Asymmetrically Timely Loss Recognition and the Accrual Anomaly. Discussion of Konstantinidi et al.

Asymmetrically timely loss recognition—also known as conditional conservatism in financial reporting—is embedded in US GAAP earnings through practices such as lower-of-cost-or-market accounting for inventories (ASC 330, FASB, 2009a), goodwill impairments (ASC 350, FASB, 2009b), and asset write-downs (ASC 360, FASB, 2009c). Such conditionally conservative accounting practices mandate the more timely recognition of losses relative to gains through transitory negative accrual items (e.g., Basu, 1997; Ball and Shivakumar, 2006; Patatoukas and Thomas, 2011; Ball et al., 2013; Patatoukas and Thomas, 2015). A direct implication of asymmetrically timely loss recognition is asymmetry in the persistence of accruals depending on whether the firm experiences a gain or a loss in the current year: accruals should be less persistent for loss years relative to profit years. 1

Asymmetry in the persistence of accruals across profit and loss years has interesting implications for research on the pricing of accruals. Sloan (1996) provides evidence that investors overestimate the persistence of accruals by naïvely fixating on total earnings and, therefore, tend to undervalue low accrual firms and overvalue high accrual firms—a phenomenon commonly referred to as the accrual anomaly. In the presence of conditional conservatism, naïve earnings fixation would imply that investors tend to overestimate the persistence of accruals, especially in loss years. It follows that investors are more likely to misprice the accruals of loss firms and, therefore, Sloan's (1996) accrual anomaly, that is, the negative association between accruals and future abnormal stock returns, is predicted to be more pronounced for loss firms relative to profit firms.

In contrast to this prediction, however, Konstantinidi et al. (2016) find that the accrual anomaly is significant for profit firms, while they do not find evidence of such an anomaly for loss firms. More specifically, Konstantinidi et al. (2016) measure accruals as the difference between accounting earnings less cash flows from operating activities, and find evidence of significantly positive hedge returns from buying/selling low/high accrual profit firms, while they find insignificant hedge returns when repeating the trading strategy for loss firms. The authors conclude that to the extent that the accrual anomaly reflects mispricing, the market failure appears to be rooted in the processing of accruals for profit firms rather than investors’ naïve earnings fixation. On the other hand, if the accrual anomaly reflects variation in risk premium, such variation would appear to be significant only for profit firms.

Given the ongoing debate among academics about the origins of the accrual anomaly as well as the practical implications for the investment community, I revisit the accrual anomaly for subsamples of profit and loss firms using a broader measure of accruals. To this end, I start by measuring accruals as the difference between accounting earnings less free cash flows, that is, the sum of cash flows from operating and investing activities as reported in the statement of cash flows. When compared to cash flows from operating activities, free cash flows match the flows in earnings more comprehensively and, therefore, the difference between earnings less free cash flows offers a broader measure of accruals and earnings quality (e.g., Richardson et al., 2005; Dechow and Ge, 2006; Dechow et al., 2011).

Separating firms based on the sign of reported earnings, I find that although the accrual anomaly extends across profit and loss firms, it appears to be stronger for loss firms. The average hedge return from buying/selling low/high accrual loss firms is 16.99%, while the hedge return from buying/selling low/high accrual profit firms is 5.82%. Evidence of accrual mispricing further increases when I separate loss firms experiencing negative contemporaneous abnormal returns (roughly 72% of all loss firms), with the hedge return from buying/selling low/high accrual loss firms rising to 21.77%.

Overall, the evidence suggests that the subsample of loss firms is more susceptible to accruals mispricing, which is consistent with the prediction that investors naïvely fixate on total earnings and, therefore, tend to overestimate the persistence of accruals, especially in loss years. The evidence presented here also highlights that inferences regarding variation in the accrual anomaly across profit and loss firms are sensitive to the measurement of the accrual component of earnings. Indeed, using a less comprehensive measure of accruals and alternative definitions of gains and losses, Konstantinidi et al. (2016) reach the opposite conclusion, namely that the accrual anomaly is stronger for profit firms and non-existent for loss firms.

Sample Construction

I obtain accounting data from the Compustat fundamental annual file and stock return data from the CRSP monthly returns file. 2 I measure accounting earnings (X_it) as income before extraordinary items per share scaled by average total assets per share. I extract the accrual portion of earnings using the cash flow statement approach, which becomes available after 1988. I measure accruals (A_it) as the difference between income before extraordinary items per share less free cash flows per share (i.e., the sum of cash flows from operating and investing activities per share) scaled by average total assets per share. Following prior research, I use size-adjusted buy-and-hold stock returns to measure abnormal stock returns (r_it+1). The size-adjusted return is computed as the 12-month buy-and-hold return including distributions and delisting returns in excess of the buy-and-hold return of firms with similar market capitalization. To mitigate the effect of missing delisting returns, if a delisting is due to liquidation I set delisting returns equal to –30% for NYSE and AMEX firms and equal to –55% for NASDAQ firms (e.g., Shumway, 1997; Shumway and Warther, 1999). The return measurement window is annual beginning with the first day of the fourth month after the fiscal year-end. The market capitalization portfolio cutoffs are determined as of the fiscal year-end.

To align portfolio returns in calendar time, I restrict my sample to include only December fiscal year-end firms. Consistent with prior studies, I further restrict my sample to include only NYSE/AMEX/NASDAQ non-financial firms. To reduce the effects of outliers, I trim observations in the top and bottom 1% of each annual cross-section of earnings and earnings components. To reduce the effect of penny stocks, I exclude firms with stock price per share less than $1 as of the fiscal year-end as well as stocks with average assets per share below $1 per share. My final sample includes 56,482 firm-year observations with non-missing earnings and earnings components over the 23-year period from 1989 to 2011. The Appendix summarizes the variables definitions.

Table 1 reports descriptive statistics. The empirical distributions of earnings and earnings components are similar to those reported in prior research (e.g., Dechow and Ge, 2006). Earnings exhibit negative skewness with a mean value of –0.012, while 32.1 of the observations report negative earnings. On average, accruals are positive with a mean value of 0.025, while free cash flows are negative with a mean value of –0.037, which suggests that the average firm is raising financing. Earnings are positively correlated with the total accrual and free cash flow components. Accruals are strongly negatively correlated with free cash flows, which is consistent with accounting matching. Finally, consistent with prior evidence on stock return predictability, accruals (free cash flows) are negatively (positively) correlated with one-year-ahead abnormal stock returns.

Revisiting the Accrual Anomaly

Prior to exploring variation in the accrual anomaly across profit and loss firms, I replicate Sloan's (1996) portfolio trading strategy. Specifically, for each year t, I sort firms into decile portfolios based on the annual cross-sectional distribution of accruals. Then, for each portfolio and each year I calculate the equal-weighted average abnormal stock return earned over the subsequent year t+1. Table 2 reports time-series average values of abnormal returns across decile portfolios based on accruals. First, I observe that there is tremendous variation in the average magnitude of accruals across portfolios, ranging from –0.265 for the low accruals portfolio to 0.308 for the high accruals portfolio. Consistent with prior evidence, I find a negative association between accruals and one-year-ahead abnormal stock returns. Moving from low to high accrual portfolios, one-year-ahead abnormal returns decrease from 8.04% for the bottom decile portfolio to –7.53% for the top decile portfolio. A trading strategy that takes a long position in low accrual firms and a short position in high accrual firms yields an average hedge return of 15.57% (t-statistic = 3.39, two-sided p-value = 0.003). 1

In summary, Table 2 provides significant evidence of stock return predictability based on accruals. Consistent with Sloan (1996), it appears that investors undervalue low accrual firms and overvalue high accrual firms. A risk-based explanation would require that unmodelled risk premia decrease with the magnitude of accruals. A comparison of my results with the evidence in Konstantinidi et al. (2016) suggests that the hedge return from buying/selling low/high accrual firms is larger when using a broader measure of accruals—a finding consistent with prior research in the pricing of accruals (e.g., Richardson et al., 2005; Dechow et al., 2011).

On the Importance of Loss Firms

With 32.1% of the observations reporting negative earnings in the current year (see Table 1, Panel A), the question becomes whether and how the mix-up of loss firms and profit firms has the potential to affect inferences with respect to the magnitude and causes of the accrual anomaly. This question is especially relevant if the frequency of losses systematically varies with the magnitude of accruals. Indeed, Table 3 provides evidence that the frequency of loss firms generally decreases, although not monotonically, with the magnitude of accruals. The frequency of loss firms is 82.3% for the bottom accrual portfolio and 25.5% for the top accrual portfolio. Table 3 also presents the frequency of firms with negative accruals and negative free cash flows across decile portfolios of accruals. Looking across portfolios, I note that the frequency of negative free cash flow firms is 38.4% for the bottom accrual portfolio and 98.1% for the top accrual portfolio. Therefore, separating firms based on the sign of free cash flows would result in a significant mix-up of profit and loss firms. 1

A key message from Table 3 is that loss firms are over-represented in the bottom accrual portfolios relative to the top accrual portfolios and, therefore, inferences with respect to variation in the accrual anomaly can be affected by pooling loss firms with profit firms. Another is that the sign of earnings provides a comprehensive way to identify economic gains from losses capturing the combined magnitudes of accruals and free cash flows. Next, I revisit evidence of variation in the accrual anomaly across subsamples of profit and loss firms.

Accrual Anomaly: Variation Across Profit and Loss Firms

Table 4 reports abnormal returns across portfolios based on accruals formed separately across subsamples of profit and loss firms. I separate firms into profit and loss subsamples based on the sign of income before extraordinary items reported in t. Effectively, Table 4 repeats Sloan's (1996) portfolio trading strategy conditioning on the sign of reported earnings. For each year t, I sort firms into portfolios based on the annual cross-sectional distribution of accruals separately for profit and loss firms (i.e., the portfolio sorts are conditional on the sign of reported earnings). While there are enough firm-year observations in the subsample of profit firms to ensure that there are no less than 100 stocks for each annual decile portfolio, this is not the case for the subsample of loss firms. Accordingly, I sort loss firms in quartile portfolios. Then, for each portfolio and each year, I calculate the equal-weighted average abnormal return earned over the subsequent year t+1 separately for profit and loss firms.

Panels A and B of Table 4 report time-series average abnormal returns across portfolios based on accruals for profit firms and loss firms. Focusing on profit firms (38,361 observations), Panel A shows that average abnormal returns decrease from –0.58% for the bottom accrual portfolio of profit firms to –6.40% for the top accrual portfolio of profit firms. The trading strategy that takes a long position in low accrual profit firms and a short position in high accrual profit firms yields an average hedge return of 5.82% (t-statistic = 2.21, two-sided p-value = 0.038). Turning to Panel B, I find that the accrual anomaly is stronger for the subsample of loss firms (18,121 observations). Average abnormal returns decrease from 9.66% for low accrual loss firms to –7.33% for high accrual loss firms, with an average hedge return of 16.99% (t-statistic = 5.26, two-sided p-value = 0.000). 0

As explained in the paper's introduction, in the presence of conditional conservatism, naïve earnings fixation would imply that investors tend to overestimate the persistence of accruals, especially in loss years. Consistent with this story, I find evidence that the accrual, anomaly is stronger for the subsample of loss firms. A potential concern, however, is that partitioning the sample based on the sign of reported earnings does not provide a clean separation of firms experiencing economic losses. To alleviate this concern, I further identify loss firms experiencing negative contemporaneous abnormal returns (13,058 observations or roughly 72% of all loss firms). Panel C of Table 4 provides evidence that the accrual anomaly is even stronger for this subsample. Average abnormal returns decrease from 14.50% for low accrual loss firms to –7.27% for high accrual loss firms, with an average hedge return of 21.77% (t-statistic = 4.56, two-sided p-value = 0.000). This finding is comforting if one believes that separating firms based on both the sign of reported earnings and the sign of contemporaneous abnormal returns provides a cleaner identification of firms experiencing economic losses in the current year.

Conclusion

A direct implication of asymmetrically timely loss recognition due to conditionally conservative accounting practices is asymmetry in the persistence of accruals depending on whether the firm experiences a gain or a loss: accruals, should be less persistent for loss years relative to profit years. If investors naïvely fixate on total earnings, however, they will tend to overestimate the persistence of accruals, especially in loss years. Consistent with naïve earnings fixation, I find that the accrual anomaly is more pronounced for loss firms relative to profit firms. The evidence highlights that inferences regarding variation in the accrual anomaly across profit and loss firms are sensitive to the measurement of the accrual component of earnings. Indeed, using a less comprehensive measure of accruals and alternative definitions of gains and losses, Konstantinidi et al. (2016) reach opposite conclusions, namely that the accrual anomaly is stronger for profit firms and non-existent for loss firms.

By virtue of standard accounting identities, the broader measure of accruals employed in this study can be disaggregated into different components, including the change in non-cash working capital (i.e., the difference between the change in current operating assets less the change in current operating liabilities) and the change in non-current net operating assets (i.e., the difference between the change in non-current operating assets less the change in non-current operating liabilities). An interesting direction for future research would be to explore sources of variation in the pricing of different accrual components across profit and loss firms.

Future research could also explore alternative, non-mutually exclusive explanations for the pricing of accruals across profit and loss firms, other than naïve earnings fixation. One possibility is that unmodelled risk premia decrease with the magnitude accruals, especially for loss firms. Another possibility is that there are systematic differences in the investor-base sophistication (e.g., percentage retail ownership) and the intensity of market frictions (e.g., short sale constraints) across profit and loss firms. 1