Earnings management and the post-earnings announcement drift

Henock Louis and Amy X. Sun

Smeal College of Business, Pennsylvania State University, University Park, PA 16802

Abstract

There is reliable evidence that managers smooth their reported earnings. If some firms manage
earnings downwards (upwards) when they experience large positive (negative) earnings shocks
and if investors have cognitive limits or are inattentive, then it is plausible that the post-earnings
announcement drift could be related to earnings management. Consistent with this conjecture, we
find that firms with large negative (positive) changes in operating cash flows manage their
accruals substantially upwards (downwards). Most importantly, we find no evidence of a positive
post-earnings announcement drift for those firms with large positive earnings changes that are
least likely to have managed earnings downward or a negative post-earnings announcement drift
for those firms with large negative earnings changes that are least likely to have managed
earnings upward. That is, for these firms, there is no evidence of an underreaction to earnings
changes. The underreaction is concentrated largely among those firms that are most likely to
have smoothed their reported earnings, although this effect has weakened in recent years as
investors started paying more attention to the anomalies and hedge funds were focusing on
exploiting them. Finally, consistent with the earnings management hypothesis, we also find that
the post-earnings announcement drift is generally associated with discretionary (or abnormal)
accruals and not with nondiscretionary accruals. These findings reconcile PEAD with the
(abnormal) accrual anomaly.

JEL Classification: G12, G14, M41, M43

Keywords: Market anomalies, post-earnings-announcement drift, (abnormal) accrual anomaly,

earnings management, earnings smoothing

*We thank Walid Al-Issa, Sung Chung, Paul Fischer, Dan Givoly, Zhaoyang Gu, Chris Jones, Bin Ke, Adam Koch,
Joan Lee, Andrew Leone, Russell Lundholm, James McKeown, Jeffrey Ng, Scott Richardson, Lakshmanan
Shivakumar, Oktay Urcan, Hal White, and workshop participants at Carnegie Mellon University, London Business
School, the third Penn State Summer Research Conference, the 2007 AAA annual conference, the Securities and
Exchange Commission, and the 2007 Norfolk Southern Excellence in Accounting Conference at the College of
William and Mary for helpful comments on an earlier draft. We are particularly grateful to Steven Huddart for an
extensive discussion of the research idea and to an anonymous referee for some very constructive comments.
 One of the most robust market anomalies is the delayed market reaction to earnings news,

referred to in the literature as the post-earnings announcement drift (PEAD).1 More specifically,

it has been documented that firms reporting large positive earnings changes experience positive

abnormal returns over the six months after the earnings announcement, whereas those that report

large negative earnings changes experience negative abnormal returns. PEAD has been subjected

to intense scrutiny and, although many previously documented anomalies have apparently

disappeared (Schwert, 2003), PEAD seems to persist to the dismay of market efficiency

proponents. PEAD is so robust that Fama (1998) concludes that it is “above suspicion.”

Many explanations have been suggested for the delayed market reaction: 1) investors’

(presumably) false assumptions about earnings properties (Bernard and Thomas, 1990; Barberis,

Shleifer, and Vishny, 1998), 2) fluctuations in overconfidence (Daniel, Hirshleifer, and

Subrahmanyam, 1998), 3) underreaction to information due to cognitive limits (Hong and Stein,

1999), 4) disposition effect (Frazzini, 2006; ,Grinblatt and Han, 2005), 5) information

uncertainty (Zhang, 2006), and 6) inattention (DellaVigna and Pollet, 2009; ,Hirshleifer, Lim,

and Teoh, 2009). However, none of these studies has considered the potential role of earnings

management.

Investors presumably have cognitive limits and limited attention (Hong and Stein, 1999;

Hirshleifer and Teoh, 2003; Peng and Xiong, 2006; Barber and Odean, 2008; DellaVigna and

Pollet, 2009; Hirshleifer et al., 2009; Loh, 2010; Louis and Sun, 2010). DellaVigna and Pollet

(2009) and Hirshleifer et al. (2009) suggest that PEAD is associated with investor inattention.

They make no assumption about the quality of the earnings reports. They simply maintain that

1
There is a long list of studies that document PEAD. They include: Ball and Brown (1968), Jones and Litzenberger
(1970), Brown and Kennelly (1972), Rendleman, Jones, and Latané (1982), Bernard and Thomas (1989, 1990),
Constantinou, Forbes, and Skerratt (2003), Battalio and Mendenhall (2005), Chordia and Shivakumar (2006), Zhang
(2006), DellaVigna and Pollet (2009), Hirshleifer et al. (2009), among others.

1
due to limited attention, investors’ responses to the reports are delayed. However, Hirshleifer,

Hou, Teoh, and Zhang (2004) suggest that because investors have limited attention, they tend to

incorrectly value abnormal accruals. Consistent with this view, extant studies suggest that many

well documented anomalous stock price behaviors are associated with (discretionary) abnormal

accruals. They conclude, for instance, that discretionary accruals drive the accrual anomaly (Xie,

2001; Chan, Chan, Jegadeesh, and Lakonishok, 2006) and are associated with anomalous long-

term returns after seasoned public offerings (Teoh, Welch, and Wong, 1998a; Teoh and Wong,

2002; DuCharme, Malatesta, and Sefcik, 2004), initial public offerings (Teoh, Welch, and Wong,

1998b; Teoh, Wong, and Rao, 1998; Teoh and Wong, 2002), stock-for-stock mergers (Louis,

2004; Gong, Louis, and Sun, 2008a), open market repurchases (Gong, Louis, and Sun, 2008b).2

Therefore, we posit that the post-earnings announcement drift could also be due, at least partly,

to managerial discretionary reporting behavior. More specifically, we speculate that earnings

smoothing could be one reason why investors are mistaken about earnings properties.

As Collins and Hribar (2000) conjecture, “firms faced with large negative unexpected

earnings shocks may attempt to smooth earnings and/or make their situation look better by using

large income increasing accruals … [whereas] firms with large positive unexpected earnings

shocks may attempt to mitigate these shocks by creating large income decreasing accruals”.3 If

some firms manage earnings downward (upward) when they experience large positive (negative)

earnings shocks and if investors have cognitive limits or are inattentive, then future returns are

2
Prior studies find that accruals are negatively associated with future stock returns. Chan et al. (2006) suggest that
some components of accruals help signal future business prospects and investors may underreact to the signal.
However, consistent with the earnings management hypothesis, they also find that the non-discretionary component
of accruals does not predict future returns and that only the discretionary components of accruals predict returns,
consistent with Xie (2001).
3
Managers generally have incentives to maximize their stock prices (even artificially). Hence, it is not always
immediately clear why managers would want to temporarily deflate their earnings. However, while smoothing
earnings downward reduces stock prices temporarily, it can also increase firm value in the long run by reducing the
cost of capital (Tucker and Zarowin, 2006). Accordingly, earnings smoothing is quite common among U.S.
corporations. Earnings smoothing became so common that the Securities and Exchange Commission (SEC) has
investigated many companies, including Microsoft, for the practice (Markoff, 1999).

2
likely to be positively correlated with current earnings changes. Because earnings management is

implemented in a direction opposite that of the earnings changes (e.g., firms manage earnings

downward (upward) when they experience large positive (negative) earnings shocks), investors

may more easily misinterpret managers’ actions. Therefore, notwithstanding the other plausible

explanations for PEAD, earnings management could be a determining factor.

Consistent with the earnings management hypothesis, we find strong evidence to suggest

that firms with large negative changes in operating cash flows manage their accruals

substantially upward, while those with large positive changes in operating cash flows manage

their accruals significantly downward.4 We also find no evidence of a positive drift for those

firms with large positive earnings changes that are least likely to have managed earnings

downward or a negative drift for those firms with large negative earnings changes that are

unlikely to have managed earnings upward.5 Furthermore, we find that most of the upward drift

after positive earnings surprises is concentrated among those firms that are most likely to have

managed earnings downward. Similarly, we find that most of the downward drift after negative

earnings surprises is concentrated among those firms that are inclined to have managed earnings

upward. We also find that this effect tends to become weaker in recent years as investors have

been paying more attention to the anomalies and hedge funds have been focusing on exploiting

them. Finally, consistent with the earnings management hypothesis, we also find that the drifts

are generally associated with discretionary (or abnormal) accruals and not with nondiscretionary

accruals.

4
Ideally, we would like to analyze the abnormal accruals of firms with large earnings changes in general. However,
because abnormal accruals are a component of earnings, it is almost impossible to totally isolate them from extreme
earnings performance. Therefore, to assess whether firms with large negative (positive) changes in operating
performance inflate (deflate) their reported earnings, we analyze the abnormal accruals of firms with large negative
(positive) changes in operating cash flows instead of firms with large negative (positive) changes in earnings.
5
We explain how we classify firms into those that are most or least likely to have managed earnings in the next
section.

3
 Taken together, the results support the earnings management hypothesis. In particular,

they suggest that PEAD is consistent with a situation where: 1) firms with large positive earnings

and large positive earnings changes manage earnings downward to create reserves and 2) firms

with large negative earnings changes manage earnings upward, particularly to avoid reporting

losses. As Kothari (2001) notes, given that accruals are a major component of earnings, the

overreaction to accruals and underreaction to earnings changes are quite intriguing. Our analysis

provides an explanation for the apparent inconsistency.

The balance of the paper is organized as follows. Section I describes the potential

correlation between earnings management and the post-earnings announcement drift. Section II

discusses the variable measurement process, some implementation issues, and our sample

selection. Section III reports abnormal accruals, earnings changes, and other descriptive statistics

for the firms. Section IV analyzes the association between operating cash flows and abnormal

accruals. Section V examines the association between abnormal accruals and the post-earnings

announcement drift. Section VI explores the drift associated with firms that are deemed least

likely to have managed earnings. Section VII investigates the drift associated with firms that are

deemed most likely to have managed earnings. Section VII provides our conclusions.

I. Analyzing the Effect of Earnings Management on the Drift

A. Abnormal Accruals as a Proxy for Earnings Management

Collins and Hribar (2000) allude to a potential link between the post-earnings

announcement drift and accruals. They posit that “the empirical fact that the market

systematically underestimates the persistence of earnings surprises might result simply from not

impounding the mean reverting tendencies of the accruals embedded within the earnings

surprise.” However, they conclude that the PEAD effect is distinct from the potential earnings

4
management effect. In particular, they observe that accruals and changes in earnings are

positively correlated and that a strategy consisting of taking long positions in firms with the most

positive changes in earnings and short positions in firms with the most negative changes in

earnings is profitable, even after removing firms with the most negative and the most positive

accruals. However, because total accruals are a very noisy proxy for earnings management, these

results do not rule out earnings management as an explanation for PEAD. In addition, the drift is

observed in the extreme earnings changes portfolios, and has substantial cross-sectional

variations within these portfolios. Hence, it is still plausible that the firms with the most positive

(negative) earnings changes that experience the drift are those that also deflate (inflate) earnings

the most.

Generally, firms that have good (poor) performance tend to have positive (negative)

accruals. Therefore, some of the firms with large positive earnings changes and large negative

accruals could be firms that simply have poor operating performance (although the changes in

earnings are positive), as opposed to firms that deflate earnings. Similarly, some of the firms

with large negative earnings changes and large positive accruals could be firms that have good

operating performance as opposed to firms that inflate earnings. To mitigate potential errors in

sorting firms into income decreasing and income increasing earnings management categories, we

use discretionary (abnormal) accruals instead of total accruals. The use of abnormal accruals as a

proxy for earnings management is consistent with Teoh et al. (1998a, 1998b), Teoh et al. (1998),

Xie (2001), Teoh and Wong (2002), DuCharme et al. (2004), Fischer and Louis (2008), and

Louis, Robinson, and Sbaraglia (2008), among others.

Although abnormal accruals estimates are less noisy than total accruals as a measure of

earnings management, they still include substantial noise. The estimated abnormal accruals could

5
be positive although the unobserved abnormal accruals are negative (and vice versa). However,

our analysis focuses on firms with the largest estimated abnormal accruals. Therefore, it is less

likely that the signs of the abnormal accrual estimates would switch because of estimation errors,

reducing the risk that we would misclassify income increasing and income decreasing abnormal

accruals.

B. Cases Most Likely Associated with Earnings Management

Kothari, Leone, and Wasley (2005) suggest that, generally, estimated discretionary

accruals are also correlated with performance. Some highly (poorly) performing firms are likely

to have positive (negative) abnormal accruals. The need to control for the effect of performance

when testing for earnings management is largely recognized in the literature (Teoh et al., 1998a,

1998b). However, extant procedures offer only partial controls for the problem. Abnormal

accruals are affected by current period earnings performance while, at the same time, they are a

component of the reported earnings. Therefore, it is practically impossible to totally separate

current period abnormal accruals from current period earnings performance. Extant procedures

that generally control for lagged earnings performance can mitigate the problem in most settings;

however, they are less effective in our setting. We are analyzing earnings management in the

context of extreme earnings changes, as opposed to stock issuances as in Teoh et al. (1998a,

1998b) or stock-for-stock mergers as in Louis (2004). Contrary to these alternative settings,

extreme earnings changes are mechanically related to the current period abnormal accruals.

To mitigate the confounding effects of performance and other potential estimation errors,

and better capture the effect of earnings management, we instead condition the analysis on the

likelihood that a firm indeed manages earnings. More specifically, we classify: 1) firms with

6
large positive earnings changes into those that are most likely and those that are least likely to

have managed earnings downward and 2) firms with large negative earnings changes into those

that are most likely and those that are least likely to have managed earnings upward. This

classification then serves as the basis for our predictions. We provide details regarding the

classification below. It is important to note that our attempt here is not to dismiss the extensive

literature on identifying earnings management using abnormal accruals. Instead, we rely on the

abnormal accrual measure and, because of the mechanical correlation between extreme earnings

changes and abnormal accruals, we try to differentiate between cases where abnormal accruals

are most likely to be associated with earnings management and cases where they are least likely

to be associated with earnings management.

1 The Likelihood of Downward Earnings Management

Fudenberg and Tirole (1995) model managers’ incentives to smooth earnings. They

suggest that in good times (i.e., when performance is high), managers “are less concerned by

their short-term prospects, and information decay gives them an incentive to save for future bad

times.” Goel and Thakor (2003) also argue that “[w]hen reported earnings are high, reporting

even higher earnings tends to elicit a relatively small positive market reaction. The company may

therefore want to “hide” some of its current earnings for reporting it in a future period when

earnings are lower and the marginal impact of a higher report is greater.” Accordingly, we

consider a firm most likely to manage earnings downward when earnings are high and least

likely to do so when earnings are negative. Loss firms can have large (estimated) negative

abnormal accruals; however, these negative abnormal accruals are more likely to be associated

with the poor earnings performance rather than downward earnings smoothing. Firms are

7
presumably unlikely to smooth earnings into negative territory. Large reported losses can result

from “big baths;” however, big baths are not consistent with the large positive earnings increases

that characterize firms in the long position of the PEAD strategy. Firms that take big baths

typically report large negative earnings changes as opposed to large positive earnings changes.

We do not suggest that firms with large positive earnings changes cannot manage

earnings upward. Obviously, some of the highly performing firms could have managed earnings

upward. We instead argue that under the earnings management hypothesis, firms with both high

positive earnings changes and high positive abnormal accruals are unlikely to be the driver of

PEAD. Our point is that when employing the earnings management hypothesis, there should not

be a positive drift for firms with high positive earnings changes and positive abnormal accruals.

Upward earnings management should lead to negative (as opposed to positive) future abnormal

returns. In addition, for firms with high positive earnings changes, abnormal accruals can be

positive even if these firms manage earnings downward as long as the performance component

of abnormal accruals is larger than the actual downward earnings management. We also note that

investors are presumably more inclined to see through earnings management when it is in the

same direction as the earnings surprise. Therefore, mispricing and subsequent corrections are less

likely for firms with high positive earnings changes and positive abnormal accruals.

2 The Likelihood of Upward Earnings Management

Previous studies also suggest that managers have strong incentives to manage earnings

upward to meet certain benchmarks, most notably analyst forecasts, seasonally lagged quarterly

earnings, and the zero earnings benchmark (Burgstahler and Dichev, 1997; DeGeorge, Patel, and

Zeckhauser, 1999; Graham, Harvey, and Rajgopal, 2005). As such, positive abnormal accruals

8
are more likely to be related to earnings inflation than to some other factors when they allow

managers to meet certain earnings benchmarks. Consequently, we consider a firm least likely to

manage earnings upward when earnings are already high and more inclined to do so when the

previous quarter earnings are above zero6 and the firm would have missed the zero earnings

benchmark if it did not report positive abnormal accruals. A firm with an extreme earnings

decline, by definition, has already missed the seasonally lagged quarterly earnings benchmark.

For such a firm, the relevant target is the zero earnings benchmark. The other benchmarks are

essentially non-binding. Accordingly, Jiang (2008) finds that beating the zero earnings

benchmark is most beneficial to a firm in terms of lowering the cost of debt.

An important advantage of using the zero earnings benchmark is that firms operating at

the breakeven point do not generally produce large abnormal accruals through their normal

operations. The abnormal accrual measurement process can underestimate “normal” accruals and

over-estimate abnormal accruals for firms with large sales increases (Kothari et al., 2005).

However, firms operating near the earnings breakeven point typically do not have large sales

increases. Therefore, their abnormal accruals are less likely to be correlated with their operating

performance. In this regard, conditioning on the likelihood of earnings management using the

zero earnings benchmark is a reasonable substitute for performance matching.

Poorly performing firms commonly have incentives to hide the extent of their losses.

However, prior studies suggest that the incentives are stronger when the earnings management

activities allow the firms to meet or beat the zero earnings benchmark.7 Firms with large positive

6
We impose this condition because, generally, the incentive and pressure to meet a benchmark weakens after a firm
has recently missed the benchmark. Yong (2007) finds that firms seldom reinitiate a second earnings string after a
break in their earnings string. He also finds that firms reduce their earnings management activities once the string is
broken, suggesting that a firm is more likely to manage earnings to meet a benchmark when it met the benchmark in
the previous period than when it missed the benchmark.
7
Prior studies find that the number of observations that fall immediately to the left (right) of the zero earnings
benchmark is abnormally low (high). This phenomenon is often interpreted as an indication that firms inflate

9
earnings can have large (estimated) positive abnormal accruals. Yet, given the mechanical

relationship between extreme earnings performance and abnormal accruals, the positive

abnormal accruals can be related to the high earnings performance as opposed to earnings

inflation. Loss firms are presumably more likely to take big baths than to inflate earnings.

However, those firms that we deem most likely to have managed earnings upward have positive

(and not negative) abnormal accruals. They are not firms that have taken big baths.

3. Summary

Based on the aforementioned considerations, we posit that if the drift is associated with

earnings management, 1) the abnormal returns associated with large positive earnings changes

will be less positive when earnings are negative and 2) the abnormal returns associated with

large negative earnings changes will be less negative when earnings are high. Conversely, we

expect future abnormal returns to be 1) positive for firms with large positive earnings changes

that also have large negative abnormal accruals and large positive earnings and 2) negative for

firms with large negative earnings changes that also have large positive abnormal accruals and

that would have missed the zero earnings benchmark if they did not report positive abnormal

accruals. These hypotheses form the bases of our analysis.

II. Variable Measurement, Implementation Issues, and Sample Selection

In this section, we describe our abnormal accrual estimation process, our abnormal return

measures, various implementation and research design choices, and the sample selection process.

earnings by small amounts to meet or beat the zero earnings benchmark. The assumption is that the observations that
fall immediately to the right of the zero earnings benchmark come from firms with small “unmanaged” losses that
would have fallen immediately to the left of the zero earnings benchmark in the absence of earnings management. A
broader interpretation of the extant evidence on the distribution of earnings, however, is that firms are less likely to
report small losses. Once they get very close to the zero earnings benchmark (whether it is through earnings
management or not), they are likely to (further) manage earnings to avoid reporting losses.

10
A. Estimating Abnormal Accruals

Following the extant literature, we proxy for discretionary accruals by the level of

abnormal accruals using the residual from a modified version of the Jones (1991) accruals

model. For each calendar quarter and two-digit SIC code industry, we estimate the following

model using all firms that have the necessary data on Compustat:

4
TAi = j-1Qj,i + 4 ∆SALEi + 5 PPEi + 6 L1TAi + 7ASSETi + 8L4NIi + i, (1)
j=1

where TA is total accruals, Qj is a binary variable taking the value of one for fiscal quarter j and

zero otherwise, ΔSALE is the quarterly change in sales, PPE is property, plant, and equipment at

the beginning of the quarter, L1TA is the lag of total accruals, ASSET is total assets at the

beginning of the quarter, L4NI is net income for the same quarter of last year, and ε is the

regression residual. We measure total accruals as the difference between net income and cash

flows from operations. All of the variables, including the indicator variables, are scaled by total

assets at the beginning of the quarter. After we deflate the model, ASSET is transformed into a

column of ones allowing us to estimate the model with the standard intercept. To mitigate the

effects of outliers and errors in the data, we delete the top and bottom one percentiles of all the

variables in the model. We also require at least 20 observations for each estimation. The model

has an average adjusted R2 of 28.81%.

The abnormal accruals model is a modification of the models used in Louis and White

(2007) and Gong et al. (2008a, 2008b). We include lagged net income (L4NI) in the model to

mitigate the effect of performance on abnormal accruals. Kothari et al. (2005) suggest that one

means of controlling for the effect of performance on abnormal accruals is to include lagged

earnings in the accrual model. They find that misspecification problems are attenuated when

11
lagged ROA is included in the model. Our results hold if we use Kothari et al.’s (2005) matching

procedure. However, trading strategies based on abnormal accruals estimated by this procedure

are not implementable. In particular, we cannot ensure that the potential matches would have

already reported their earnings by the time of the portfolio formation, which would have been

necessary to implement the trading strategies.

B. Estimating Abnormal Returns

Following Collins and Hribar (2000), we measure abnormal returns over the period from

18 trading days after the earnings announcement for Quarter 0 (the current quarter) to 17 trading

days after the earnings announcement for Quarter +2 (the second quarter after Quarter 0). There

is no consensus in the literature regarding how to estimate long-term abnormal returns (see

Barber and Lyon, 1997; Kothari and Warner, 1997; Fama, 1998; Ikenberry, Shockley, and

Womack, 1999; Lyon, Barber, and Tsai, 1999; Brav, Geczy, and Gompers, 2000; and Mitchell

and Stafford, 2000 for issues related to estimating long-term abnormal returns). However, in the

case of the post-earnings announcement drift, the return horizon is relatively short – generally six

months. Issues related to long-term abnormal return measurements are more serious for return

horizons of one year or more (Ikenberry et al., 1999). Accordingly, many studies on post-

earnings announcement drifts use size-adjusted buy-and-hold returns.

To ensure that our results are robust to the concerns expressed by Fama (1998) and Brav

et al. (2000), among others, about inferences based on long-term buy-and-hold returns, we also

use Fama and French’s (1993) three-factor model. The three-factor model controls for the market

factor, the size factor, and the book-to-market factor. Carhart (1997) extends the three-factor

model by including a momentum factor. We do not use the four-factor model since Chordia and

Shivakumar (2006) suggest that the momentum effect is related to the earnings change effect.

12
C. Implementing the Trading Strategies

The execution of the study involves some important implementation issues. This section

discusses these issues and the design choices that we make.

First, to ensure that the hedge portfolio strategies are implementable, the entire

distribution of earnings, earnings changes, and abnormal accruals must be known prior to the

portfolio formation date. However, because firms do not announce earnings or file proxy reports

at the same time, the variables in the abnormal accrual model are not available at the same time

for all the firms. To address this issue, we estimate the accrual model four quarters prior to the

quarter of the portfolio formation and then apply the estimated parameters of the model to the

data for the individual firms in the quarter of the portfolio formation.

Additionally, the study involves sorting the sample firms into quintiles of earnings

changes, abnormal accruals, and earnings. We determine the cut-off points for abnormal

accruals, earnings changes, and earnings in Quarter -4. Since we compute abnormal accruals for

the quarter of the portfolio formation (Quarter 0) using parameters estimated in the previous year

(Quarter -4), we also compute abnormal accruals for Quarter -4 using parameter estimates in

Quarter -8. Note, however, that our results are robust to the timing of the abnormal accrual

estimation. We obtain similar results whether we estimate the model in Quarter 0, Quarter -4, or

Quarter -8.

Furthermore, firms in the short position typically have relatively high earnings in the

previous year (they have large earnings declines and still report positive earnings). Considering

their previous performance, many of these firms are likely to have incentives to keep their

reported earnings at a certain distance above the zero earnings benchmark. However, it could be

argued that firms are more likely to have inflated earnings when they marginally beat their

benchmarks than when they beat the benchmarks by large amounts. To address this issue, we

13
also require that the firms in the short position have earnings below 1% of assets (as opposed to

below the top earnings quintile). The tighter restriction also recognizes that the magnitude of the

estimated abnormal accruals does not correspond to the actual amount of earnings management.

Firms sometimes manage earnings by large amounts, depending on the managers’ motivation.

Hennes, Leone, and Miller (2008) report a net amount of earnings overstatement of about 12% of

the total assets for firms that restate earnings due to irregularities.8 However, due to the potential

noise in abnormal accrual estimates, we use large positive (negative) abnormal accruals only as

indications of upward (downward) earnings management. Therefore, by requiring that reported

earnings be close to zero, we increase the probability that the firms with large estimated

abnormal accruals would have reported losses if they did not inflate their earnings.

Moreover, since abnormal accruals are deflated by total assets, we also use total assets to

deflate earnings changes.9 We compute earnings changes as the seasonal change in the ratio of

net income to beginning total assets. We obtain qualitatively similar results to those reported in

our paper if we use change in net income divided by beginning total assets. We use the change in

the ratio of net income to beginning total assets as it better adjusts the performance measure for

changes in the size of the firm’s operation. Consider, for instance, the case where a firm issues

stock during the quarter. Net income for the current quarter and net income for the same quarter

of the previous year are not directly comparable due to the change in the size of the firm’s

operations. However, deflating net income by the corresponding beginning total assets adjusts

for the difference in the size of the firm’s operations. This issue does not exist when using

8
The 12% is the net average of overstatement. The sample includes both overstatement and understatement.
Therefore, the average amount of overstatement (excluding the cases of understatement) is more than 12%. Some of
the irregularities cover multiple periods; however, they are unlikely to be evenly distributed over the quarters and,
for many firms, the irregularities take place in a single period.
9
Note, however, that the hedge portfolio returns are qualitatively similar if we define earnings changes as the
seasonal change in earnings per share deflated by price at the beginning of the current quarter.

14
change in earnings per share deflated by price as both earnings per share and price are adjusted

for size. Firms with negative assets are deleted.

Finally, to determine the level of earnings, we deflate net income by beginning total

assets. We deflate by assets instead of market value to differentiate the earnings management

effect from the value strategy effect. Earnings-to-price can be high not because earnings are high,

but because the deflator (price) is low. Under the earnings management explanation, the drift

should be associated with earnings, whether earnings are deflated by market value or total assets.

However, the earnings management effect is confounded with the value strategy effect when

market value is used as the deflator.

D. Sample Selection

Our sample period extends from the second quarter of 1990 to the fourth quarter of 2006.

Cash flow (Compustat Data Item #108 - Compustat Data Item #78), used to estimate accruals, is

available starting in 1987. However, because we sort the sample on three different dimensions

and the cash flow variable on Compustat is sparsely populated in 1987, there are not enough data

points in 1987 to conduct our analyses. Since we use lagged accruals in our accrual regression

model, the first period for which we can estimate abnormal accruals is the second quarter of

1988. Due to the restrictions imposed by the implementation of the trading strategy, the data

necessary to conduct the analysis are only available from the second quarter of 1990 forward.

The sample period ends in the fourth quarter of 2006. We measure abnormal returns over the

months after the earnings announcement. Completing the sample in 2006 ensures that our results

are not affected by the recent financial crisis that started in 2008.

We require that a firm release its earnings report within 45 (90) days of the end of the

three interim (the fourth) fiscal quarters, which is the Securities and Exchange Commission

15
(SEC) filing deadline for most of our sample period. Since late announcers tend to face some

special circumstances (e.g., SEC investigation), the post-earnings announcement returns of late

announcers are less likely to be associated with managers’ manipulation of the current fiscal

quarter earnings report.

III. Abnormal Accruals, Earnings Changes, and Other Descriptive Statistics

Table I analyzes the abnormal accruals and earnings changes of the sample firms. The

average abnormal accrual for firms with the most negative earnings changes that are deemed

most likely to have managed earnings upward is 0.066 (Panel A). We will later demonstrate that

these firms experience the most negative drift. Similarly, the average abnormal accrual for firms

with the most positive earnings changes that are considered most likely to have managed

earnings downward is -0.069 (Panel B). These firms experience the most positive drift. In

comparison, the average abnormal accrual for firms with the most negative earnings changes that

are deemed least likely to have managed earnings upward is 0.046 (Panel A), while the average

abnormal accrual for firms with the most positive earnings changes that are least likely to have

managed earnings downward is -0.074 (Panel B). The average abnormal accruals for the other

firms with the most negative earnings changes and the most positive abnormal accruals is 0.073.

The average abnormal accruals for the other firms with the most positive earnings changes and

the most negative abnormal accruals is -0.063. Therefore, there is no evidence that the extreme

abnormal accruals are concentrated among the firms in our hedge portfolios as opposed to among

those firms that we believe least likely to have managed earnings. This observation is important

as our sample partition could yield larger hedge portfolio returns simply because it concentrates

on the long position firms with the largest negative abnormal accruals and on the short position

16
firms with the largest positive abnormal accruals. However, the statistics in Table I demonstrate

that this is not the case.

Insert Table I about here.

The same argument is valid for change in earnings. The statistics in Table I provide no

evidence that the extreme changes in earnings are concentrated among the firms in our hedge

portfolios as opposed to those firms that we deem least likely to have managed earnings. The

average earnings change for firms with the most negative earnings changes that are deemed most

likely to have managed earnings upward is only -0.031 (Panel A). The average earnings change

for firms with the most positive earnings changes that are considered most likely to have

managed earnings downward is only 0.044 (Panel B). In comparison, the average earnings

change for firms with the most negative earnings changes that are deemed least likely to have

managed earnings upward is -0.033 (Panel A). The average earnings changes for firms with the

most positive earnings changes that are considered least likely to have managed earnings

downward is 0.086 (Panel B). The average earnings change for the other firms with the most

negative earnings changes and the most positive abnormal accruals is -0.054. The average

earnings change for the other firms with the most positive earnings changes and the most

negative abnormal accruals is 0.036. To facilitate the comparison, we also report the distributions

of earnings changes and abnormal accruals by quintiles of earnings changes (Panel C) and

quintiles of abnormal accruals (Panel D).

Table II presents the average risk characteristics for firms in the top and bottom quintiles

of earnings changes. The statistics in Panel A of Table II report that firms in the top and bottom

quintiles of earnings changes differ on many potential risk factors such as SIZE, BM (book-to-

market value), MM (momentum), ARBRISK (arbitrage risk), PRICE, VOLUME, and IO

17
(institutional ownership). The statistics in Panel B of Table II indicate that these variables are

also significantly different across firms in the top quintiles of earnings changes that are deemed

most likely to have managed earnings downward and those in the bottom quintiles of earnings

changes that are considered most likely to have managed earnings upward. We control for the

effect of these variables in our regression models.

Insert Table II about here.

IV. Association Between Operating Cash Flow and Abnormal Accruals

In Table I, we find that the abnormal accruals in the short position are positive, while

those in the long position are negative. One reason for this observation is that the short position

has only SUE1/ABAC5 firms and the long position has only SUE5/ABAC1 firms. By

construction, the ABAC5 firms are companies with positive abnormal accruals and the ABAC1

firms are companies with negative abnormal accruals. Therefore, the statistics in Table I are not

informative regarding the smoothing activities of the sample firms.

As explained earlier, we would prefer to analyze the abnormal accruals of firms with

large earnings changes in general. However, since abnormal accruals are a component of

earnings, it is practically impossible to totally separate current period abnormal accruals from

extreme current period earnings performance. Extreme earnings changes and estimated abnormal

accruals have a mechanical positive correlation. Therefore, to assess whether firms with large

negative (positive) changes in operating performance inflate (deflate) their reported earnings, we

analyze the abnormal accruals of firms with large negative (positive) changes in operating cash

flows.

Insert Table III about here.

18
 The results, reported in Table III, are consistent with the earnings smoothing hypothesis.

We find that, on average, firms with large negative changes in operating cash flows manage their

accruals upward, whereas those with large positive changes in operating cash flows manage their

accruals downward. More specifically, firms in the bottom quintile of changes in operating cash

flows (SUCF1) report an average earnings change of -1.8% and an average abnormal accrual of

2.8% of assets. In contrast, firms in the top quintile of changes in operating cash flows (SUCF5)

report an average earnings change of 2.3% and an average abnormal accrual of -2.1% of assets.

These abnormal accruals are both statistically significant and economically substantial.

V. Association Between Abnormal Accruals and the Post-Earnings Announcement Drift

We first analyze the association between abnormal accruals and the post-earnings

announcement drift after controlling for other factors that might affect future returns. We model

the post-announcement drift as follows:

ABRETit = 1SUEQ1it + 2SUEQ5it + 3SUEQ1it*ABACQit + 4SUEQ5it*ABACQit

+ 5SIZEDit + 6BMit + 7MMit + 8ARBRISKit + 9PRICEDit
+ 10VOLUMEDit + 11IOit + it (2)

ABRET is the size-adjusted return measured over the period from 18 trading days after the

earnings announcement for Quarter 0 (the current quarter) to 17 trading days after the earnings

announcement for Quarter +2 (the second quarter after Quarter 0). SUEQ1 is a binary variable

taking the value one if the standardized earnings change is in the bottom quintile and zero if it is

in the top quintile. SUEQ5 is a binary variable taking the value one if the standardized earnings

change is in the top quintile and zero if it is in the bottom quintile. ABACQ is quintile of

abnormal accruals (the residual from the accrual model). SIZED is the decile of the market value

of common equity at the beginning of the current quarter and BM is the ratio of the book value of

equity to the market value of equity at the beginning of the current quarter. MM, momentum, is

19
the stock return from twelve to two months prior to the earnings announcement month. ARBRISK

is arbitrage risk, defined as the residual variance from a standard market model regression of a

firm’s return on the returns of the CRSP S&P 500 equal-weighted market index over a 48-month

period ending one month prior to the earnings announcement month. PRICED is the decile of

stock price at the beginning of the current quarter. VOLUMED is the decile of dollar trading

volume over the 12-month period ending at the end of the month immediately preceding the

earnings announcement quarter and IO is the percentage of institutional ownership at the end of

the calendar quarter prior to the end of the earnings announcement quarter.

All five quintiles of abnormal accruals are included in the analysis. We do not have a

main effect for ABACQ as the sample includes only those firms in either the top or the bottom

quintile of earnings changes. We allow for two separate coefficients on ABACQ: 1) one for those

firms in the bottom quintile of earnings changes (the SUE1 firms) and 2) one for those firms in

the top quintiles of earnings changes (the SUE5 firms). We scale ABACQ to range from -0.5 to

0.5, such that the coefficients on SUEQ1 and SUEQ5 (1 and 2) represent the average negative

and positive drifts for firms in the median quintile of abnormal accruals, respectively. We

include the control variables in the models because, as reported in Panel A of Table II, firms in

the top and bottom quintiles of earnings changes differ on many of the potential risk factors. Due

to inflation, the most recent observations of size, stock price, and trading volume generally have

the largest values and the earliest observations generally contain the lowest values. To remove

the effects of the time trends in these variables each quarter, consistent with Bartov,

Radhakrishnan, and Krinsky (2000), we rank them into deciles. We also rescale them to range

from -0.5 to 0.5, consistent with Mendenhall (2004).

The results are reported in Table IV. Consistent with the notion that PEAD is related to

firms with large negative earnings changes that manage earnings upward, we find that the

20
negative return associated with large negative earnings changes increases in abnormal accruals.

That is, the more a firm manages earnings upward, the higher the return from the short position

of a trading strategy based on earnings changes. Similarly, consistent with the notion that PEAD

is related to firms with large positive earnings changes that manage earnings downward, we find

that the positive return associated with large positive earnings changes decreases in abnormal

accruals. That is, the more a firm manages earnings downward, the higher the return from the

long position of a trading strategy based on earnings changes.

Insert Table IV about here.

There are some other important observations to be noted from the results in Table IV. We

find no evidence that the coefficient on SUEQ1 (1) is significantly negative. In addition, the

coefficient on SUEQ1 for those firms in the bottom quintile of abnormal accruals (1 - 0.5*3) is

actually positive, although it is not statistically significant. These results indicate that the

negative return associated with SUEQ1 is observed mainly for those firms that also report large

positive abnormal accruals. Moreover, the coefficient on SUEQ5 (2) is significantly positive.

Specifically, there is a positive drift even for the SUEQ5 firms that have abnormal accruals in the

middle quintile. One potential explanation is that firms that have large earnings increases tend to

also have large increases in accruals (a component of earnings). Some of these firms can still

report positive abnormal accruals even after smoothing earnings downward. In the next section,

we control for this eventuality by conditioning the analysis on the likelihood that a firm manages

earnings.

The coefficient on the interaction between SUEQ5 and ABACQ is -0.05 implying that the

average abnormal return is 0.015 (0.04 - 0.05*0.5) for firms that fall in both the top quintile of

earnings changes and the top quintile of abnormal accruals. At first glance, this finding might

21
seem inconsistent with the earnings management hypothesis. Note, however, that the results in

Table IV are simply a replication of the basic results in Collins and Hribar (2000). Our argument

is that for firms with extreme earnings, abnormal accruals often reflect the effect of performance.

That is, firms that have high SUEs tend to naturally have high accruals. Therefore, further

refinements are necessary in order to make inferences regarding earnings management in such a

setting. It is also important to note that the average abnormal return of 0.015 is neither

economically nor statistically significant. It is practically zero. As we demonstrate in Tables V

and VI, there is no evidence of a drift for: 1) firms with large positive earnings changes that are

deemed least likely to have managed earnings downward or 2) firms with large negative earnings

changes that are considered least likely to have managed earnings upward.

VI. The Drift for Those Firms that are Deemed Least Likely to Have Managed Earnings

We analyze the drift associated with those firms that are deemed least likely to have

managed earnings. Essentially, we implement a strategy consisting of taking long positions in

firms with large positive earnings changes that are least likely to have managed earnings

downward and short positions in firms with large negative earnings changes that are least likely

to have managed earnings upward. If the drift is associated with earnings smoothing, then we

expect it to disappear or, at least, to weaken for those firms that are deemed least likely to have

smoothed their earnings.

The long position includes firms with earnings changes in the top quintile and earnings

below zero, while the short position includes firms with earnings changes in the bottom quintile

and earnings in the top quintile. We impose no restriction on the abnormal accruals estimates. It

would be easier to demonstrate that there is no evidence of a drift if we take: 1) long positions in

firms with large positive earnings changes, large positive abnormal accruals, and negative

22
earnings and 2) short positions in firms with large negative earnings changes, large negative

abnormal accruals, and large positive earnings. However, there is a potential confounding effect

when analyzing the returns of firms with large positive earnings changes that also have positive

abnormal accruals or the returns of firms with large negative earnings changes that also have

negative abnormal accruals. The positive drift associated with undervaluation of positive

earnings changes could be offset by the negative drift associated with overvaluation of positive

abnormal accruals. Similarly, the negative drift associated with undervaluation of negative

earnings changes could be offset by the positive drift associated with overvaluation of negative

abnormal accruals. Therefore, we could fail to observe a drift, although the market could have

undervalued the earnings changes. We mitigate this problem by not requiring that firms with

large positive earnings changes also have large positive abnormal accruals or that firms with

large negative earnings changes also have large negative abnormal accruals.

The results of our analysis are reported in Table V. Consistent with the earnings

management hypothesis, the buy-and-hold hedge portfolio return is essentially zero (and, if

anything, it is negative) for those firms with large positive earnings changes that are least likely

to have managed earnings downward and those with large negative earnings changes that are

least likely to have managed earnings upward. A long position in firms with the most positive

earnings changes yields an average buy-and-hold abnormal return of -1.2%. A short position in

firms with the most negative earnings changes yields an average abnormal return of -1.0%, for a

hedge portfolio return of -2.2% over the two quarters after the earnings announcement. We

obtain similar results when we use Fama and French’s (1993) three-factor model. The average

monthly abnormal returns are 0.4% for the long position and -0.1% for the short potion. These

are not statistically significant at conventional levels.

Insert Table V about here.

23
 It is plausible that we do not fully control for the offsetting effects of positive earnings

changes and positive abnormal accruals, or the offsetting effects of negative earnings changes

and negative abnormal accruals. To control for the effect of abnormal accruals and other factors

that could affect the abnormal returns, we regress abnormal returns on an indicator variable for

large earnings changes, abnormal accruals, and other control variables. More specifically, we

estimate the following regression model using those firms with earnings changes in either the top

or the bottom quintile that are deemed least likely to have managed earnings:

ABRETit = 0 + 1SUEQ5it + 2ABACQit + 3SIZEDit + 4BMit + 5MMit

+ 6ARBRISKit + 7PRICEDit + 8VOLUMEDit + 9IOit + it, (3)

where SUE5 is a binary variable taking a value one if an earnings change is in the top quintile

and zero if it is in the bottom quintile. ABACQ is the quintile of abnormal accruals. We assign to

the abnormal accrual quintiles values ranging from -0.5 to 0.5. Again, there is no restriction on

the abnormal accruals estimates. The other variables are as defined previously.

The results are reported in Table VI. Under Column (1), we control only for ABACQ.

Consistent with the results in Table V, the coefficient on SUE5 is not significantly positive. In

fact, the sign of the coefficient on SUE5 is negative. The average abnormal return for firms with

the most positive earnings changes, 0 + 1, is virtually zero. Hence, there is no evidence that, on

average, the firms with large positive earnings changes that we deem least likely to have

managed earnings downward experience positive abnormal returns after the earnings

announcement or the firms with large negative earnings changes that are considered least likely

to have managed earnings upward experience negative abnormal returns after the earnings

announcement. The coefficient on ABACQ is not statistically negative either. This is not

consistent with the predictions of the abnormal accrual anomaly. However, this result is not

surprising given that we limit the sample to firms that are deemed least likely to have managed

24
earnings. Xie (2001) suggests that the abnormal accrual anomaly is related to earnings

management.10

Insert Table VI about here.

Under Column (2), we control for various factors that could affect returns. We still find

no evidence that the average abnormal return for firms with the most negative earnings changes,

0, is negative. In fact, 0 is highly positive (0.047). There is also no evidence that the average

abnormal return for firms with the most positive earnings changes, 0 + 1, is positive. The

average return for these firms is essentially zero for a hedge portfolio return of approximately -

0.016. Therefore, there is no evidence that the disappearance of the drift (documented in Table

V) for those firms with the most negative earnings changes that are least likely to have managed

earnings upward and those with the most positive earnings changes that are least likely to have

managed earnings downward is due to a failure to control for correlated omitted variables. If

anything, the large positive intercept (0) suggests that, typically, those firms with large negative

earnings changes that are considered least likely to have managed earnings upward might have

actually managed earnings downward. Hence, the best strategy, if any, would be to take a long

position (instead of a short position) in those firms with large negative earnings changes that are

least likely to have managed earnings upward.

One potential explanation for the disappearance of the drifts is that the extreme earnings

changes and abnormal accruals might be concentrated among those cases that are more likely to

be associated with earnings management as opposed to those least likely to be associated with

earnings management. However, the results in Table V indicate that, among the firms that are

least likely to have managed earnings, the average earnings changes are -0.032 for those with the

10
The coefficient on abnormal accruals is nonetheless negative, albeit insignificantly so. Some firms with large
positive earnings changes, negative earnings, and positive abnormal accruals could have managed earnings upward
depending on the managers’ incentives. Similarly, some firms with large negative earnings changes, large positive
earnings, and negative abnormal accruals could have managed earnings downward.

25
most negative earnings changes and 0.085 for those with the most positive earnings changes.

These are not substantially different from the average earnings changes for the other subgroups

(see Table I). As discussed earlier, the results in Table I provide no evidence that the extreme

abnormal accruals or extreme earnings changes are concentrated among the firms in our hedge

portfolios as opposed to those that are considered least likely to have managed earnings.

VII. The Drift for Those Firms that are Deemed Most Likely to Have Managed Earnings

Thus far, we demonstrate that there is no evidence of PEAD for those firms that are

deemed least likely to have managed earnings. We now analyze the drift for those firms that are

most likely to have managed earnings. Firms with large positive earnings changes are most likely

to manage earnings downward (and negative estimated abnormal accruals most likely related to

earnings smoothing) when earnings are high. Firms with large negative earnings changes are

most likely to have managed earnings upward (and positive estimated abnormal accruals most

likely related to earnings smoothing) when the previous quarter earnings are above zero and the

firms would have missed the zero earnings benchmark in the current quarter if they did not report

positive abnormal accruals.

A. Unconditional Hedge Portfolio Returns

Consistent with our expectations, the results reported in Panel A of Table VII

demonstrate a substantial increase in the portfolio returns. The long (short) positions yield

average abnormal returns of 9.1% (10.4%), for a total hedge portfolio return of about 19.5%. In

contrast, untabulated results indicate that a trading strategy based on earnings changes alone

yields a hedge portfolio return of only 5.0%, while a trading strategy based on earnings changes

26
and abnormal accruals alone yields a hedge portfolio return of only 12.6%. We obtain a similar

pattern when we use Fama and French’s (1993) three-factor model.

Insert Table VII about here.

The numbers of observations in our portfolios are relatively small. Assuming that an

investor holds these positions for the entire return horizon (two quarters), he will typically hold

approximately 80 stocks in the long position and 30 stocks in the short position at a given time.

However, provided that there are a sufficient number of liquid stocks available in these

portfolios, the small number of observations may not impede implementing our strategy. In a

given quarter, the number of firms in our long and short positions are closer to the number of

firms that some arbitrageurs apparently hold in their portfolios. For example, Mendenhall (2004)

mentions the case of the manager of a $1 billion portfolio who concentrates his holdings among

40 stocks. Mendenhall (2004) also reports that “one partner of a hedge fund advisory service

confided that the firm has several clients who hold fewer than 25 long (and short) positions.”

Untabulated results also report a striking difference in the abnormal returns between

firms that slightly miss the zero earnings benchmark and those that meet or slightly beat the

benchmark. The average abnormal return for firms with large negative earnings changes and

positive abnormal accruals is only -5.4% for firms with ROAs between -0.5% and 0.0% (i.e.,

firms that slightly miss the zero earnings benchmark). The average abnormal return is

dramatically more negative (-11.3%) for firms with ROAs between 0.0% and 0.5% (i.e., firms

that meet or slightly beat the zero earnings benchmark) and -9.5% for those with ROAs between

0.5% and 1.0%. Limiting the short position to firms with positive ROAs below 0.5% (instead of

positive ROAs below the top quintile) increases the abnormal return from the short position by

4.6% (from 10.4% to 15.0%) for a total hedge portfolio return of 24.1%. Therefore, the average

abnormal return from the hedge portfolio is larger when we limit the short position to firms with

27
positive ROAs below 0.5%. However, the number of observations becomes relatively small,

which may further impede the implementation of the strategy.

There is some profit from going short in firms with large negative earnings changes and

large positive abnormal accruals that report losses. Untabulated results indicate that such a

strategy yields an average abnormal return of 4.4% over the two quarters consistent with the

notion that some firms could have incentives to inflate earnings even if they cannot reach the

zero earnings benchmark. However, the drift is much larger (10.4%) for firms that report positive

earnings, but would have reported losses if they did not also report positive abnormal accruals.

This is consistent with the conjecture that positive abnormal accruals are most likely to be

associated with earnings management when they allow a poorly performing firm to meet the zero

earnings benchmark.

The long position of our trading strategy includes firms with large negative abnormal

accruals and large positive earnings, and the short position includes firms with large positive

abnormal accruals that would have reported losses if they did not report positive abnormal

accruals. Therefore, it is unlikely that the large abnormal accruals in the long and short positions

are driven by performance.11 Nonetheless, to further assess the extent to which the abnormal

accrual effect is related to managerial discretionary financial reporting behavior, we repeat the

analysis using estimated nondiscretionary (normal) as opposed to estimated discretionary

(abnormal) accruals.

Consistent with the earnings management hypothesis, we find that the improvement in

abnormal returns is largely due to the discretionary (or abnormal) component of accruals. The

results reported in Panel B of Table VII indicate that our trading strategy yields an average

11
Furthermore, since our analyses are based on the top and bottom quintile ranking of abnormal accruals (instead of
the actual abnormal accrual estimates), it is less likely that our inferences are driven by potential noise in our
abnormal accrual measure.

28
abnormal return of 12.6% over the two quarters after the earnings announcement when we use

nondiscretionary (normal) accruals instead of abnormal accruals as a partitioning variable. 12 This

return is actually lower than the return (13.7%) that we obtain when we do not partition on

accruals at all, and substantially lower than the 19.5% return that we obtain when we partition on

abnormal accruals. The main difference between the return from the hedge portfolio based on

abnormal accruals and the return from the hedge portfolio based on nondiscretionary (normal)

accruals comes from the short position. The average abnormal return is 0.6% higher when

sorting on discretionary accruals than when sorting on nondiscretionary accruals (9.1% vs. 8.5%)

for the long position and 6.3% higher (10.4% vs. 4.1%) for the short position. These results

suggest that abnormal accruals are a less noisy measure of earnings management for firms in the

short position than for those in the long position. We also note that the strategy involving

nondiscretionary accruals is profitable. One reason is that the portfolios are formed on the basis

of the likelihood that a firm manages earnings, and those firms with large negative (positive)

earnings changes that are deemed to have inflated (deflated) earnings the most can also have

large positive (negative) estimated nondiscretionary accruals.13

Insert Figure I about here.

Figure I presents a synoptic view of the drifts associated with various trading strategies.

The results indicate that a strategy based on earnings changes yields a short (long) position return

of 2.1% (2.9%) over the 120 trading days starting 18 trading days after the earnings

announcement date. Using abnormal accruals as an additional partitioning variable increases the

12
Nondiscretionary (normal) accrual is the difference between total accrual and discretionary (abnormal) accrual
(i.e., the predicted value from the abnormal accrual model).
13
Note that we do not have a perfect proxy for discretionary/non-discretionary accruals and that abnormal accruals
can be affected by factors other than earnings management. In addition, firms can use real earnings management that
does not involve accruals. Therefore, considering that we classify firms on their incentives to smooth earnings, it is
not surprising to observe abnormal returns whether we further classify the firms on the estimated abnormal accruals
or not. However, it is worth noting the substantial difference in the abnormal returns when we sort on the estimated
discretionary (abnormal) as opposed to the non-discretionary accruals.

29
abnormal returns to the short (long) position to 5.8% (6.8%). If we further condition on the

likelihood of earnings management, the abnormal returns increase to 9.7% and 9.1% for the short

and long positions, respectively, for an average hedge portfolio return of about 18.8%. These

results indicate that our findings are robust to measuring abnormal returns over a fixed horizon.14

A casual inspection of Figure I also indicates that our results are not sensitive to alternative

starting dates. The drift is quite smooth over the return horizon. Therefore, an investor does not

have to open the positions 18 trading days after the earnings announcement. He can actually

exploit our strategy by opening the positions almost anytime during the return horizon.

B. Conditional Hedge Portfolio Returns

As reported in Panel B of Table II, the hedge portfolio firms are different from the

population. Firms in the long and short positions are also varying in many important factors,

although the disparities are not extreme. We control for the potential effects of these differences

by estimating the following regression models for: 1) firms with large positive earnings changes

and large negative abnormal accruals (Model 4A) and 2) firms with large negative earnings

changes and large positive abnormal accruals (Model 4B):

ABRETit = 1MOST_DOWNit + 2LEAST_DOWNit + 3OTHER_DOWNit + 4SIZEDit

+ 5BMit + 6MMit + 7ARBRISKit + 8PRICEDit + 9VOLUMEDit
+ 10IOit + it, (4A)

ABRETit = 1MOST_UPit + 2LEAST_UPit + 3OTHER_UPit + 4SIZEDit + 5BMit

+ 6MMit + 7ARBRISKit + 8PRICEDit + 9VOLUMEDit + 10IOit + it, (4B)

14
The returns in the graph are cumulated over the 120 days starting 18 trading days after the earnings announcement
(Quarter 0) whereas those in the tables are measured over the period from 18 trading days after the earnings
announcement for Quarter 0 (the current quarter) to 17 trading days after the earnings announcement for Quarter +2
(the second quarter after Quarter 0).

30
where ABRET is the size-adjusted return over the period from 18 trading days after the earnings

announcement for Quarter 0 (the current quarter) to 17 trading days after the earnings

announcement for Quarter +2 (the second quarter after Quarter 0) (the expected return is the

compound return of the respective firm’s CRSP NYSE/AMEX/NASDAQ size-decile portfolio

assignment at the beginning of the year. MOST_DOWN is a binary variable taking a value of one

if (standardized) earnings are high (in the top quintile), and zero otherwise. LEAST_DOWN is a

binary variable taking a value of one if earnings are negative, and zero otherwise.

OTHER_DOWN is a binary variable taking a value of one if both MOST_DOWN and

LEAST_DOWN are zero, and zero otherwise. MOST_UP is a binary variable taking a value of

one if the previous quarter earnings are above zero and the firm would have missed the zero-

earnings benchmark in the current quarter if it did not report positive abnormal accruals, and zero

otherwise. LEAST_UP is a binary variable taking a value of one if (standardized) earnings are

high (in the top quintile), and zero otherwise. OTHER_UP is a binary variable taking a value of

one if both MOST_UP and LEAST_UP are zero, and zero otherwise. The other variables are as

previously defined.

The results are reported in Table VIII. Consistent with our expectations, the coefficient

on MOST_DOWN for the long position in Panel A is the highest (0.091). That is, after accounting

for the control variables in the regression, the long position of the hedge portfolio earns an

abnormal return of 9.1% when it is limited to those firms that are deemed most likely to have

managed earnings downward. The coefficient on MOST_UP for the short position in Panel B is

−0.083. Specifically, after accounting for the control variables, the short position of the hedge

portfolio earns an abnormal return of 8.3% when it is limited to those firms that are deemed most

likely to have managed earnings upward, for a total hedge portfolio return of 17.4%. The results

31
are generally the same whether we use a pooled regression or the Fama and MacBeth (1973)

procedure. They are also quite close to those reported in Table VII. Therefore, it does not appear

that the hedge portfolio returns are due to the documented differences in the sample firms’

characteristics.15

Insert Table VIII about here.

C. Other Sensitivity Analyses

1. Controlling for the Lag Between 10-Q Filing and 10-K Filing

The lag between the earnings announcement date and the 10-K filing date is generally

different from the lag between the earnings announcement date and the 10-Q filing date. Easton

and Zmijewski (1993) indicate that the average lag between the end of the fiscal quarter and the

filing date is 94.2 (46.8) days for the fourth quarter (the three interim quarters) for NASDAQ

firms and 97.2 (44.7) days for NYSE/AMEX firms. Accordingly, to assess the extent to which

the difference in the lag between the end of the fiscal quarter and the filing date could affect our

results, for the fourth quarter, we measure the abnormal returns from 98 calendar days (or 14

weeks) after the fiscal quarter end to 97 days after the second fiscal quarter end subsequent to the

earnings announcement. For the interim quarters, we measure the abnormal returns from 49

calendar days (or 7 weeks) after the fiscal quarter end to 48 days after the second fiscal quarter

end subsequent to the earnings announcement. We obtain an average hedge portfolio return of

19.5%. Consequently, it does not appear that controlling for the lag between the earnings

announcement date and the 10-Q filing date affects our results.

15
The coefficient on momentum is negative and marginally significant for the short position, which might seem
surprising given that momentum is generally positively associated with future returns. However, Chordia and
Shivakumar (2006) suggest that the momentum effect could be associated with the change in earnings effect.
Therefore, there is not a strong reason why momentum should be positively correlated with future returns within the
change in earnings quintiles.

32
2. The Potential Effect of Special Items

Prior studies suggest that the market under reacts less to special items than to other

earnings components (Burgstahler, Jiambalvo, and Shevlin, 2002). It is also possible that the loss

firms with the most positive earnings changes and the most negative abnormal accruals have

large asset write-downs and report earnings with large negative and transitory components,

which could explain the disappearance of the drift for these firms. To assess the potential effect

of special items, we replicate the findings in Table VI after adjusting all the earnings variables

for special items (Compustat Quarterly Data Item 32). We set the tax rate to the income tax

expense (Compustat Quarterly Data Item 6) divided by the pre-tax earnings (Compustat

Quarterly Data Item 23) if both the income tax expense and the pre-tax earnings are positive and

to zero if either the income tax expense or the pre-tax earnings is non-positive. We obtain an

average hedge portfolio return of 19.9%. As a result, special items are not the drivers of our

hedge portfolio returns.

3. The Potential Effect of Recently Heightened Attention to the Anomalies

If earnings management affects the delayed market reaction to earnings announcements

and if this effect is due to investor inattention, we would expect the effect to at least decrease as

investors pay more attention to the anomalies. Consistent with this view, Schwert (2003) finds

that many well-documented anomalies tend to disappear as they become known and investors

start paying attention to them. Using accruals up to 2002, Lev and Nissim (2006) find no

evidence that the returns from the accrual anomaly have declined over time. However, since the

early 2000s, there has been an explosion in the number of hedge funds and the size of the assets

under their management (Stultz, 2007). Many of these funds specialized in exploiting anomalies

33
documented in the academic literature and have been actively recruiting analysts with Ph.D.

degrees and prominent accounting professors. As such, there has been much more attention

brought to the anomalies after 2002 and the attention effect is more likely to be felt since then.

Green, Hand, and Soliman (2010) provide an extensive discussion of the effect of hedge funds

on the potential disappearance of the accrual anomaly.

To assess the extent to which the recent increase in attention to the anomalies affects our

results, we analyze the time series of the abnormal returns from our trading strategies. As

reported in Table IX, our trading strategy is profitable in every single year from 1990-2006.

However, consistent with the inattention explanation, we find that the returns from the strategy

are much smaller in recent years (2003, 2004, 2005, and 2006). Over these four years, the

average six-month abnormal return is 7.7% compared to 20.5% over the earlier years. It is

important to note that our results are not affected by the recent financial crisis. The years in our

sample refer to the fiscal years for which the earnings reports are prepared, and we measure

abnormal returns over a six-month period after the earnings announcements. Therefore, the

return periods in our sample do not extend to 2008, the year when the financial crisis started.

Insert Table IX about here.

VIII. Conclusion

We posit that the delayed market response to earnings news could be related to earnings

management. Prior studies suggest that investors have cognitive limits and limited attention. In

particular, they find that the delayed market response to earnings news is partly related to

investor inattention. However, there is also evidence that many well known market anomalies are

related to earnings management. Hirshleifer et al. (2004) also suggest that because investors have

34
limited attention, they tend to misvalue firms with large abnormal accruals. Furthermore, there is

reliable evidence that managers smooth their reported earnings. If firms manage earnings

downward (upward) when they experience large positive (negative) earnings shocks and if

investors have cognitive limits or are inattentive, then the post-earnings announcement drift

could also be related to earnings management.

Our results support our conjecture. Consistent with the earnings management hypothesis,

we find strong evidence suggesting that firms with large negative (positive) changes in operating

cash flows manage their accruals substantially upward (downward). Moreover, we find no

evidence of a positive drift for those firms with large positive earnings changes that are least

likely to have managed earnings downward or a negative drift for those firms with large negative

earnings changes that are least likely to have managed earnings upward. That is, for these firms,

there is no evidence of an underreaction to earnings changes. In contrast, we find that most of the

upward drift after positive earnings changes is concentrated among those firms that are most

likely to have managed earnings downward. Similarly, we find that most of the downward drift

after negative earnings changes is concentrated among those firms that are most likely to have

managed earnings upward. We also find that this effect tends to become weaker in recent years

as investors started paying more attention to the anomalies and hedge funds were focusing on

exploiting them. Finally, consistent with the earnings management hypothesis, we also find that

the drifts are generally associated with discretionary (or abnormal) accruals and not with

nondiscretionary accruals.

In sum, our study suggests that earnings management is likely one determinant of the

delayed market response to earnings news. It demonstrates that the anomaly is consistent with a

situation where firms with large positive earnings and large positive earnings changes manage

35
earnings downward, while those with large negative earnings changes manage earnings upward,

particularly to stay above the zero-earnings benchmark.

36
Reference

Ball, R. and P. Brown, 1968, “An Empirical Evaluation of Accounting Numbers,” Journal of
Accounting Research 6, 159-178.

Barber, B. and J. Lyon, 1997, “Detecting Long-Run Abnormal Stock Returns: The Empirical
Power and Specification of Test Statistics,” Journal of Financial Economics 43, 341-372.

Barber, B. and T. Odean, 2008, “All that Glitters: The Effect of Attention and News on the
Buying Behavior of Individual and Institutional Investors,” Review of Financial Studies 21, 785-
818.

Barberis, N., A. Shleifer, and R. Vishny, 1998, “A Model of Investor Sentiment,” Journal of
Financial Economics 49, 307-343.

Bartov, E., S. Radhakrishnan, and I. Krinsky, 2000, “Investor Sophistication and Patterns in
Stock Returns After Earnings Announcements,” The Accounting Review 75, 43-63.

Battalio, R. and R. Mendenhall, 2005, “Earnings Expectations, Investor Trade Size, and
Anomalous Returns Around Earnings Announcements,” Journal of Financial Economics 77,
289-319.

Bernard, V. and J. Thomas, 1989, “Post-Earnings-Announcement Drift: Delayed Price Response

or Risk Premium?” Journal of Accounting Research 27, 1-36.

Bernard, V. and J. Thomas, 1990, “Evidence That Stock Prices Do Not Fully Reflect the
Implications of Current Earnings for Future Earnings,” Journal of Accounting and Economics
13, 305-340.

Brav, A., C. Geczy, and P. Gompers, 2000, “Is the Abnormal Return Following Equity Issuances
Anomalous?” Journal of Financial Economics 56, 209-240.

Brown, P. and J. Kennelly, 1972, “The Information Content of Quarterly Earnings: An Extension
and Some Further Evidence,” Journal of Business 45, 403-415.

Burgstahler, D. and I. Dichev, 1997, “Earnings Management to Avoid Earnings Decreases and
Losses,” Journal of Accounting and Economics 24, 99-126.

Burgstahler, D., J. Jiambalvo, and T. Shevlin, 2002, “Do Stock Prices Fully Reflect the
Implications of Special Items for Future Earnings?” Journal of Accounting Research 40, 585-
612.

Carhart, M., 1997, “On Persistence in Mutual Fund Performance,” Journal of Finance 52, 57-82.

Chan, K., L. Chan, N. Jegadeesh, and J. Lakonishok, 2006, “Earnings Quality and Stock
Returns,” Journal of Business 79, 1,041-1,082.

37
Chordia, T. and L. Shivakumar, 2006, “Earnings and Price Momentum,” Journal of Financial
Economics 80, 627-656.

Collins, D. and P. Hribar, 2000, “Earnings-Based and Accrual-Based Market Anomalies: One
Effect or Two?” Journal of Accounting & Economics 29, 101-123.

Constantinou, C., W. Forbes, and L. Skerratt, 2003, “Analyst Underreaction in the United
Kingdom,” Financial Management 32, 93-106.

Daniel, K., D. Hirshleifer, and A. Subrahmanyam, 1998, “Investor Psychology and Security
Market Under- and Overreactions,” Journal of Finance 53, 1,839-1,885.

DeGeorge, F., J. Patel, and R. Zeckhauser, 1999, “Earnings Management to Exceed Thresholds,”
Journal of Business 72, 1-33.

DellaVigna, S. and J. Pollet, 2009, “Investor Inattention and Friday Earnings Announcements,”
Journal of Finance 64, 709-749.

DuCharme, L., P. Malatesta, and S. Sefcik, 2004, “Earnings Management, Stock Issues, and
Shareholder Lawsuits,” Journal of Financial Economics 71, 27-49.

Easton, P. and M. Zmijewski, 1993, “SEC Form 10K/10Q Reports and Annual Reports to
Shareholders: Reporting Lags and Squared Market Model Prediction Errors,” Journal of
Accounting Research 31, 113-129.

Fama, E., 1998, “Market Efficiency, Long-Term Returns, and Behavioral Finance,” Journal of
Financial Economics 49, 283-306.

Fama, E. and K. French, 1993, “Common Risk Factors in the Returns on Stocks and Bonds,”
Journal of Financial Economics 33, 3-56.

Fama, E. and J. MacBeth, 1973, “Risk, Return, and Equilibrium: Empirical Tests,” Journal of
Political Economy 81, 607-636.

Fischer, P. and H. Louis, 2008, “Financial Reporting and Conflicting Managerial Incentives: The
Case of Management Buyouts,” Management Science 54, 1,700-1,714.

Frazzini, A., 2006, “The Disposition Effect and Under-Reaction to News,” Journal of Finance
61, 2,017-2,046.

Fudenberg, D. and J. Tirole, 1995, “A Theory of Income and Dividend Smoothing Based on
Incumbency Rents,” Journal of Political Economy 103, 75-93.

Goel, A. and A. Thakor, 2003, “Why Do Firms Smooth Earnings?” Journal of Business 76, 151-
192.

38
Gong, G., H. Louis, and A. Sun, 2008a, “Earnings Management, Lawsuits, and Stock-for-Stock
Acquirers’ Market Performance,” Journal of Accounting and Economics 46, 62-77.

Gong, G., H. Louis, and A. Sun, 2008b, “Earnings Management and Firm Performance
Following Open-Market Repurchases,” Journal of Finance 63, 947-986.

Graham, J., C. Harvey, and S. Rajgopal, 2005, “The Economic Implications of Corporate
Financial Reporting,” Journal of Accounting and Economics 40, 3-73.

Green, J., J. Hand, and M. Soliman, 2010, “Going, Going, Gone? The Death of the U.S. Accruals
Anomaly,” University of North Carolina - Chapel Hill Working Paper.

Grinblatt, M. and B. Han, 2005, “Prospect Theory, Mental Accounting, and Momentum,”
Journal of Financial Economics 78, 311-333.

Hennes, K., A. Leone, and B. Miller, 2008, “The Importance of Distinguishing Errors from
Irregularities in Restatement Research: The Case of Restatements and CEO/CFO Turnover,” The
Accounting Review 83, 1,487-1,519.

Hirshleifer, D., K. Hou, S. Teoh, and Y. Zhang, 2004, “Do Investors Overvalue Firms with
Bloated Balance Sheets?” Journal of Accounting & Economics 38, 297-331.

Hirshleifer, D., S. Lim, and S. Teoh, 2009, “Driven to Distraction: Extraneous Events and
Underreaction to Earnings News,” Journal of Finance 64, 2,289-2,325.

Hirshleifer, D. and S. Teoh, 2003, “Limited Attention, Information Disclosure, and Financial
Reporting,” Journal of Accounting and Economics 36, 337-386.

Hong, H. and J. Stein, 1999, “A Unified Theory of Underreaction, Momentum Trading, and
Overreaction in Asset Markets,” Journal of Finance 54, 2,143-2,184.

Ikenberry, D., R. Shockley, and K. Womack, 1999, “Why Active Fund Managers Often
Underperform the S&P 500: The Impact of Size and Skewness,” Journal of Private Portfolio
Management 1, 13-26.

Jiang, J., 2008, “Beating Earnings Benchmarks and the Cost of Debt,” The Accounting Review
83, 377-416.

Jones, J., 1991, “Earnings Management During Import Relief Investigations,” Journal of
Accounting Research 29, 193-228.

Jones, C. and R. Litzenberger, 1970, “Quarterly Earnings Reports and Intermediate Stock Price
Trends,” Journal of Finance 25, 143-148.

39
Kothari, S., 2001, “Capital Markets Research in Accounting,” Journal of Accounting and
Economics 31, 105-231.

Kothari, S., A. Leone, and C. Wasley, 2005, “Performance Matched Discretionary Accruals,”
Journal of Accounting and Economics 39, 163-197.

Kothari, S. and J. Warner, 1997, “Measuring Long-Horizon Security Price Performance,”

Journal of Financial Economics 43, 301-339.

Lev, B. and D. Nissim, 2006, “The Persistence of the Accruals Anomaly,” Contemporary
Accounting Research 23, 193-226.

Loh, R., 2010, “Investor Inattention and the Underreaction to Stock Recommendations,”
Financial Management 39, 1,223-1,252.

Louis, H., 2004, “Earnings Management and the Market Performance of Acquiring Firms,”
Journal of Financial Economics 74, 121-148.

Louis, H., D. Robinson, and A. Sbaraglia, 2008, “An Integrated Analysis of the Association
Between Accrual Disclosure and the Abnormal Accrual Anomaly,” Review of Accounting
Studies 13, 23-34.

Louis, H. and A. Sun, 2010, "Investor Inattention and the Market Reaction to Merger
Announcements," Management Science 56, 1,781-1,793.

Louis, H. and H. White, 2007, “Do Managers Intentionally Use Repurchase Tender Offers to
Signal Private Information? Evidence from Firm Financial Reporting Behavior,” Journal of
Financial Economics 85, 205-233.

Lyon, J., B. Barber, and C. Tsai, 1999, “Improved Methods for Tests of Long-Run Abnormal
Stock Returns," Journal of Finance 54, 165-201.

Markoff, J., 1999, “Microsoft’s Accounting Under Scrutiny,” The New York Times (July 1).
http://www.nytimes.com/1999/07/01/business/microsoft-s-accounting-under-scrutiny.html
(accessed October 7, 2009).

Mendenhall, R., 2004, “Arbitrage Risk and Post-Earnings-Announcement Drift,” Journal of

Business 77, 875-894.

Mitchell, M. and E. Stafford, 2000, “Managerial Decisions and Long-Term Stock Price
Performance,” Journal of Business 73, 287-329.

Peng, L. and W. Xiong, 2006, “Investor Attention, Overconfidence and Category Learning,”
Journal of Financial Economics 80, 563-602.

40
Rendleman, R., C. Jones, and H. Latané, 1982, “Empirical Anomalies Based on Unexpected
Earnings and the Importance of the Risk Adjustments,” Journal of Financial Economics 10, 269-
287.

Schwert, W., 2003, “Anomalies and Market Efficiency,” in Handbook of the Economics of
Finance, 21, Amsterdam, London, and New York, Elsevier, North Holland.

Stultz, 2007, “Hedge Funds: Past, Present, and Future,” Journal of Economic Perspectives 21,
175-194.

Teoh, S., I. Welch, and T. Wong, 1998a, “Earnings Management and the Underperformance of
Seasoned Equity Offerings,” Journal of Financial Economics 50, 63-100.

Teoh, S., I. Welch, and T. Wong, 1998b, “Earnings Management and the Long-Run Market
Performance of Initial Public Offerings,” Journal of Finance 53, 1,935-1,974.

Teoh, S. and T. Wong, 2002, “Why New Issues and High-Accrual Firms Underperform: The
Role of Analysts’ Credulity,” The Review of Financial Studies 15, 869-900.

Teoh, S., T. Wong, and G. Rao, 1998, “Are Accruals During Initial Public Offerings
Opportunistic?” Review of Accounting Studies 3, 175-208.

Tucker, J. and P. Zarowin, 2006, “Does Income Smoothing Improve Earnings Informativeness?”
The Accounting Review 81, 251-270.

Xie, H., 2001, “The Mispricing of Abnormal Accruals,” The Accounting Review 76, 357-373.

Yong, K., 2007, “Why Do Firms Not Resume Their Earnings Strings After an Earnings Break?”
Duke University Working Paper.

Zhang, F., 2006, “Information Uncertainty and Stock Returns,” Journal of Finance 61, 105-136.

41
 Table I. Average Abnormal Accruals and Earnings Changes for Our Hedge Portfolios

Standardized earnings changes (SUE) are defined as the seasonal change in the ratio of net income to total assets. ABAC,
abnormal accruals, are computed as the residual from a modified accrual model. SUE1, SUE2, SUE3, SUE4, and SUE5 are the
first, second, third, fourth, and fifth quintiles of (standardized) earnings changes. Similarly, ABAC1, ABAC2, ABAC3, ABAC4,
and ABAC5 are the first, second, third, fourth, and fifth quintiles of abnormal accruals. For the short position, a firm with large
negative earnings changes is deemed most likely to have managed earnings upward (and (estimated) positive abnormal accruals
most likely to be related to earnings smoothing) when the previous quarter earnings are above zero and the firms would have
missed the zero earnings benchmark in the current quarter if they did not report the income increasing abnormal accruals. It is
deemed least likely to manage earnings upward when (standardized) earnings are high (in the top quintile). For the long position,
a firm with large positive earnings changes is deemed most likely to have managed earnings downward (and (estimated) negative
abnormal accruals more likely to be related to earnings smoothing) when reported ROA is high (top quintile). It is considered
least likely to manage earnings downward when reported earnings are negative. We winsorize the quarterly earnings changes and
abnormal accruals at the top and bottom one percentiles. We report t-statistics in parentheses and the number of observations in
brackets.

Panel A. Short Position - SUE1 / ABAC5

Most Likely Cases of Least Likely Cases of
Upward Earnings Upward Earnings Others
Management Management

-0.031+++ -0.033+++ -0.054+++

Earnings Change (-42.92) (-14.47) (-74.63)
[n=1,010] [n=121] [n=4,446]
0.066+++ 0.046+++ 0.073+++
Abnormal Accruals (68.83) (38.62) (133.51)
[n=1,010] [n=121] [n=4,446]
Panel B. Long Position - SUE5 / ABAC1
Most Likely Cases of Least Likely Cases of
Downward Earnings Downward Earnings Others
Management Management
0.044+++ 0.086+++ 0.036+++
Earnings Change (48.30) (40.26) (39.59)
[n=2,666] [n=1,866] [n=1,565]
-0.067+++ -0.074+++ -0.063+++
Abnormal Accruals (-89.42) (-66.99) (-71.34)
[n=2,666] [n=1,866] [n=1,565]
Panel C. SUE Quintiles

SUE1 SUE2 SUE3 SUE4 SUE5

+++ +++ *** +++
-0.058 -0.007 -0.000 0.005 0.058+++
Earnings Change (-212.4) (-336.06) (-39.37) (357.19) (181.23)
[n=41,889] [n=44,877] [n=27,792] [n=45,619] [n=45,292]

-0.018*** 0.004*** 0.006*** 0.006*** 0.016***

Abnormal Accruals (-53.83) (23.3) (26.4) (29.8) (59.03)
[n=41,889] [n=44,877] [n=27,792] [n=45,619] [n=45,292]

Panel D. ABAC Quintiles

ABAC1 ABAC2 ABAC3 ABAC4 ABAC5

-0.022*** -0.002*** 0.001*** 0.005*** 0.019***
Earnings Change (-55.23) (-9.27) (6.06) (22.83) (51.89)
[n=36,328] [n=426,44] [n=43,503] [n=43,100] [n=39,894]
-0.072+++ -0.016+++ 0.003*** 0.022+++ 0.071+++
Abnormal Accruals (-280.98) (-397.38) (98.48) (544.44) (390.81)
[n=36,328] [n=42,644] [n=43,503] [n=43,100] [n=39,894]
*** +++
( ) indicates significance at the 1% level in a two- (one-) tailed test.

42
 Table II. Average Risk Characteristics for the Full Sample of Firms in the Top and Bottom Quintiles of Earnings Changes

SUE1 and SUE5 are the first and the fifth quintiles of (standardized) earnings changes defined as the seasonal change in the ratio of net income to total assets. SIZE is the market
value of common equity at the beginning of the current quarter (in millions of dollars). BM is the ratio of book value of equity to market value of equity at the beginning of the
current quarter. MM, momentum, is the stock return from twelve to two months prior to the earnings announcement month. ARBRISK is the arbitrage risk defined as the residual
variance from a standard market model regression of a firm’s return on the returns of the CRSP S&P 500 equal-weighted market index over a 48-month period ending one month
prior to the earnings announcement month. PRICE is the stock price (Compustat quarterly Data Item #14) at the beginning of the current quarter. VOLUME is the monthly trading
volume over the 12-month period ending at the end of the month immediately proceeding the earnings announcement quarter (in millions of dollars). IO is the percentage of
institutional ownership at the end of the calendar quarter immediately preceding the end of the earnings announcement quarter. The short position consists of firms with the most
negative earnings changes and the most positive abnormal accruals that are most likely to manage earnings upward. The long position consists of firms with the most positive
earnings changes and the most negative abnormal accruals that are most likely to manage earnings downward. The population includes all NYSE/AMEX/NASDAQ firms in our
sample period that have enough data to compute the variables in the regression analysis in Table IV.

Panel A. Full Sample of Firms in the Top and Bottom Quintiles of Earnings Changes
T-value for Testing the T-value for Testing the
T-value for Testing for
SUE5 Firms SUE1 Firms Mean of the SUE5 Firms Mean of SUE1 Firms
Variables Population Mean Difference Between
(N = 37,525) (N =34,306) Against the Population Against the Population
SUE5 and SUE1 Firms
Mean Mean
SIZE 1,756.000 1,201.900 1,097.000 -21.40*** -25.03*** 2.84***
BM 0.594 0.516 0.577 -24.47*** -5.02*** -13.04***

MM 1.181 1.422 0.977 44.21*** -48.38*** 64.56***

ARBRISK 0.033 0.049 0.047 58.07*** 49.97*** 5.56***

PRICE 18.205 14.319 13.212 -44.69*** -64.06*** 9.49***

VOLUME 25.800 21.900 26.100 -9.29*** 0.64 -6.23***

IO 39.161 32.446 33.215 -48.04*** -41.80*** -3.85***

43
 Table II. Average Risk Characteristics for the Full Sample of Firms in the Top and Bottom Quintiles of Earnings Changes (Continued)

Panel B, Firms in the Top (Bottom) Quintiles of Earnings Changes that are Deemed Most Likely to Have Managed Earnings Downward (Upward)

T-value for Testing the T-value for Testing the T-value for Testing for
Long Position Short Position Mean for the Firms in the Mean for the Firms in the Mean Difference Between
Variables Population
(N = 2,561) (N = 941) Long Position Against the Short Position Against the Firms in the Long and
Population Mean Population Mean Short Positions

SIZE 1,756.000 1,342.200 493.980 -4.18*** -24.12*** -7.58***

BM 0.594 0.393 0.484 -25.12*** -10.62*** 6.99***

MM 1.181 1.733 1.010 26.96*** -9.63*** -26.69***

ARBRISK 0.033 0.038 0.035 5.98*** 1.57 -2.2**

PRICE 18.205 19.370 14.605 3.12*** -9.19*** -8.81***

VOLUME 25.800 25.100 13.300 -0.40 -7.13*** -4.92***

IO 39.161 36.340 31.400 -4.94*** -9.37*** -4.92***

*** **
/ indicates significance at the 1%/5% level using a two-tail t test.

44
 Table III. Average (Standardized) Earnings Changes and Abnormal Accruals by Quintile of
(Standardized) Changes in Operating Cash Flows

Change in earnings and change in operating cash flows are seasonal differences standardized by beginning total assets. SUCF1,
SUCF2, SUCF3, SUCF4, and SUCF5 are the first, second, third, fourth, and fifth quintiles of changes in operating cash flows.
We winsorize the quarterly earnings changes and abnormal accruals at the top and bottom one percentiles. We report t-statistics
in parentheses and the number of observations in brackets.

SUCF1 SUCF2 SUCF3 SUCF4 SUCF5

-0.018*** -0.003*** 0.000 0.003*** 0.023***

Earnings Change (-54.06) (-17.98) (0.86) (16.75) (58.15)
[n=34,430] [n=39,441] [n=34,402] [n=39,588] [n=35,080]

0.028*** 0.010*** 0.003*** -0.004*** -0.021***

Abnormal Accruals (88.67) (50.61) (13.98) (-19.77) (-64.23)
[n=34,430] [n=39,441] [n=34,402] [n=39,588] [n=35,080]
***
indicates significance at the 1% level in a two-tailed test.

45
Table IV. The Association Between Abnormal Accruals and the Post-Earnings-Announcement Drift
(N = 71,831)

ABRETit = 1SUEQ1it + 2SUEQ5it + 3SUEQ1it*ABACQit + 4SUEQ5it*ABACQit + 5SIZEDit + 6BMit + 7MMit +

8ARBRISKit + 9PRICEDit + 10VOLUMEDit + 11IOit + it. The sample includes firms that are in either the top or bottom
quintile of earnings changes. ABRET is the size-adjusted return measured over the period from 18 trading days after the earnings
announcement for Quarter 0 (the current quarter) to 17 trading days after the earnings announcement for Quarter +2 (the second
quarter after Quarter 0). SUEQ1 is a binary variable taking a value of one if the standardized earnings change is in the bottom
quintile and zero if it is in the top quintile. SUEQ5 is a binary variable taking a value of one if the standardized earnings change is
in the top quintile and zero if it is in the bottom quintile. ABACQ is the quintile of abnormal accruals (the residual from the
accrual model). We impose no restriction on abnormal accruals. SIZED is the decile of the market value of common equity at the
beginning of the current quarter. BM is the ratio of the book value of equity to the market value of equity at the beginning of the
current quarter. MM, momentum, is the stock return from twelve to two months prior to the earnings announcement month.
ARBRISK is arbitrage risk defined as the residual variance from a standard market model regression of a firm’s return on the
returns of the CRSP S&P 500 equal-weighted market index over a 48-month period ending one month prior to the earnings
announcement month. PRICED is the decile of the stock price at the beginning of the current quarter. VOLUMED is the decile of
the dollar trading volume over the 12-month period ending at the end of the month immediately preceding the earnings
announcement quarter. IO is the percentage of institutional ownership at the end of the calendar quarter immediately preceding
the end of the earnings announcement quarter. Each quarter, we rank size, stock price, and trading volume into deciles. Then, we
assign values ranging from –0.5 to 0.5 to the rank deciles. T-statistics are reported in parentheses. They are adjusted using one-
way clustering at the quarter (time) level. We use one-way clustering because many firms are in the sample only once.

Column (1) Column (2)

-0.021 -0.021
SUEQ1
(-1.27) (-0.99)
0.040+++ 0.044++
SUEQ5
(2.88) (2.21)
-0.062+++ -0.059+++
SUEQ1*ABACQ
(-5.84) (-5.55)
-0.050+++ -0.049+++
SUEQ5*ABACQ
(-3.81) (-3.61)
0.026
SIZED _
(0.83)
0.024+
BM _
(1.34)
-0.007
MM _
(-0.92)
-0.540+++
ARBRISK _
(-2.40)
-0.124+++
PRICED _
(-2.58)
-0.060+
VOLUMED _
(-1.43)
0.001++
IO _
(1.79)
Adj. R2 0.003 0.009
+++ ++
/ indicates significance at the 1%/5% level using a one-tail t test.

46
 Table V. Trading Strategies Based on Earnings Changes Using those Firms That Are Deemed
Least Likely to Have Managed Earnings: Univariate Analysis

We take long positions in those firms with large positive earnings changes (SUE5) that are considered least likely to have
managed earnings downward and short positions in those firms with large negative earnings changes (SUE1) that are deemed
least likely to have managed earnings upward. Firms in the long position are judged least likely to have managed earnings
downward when earnings are negative, while firms in the short position are deemed least likely to have managed earnings
upward when earnings are in the top quintile. We impose no restriction on abnormal accruals. That is, abnormal accruals can take
any value for either the short or the long position. We report t-statistics in parentheses.

Long Position Short Position

SUE5 Firms That Are Least Likely SUE1 Firms That Are Least
to Have Managed Earnings Likely to Have Managed
Downward Earnings Upward

-0.012 0.010
Average Buy-and-Hold Abnormal Return
(-1.82) (1.36)

Alpha from Fama and French’s (1993) 0.004 0.001

Three-Factor Model (1.17) (0.60)

0.085+++ -0.032+++
Average Earnings Change
(105.18) (-71.64)

N 12,436 3,074

***
indicates significance at the 1% level in a one-tailed test.

47
 Table VI. Trading Strategies Based on Earnings Changes Using those Firms That Are Deemed
Least Likely to Have Managed Earnings: Multivariate Analysis (N = 14,331)

ABRETit = 0 + 1SUEQ5it + 2ABACQit + 3SIZEDit + 4BMit + 5MMi + 6ARBRISKit + 7PRICEDit + 8VOLUMEDit +

9IOit + it. The sample includes firms that are in either the top or the bottom quintile of earnings changes. ABRET is the size-
adjusted return measured over the period from 18 trading days after the earnings announcement for Quarter 0 (the current
quarter) to 17 trading days after the earnings announcement for Quarter +2 (the second quarter after Quarter 0). SUEQ5 is a
binary variable taking a value of one if the standardized earnings change is in the top quintile and zero if it is in the bottom
quintile. ABACQ is the quintile of abnormal accruals (the residual from the accrual model). We impose no restriction on
abnormal accruals. SIZED is the decile of the market value of common equity at the beginning of the current quarter. BM is the
ratio of the book value of equity to the market value of equity at the beginning of the current quarter. MM, momentum, is the
stock return from twelve to two months prior to the earnings announcement month. ARBRISK is arbitrage risk defined as the
residual variance from a standard market model regression of a firm’s return on the returns of the CRSP S&P 500 equal-weighted
market index over a 48-month period ending one month prior to the earnings announcement month. PRICED is the decile of the
stock price at the beginning of the current quarter. VOLUMED is the decile of the dollar trading volume over the 12-month period
ending at the end of the month immediately preceding the earnings announcement quarter. IO is the percentage of institutional
ownership at the end of the calendar quarter immediately preceding the end of the earnings announcement quarter. Each quarter,
we rank size, stock price, and trading volume into deciles. Then, we assign values ranging from –0.5 to 0.5 to the rank deciles.
The numbers of observations in Table V do not sum up to the number of observations in Table VI as some of the control
variables in Table VI have missing observations. T-statistics are reported in parentheses. They are adjusted using one-way
clustering at the quarter (time) level. We use one-way clustering because many firms are in the sample only once.

Column (1) Column (2)

0.007 0.047
Intercept
(0.73) (2.13)
-0.007 -0.063
SUEQ5
(-0.17) (-1.79)
-0.032 -0.031
ABACQ
(-1.19) (-1.12)
0.091++
SIZED _
(1.76)
0.039++
BM _
(1.92)
-0.012
MM _
(-1.03)
-0.385++
ARBRISK _
(-2.26)
-0.198+++
PRICED _
(-3.02)
-0.136+++
VOLUMED _
(-2.66)
0.001
IO _
(1.23)
Adj. R2 0.000 0.011
+++ ++
/ indicates significance at the 1%/5% level using a one-tail t test.

48
 Table VII. The Effect of Earnings Management on the Post-Earnings Announcement Drift

In Panel A, the short position includes firms with large negative earnings changes (bottom quintile) and large positive abnormal
accruals (top quintile) that are most likely to manage earnings upward (ABAC1/SUE5/MOST_UP). The long position incorporates
firms with large positive earnings changes (top quintile) and large negative abnormal accruals (bottom quintile) that are most likely
to manage earnings downward (ABAC5/SUE1/MOST_DOWN). In Panel B, abnormal accruals are replaced by normal accruals.
Normal accrual is the difference between total accrual and abnormal accrual. NAC1 and NAC5 are the first and the fifth quintiles of
normal accruals, respectively. Firms with large negative earnings changes are deemed most likely to have managed earnings
upward (and (estimated) positive abnormal accruals most likely to be related to earnings smoothing) when the previous quarter
earnings are above zero and the firms would have missed the zero-earnings benchmark in the current quarter if they did not report
the income increasing abnormal accruals. Firms with large positive earnings changes are considered most likely to have managed
earnings downward (and (estimated) negative abnormal accruals more likely to be related to earnings smoothing) when reported
ROA is high (top quintile). We report t-statistics in parentheses and the number of observations in brackets. Note that the abnormal
returns from Fama and French’s (1993) three-factor model are monthly abnormal returns (and not buy-and-hold abnormal returns).

Panel A. Trading Strategy Based on Abnormal Accruals, Earnings Changes, and the Likelihood of Earnings Management

Long Position Short Position Hedge-Portfolio

ABAC1/SUE5/MOST_DOWN ABAC5/SUE1/MOST_UP Return
0.091+++ -0.104+++
Average Buy-and-Hold Abnormal 0.195+++
(8.23) (-8.34)
Returns (9.99)
[n=2,666] [n=1,010]
-0.011+++ 0.027+++
Fama and French’s (1993) Three- 0.016+++
(-3.55) (7.90)
Factor Model (6.83)

Panel B. Trading Strategy Based on Normal Accruals, Earnings Changes, and the Likelihood of Earnings Management

Long Position Short Position Hedge-Portfolio

NAC1/SUE5/MOST_DOWN NAC5/SUE1/MOST_UP Return
0.085+++ -0.041+++
Average Buy-and-Hold Abnormal 0.126+++
(6.89) (-2.68)
Returns (5.58)
[n=2,308) (n=827)
Fama and French’s (1993) Three- 0.010+++ -0.007++ 0.016+++
Factor Model (3.56) (-1.98) (4.02)
+++ ++
/ indicates significance at the 1%/5% level using a one-tail test.

49
 Table VIII. Association Between Future Returns and the Likelihood of Earnings Management
Conditional on Abnormal Accruals and Earnings Changes

ABRET is the size-adjusted return over the period from 18 trading days after the earnings announcement for Quarter 0 (the
current quarter) to 17 trading days after the earnings announcement for Quarter+2 (the second quarter after Quarter 0). The
expected return is the compound return of the respective firm’s CRSP NYSE/AMEX/NASDAQ size-decile portfolio assignment
at the beginning of the year. MOST_DOWN is a binary variable taking a value of one if (standardized) earnings are high (in the
top quintile), and zero otherwise. LEAST_DOWN is a binary variable taking a value of one if earnings are negative, and zero
otherwise. OTHER_DOWN is a binary variable taking a value of one if both MOST_DOWN and LEAST_DOWN are zero, and
zero otherwise. MOST_UP is a binary variable taking a value of one if the previous quarter earnings are above zero and the firm
would have missed the zero earnings benchmark in the current quarter if it did not report positive abnormal accruals, and zero
otherwise. LEAST_UP is a binary variable taking a value of one if (standardized) earnings are high (in the top quintile), and zero
otherwise. OTHER_UP is a binary variable taking a value of one if both MOST_UP and LEAST_UP are zero, and zero otherwise.
SIZED is the decile of the market value of common equity at the beginning of the current quarter. BM is the ratio of book value of
equity to the market value of equity at the beginning of the current quarter. MM, momentum, is the stock return from twelve to
two months prior to the earnings announcement month. ARBRISK is arbitrage risk defined as the residual variance from a
standard market model regression of a firm’s return on the returns of the CRSP S&P 500 equal-weighted market index over a 48-
month period ending one month prior to the earnings announcement month. PRICED is the decile of the stock price at the
beginning of the current quarter. VOLUMED is the decile of the dollar trading volume over the 12-month period ending at the
end of the month immediately preceding the earnings announcement quarter. IO is the percentage of institutional ownership at
the end of the calendar quarter immediately preceding the end of the earnings announcement quarter. Each quarter, we rank size,
stock price, and trading volume into deciles. Then, we assign values ranging from –0.5 to 0.5 to the rank deciles. The t-values are
adjusted using one-way clustering at the quarter (time) level. We use one-way clustering as many firms are in the sample only
once.

Panel A. Long Position - SUE5 / ABAC1 (N = 5,787)

ABRETit = 1MOST_DOWNit + 2LEAST_DOWNit + 3OTHER_DOWNit + 4SIZEDit + 5BMit + 6MMit +7ARBRISKit +

8PRICEDit + 9VOLUMEDit + 10IOit + it
Coefficient t-value
+++
MOST_DOWN 0.091 2.93
LEAST_DOWN 0.015 0.31
++
OTHER_DOWN 0.051 1.81
SIZED -0.046 -0.75
BM 0.002 0.09
MM 0.000 0.03
+
ARBRISK -0.502 -1.66
+++
PRICED -0.199 -3.64
VOLUMED 0.001 0.02
+
IO 0.001 1.54
Adjusted R2 0.023

50
 Table VIII. Association Between Future Returns and the Likelihood of Earnings Management
Conditional on Abnormal Accruals and Earnings Changes (Continued)

Panel B. Short Position - SUE1 / ABAC5 (N = 5,278)

ABRETit = 1MOST_UPit + 2LEAST_UPit + 3OTHER_UPit + 4SIZEDit + 5BMit + 6MMit + 7ARBRISKit + 8PRICEDit

+ 9VOLUMEDit + 10IOit + i

Coefficient t-value
++
MOST_UP -0.083 -2.31
LEAST_UP -0.017 -0.31
OTHER_UP -0.023 -0.50
++
SIZED 0.140 1.78
BM 0.004 0.12
MM -0.011 -0.63
+
ARBRISD -0.477 -1.62
+
PRICED -0.143 -1.62
++
VOLUMED -0.122 -1.89
+
IO 0.001 1.43
Adjusted R2 0.011
+++ ++
/ indicates significance at the 1%/5% level using a one-tail test.

51
 Table IX. Year by Year Average Abnormal Return from the Trading Strategy Based on Abnormal
Accruals, Earnings Changes, and the Likelihood of Earnings Management

The abnormal returns are the average size-adjusted buy-and-hold returns for the four quarters of the year. The sample starts in the
second calendar quarter of 1990 and ends in the fourth calendar quarter of 2006. The abnormal returns are computed daily over
the 120 trading days starting 18 trading days after the earnings announcement. The short position includes firms with large
negative earnings changes (bottom quintile) and large positive abnormal accruals (top quintile) that are most likely to manage
earnings upward, while the long position includes firms with large positive earnings changes (top quintile) and large negative
abnormal accruals (bottom quintile) that are most likely to manage earnings downward.

Year Average Hedge-Portfolio Return

1990 0.168

1991 0.217

1992 0.182

1993 0.101

1994 0.154

1995 0.259

1996 0.064

1997 0.187

1998 0.301

1999 0.505

2000 0.195

2001 0.128

2002 0.234

2003 0.069

2004 0.052

2005 0.110

2006 0.076

52
 Figure I. The Drift from Various Alternative Trading Strategies Based on Abnormal Accruals,
Earnings Changes, and the Likelihood of Earnings Management

The abnormal returns are the average size-adjusted returns. SUE1 and SUE5 are the first and the fifth quintiles of (standardized)
earnings changes defined as the seasonal change in the ratio of net income to total assets. ABAC1 and ABAC5 are the first and the
fifth quintiles of abnormal accruals computed as the residual from a modified accrual model. The observations are pooled across
all the quarters from the second calendar quarter of 1990 to the fourth calendar quarter of 2006. The abnormal returns are
computed daily over the 120 trading days starting 18 trading days after the earnings announcement. MOSTUP designates
SUE1/ABAC5 firms that are most likely to have managed earnings upward. MOSTDOWN designates SUE5/ABAC1 firms that are
most likely to have managed earnings downward.

0.12 SUE5/ABAC1/
0.1 MOSTDOWN
(n=2795)
0.08
SUE5/ABAC1
0.06 (n=6497)
Abnormal return

0.04
SUE5
0.02 (n=41982)
0
-0.02 SUE1
(n=38060)
-0.04
-0.06 SUE1/ABAC5
(n=5983)
-0.08
-0.1 SUE1/ABAC5/
-0.12 MOSTUP
(n=1063)
17 47 77 107 137
Days relative to the quarterly earnings
announcement date

53