An integrated analysis of the association between accrual disclosure and the abnormal accrual anomaly

Because of the probability of misclassifying the earnings components, our results could be driven by measurement errors. Bernard and Skinner (1996) suggest that special items are often unrelated to managerial discretion. To assess the effect of special items on our results, we remove special items (Compustat data item 32) from total accruals. Following Xie (2001), we assume that the marginal tax rate is 40% for all firms in the sample over the sample period. Adjusting accruals for special items does not qualitatively change our inferences.

The literature generally assumes that managers use their reporting discretion to manage accruals. However, generally, investors can only assess whether accruals are abnormal or not. Determining whether they are discretionary may not be possible because a firm’s accruals may be affected by factors unrelated to managerial discretion. Because of the difficulty in directly estimating discretionary accruals, we use investors’ pricing of abnormal accruals to proxy for the pricing of discretionary earnings and, at times, use abnormal accruals, discretionary accruals, and discretionary earnings interchangeably. We assume that, while discretionary accruals are not directly observable by investors, they can be approximated by the level of abnormal accruals.

At first glance, the notion that investors would discount managed earnings might seem contrary to prior evidence suggesting that investors overprice (discretionary) accruals (Sloan, 1996; Xie, 2001). However, the overpricing and the discounting of accruals are not mutually exclusive. Assume, for instance, that the market should have put a weight of zero on the discretionary component of earnings and one on the nondiscretionary component. If, at the earnings announcement or at the filing date, it instead puts a weight of 0.40 on the discretionary accruals, the discretionary accruals are then both overpriced and discounted (relative to nondiscretionary earnings).

Guay, Kothari, and Watts (1996) and Subramanyam (1996) analyze the association between return and abnormal accruals using long-window return/earnings associations. They suggest that abnormal accruals are priced at a discount. However, because they analyze long-window associations, the documented discounting may have taken place long after the earnings announcement or the filing dates. Therefore, these studies are not informative about the market pricing of abnormal accruals around either earnings announcement or the filing dates.

Other studies also analyze the market reaction to the information in 8-K filings (Fried & Schiff, 1981; Johnson & Lys, 1990; Klock, 1994; Pastena, 1979; Schwartz & Soo, 1995). In general, these studies find the market does not react to the information in 8-K filings. Johnson and Lys (1990), for example, analyzing the market reaction to auditor changes, find no significant market reaction over the 21 trading days centered on the 8-K filing. They fail to find a significant market reaction even when the switch is due to auditor disagreements. The failure to detect a significant market reaction may be a result of firms announcing the auditor changes through press releases before filing 8-K forms; however, Johnson and Lys (1990) report that an examination of Wall Street Index citations provides no evidence that the changes were pre-announced. Furthermore, Klock (1994) analyzes the market reaction around the date the auditor is formally terminated, the filing date, and the on-file date. He finds no significant market reaction around any of these dates. It seems that auditor switches, particularly if they are due to disagreements, are such important events that they would likely induce significant market reactions. A failure to observe such a movement around these events seems to suggest that the market does not pay enough attention to SEC filings. For instance, while Johnson and Lys (1990) document an insignificantly abnormal return of about −0.72% from day −1 to day +10 relative to the 8-K filing date, they find a significantly abnormal return of −3.25% from day +11 to day +60, which is consistent with the conjecture that the market is slow in processing the information in SEC filings.

Evidence in the human behavioral literature suggests that decision makers tend to anchor on the earliest information provided (Hogarth & Einhorn, 1992; Kahneman, Slovic, & Tversky, 1982; Schlottmann & Anderson, 1995; Slovic & Lichtenstein, 1971; Zhang, Johnson, & Wang, 1997). That is, all else equal, investors may overemphasize the information provided in the earnings report press release and underemphasize the information provided in the SEC filing. However, the evidence provided by Stice (1991) and Chung et al. (2003) that the market inefficiently processes the information contained in SEC filings even if the reports are filed before the earnings announcement press release suggests that anchoring is, at best, a partial explanation for investors’ apparent lack of attention to SEC filings.

For discussions on the use of accruals as a signaling device, see Watts and Zimmerman (1986), Subramanyam (1996), Guay et al. (1996), Demski (1998), Arya, Glover, and Sunder (2003), Louis and Robinson (2005), and Louis and White (2007).

We deflate the change in earnings by lagged assets because this is the deflator used to scale the discretionary earnings measure.

Delisting returns are included if available. If a stock disappears from CRSP, we assume the stock earns zero abnormal returns for the missing months.

The benchmark portfolio returns are value weighted because equally-weighted portfolios generally place too much weight on smaller firms. However, abnormal returns do not sum to zero when the benchmark portfolio returns are value weighted; they are actually positive in our sample. This has almost no effect on our hedge-portfolio returns but does bias in favor of the null hypothesis that the hedge-portfolio returns come from the high accrual side of the hedge portfolio. We address this problem by subtracting the average abnormal return of the population from each firm’s abnormal return.

Our accrual measure might be misestimated because of corporate transactions such as mergers and divestitures (see Hribar & Collins, 2002; Revsine, Collins, & Johnson 1999). One particular source of complication is that a merger can be finalized at any point during the quarter, creating the need to (arbitrarily) allocate the earnings (cash flows and accruals) of the target over the quarter, depending on the existing relation between the merging partners and the acquisition method (purchase vs. pooling). In addition, when an acquisition or discontinuation of operations occurs in the first quarter, Compustat adjusts only the data for the first quarter of the prior year. Hence, total assets at the beginning of the quarter and sales for the fourth quarter of the prior year are on a different basis, biasing the estimated discretionary accruals. To assess the effect of mergers and divestitures on our results, we delete firms that have absolute value of cash flows related to acquisitions (quarterly Compustat data item 94) scaled by total assets greater than 0.001 or absolute value of discontinued operations (quarterly Compustat data item 33) scaled by total assets greater than 0.001. This restriction does not qualitatively alter any of our inferences.

See Greenberg, Johnson, and Ramesh (1986), Finger (1994), Lorek and Willinger (1996), Dechow, Kothari, and Watts (1998), and Barth, Cram, and Nelson (2001) for details on the accrual time-series correlation. Prior studies that model current accruals as a function of lagged accruals include Dechow, Richardson, and Tuna (2003), Louis and White (2007), and Gong, Louis, and Sun (2006).

The results reported in Tables 4 and 5 are qualitatively similar when we exclude lagged accruals from the unexpected discretionary accrual model. Although the hedge portfolio returns reported in Table 6 are lower for Nondisclosers, our inference about the effect of accrual disclosure on the long-term abnormal returns has not changed.

Nontabulated results show that our discretionary accrual model is quite powerful, with the explanatory variables having generally consistent signs across time and industry. The average of the 16 cross-industry average adjusted R ² for the period from the fourth calendar quarter of 1998 to the third calendar quarter of 2002 (our sample includes earnings announcements that were made over the period 1999–2002) is 0.297. The cross-industry average coefficient on change on sales (ΔSALE) is consistently positive over time and the cross-industry average coefficient on property, plant, and equipment (PPE) is consistently negative over time. The cross-industry average coefficient on lagged total accruals (LTA) is negative in 15 out of the 16 quarters. The averages of the 16 cross-industry average coefficient estimates are very significant for all three variables, with p-values of 0.000. We find similar results across industries. The averages of the 48 time-series average adjusted R ² is 0.310. The time-series average coefficient on change on sales (ΔSALE) is positive for 38 of the 48 two-digit SIC code industries. The time-series average coefficient on property, plant, and equipment (PPE) is negative across the 48 industries. The time-series average coefficient on lagged total accruals (LTA) is negative for 37 out of the 48 industries. The averages of the 48 time-series average coefficient estimates are very significant for all three variables, with p-values of 0.000.

In estimating model (4), we require at least 20 observations in each industry in a given quarter. The results of the study are not statistically sensitive to this restriction or the inclusion of the additional variables in the accrual model.

The results of the study are generally similar if we do not adjust for performance.

Note that we sort the sample into low and high discretionary accrual deciles using all nonfinancial firms that have the necessary data on Compustat to estimate discretionary accruals for the current quarter. Kraft, Leone, and Wasley (2006) argue that requiring observations on forthcoming data induces a hindsight bias.

We use value-weighted returns because equally weighted portfolios generally place too much weight on smaller firms. However, abnormal returns do not sum to zero when the benchmark portfolio returns are value weighted. To address this problem, we subtract the average abnormal return of the population from each firm’s abnormal return. To assess the effects of this adjustment, we repeat the analysis without the adjustment. Our inferences are qualitatively unchanged. The difference in the abnormal returns of Nondisclosers with low and high abnormal accruals is 20.5% when we winsorize the top and bottom one-percentiles of the abnormal returns and 23.5% when we do not winsorize the returns. In contrast, we find no evidence that Disclosers with low abnormal accruals outperform those with high abnormal accruals.