional conservatism as measured by the accruals-CFO-based measure:

image(10)

where notation is as previously defined. β10,8 indicates whether conditional conservatism measured by reference to the response of accruals to cash flows is associated with financial distress: β10,8 > 0 implies that it is more pronounced for relatively financially distressed firms.

Measures of Financial Distress

In estimating regression models (3), (4), (6), (7), (8) and (10), we use two measures of financial distress. The first of these is based on recent option-pricing-model-based developments in the measurement of financial distress, and the second is a more longstanding measure. First, we use the annually computed percentile rank of the Hillegeist et al. (2004) BSM Score for firm i as at the beginning of the accounting period ended at time t. The BSM Score is derived from the option-pricing literature pioneered by Black and Scholes (1973) and Merton (1974), hence the acronym ‘BSM’. It relies in particular on the insight developed in Vassalou and Xing (2004) that the equity of a firm can be viewed as a call option on the assets of the firm, where the exercise price of the option is equal to the face value of the firm's liabilities and the probability that the call option is not exercised by the shareholders is interpreted as a probability of bankruptcy. The BSM Score is derived from the following adaptation of the Black–Scholes–Merton option pricing model, which values equity, denoted by VE, as a European call option on the firm's assets:

image((11.a))

where

image((11.b))

and

image((11.c))

In (11.a), (11.b) and (11.c): VA denotes the current market value of assets, treated as the underlying asset on which the call option is written; N(.) denotes the cumulative density function of the standard normal distribution; L denotes the value of liabilities maturing at time T, treated as the exercise price of the call option; r is the continuously compounded risk-free rate of return; δ is the continuous dividend rate, expressed as a proportion of the current market value of assets; σA is the standard deviation of asset returns. From the above, Hillegeist et al. (2004) define as in (12) below the ‘BSM Prob’. This is derived from d2 in (11.c), which is the probability that the value of the liabilities at their maturity will exceed the value of the assets at that date and that the call option will not be exercised, and which is treated as a measure of the probability of default:

image(12)

where µ is the continuously compounded expected rate of return on the assets, which is used in place of the risk-free rate of return from (11.c). For the purpose of their empirical tests, Hillegeist et al. find it convenient to re-express their BSM Prob as a BSM Score as follows3:

image(13)

As Hillegeist et al. observe, the negatively signed term within brackets on the right-hand side of (12) comprises the logarithm of the ratio of the firm's assets to its liabilities (=ln(VA/L)) plus the expected annual growth in asset values (inline image), deflated by a term reflecting the volatility in asset returns. Recently developed option-pricing-model-based indicators of financial distress such as the Hillegeist et al. BSM Score differ from more traditional financial-distress measures, such as the Z Score (Altman, 1968) and the O Score (Ohlson, 1980), in the important regard that they reflect volatility of return. Evidence in Hillegeist et al. (2004) suggests that the BSM Score is more effective than the Z Score and the O Score in predicting bankruptcy.

Second, we also use as a measure of financial distress the annually computed percentile rank of the Altman (1968) Z Score for firm i as at the beginning of the accounting period ended at balance-sheet date t. As higher values of the Z Score are likely to be associated with lower levels of financial distress, we multiply the Z Score by −1. The Altman (1968) Z Score is defined as follows:

image(14)

where Zit is the Z Score for firm i at balance-sheet date t, B1,it is current assets less current liabilities all divided by total assets, B2,it is retained earnings divided by total assets, B3,it is earnings before interest and taxes divided by total assets, B4,it is market value of equity divided by book value of total liabilities and B5,it is sales divided by total assets.

DATA AND RESULTS

Data

Our data are drawn from non-financial firms listed on the New York Stock Exchange and the American Stock Exchange for any part of the period from 1989 to 2005. In using New York Stock Exchange and American Stock Exchange firms, we are consistent with Basu (1997). Our data start in 1989, which is the first full year for which U.S. firms were required to publish a Statement of Cash Flows in accordance with SFAS 95, Statement of Cash Flows (FASB, 1987).4 This avoids the need to estimate CFO indirectly from income-statement and balance-sheet data, and the consequent danger that the CFO measure might be contaminated by accruals. Our data predate the introduction of SFAS 159, The Fair Value Option for Financial Assets and Financial Liabilities (FASB, 2007), so the influence of the fair-value option in allowing the carrying value of firms' liabilities to reflect periodic changes in default risk is absent in our data.

Annual equity returns are obtained from CRSP. For EBEI we use Compustat data item 18 (Income Before Extraordinary Items). On the basis of our own comparison of Compustat data with a sample of published financial statements, we follow the approach recommended by Hribar and Collins (2002) in order to measure CFO and accruals. For CFO, we use Compustat data item 308 (Operating Activities-Net Cash Flow (Statement of Cash Flows)) less Compustat data item 124 (Extraordinary Items and Discontinued Operations (Statement of Cash Flows)). Accruals are then defined as EBEI less CFO.5 In regression models (1), (2), (3) and (4), EBEI, accrualsand CFO are all scaled by beginning-of-period market value of equity, equal to the product of the number of common shares outstanding and the closing share price as reported by Compustat (Compustat data item 25 (Common Shares Outstanding) times Compustat data item 199 (Price-Fiscal Year-Close)). In regression models (5) to (10), EBEI items, accruals items and cash-flow items are scaled by Compustat data item 6 (Total Assets) at time t−1. The book-to-market ratio is equal to the book value of equity (Compustat data item 60 (Common Equity-Total)) divided by the market value of equity as defined above. In order to calculate the Hillegeist et al. (2004) BSM Score, we use the following inputs for the accounting period ended at time t−1: for total liabilities, denoted L, Compustat data item 181 (Liabilities-Total); market value of equity at the balance-sheet date (as defined above); dividends as reported by Compustat (Compustat data item 21 (Dividends-Common) plus Compustat data item 19 (Dividends-Preferred)), which are divided by the sum of total liabilities and the market value of equity to give an estimate of dividend expressed as a proportion of the current market value of assets, denoted δ. The maturity of the liabilities, denoted T, is set to one year as in Hillegeist et al. (2004). The estimates of the value of assets (VA) and of the annualized standard deviation of assets (σA) used for the BSM Score are given by the simultaneous-equation procedure that is documented in detail in Hillegeist et al. (2004, Appendix). This procedure uses the inputs referred to above plus the annualized standard deviation of equity returns for the year ended at balance-sheet date t−1. The estimate of the continuously compounded return on assets, denoted µ, is derived from the estimates of the value of assets given by the simultaneous-equation procedure, together with observed dividends. As in Hillegeist et al., µ is winsorized such that it has a minimum possible value equal to the one-year Treasury Bill rate and a maximum possible value of 100%.6 In order to calculate the Altman (1968) Z Score, we use the following inputs for the accounting period ended at time t−1: current assets (Compustat data item 4 (Current Assets—Total)); current liabilities (Compustat data item 5 (Current Liabilities—Total)); total assets (as defined above); retained earnings (Compustat data item 36 (Retained Earnings)); earnings before interest and taxes (Compustat data item 13 (Operating Income Before Depreciation) less Compustat data item 14 (Depreciation and Amortization)); market value of equity (as defined above); total liabilities (as defined above); sales (Compustat data item 12 (Sales—Net)).

We collect data for all firm-year cases for which the data necessary for the construction of our measures of EBEI, accruals, CFO, equity return, the Hillegeist et al. BSM Score, the Altman Z Score and the book-to-market ratio are available from Compustat or CRSP, as applicable. This provides 22,778 cases. Observations in the top and bottom 1% of the distributions of EBEI, accruals and CFO, all scaled by beginning-of period market value of equity, and equity return are eliminated as outliers. This leaves 21,513 cases, as described in Table 1 Panel A. Details of the distribution of these 21,513 cases by year are given in Table 1 Panel B. Details of the distribution by industry group are given in Table 1 Panel C.

Table 1.  DETAILS OF FIRM-YEAR CASES USED IN THE STUDY
Panel A: Construction of sample
Firm-year cases for non-financial firms listed on the New York Stock Exchange and the American Stock Exchange for any part of the period from 1989 to 2005 for which necessary data are available from Compustat or CRSP, as applicable22,778
Less: observations in the top and bottom 1% of the distributions of equity return, earnings before extraordinary items (EBEI), accruals and cash flow from operating activities (CFO), deleted as outliers(1,265)
Firm-year cases used in the study21,513
Panel B: Firm-year cases by year
Year
19891,103
1990947
19911,186
19921,168
19931,181
19941,193
19951,315
19961,201
19971,288
19981,339
19991,473
20001,416
20011,447
20021,413
20031,480
20041,272
20051,091
Firm-year cases used in the study21,513
Panel C: Composition of sample by industry group
Industry group (SIC codes)
Metal Mining (10)468
Coal Mining (12), Oil & Gas Extraction (13), Pipelines Except Natural Gas (46)1,405
Forestry (08), Nonmetallic Minerals Except Fuels (14), General Building Contractors (15), Special Trade Contractors (17), Buiding Materials & Garden Supplies (52)178
Heavy Construction Except Building (16)126
Agriculture (01), Food & Kindred Products (20), Tobacco (21)801
Textile Mill Products (22)159
Apparel & Other Textile Products (23)281
Lumber & Wood Products (24)120
Furniture & Fixtures (25)190
Paper & Allied Products (26)500
Printing & Publishing (27)416
Chemicals & Allied Products (28)1,697
Petroleum & Coal (29)370
Rubber & Miscellaneous Plastics Products (30)305
Leather & Leather Products (31)136
Stone, Clay & Glass Products (32)274
Primary Metal Industries (33)603
Fabricated Metal Products (34)566
Industrial Machinery & Equipment (35)1,432
Electronic & Other Electric Equipment (36)1,493
Transportation Equipment (37)777
Instruments & Related Products (38)1,046
Miscellaneous Manufacturing Industries (39)243
Railroad Transportation (40), Trucking & Warehousing (42), Transportation Services (47)233
Water Transportation (44)162
Transportation by Air (45)187
Communications (48)763
Electric, Gas & Sanitary Services (49)1,242
Wholesale Trade-Durable Goods (50)515
Wholesale Trade-Nondurable Goods (51)280
General Merchandise Stores (53)281
Food Stores (54)274
Apparel & Accessory Stores (56)266
Furniture & Home Furnishings Stores (57)161
Eating & Drinking Places (58)288
Miscellaneous Retail (59)396
Hotels & Other Lodging Places (70)141
Business Services (73)1,178
Automotive Dealers & Service Stations (55), Auto-Repair Services & Parking (75)171
Motion Pictures (78)113
Amusement & Recreation Services (79)269
Health Services (80)417
Engineering & Management Services (87)315
Personal Services (72), Educational Services (82), Social Services (83), Miscellaneous275
Firm-year cases used in the study21,513

Table 2 reports descriptive statistics for equity return, EBEI scaled by beginning-of period market value of equity, accruals scaled by beginning-of period market value of equity, CFO scaled by beginning-of period market value of equity, the beginning-of-period BSM Score, the beginning-of-period Z Score times −1 and the beginning-of-period book-to-market ratio. For the BSM score and Z Score financial-distress measures and the book-to-market ratio, we use annually computed rank data in our regression models, but the descriptive statistics for these items in Table 2 are for raw data. Table 3 reports the Pearson and Spearman correlation matrices for the items reported in Table 2. Here, the correlation coefficients for the financial-distress measures and the book-to-market measure are for rank data as used in our regression models. We note that the beginning-of-period financial-distress measures are negatively correlated with EBEI and that this negative correlation is due to negative correlation between the financial-distress measures and accruals, consistent with beginning-of-period financial distress being associated with conservative accounting.

Table 2.  DESCRIPTIVE STATISTICS
VariableNMSDPercentile 1Quartile 1MedianQuartile 3Percentile 99
  1. Note: Data are for U.S. non-financial firms from 1989 to 2005. Equity return is the annual equity return for firm i for the accounting period ended at balance-sheet date t (expressed as a proportion); earnings before extraordinary items (EBEI), accruals and cash flow from operating activities (CFO) are deflated by beginning-of-period market value of equity; the BSM Score is the Hillegeist et al. BSM Score at the beginning of the accounting period; Z Score (times −1) is the Altman Z Score at the beginning of the accounting period times −1; the book-to-market ratio is measured at the beginning of the period. Observations falling in the top and bottom 1% for equity return, EBEI, accruals and CFO are excluded as outliers.

Equity return21,5130.1610.497−0.680−0.1540.0970.3731.933
Earnings before extraordinary items (EBEI)21,5130.0320.130−0.5090.0160.0550.0860.252
Accruals21,513−0.1000.173−0.770−0.141−0.062−0.0160.207
Cash flow from operating activities (CFO)21,5130.1320.151−0.2120.0550.1110.1870.671
BSM Score21,513−14.2989.309−36.044−20.734−12.392−6.729−0.389
Z Score (times −1)21,513−3.9599.244−24.106−4.383−2.915−1.8861.210
Book-to-market ratio21,5130.6200.555−0.3290.3260.5290.7962.540
Table 3.  CORRELATION MATRIX
 Equity returnEBEIAccrualsCFOBSM ScoreZ Score (times −1)Book-to-market ratio
  1. Note: Pearson (Spearman) correlation coefficients are below (above) the diagonal. Data are for U.S. non-financial firms from 1989 to 2005 (21,513 observations). The correlation coefficients are for the data used in our regression models, including annually computed rank data for the Hillegeist et al. beginning-of-period BSM Score, the Altman beginning-of-period Z Score (times −1) and the beginning-of-period book-to-market ratio. EBEI (earnings before extraordinary items), accruals and CFO (cash flow from operating activities) are all scaled by the beginning-of-period market value of equity.

Equity return1.000.42−0.010.26−0.010.030.07
EBEI0.241.000.180.41−0.15−0.080.06
Accruals−0.010.541.00−0.72−0.24−0.42−0.35
CFO0.230.25−0.691.000.080.300.33
BSM Score0.05−0.22−0.270.111.000.330.28
Z Score (times −1)0.04−0.16−0.330.240.331.000.30
Book-to-market ratio0.07−0.06−0.310.300.280.301.00

Financial Distress and the Basu (1997) Earnings-Sensitivity Difference

Before examining the effect of financial distress on the ESD, we first report in Table 4 the parameter estimates for EBEI, accruals and CFO from the ESD-measurement regression models that do not include a financial-distress term. Model (1) is the standard Basu (1997) model; model (2) includes a control for the beginning-of-period book-to-market ratio. Consistent with numerous previous studies, the β1,4 sensitivity-difference coefficient from model (1) for EBEI is positive and significantly different from zero (β1,4= 0.209, t-statistic 7.51). This is conventionally interpreted as indicating that EBEI is more timely in its recognition of bad news than in its recognition of good news, consistent with conditional conservatism. A significant sensitivity difference is observed in both the accruals component of EBEI (β1,4= 0.136, t-statistic 4.56) and the CFO component of EBEI (β1,4= 0.073, t-statistic 5.35). The existence of a positive sensitivity difference for CFO is consistent with results reported in Basu (1997), Ball et al. (2000) and Dietrich et al. (2007). Since CFO is measured directly from the Statement of Cash Flows and should not therefore be directly affected by conditional accounting conservatism, this indicates that a substantial proportion of the sensitivity difference for EBEI is due to factors other than conditional accounting conservatism.7 From model (2), the introduction of a control for the beginning-of-period book-to-market ratio as a proxy for the cumulative conservatism of prior periods causes the sensitivity difference for accruals to become insignificantly different from zero (β2,4= 0.015, t-statistic 0.76), consistent with a conditional-conservatism interpretation of the ESD. The corresponding coefficient for CFO remains positive and significantly different from zero (β2,4= 0.101, t-statistic 6.34).

Table 4.  SENSITIVITY DIFFERENCES FOR EARNINGS BEFORE EXTRAORDINARY ITEMS (EBEI), ACCRUALS AND CASH FLOW FROM OPERATING ACTIVITIES (CFO): WITHOUT FINANCIAL-DISTRESS MEASURE
 Model (1)Model (2)
 EBEIAccrualsCFO EBEIAccrualsCFO
  1. Notes:

  2. 1. Regression models (1) and (2) are estimated using pooled cross-section and time-series data for U.S. non-financial firms from 1989 to 2005. The number of observations is 21,513. The models are as follows:

    image(1)
    image(2)

    where Xit is either (i) EBEI, (ii) Accruals or (iii) CFO, deflated by beginning-of-period market value of equity, for firm i for the accounting period ended at balance-sheet date t; Rit is the equity return for firm i for the accounting period ended at balance-sheet date t; DRit is a dummy variable that takes the value of one when Rit is negative, and 0 otherwise: BMt−1 is the annually computed percentile rank of the book-to-market ratio of firm i at the beginning of the accounting period ended at time t; the β terms are regression coefficients; the ε terms are error terms.

  3. 2. t-statistics are given in parentheses. * (**) denotes significance at the 5% (1%) level in a two-tailed test. t-statistics are based on standard errors clustered by firm and by year.

Intercept(β1,1)0.052−0.0830.135(β2,1)0.063−0.0060.069
(16.73)**(−13.48)**(26.48)** (10.39)**(−1.38)(12.00)**
DRit(β1,2)−0.0030.004−0.007(β2,2)−0.0010.001−0.002
(−0.54)(0.65)(−2.56)* (−0.14)(0.20)(−0.31)
Rit(β1,3)0.015−0.0330.048(β2,3)−0.005−0.0080.003
(1.75)(−3.03)**(7.06)** (−0.52)(−0.76)(0.33)
Rit×DRit(β1,4)0.2090.1360.073(β2,4)0.1160.0150.101
(7.51)**(4.56)**(5.35)** (6.91)**(0.76)(6.34)**
BMi,t−1    (β2,5)−0.021−0.1520.131
(−1.75)(−8.34)**(9.68)**
DRit×BMi,t−1    (β2,6)−0.0030.006−0.009
(−0.22)(0.58)(−0.95)
Rit×BMi,t−1    (β2,7)0.038−0.0390.077
(2.95)**(−2.04)*(4.03)**
Rit×DRit×BMi,t−1    (β2,8)0.2270.290−0.063
(3.62)**(3.62)**(−1.63)
Adjusted R2 10.5%1.0%5.6% 12.9%11.2%14.2%

Table 5 reports the parameter estimates from models (3) and (4), which include financial-distress terms. The results from use of the BSM Score as the financial-distress measure are reported in Panel A, and those from use of the Z Score are reported in Panel B. The objects of interest here are (a) whether the ESD for EBEI is associated with financial distress and (b) whether any such association arises from the accruals component of EBEI and/or from the CFO component. For models (3) and (4) for both the BSM Score and the Z Score, all of the coefficients on the interaction term for the ESD and financial distress (Rit×FDit×DRit for EBEI) are positive and significantly different from zero at the 5% level: β3,8 (BSM Score, EBEI) = 0.297, t-statistic 4.87; β4,8 (BSM Score, EBEI) = 0.254, t-statistic 5.13; β3,8 (Z Score, EBEI) = 0.308, t-statistic 5.36; β4,8 (Z Score, EBEI) = 0.261, t-statistic 5.67. The breakdown between accruals and CFO of the association between financial distress and the ESD suggests that the association arises predominantly from accruals. For both the BSM Score and the Z Score, all of the coefficients on the interaction term for the sensitivity difference for accruals (Rit×FDit×DRit for accruals) are positive and significantly different from zero at the 5% level: β3,8 (BSM Score, accruals) = 0.273, t-statistic 4.50; β4,8 (BSM Score, accruals) = 0.195, t-statistic 4.32; β3,8 (Z Score, accruals) = 0.282, t-statistic 5.21; β4,8 (Z Score, accruals) = 0.213, t-statistic 4.94. Although all of the four corresponding coefficients for CFO are positive, only one of them is significantly different from zero at the 5% level: β3,8 (BSM Score, CFO) = 0.024, t-statistic 0.74; β4,8 (BSM Score, CFO) = 0.059, t-statistic 2.09; β3,8 (Z Score, CFO) = 0.026, t-statistic 0.75; β4,8 (Z Score, CFO) = 0.048, t-statistic 1.56. As we report in our later subsection on sensitivity tests, the significance at the 5% level of the β4,8 (BSM Score, CFO) coefficient is not robust to any of our sensitivity tests. We also note that, after inclusion of the financial-distress interaction term, none of the four coefficients on Rit×DRit for accruals reported in Table 5 is positive and significantly different from zero but all of the corresponding coefficients for CFO remain positive and significantly different from zero. This provides further indication that financial distress is influential with regard to the positive sensitivity difference for accruals but not with regard to the positive sensitivity difference for CFO.

Table 5.  SENSITIVITY DIFFERENCES FOR EARNINGS BEFORE EXTRAORDINARY ITEMS (EBEI), ACCRUALS AND CASH FLOW FROM OPERATING ACTIVITIES (CFO): WITH FINANCIAL-DISTRESS MEASURE
Panel A: BSM Score as the financial-distress measure
 Model (3)Model (4)
 EBEIAccrualsCFO EBEIAccrualsCFO
Intercept(β3,1)0.070−0.0380.108(β4,1)0.0720.0120.060
(22.49)**(−11.35)**(26.79)** (13.21)**(1.97)*(11.22)**
DRit(β3,2)0.0010.005−0.004(β4,2)0.0020.0000.002
(0.28)(1.02)(−0.64) (0.31)(0.04)(0.33)
Rit(β3,3)0.0440.0290.015(β4,3)0.0280.028−0.000
(4.99)**(3.76)**(1.00) (3.90)**(3.01)**(−0.02)
Rit×DRit(β3,4)−0.015−0.1000.085(β4,4)−0.059−0.1390.080
(−0.59)(−3.54)**(3.82)** (−2.13)*(−4.40)**(4.00)**
FDi,t−1(β3,5)−0.046−0.1140.068(β4,5)−0.043−0.0780.035
(−5.19)**(−6.03)**(5.25)** (−5.20)**(−5.70)**(3.49)**
DRit×FDi,t−1(β3,6)−0.019−0.003−0.016(β4,6)−0.020−0.001−0.019
(−1.22)(−0.17)(−1.47) (−1.32)(−0.12)(−1.40)
Rit×FDi,t−1(β3,7)−0.033−0.0670.034(β4,7)−0.053−0.049−0.004
(−1.61)(−6.26)**(1.67) (−2.52)*(−4.91)**(−0.20)
Rit×DRit×FDi,t−1(β3,8)0.2970.2730.024(β4,8)0.2540.1950.059
(4.87)**(4.50)**(0.74) (5.13)**(4.32)**(2.09)*
BMi,t−1    (β4,9)−0.008−0.1280.120
(−0.76)(−9.05)**(9.38)**
DRit×BMi,t−1    (β4,10)0.0070.010−0.003
(0.59)(0.82)(−0.28)
Rit×BMi,t−1    (β4,11)0.052−0.0270.079
(3.91)**(−1.30)(3.85)**
Rit×DRit×BMi,t−1    (β4,12)0.1790.258−0.079
(3.31)**(3.41)**(−2.00)*
Adjusted R2 15.3%7.5%7.3% 16.7%14.1%14.4%
Panel B: Z Score as the financial-distress measure
 Model (3)Model (4)
 EBEIAccrualsCFO EBEIAccrualsCFO
  1. Notes:

  2. 1. Regression models (3) and (4) are estimated using pooled cross-section and time-series data for U.S. non-financial firms from 1989 to 2005. The number of observations is 21,513. The models are as follows:

    image(3)
    image(4)

    where Xit is either (i) EBEI, (ii) Accruals or (iii) CFO, deflated by beginning-of-period market value of equity, for firm i for the accounting period ended at balance-sheet date t; Rit is the equity return for firm i for the accounting period ended at balance-sheet date t; DRit is a dummy variable that takes the value of one when Rit is negative, and 0 otherwise: FDi,t−1 is a measure of financial distress, equal to either (i) the annually computed percentile rank of the Hillegeist et al. BSM Score for firm i at the beginning of the period ended at time t or (ii) the annually computed percentile rank of the Altman Z Score (times −1) at the beginning of the period ended at time t; BMt−1 is the annually computed percentile rank of the book-to-market ratio of firm i at the beginning of the period ended at time t; the β terms are regression coefficients; the ε terms are error terms.

  3. 2. t-statistics are given in parentheses. * (**) denotes significance at the 5% (1%) level in a two-tailed test. t-statistics are based on standard errors clustered by firm and by year.

Intercept(β3,1)0.065−0.0060.071(β4,1)0.0700.0340.036
(14.32)**(−1.29)(14.29)** (11.96)**(5.39)**(6.35)**
DRit(β3,2)0.003−0.0010.004(β4,2)0.0050.0020.003
(0.63)(−0.15)(0.73) (0.99)(0.34)(0.78)
Rit(β3,3)0.0420.0140.028(β4,3)0.0220.028−0.006
(3.61)**(2.80)**(2.45)* (1.98)*(2.21)*(−0.58)
Rit×DRit(β3,4)0.045−0.0220.067(β4,4)0.013−0.0770.090
(2.59)**(−1.14)(4.67)** (0.61)(−2.76)**(5.36)**
FDi,t−1(β3,5)−0.027−0.1560.129(β4,5)−0.023−0.1210.098
(−3.13)**(−14.90)**(15.19)** (−3.05)**(−12.94)**(11.07)**
DRit×FDi,t−1(β3,6)−0.0170.001−0.018(β4,6)−0.016−0.002−0.014
(−1.08)(0.10)(−1.66) (−1.01)(−0.19)(−1.16)
Rit×FDi,t−1(β3,7)−0.049−0.0790.030(β4,7)−0.063−0.0750.012
(−1.70)(−4.34)**(1.39) (−2.28)*(−3.74)**(0.68)
Rit×DRit×FDi,t−1(β3,8)0.3080.2820.026(β4,8)0.2610.2130.048
(5.36)**(5.21)**(0.75) (5.67)**(4.94)**(1.56)
BMi,t−1    (β4,9)−0.012−0.1120.100
     (−1.09)(−6.43)**(6.90)**
DRit×BMi,t−1    (β4,10)−0.0010.002−0.003
(−0.07)(0.17)(−0.23)
Rit×BMi,t−1    (β4,11)0.050−0.0290.079
(4.46)**(−1.58)(3.99)**
Rit×DRit×BMi,t−1    (β4,12)0.1440.222−0.078
(2.69)**(2.82)**(−1.85)
Adjusted R2 14.6%12.5%11.5% 15.8%17.8%16.9%

Financial Distress and Other Measures of Conditional Conservatism not Based on Equity Returns

Table 6 reports the parameter estimates from regression models (5) and (6). Model (5) gives the earnings-change-reversal measure of conditional conservatism, and model (6) shows the effect of financial distress on this measure. The results from model (5) indicate that earnings-change reversal is greater for negative earnings changes than for positive earnings changes (β5,4=−0.503, t-statistic −6.30), indicative of the existence of conditional conservatism by this measure and consistent with Basu (1997) and Ball and Shivakumar (2005). From model (6), the asymmetry in earnings-change reversal is greater for relatively financially distressed firms both for the BSM Score financial-distress measure and the Z Score financial-distress measure: β6,8(BSM Score) =−0.446, t-statistic −3.55; β6,8(Z Score) =−0.433, t-statistic −3.74. This result is consistent with our earlier inference that financial distress is positively associated with conditional accounting conservatism. Table (7) reports the parameter estimates from regression models (7) and (8), which provide evidence as to whether the financial-distress effect on the earnings-change-reversal measure of conditional conservatism from model (6) is associated with the accrual component of earnings and/or the CFO component of earnings. The results suggest that it is associated with the accruals component, but not the CFO component. The relevant coefficients for accruals are negative and significantly different from zero at the 5% and 10% levels, respectively: β7,8(BSM Score) =−0.001, t-statistic −3.65; β7,8(Z Score) =−0.002, t-statistic −1.84. The relevant coefficients for CFO are positive and not significantly different from zero at the 10% level: β8,8(BSM Score) = 0.001, t-statistic 1.27; β8,8(Z Score) = 0.002, t-statistic 1.08. This is consistent with the financial-distress effect observed for model (6) arising from accruals and not from CFO, and is therefore consistent with it arising from conservative accounting rather than from other sources.8

Table 6.  TESTS OF THE ASSOCIATION BETWEEN FINANCIAL DISTRESS AND A MEASURE OF CONDITIONAL CONSERVATISM BASED ON EARNINGS-CHANGE REVERSAL
 Model (5)Model (6)
  Financial-distress measure
BSM ScoreZ Score
  1. Notes:

  2. 1. Regression models (5) and (6) are estimated using pooled cross-section and time-series data for U.S. non-financial firms from 1989 to 2005. The number of observations is 21,513. The models are as follows:

    image(5)
    image(6)

    where ΔEBEIit is the change in EBEI of firm i from the accounting period ended at balance-sheet date t−1 to the period ended at balance-sheet date t, scaled by total assets at t−1; DΔEBEIi,t−1 is a dummy variable that takes the value of one where ΔEBEIi,t−1 is negative and 0 otherwise; FDi,t−1 is a measure of financial distress, equal to either (i) the annually computed percentile rank of the Hillegeist et al. BSM Score for firm i at the beginning of the period ended at time t or (ii) the annually computed percentile rank of the Altman Z Score (times −1) at the beginning of the period ended at time t; the β terms are regression coefficients; the ε terms are error terms.

  3. 2. t-statistics are given in parentheses. * (**) denotes significance at the 5% (1%) level in a two-tailed test. t-statistics are based on standard errors clustered by firm and by year.

Intercept(β5,1)−0.010(β6,1)−0.001−0.011
(−4.75)** (−0.46)(−4.34)**
DΔEBEIi,t−1(β5,2)−0.009(β6,2)−0.004−0.006
(−3.77)** (−1.77)(−1.95)
ΔEBEIi,t−1(β5,3)−0.075(β6,3)−0.119−0.139
(−2.56)* (−2.16)*(−2.72)**
ΔEBEIi,t−1×DΔEBEIi,t−1(β5,4)−0.503(β6,4)−0.224−0.265
(−6.30)** (−1.94)(−2.35)*
FDi,t−1  (β6,5)−0.0190.002
(−3.50)**(0.54)
DΔEBEIi,t−1×FDi,t−1  (β6,6)−0.005−0.003
(−0.72)(−0.40)
ΔEBEIi,t−1× FDi,t−1  (β6,7)0.0900.145
(0.99)(1.84)
ΔEBEIi,t−1×DΔEBEIi,t−1×FDi,t−1  (β6,8)−0.446−0.433
(−3.55)**(−3.74)**
Adjusted R2 0.140 0.1470.152
Table 7.  TESTS OF THE ASSOCIATION BETWEEN FINANCIAL DISTRESS AND A MEASURE OF CONDITIONAL CONSERVATISM BASED ON EARNINGS-CHANGE REVERSAL: PARTITIONING BY ACCRUALS AND CASH FLOW FROM OPERATIONS (CFO)
Panel A: Accruals
 Model (7)
 Financial distress measure
BSM ScoreZ Score
Intercept(β7,1)−0.024−0.050
(−5.98)**(−6.47)**
DΔACCi,t−1(β7,2)0.0520.087
(7.28)**(7.07)**
ΔACCi,t−1(β7,3)−0.000−0.000
(−1.09)(−0.85)
ΔACCi,t−1×DΔACCi,t−1(β7,4)0.0000.001
(1.60)(1.52)
FDi,t−1(β7,5)−0.0210.033
(−2.04)*(2.50)*
DΔACCi,t−1×FDi,t−1(β7,6)0.033−0.037
(2.95)**(−1.40)
ΔACCi,t−1×FDi,t−1(β7,7)0.0010.001
(2.05)*(1.02)
ΔACCi,t−1×DΔACCi,t−1×FDi,t−1(β7,8)−0.001−0.002
(−3.65)**(−1.84)*
Adjusted R2 0.0140.015
Panel B: Cash flow from operations (CFO)
 Model (8)
 Financial distress measure
BSM ScoreZ Score
  1. Notes:

  2. 1. Regression models (7) and (8) are estimated using pooled cross-section and time-series data for U.S. non-financial firms from 1989 to 2005. The number of observations is 21,513. The models are as follows:

    image(7)
    image(8)

    where ΔACCitCFOit) is the change in accruals (CFO) of firm i from the accounting period ended at balance-sheet date t−1 to the period ended at balance-sheet date t, scaled by total assets at t−1; DΔACCi,t−1 (DΔCFOi,t−1) is a dummy variable that takes the value of one where ΔACCi,t−1CFOi,t−1) is negative and 0 otherwise; FDi,t−1 is a measure of financial distress, equal to either (i) the annually computed percentile rank of the Hillegeist et al. BSM Score for firm i at the beginning of the period ended at time t or (ii) the annually computed percentile rank of the Altman Z Score (times −1) at the beginning of the period ended at time t; the β terms are regression coefficients; the ε terms are error terms.

  3. 2. t-statistics are given in parentheses. * (**) denotes significance at the 10% (1%) level in a two-tailed test. t-statistics are based on standard errors clustered by firm and by year.

Intercept(β8,1)−0.028−0.033
(−4.54)**(−5.14)**
DΔCFOi,t−1(β8,2)0.0440.058
(6.56)**(7.31)**
ΔCFOi,t−1(β8,3)0.0000.001
(1.26)(1.20)
ΔCFOi,t−1×DΔCFOi,t−1(β8,4)−0.000−0.001
(−0.95)(−1.00)
FDi,t−1(β8,5)0.0040.011
(0.48)(0.87)
DΔCFOi,t−1×FDi,t−1(β8,6)0.014−0.010
(1.40)(−0.64)
ΔCFOi,t−1×FDi,t−1(β8,7)−0.001−0.002
(−1.33)(−1.21)
ΔCFOi,t−1×DΔCFOi,t−1×FDi,t−1(β8,8)0.0010.002
(1.27)(1.08)
Adjusted R2 0.0090.009

Table 8 reports the parameter estimates from regression models (9) and (10). Model (9) gives the measure of conditional conservatism based on response of accruals to cash flows, and model (10) shows the effect of financial distress on this measure. The results from model (9) indicate that the association between CFO and accruals is less negative in negative-CFO periods than in positive-CFO periods (β9,4= 0.867, t-statistic 8.67), indicative of the existence of conditional conservatism by this measure and consistent with Ball and Shivakumar (2005). From model (10), the asymmetry in the response of accruals to CFO is greater for relatively financially distressed firms both for the BSM Score financial-distress measure and the Z Score financial-distress measure: β10,8(BSM Score) = 4.732, t-statistic 3.44; β10,8(Z Score) = 4.648, t-statistic 2.22.9 This result is again consistent with our earlier inference that financial distress is positively associated with conditional accounting conservatism.

Table 8.  TEST OF THE ASSOCIATION BETWEEN FINANCIAL DISTRESS AND A MEASURE OF CONDITIONAL CONSERVATISM BASED ON RESPONSE OF ACCRUALS TO CASH FLOW FROM OPERATIONS (CFO)
 Model (9)Model (10)
  Financial-distress measure
BSM ScoreZ Score
  1. Notes:

  2. 1. Regression models (9) and (10) are estimated using pooled cross-section and time-series data for U.S. non-financial firms from 1989 to 2005. The number of observations is 21,513. The models are as follows:

    image(9)
    image(10)

    where ACCit and CFOit are accruals and CFO, respectively, of firm i for the accounting period ended at balance-sheet date t, scaled by beginning-of-period total assets; DCFOit takes the value of one where CFOit is negative and 0 otherwise; FDi,t−1 is a measure of financial distress, equal to either (i) the annually computed percentile rank of the Hillegeist et al. BSM Score for firm i at the beginning of the period ended at time t or (ii) the annually computed percentile rank of the Altman Z Score (times -1) at the beginning of the period ended at time t; the β terms are regression coefficients; the ε terms are error terms.

  3. 2. t-statistics are given in parentheses. * (**) denotes significance at the 5% (1%) level in a two-tailed test. t-statistics are based on standard errors clustered by firm and by year.

Intercept(β9,1)0.032(β10,1)0.010−0.144
(10.89)** (0.26)(−0.74)
DCFOit(β9,2)−0.028(β10,2)−0.302−0.119
(−4.18)** (−2.43)*(−0.52)
CFOit(β9,3)−0.921(β10,3)−0.3170.355
(−30.86)** (−1.98)*(0.41)
CFOit×DCFOit(β9,4)0.867(β10,4)−2.918−2.341
(8.67)** (−2.68)**(−1.93)
FDi,t−1  (β10,5)−0.0820.202
(−0.48)(0.58)
DCFOit×FDi,t−1  (β10,6)0.4240.151
(2.13)*(0.40)
CFOit×FDi,t−1  (β10,7)−0.733−2.065
(−0.91)(−1.16)
CFOit×DCFOit×FDi,t−1  (β10,8)4.7324.648
(3.44)**(2.22)*
Adjusted R2 0.511 0.4460.556

Overall, the results from our additional tests of asymmetric timeliness that are not based on equity returns are consistent with our earlier inference that financial distress is positively associated with conditional accounting conservatism.

Sensitivity Tests

We test the sensitivity of our results to a number of variations in methodology. First, we re-estimate regression models using as our measure of CFO Compustat data item 308 (Operating Activities-Net Cash Flow (Statement of Cash Flows)), without subtraction of Compustat data item 124 (Extraordinary Items and Discontinued Operations (Statement of Cash Flows)), with a corresponding accruals measure equal to EBEI less this measure of CFO. Second, we recognize the possibility that CFO might be affected indirectly by conditional conservatism through conservatism-related decisions to expense items, in which case they would appear as part of CFO, or to capitalize them, in which case they would appear as part as cash flows from investing activities. Therefore, we re-estimate regression models using as our dependent variables cash flow from operating and investing activities (CFOI), defined as CFO less Compustat data item 311 (Investing Activities-Net Cash Flow (Statement of Cash Flows)), and an accruals measure defined as EBEI less CFOI. Third, we estimate regression models separately for each of the seventeen years covered by the study without allowing for clustering, and base our t-statistics on the seventeen-year averages of the regression coefficients. Fourth, we use the raw measures of financial distress instead of percentile ranks thereof. Fifth, for the measures of conditional conservatism not based on equity returns, we re-estimate the relevant regression models using net income, as used by Ball and Shivakumar (2005), instead of EBEI. Sixth, because our financial-distress results may have arisen because the financial-distress measures reflect cross-sector differences in underlying debt-contracting-related demand for accounting conservatism, we add a control for industry leverage. Seventh, we perform our tests including financial firms. Our inference that association between financial distress and the ESD arises primarily because relatively financially distressed firms are relatively conditionally conservative is robust to all of these variations in methodology. The only notable finding from the sensitivity tests is that the significance at the 5% level of the effect of the BSM Score measure of financial distress on the sensitivity difference for CFO in model (4) (coefficient β4,8 (BSM Score, CFO)) is not robust to any of the variations in methodology. This is supportive of the finding that association between financial distress and the ESD arises predominantly through accruals rather than through CFO.

Summary

In common with many other studies, we observe a significant sensitivity difference for EBEI. This comprises significant sensitivity differences for both the accruals component of EBEI and the CFO component thereof. The existence of a sensitivity difference for CFO suggests that the ESD arises to a significant degree from factors other than conditional conservatism, and may be indicative of the effect suggested by Ball et al. (2009) whereby operating-cash-outflow responses to adverse events are relatively immediate. With regard to our principal focus of interest, we report a positive association between financial distress and the ESD. There is some weak evidence that this arises in part from the CFO component of earnings, but it arises predominantly from the accruals component. This result is consistent with the effect of financial distress on the ESD arising primarily because relatively financially distressed firms adopt a relatively high degree of conditional conservatism rather than because of other non-conservatism sources of association between financial distress and the ESD. Additional evidence from non-equity-return-based measures of conditional conservatism and a number of sensitivity tests is consistent with this. Our inference is consistent with the inference drawn by Basu (1997) in his brief examination of the possibility that the ESD was due to termination-option-related non-linearity in the relationship between equity earnings and equity return, but contrasts with the inference drawn by Wang et al. (2009) from ESDs that are based on estimates of entity return. Although we cannot rule out the possibility that association between financial distress and the ESD arises in part through a non-conditional-conservatism route, our evidence is consistent with such association arising primarily through a conditional-conservatism route.

CONCLUSION

In this paper, we seek evidence relevant to the conditional-conservatism interpretation of the excess of the sensitivity of accounting earnings to negative equity returns over its sensitivity to positive equity returns, which we term the earnings-sensitivity difference (ESD). Motivated by concern about the conditional-conservatism interpretation of the ESD and by recognition that the effect of financial distress could contribute to the ESD through a conditional-conservatism route and/or through a non-conditional-conservatism route, we examine the association between measures of financial distress and the ESD. In particular, we examine whether such association arises through the accruals component of earnings, which can reflect conditional accounting conservatism, or through the CFO component, which cannot directly reflect conditional accounting conservatism. We also provide additional evidence on the association between financial distress and conditional conservatism using other non-equity-return-based measures of conditional conservatism.

In common with many other studies, we find a positive sensitivity difference for EBEI. This comprises significant sensitivity differences for both the accruals component of EBEI and the CFO component. The sensitivity difference for CFO suggests that the ESD arises to a significant degree from factors other than conditional conservatism. We find a positive association between each of our measures of financial distress and the ESD. There is some weak evidence that this association arises in part from the CFO component of earnings, but it arises predominantly from the accruals component. This is consistent with the association arising primarily from a relatively high degree of conditional conservatism in relatively financially distressed firms rather than from other sources of financial-distress-related non-linearity in the relationship between equity return and earnings. Our results are consistent with prior evidence that debt-related factors are an important source of conditional-conservatism-related asymmetric timeliness in earnings. The inference that there is a positive association between financial distress and conditional conservatism is supported by evidence from measures of conditional conservatism other than the ESD. Other studies have cast doubt on the conditional-conservatism interpretation of the ESD by arguing that it reflects phenomena other than conditional conservatism. The results of this study do not suggest that the effect of financial distress gives significant additional cause to doubt the reliability of the ESD as an indicator of conditional accounting conservatism.

Footnotes

  • 1

    We acknowledge the possibility that CFO might be affected by conditional conservatism through conservatism-related decisions to expense items, in which case they would appear as part of CFO, or to capitalize them, in which case they would appear as part of cash flows from investing activities. In light of this, we perform a sensitivity test in which we use cash flow from operating and investing activities (CFOI), with a corresponding accruals measure, in place of CFO. As reported later in the subsection on sensitivity tests, our inference is not sensitive to the use of CFOI instead of CFO.

  • 2

    We obtain t-statistics based on standard errors clustered by firm and by year using a Stata program written by Mitchell Petersen.

  • 3

    As we use percentile ranks of our financial-distress measures, it makes no difference for our reported results whether or not we apply the transformation in (13). We do, however, use the raw financial-distress measures in a sensitivity test.

  • 4

    SFAS 95 was effective for fiscal years ending after 15 July 1988.

  • 5

    Our comparison of Compustat-reported data with published financial-statement data showed that extraordinary-items-related cash flows that are excluded from CFO in the financial statements are often reported as part of Compustat data item 308. In such cases, it is appropriate to subtract Compustat data item 124 from Compustat data item 308 in order to give a CFO measure that can be subtracted from EBEI to give a measure of accruals. In a sensitivity test, we also use the unadjusted data item 308 as the measure of operating cash flow and data item 18 less data item 308 as the measure of accruals. (Hribar and Collins, 2002, use Compustat data item 123 (Income Before Extraordinary Items (Statement of Cash Flows)), whereas we use Compustat data item 18, taken from the Statement of Income, and eliminate the approximately 2% of cases where Compustat data item 18 is not equal to Compustat data item 123.)

  • 6

    SAS code for the calculation of the BSM Score is helpfully provided in an appendix to Hillegeist et al. (2004). This code was used in our measurement of the BSM Score.

  • 7

    We also use the data for model (1) to replicate a test in Hayn (1995), in which return is regressed on earnings levels separately for profit cases and for loss cases. The untabulated results from this test are very similar to those reported by Hayn (p. 135), with the return-earnings slope coefficient being significantly steeper for profit cases than for loss cases.

  • 8

    When the dummy variable for prior-period negative change in aggregate earnings (DΔEBEIi,t−1) that is used in model (6) is used in place of DΔACCi,t−1 in model (7) (for accruals), neither of the resultant β7,8 coefficients is significantly different from zero. So, the financial-distress effect in accruals reversal that is observed when the prior-period bad-news indicator is negative change in accruals is not observed when the prior-period bad-news indicator is negative change in aggregate earnings. When DΔEBEIi,t−1 is used in place of DΔCFOi,t−1 in model (8) (for CFO), the resultant β8,8 coefficients remain insignificantly different from zero.

  • 9

    In a similar test using a shorter data period, Wang et al. (2009) report that coefficients corresponding to our β10,8 coefficients are insignificant.

Ancillary