Make a summary and PowerPoint and answer case study (A,B) (Please use easy words and easy sentences)
- prepare a 3 or 4 pages summary of each paper (3 Article) (including the Libby box)
Focus on summarizing the most salient points of the article.
. After that make sure the summaries address the following questions (I will send them).
B.and make PowerPoint slides for each study.
Three Articles in the attachments
THE ACCOUNTING REVIEW
Vol. 83, No. 3
2008
pp. 757-787
Real and Accrual-Based Earnings
Management in the Pre- and
Post-Sarbanes-Oxley Periods
Daniel A. Cohen
New York University
Aiyesha Dey
University of Chicago
Thomas Z. Lys
Northwestern University
ABSTRACT: We document that accrual-based earnings management increased steadily from 1987 until the passage of the Sarbanes-Oxley Act (SOX) in 2002, followed by
a significant decline after the passage of SOX. Conversely, the level of real earnings
management activities declined prior to SOX and increased significantly after the passage of SOX, suggesting that firms switched from accrual-based to real earnings management methods after the passage of SOX. We also document that the accrual-based
earnings management activities were particularly high in the period immediately preceding SOX. Consistent with these results, we find that firms that just achieved important earnings benchmarks used less accruals and more real earnings management
after SOX when compared to similar firms before SOX. In addition, our analysis provides evidence that the increases in accrual-based earnings management in the period
preceding SOX were concurrent with increases in equity-based compensation. Our
results suggest that stock-option components provide a differential set of incentives
with regard to accrual-based earnings management. We document that while new
options granted during the current period are negatively associated with incomeincreasing accrual-based earnings management, unexercised options are positively associated with income-increasing accrual-based earnings management.
We acknowledge the financial support from the Accounting Research Center at the Kellogg School and helpful
comments from Dan Dhaliwal, April Klein, Krishna Kumar, Eddie Riedl, Suraj Srinivasan, Ira Weiss, Jerry
Zimmermann, two anonymous reviewers, participants at the HBS Accounting and Control seminar, the 2006 FARS
Meeting, the 2004 AAA Annual Meeting, the 2004 Corporate Governance Conference at the University of Washington in St. Louis, and the 2004 LBS Accounting Symposium. All remaining errors are our own.
Editor’s note: This paper was accepted by Dan Dhaliwal.
Submitted February 2006
Accepted October 2007
757
758
Cohen, Dey, and Lys
I. INTRODUCTION
he recent
wave
of corporate
governance
failures
has raisedand
concerns
the integrity
of the
accounting
information
provided
to investors
resultedabout
in a drop
in
investor confidence (Jain et al. 2003; Jain and Rezaee 2006; Rezaee 2004). These
failures were highly publicized and ultimately led to the passage of the Sarbanes-Oxley
Act (SOX, July 30, 2002). The changes mandated by SOX were extensive, with President
George W. Bush commenting that this Act constitutes “the most far-reaching reforms of
American business practices since the time of Franklin D. Roosevelt.”‘ Similarly, the head
of the AICPA commented that SOX “contains some of the most far-reaching changes that
Congress has ever introduced to the business world'” 2 including an unprecedented shift in
the regulation of corporate governance from the states to the federal government.3
Although SOX proposed sweeping changes, the scope of the events that led to the
passage of the Act and the consequences of the resulting regulatory changes have yet to
be systematically studied. Specifically, it is unclear whether there really was a widespread
breakdown of the reliability of financial reporting prior to the passage of SOX or whether
the highly publicized scandals were isolated instances of individuals engaging in blatant financial manipulations. If it were the former, then how did the passage of SOX affect firms’
financial reporting practices? Moreover, some argue that these frauds occurred after 70
years of ever-increasing securities regulation, suggesting that more regulation may not be
the answer (Ribstein 2002).
We investigate the prevalence of both accrual-based and real earnings management
activities in the period leading to the passage of SOX and in the period following the
passage of SOX. Our primary motivation for conducting this analysis is to investigate
whether the period leading to the passage of SOX was characterized by a widespread
increase in earnings management rather than by a few highly publicized events, and whether
the passage of SOX resulted in a reduction in earnings management.
We carry out our investigation by dividing the sample period into two time periods:
the period prior to the passage of SOX (the pre-SOX period: 1987 through 2001), and the
period after the passage of SOX (the post-SOX period: 2002 through 2005). We further
subdivide the pre-SOX period into two subperiods: the period prior to the major corporate
scandals (the pre-SCA period: 1987 through 1999) and the period immediately preceding
the passage of SOX when the major scandals occurred (the SCA period: 2000 and 2001).
We document that the pre-SOX period was characterized by increasing accrual-based
earnings management, culminating in even larger increases in the SCA period, but declining
real earnings management. We also document that the increase in accrual-based earnings
management in the SCA period was associated with a contemporaneous increase in optionbased compensation. However, option-based compensation exhibits two opposing incentives
with respect to current stock prices (and thus earnings management). On the one hand,
managers have an incentive to lower the current stock price around stock option award
(e.g., Yermack 1997; Aboody and Kasznik 2000). On the other hand, for unexercised options (excluding new option grants), managers have an incentive to increase current stock
prices and, hence, to manage earnings upward. To capture these two effects, we partition
overall options held at the end of the year into unexercised options (excluding new options
‘ Bumiller (2002).
2 Melancon (2002).
Traditionally, the federal government has focused on regulating disclosure, public trading, and antitrust, while
regulating corporate governance has been the focus of the states.
The Accounting Review, May 2008
Real and Accrual-Based EarningsManagement
759
grants), new options granted in the current period, and exercisable vested options. Consistent with the notion that these components provide a differential set of incentives, we find
that while new options granted during the current period are negatively associated with
income-increasing discretionary accruals, unexercised options are positively associated
with income-increasing discretionary accruals. However, consistent with Cheng and
Warfield (2005), we do not find a significant association between accrual-based earnings
management and exercisable vested options. Furthermore, we document that the relation
between income-increasing discretionary accruals and unexercised options (excluding new
option grants) declined in the post-SOX period. To the best of our knowledge, we are the
first to document this evidence empirically.
Following the passage of SOX accrual-based earnings management declined significantly, while real earnings management increased significantly. At the same time, optionbased compensation decreased. Consistent with the results of a recent survey by Graham
et al. (2005), this suggests that firms switched to managing earnings using real methods,
possibly because these techniques, while more costly, are likely to be harder to detect.
In additional analyses, we examine both the pre-SOX and post-SOX accrual-based and
real earnings management activities for a subset of firms that are more likely to have
managed earnings (we refer to these firms as the SUSPECT firms). Specifically, we examine
three incentives for managing earnings, namely, meeting or beating last year’s earnings,
meeting or beating the consensus analysts’ forecast and avoiding reporting losses. We focus
on these incentives because Graham et al. (2005) document that these specific motives are
among the most important reasons for earnings management behavior.
Consistent with our full sample results, we find that both before and after SOX, SUSPECT firms had significantly higher discretionary accruals when compared to firms that
either just missed those benchmarks or firms that were not close to either making or missing those benchmarks. However, these SUSPECT firms used significantly less incomeincreasing accrual-management after SOX when compared to firms in similar circumstances
before SOX. An analysis of the real earnings management behavior of these firms indicates
that the SUSPECT firms had significantly higher real earnings management activities after
SOX when compared to the firms in similar circumstances prior to SOX.
Our results contribute to the current debate on the pervasiveness of earnings management prior to the passage of SOX, and the impact of SOX on such behavior. While we find
an increase in accrual-based earnings management prior to SOX, our evidence suggests that
firms are likely to have switched to earnings-management techniques that, while likely to
be more costly to shareholders, are harder to detect. This evidence forms an important
consideration in the debate on the costs and benefits of the new regulation.
The remainder of the paper proceeds as follows. Section II provides a discussion of
the research questions and hypotheses. Section III discusses the empirical methodology,
including the data and sample selection and the various measures of earnings management
used in the study. The tests and results are discussed in Section IV, and Section V concludes.
II. MOTIVATION, RESEARCH QUESTIONS, AND HYPOTHESES
In a recent commentary, U.S. Treasury Secretary Henry Paulson emphasized the importance of strong capital markets, and pointed out that capital markets rely on trust and
that trust is based on financial information presumed to be accurate and to reflect economic
reality.’ The series of corporate scandals occurring in 2000-2001 eroded that trust in financial reports. Indeed, one of the main objectives of SOX was to restore the integrity of
” Paulson (2007).
The Accounting Review, May 2008
760
Cohen, Dey, and Lys
financial statements by curbing earnings management and accounting fraud. Therefore, the
extent of earnings management prior to SOX and the effect of SOX on earnings management is an important research topic.
The primary purpose of this paper is to examine the extent of earnings management
in the period leading to the scandals and prior to SOX, and the changes in such activities
after the passage of SOX. 5 Our examination of changes in firms’ earnings management
activities is motivated in part by the literature documenting that managerial propensity to
manage earnings and to avoid negative earnings surprises has increased significantly over
time (e.g., Brown 2001; Bartov et. al. 2002; Lopez and Rees 2001; Matsumoto 2002; Brown
and Caylor 2005). Our main objective is to examine whether the degree of earnings management increased over time and reached a zenith in the period surrounding the corporate
accounting scandals, and declined after the passage of SOX.
Consistent with the literature, we examine earnings management activities using discretionary accruals. However, in addition to using accrual-based accounting estimates and
methods, firms are likely to employ real operational activities to manipulate earnings
numbers as well (e.g., Healy and Wahlen 1999; Dechow and Skinner 2000). In fact, in their
survey Graham et al. (2005) report the following:6
[W]e find strong evidence that managers take real economic actions to maintain accounting appearances. In particular, 80% of survey participants report that they would
decrease discretionary spending on R&D, advertising, and maintenance to meet an
earnings target. More than half (55.3%) state that they would delay starting a new
project to meet an earnings target, even if such a delay entailed a small sacrifice in
value.
Thus, to provide a more complete study of the trends in earnings management activities in
the periods before and after SOX, we also examine real earnings management activities
over the sample period.
Next, we examine possible explanations for any changes in earnings management activities over the sample period. We focus on the hypothesis that managers’ choices of
accounting practices are influenced by the impact of these accounting methods on their
compensation. Managers with higher stock- and option-based compensation are more sensitive to short-term stock prices, and can use their discretion to affect reported earnings if
capital markets have difficulty detecting earnings management (see Fields et al. [2001] for
a discussion on the pricing of earnings management.)7
5 In a related study, Lobo and Zhou (2006) investigate whether the SOX mandate that financial statements be
certified by firms’ CEOs and CFOs resulted in an increase in the conservatism in financial reporting, and find
evidence of greater conservatism in reported earnings.
6 In a recent working paper, Zang (2006) investigates whether managers use real and accrual manipulations in
managing earnings as substitutes. Based on a model she develops, she provides evidence consistent with managers using real and accrual manipulations as substitutes.
For instance, Coffee (2003) asserts that the increase in stock-based executive compensation created an environment where managers became very sensitive to short-term stock performance. Greenspan (2002) opines that
“the highly desirable spread of shareholding and options among business managers perversely created incentives
to artificially inflate earnings to keep stock prices high and rising.” Fuller and Jensen (2002, 42) also state that
“[a]s stock options became an increasing part of executive compensation, and managers who made great fortunes
on options became the stuff of legends, the preservation or enhancement of short-term stock prices became a
personal (and damaging) priority for many CEOs and CFOs. High share prices and earnings multiples stoked
already amply endowed managerial egos, and management teams proved reluctant to undermine their own stature
by surrendering hard won records of quarter-over-quarter earnings growth.”
The Accounting Review, May 2008
Real and Accrual-Based Earnings Management
761
Prior studies (e.g., Cheng and Warfield 2005; Bergstresser and Philippon 2006) provide
evidence suggesting that equity incentives derived from stock-option compensation are positively associated with managements’ likelihood to engage in accrual-based earnings management activities. However, Johnson et al. (2005) conclude that only unrestricted stock
holdings are associated with the occurrence of accounting fraud; the stock option grants
are not. Further, Erickson et al. (2006) find no consistent evidence that executive equity
incentives are associated with fraud.
In addition to equity-based compensation, executives are also rewarded based on explicit bonus-linked targets for reported income. Healy (1985) presents evidence that the
accruals policies of managers are related to the nonlinear incentives inherent in their bonus
contracts. Therefore, we investigate whether earnings-based compensation contracts are
associated with earnings management.
Our focus on the compensation structure is motivated by the current debate whether
option-based compensation and bonus grants are associated with earnings management.
Further, there has been a significant increase in the grant of stock options in the past decade.
We examine whether those increases (prior to SOX) and decreases (after SOX) are related
to the level of earnings management during that period.
Specifically, our hypothesis that managers behave opportunistically due to
compensation-related incentives has two empirical predictions. First, changes in reported
earnings are affected by changes in the compensation and incentives of managers. Second,
even after controlling for managerial incentives, earnings management would decline after
the passage of SOX, either because of the sanctions imposed on managers by SOX or
because of the adverse publicity and legal costs imposed on executives and firms who were
accused of questionable or even fraudulent reporting practices.
Our final objective is to investigate whether after the passage of SOX corporations
replaced some accrual-based earnings management with real earnings management, which
is not only likely to be harder to detect, but also likely to be more costly to the firm
(Graham et al. 2005).8 We based this expectation on the premise that after the passage of
SOX accrual manipulations were more likely to draw auditors’ or regulators’ scrutiny than
real earnings management. If firms were more wary after the passage of SOX, then they
would be more likely to substitute real earnings management for accrual-based earnings
management after SOX. This conjecture is also suggested by Graham et al. (2005):
[W]e acknowledge that the aftermath of accounting scandals at Enron and WorldCom
and the certification requirements imposed by the Sarbanes-Oxley Act may have
changed managers’ preferences for the mix between taking accounting versus real actions to manage earnings.
Given the above argument, investigating the trends in both real and accrual-based earnings management after SOX is important. Evidence of a decline in one type of earnings
management may lead one to conclude that such activities have decreased in response to
regulators or other events, when in fact a substitution of one earnings management method
for another has occurred. We thus hypothesize and test whether the level of real earnings
management increased after the passage of SOX, i.e., whether firms substituted between
8 In their survey, Graham et al. (2005, 66) report: “Managers candidly admit that they would take real economic
actions such as delaying maintenance or advertising expenditure, and would even give up positive NPV projects,
to meet short-term earnings benchmarks.”
The Accounting Review, May 2008
762
Cohen, Dey, and Lys
real and accrual-based methods after SOX. The next section discusses the empirical methodology employed in the study.
III. EMPIRICAL METHODOLOGY
Data and Sample Description
We collect our sample from the Compustat annual industrial and research files for the
period 1987-2005. We restrict our sample to all nonfinancial firms with available data, and
require at least eight observations in each two-digit SIC grouping per year. Further, we require that each firm-year observation has the data necessary to calculate the discretionary
accruals metrics and real earnings management proxies we employ in our analysis. This
restriction likely introduces a survivorship bias, biasing the sample toward larger and more
successful firms. However, we expect that this bias will reduce the variation in our earnings
management metric, resulting in a more conservative test of our research questions.
Following Collins and Hribar (2002), we use cash flows from operations obtained from
the Statement of Cash Flows reported under the Statement of Financial Accounting Stan-
dards No. 95 (SFAS No. 95, FASB 1987).9 The sample period of 1987-2005 permits us
to use SFAS No. 95 statement of cash flow data to estimate accruals, rather than a balance
sheet approach.
The sample obtained from Compustat consists of 8,157 firms representing 87,217
firm-year observations. To test the compensation hypothesis, we use data from
ExecuComp, which is available only from 1992 onward. Thus, merging the full sample
with ExecuComp results in a second (and smaller) sample consisting of 2,018 firms and
31,668 firm-year observations (the ExecuComp sample) for the 1992 through 2005 period.
Event Periods
We focus in our analysis on earnings management across two main time periods-the
pre-SOX period (further classified into the pre-SCA and the SCA periods), and the post-
SOX period. The pre-SOX period extends from 1987 through 2001, and the post-SOX
period extends from 2002 through the end of 2005. Within the pre-SOX period, we classify
the period from 1987 through 1999 as the pre-SCA period, and the period from 2000
through 2001 as the SCA period (i.e., the period that purportedly lead to the passage of
SOX).`° Figure 1 depicts these different time periods analyzed.”1
9 SFAS No. 95 requires firms to present a statement of cash flows for fiscal years ending after July 15, 1988.
Some firms early-adopted SFAS No. 95, so our sample begins in 1987.
10 We acknowledge that the subdivision into the pre-SCA and SCA periods may induce hindsight bias into the
analysis. We thus repeat our analysis by only dividing the entire sample period into the pre-SOX and the postSOX periods. Our conclusions on earnings management activities before and after SOX are unchanged and
become stronger. We do not report these results in the paper for the sake of brevity, but will provide them upon
request.
SThe use of annual data determines how we subdivide the sample period to some extent. Although some scandals
took place in the beginning of 2002, we include 2002 in the post-SOX period, since SOX was passed in 2002.
Also, even though the most public phase of the scandals began with Enron in 2001, we include year 2000 in
the SCA period since some frauds also occurred in 2000 (e.g., Xerox). Further, as a robustness check, we also
repeat our analysis using quarterly data and subdivide the sample period using the “Corporate Scandal Sheet”
developed by Forbes (2002). Specifically, we define the SCA period as extending from Q3, 2001 through Q2,
2002, and the post-SOX period as extending from Q3, 2002 onward. Our main conclusions remain unchanged,
which provides added confidence in the results obtained using annual data. Finally, as suggested by the referee
we repeat all analyses by defining the post-SOX period as the years 2003 through 2005 and this did not materially
alter any of our reported results.
The Accounting Review, May 2008
763
Real and Accrual-Based EarningsManagement
FIGURE 1
Time Periods Analyzed
1987
2000
10
Pre-SCA Period
2005
2002
1
SCA Period
Pre-SOX Period
Post-SOX Period
Earnings Management Metrics
Accrual-Based Earnings Management
We use a cross-sectional model of discretionary accruals, where for each year we
estimate the model for every industry classified by its two-digit SIC code. Thus, our approach partially controls for industry-wide changes in economic conditions that affect total
accruals while allowing the coefficients to vary across time (e.g., Kasznik 1999; DeFond
and Jiambalvo 1994).12
Our primary model is the modified cross-sectional Jones model (Jones 1991) as described in Dechow et al. (1995).”3 The modified Jones model is estimated for each twodigit SIC-year grouping as follows:
TAit
Assetsi
A st i,t- -1
1
” Assets3
A s t i,,
AREV.
PPE,i
k A
A-sets2
sse ts i a- 1
k A
A sse ts
t i ,t- l + eit
(1)
where, for fiscal year t and firm i, TA represents total accruals defined as:
TAi, = EBXIi, – CFOj,, where EBXI is the earnings before extraordinary items and
discontinued operations (annual Compustat data item 123) and CFO is the
operating cash flows (from continuing operations) taken from the statement
of cash flows (annual Compustat data item 308 – annual Compustat data
item 124);
Assetsi,-, = total assets (annual Compustat data item 6);
AREVi, = change in revenues (annual Compustat data item 12) from the preceding
year; and
12
13
We obtain qualitatively the same results when we use a time-series approach that assumes temporal stationarity
of the parameters for each firm.
Caveat: Various studies in the literature raise the concern that discretionary accruals measured using the Jones
model might be capturing nondiscretionary components and these errors in discretionary accruals are likely to
be correlated with stock prices and performance measures in general. While this concern is valid and we
acknowledge this limitation in measuring discretionary accruals, note that we use discretionary accruals as a
dependent variable and not as an explanatory variable. If indeed discretionary accruals are measured with error,
then the only consequence in our case will be a lower explanatory power of the model, i.e., we will obtain lower
R’s. Otherwise, using discretionary accruals measured using the Jones model as a dependent variable is not
likely to introduce any bias in our results.
The Accounting Review, May 2008
764
Cohen, Dey, and Lys
PPEi, = gross value of property, plant, and equipment (annual Compustat data item
7).
The coefficient estimates from Equation (1) are used to estimate the firm-specific normal accruals (NAi,) for our sample firms:
NAit = k
1
+ k2 (AREVi, – AARi) + k,
Assetsi,,_
Assetsi, 1
PPEi
Assetsi—
(2)
(
where AARi, is the change in accounts receivable (annual Compustat data item 2) from the
preceding year. Following the methodology used in the literature, we estimate the industryspecific regressions using the change in reported revenues, implicitly assuming no discretionary choices with respect to revenue recognition. However, while computing the normal
accruals, we adjust the reported revenues of the sample firms for the change in accounts
receivable to capture any potential accounting discretion arising from credit sales. Our
measure of discretionary accruals is the difference between total accruals and the fitted
normal accruals, defined as DAit = (TAzilAssetsit-,) – NAit.
In contrast to studies that focus on a specific corporate event, our analysis using the
full sample is conducted in calendar time. Consequently, because accruals reverse over time
and we cannot condition the analysis on events that are hypothesized to provide managers
with incentives to manage reported earnings in any given direction (e.g., inflate reported
earnings) we compute the absolute value of discretionary accruals to proxy for earnings
management and refer to it as ABS-DA throughout the analysis.’ 4 In contrast, our test of
the SUSPECT firms (that is, firms that were just able to meet or beat earnings benchmarks)
is based on a directional test.
In our robustness tests, we used two alternative measures of discretionary accruals. In
one alternative measure we estimated the following in the first stage:
TAi,
– k1t
1
Assets
Assetsi,,_ 1
+ k, (AREVi, – AAR,) + k, PPE_, + Ei,
+
Assetsi,,+1
1 +tAssets,i, –
(3)
Using the coefficient estimates obtained from Equation (3), we calculated the level of
normal accruals (NA,t) as a percent of lagged total assets. We also repeat our tests by using
a measure based on the performance-matched discretionary accruals advanced in Kothari
et al. (2005). As suggested by Kothari et al. (2005), we match each firm-year observation
with another from the same two-digit SIC code and year with the closest return on assets
in the current year, ROAi, (net income divided by total assets).’ 5 Our results using these
alternate measures of accruals are consistent with those reported in the paper.
Real Earnings Management
We rely on prior studies to develop our proxies for real earnings management. As in
Roychowdhury (2006) we consider the abnormal levels of cash flow from operations (CFO),
discretionary expenses and production costs to study the level of real activities manipulations. Subsequent studies, such as Zang (2006) and Gunny (2005), provide evidence of the
“4We repeat our analysis using the square of discretionary accruals and the results are somewhat stronger. Although
the squared discretionary accruals have more desirable distributional properties, we report the results using the
absolute values of discretionary accruals to allow comparison with other research.
15 We also carry out performance matching based on two-digit SIC code, year, and ROA (both current ROA and
lagged ROA) and obtain results similar to those reported in the paper.
The Accounting Review, May 2008
765
Real and Accrual-Based Earnings Management
construct validity of these proxies. We focus on three manipulation methods and their
impact on the above three variables:
1. Acceleration of the timing of sales through increased price discounts or more lenient credit terms. Such discounts and lenient credit terms will temporarily increase
sales volumes, but these are likely to disappear once the firm reverts to old prices.
The additional sales will boost current period earnings, assuming the margins are
positive. However, both price discounts and more lenient credit terms will result in
lower cash flows in the current period.
2. Reporting of lower cost of goods sold through increased production. Managers can
increase production more than necessary in order to increase earnings. When managers produce more units, they can spread the fixed overhead costs over a larger
number of units, thus lowering fixed costs per unit. As long as the reduction in
fixed costs per unit is not offset by any increase in marginal cost per unit, total
cost per unit declines. This decreases reported cost of goods sold (COGS) and the
firm can report higher operating margins. However, the firm will still incur other
production and holding costs that will lead to higher annual production costs relative to sales, and lower cash flows from operations given sales levels.
3. Decreases in discretionary expenses that include advertising expense, research and
development, and SG&A expenses. Reducing such expenses will boost current
period earnings. It could also lead to higher current period cash flows (at the risk
of lower future cash flows) if the firm generally paid for such expenses in cash.
We first generate the normal levels of CFO, discretionary expenses, and production
costs using the model developed by Dechow et al. (1998) as implemented in Roychowdhury
(2006). We express normal CFO as a linear function of sales and change in sales. To
estimate this model, we run the following cross-sectional regression for each industry and
year:
CFOiAssetsi.,_1
k ,
+ k
Salesit
= 1
+
-Assetsi,t I
2 AssetsA
ASalesi, +t.
+ k3Asset.s(4
(4)
ts,,
Abnormal CFO is actual CFO minus the normal level of CFO calculated using the
estimated coefficient from Equation (4).
Production costs are defined as the sum of COGS and change in inventory during the
year. We model COGS as a linear function of contemporaneous sales:
COGSit – kit
Assetsij-
I
I
Assetsi,t-
+ k2
(5)
Sales i + F
Assetsi,A
,
Next, we model inventory growth by the following:
AINVt _= k,t
I
Assetsi,t_ 1
Assets.,,
+ k2 ASalest
Assetsi., 1k
ASalesi.,_
Assets,t-
i
i
The Accounting Review, May 2008
766
Cohen, Dey, and Lys
Using Equations (5) and (6), we estimate the normal level of production costs as:
Prod,
Assetsi_,
1
1
Salesi.
ASales,t
ASalesi,-,
Assetsit
,
, 1 + k2 Assetsij
+ k 3 Assetsi,t
s
+ k4 A
Assetsit-
11
+
i,. (7)
We model the normal level of discretionary expenses as:
DiscExpi, = k ,
Assetsi,,- 1
1
si,_1 ,
+ k,
S e
+ Ej,
(8)
Assetsi, 1
Modeling discretionary expenses as a function of current sales creates a mechanical problem
if firms manage sales upward to increase reported earnings in a certain year, resulting in
significantly lower residuals from running a regression as derived in Equation (8). To address this issue, we model discretionary expenses as a function of lagged sales and estimate
the following model to derive ‘normal’ levels of discretionary expenses:
DiscExp j, =k
Assetsi,t1
1
Assetsi,
k+
k2 Salesit”-
+
e
(9)
Assetsia,
In the above equations CFO is cash flow from operations in period t (Compustat data
item 308 – annual Compustat data item 124); Prod represents the production costs in period
t, defined as the sum of COGS (annual Compustat data item 41) and the change in inventories (annual Compustat data item 3); and DiscExp represents the discretionary expenditures in period t, defined as the sum of advertising expenses (annual Compustat data item
45), R&D expenses (annual Compustat data item 46),16 and SG&A (annual Compustat data
item 189). The abnormal CFO (R-CFO), abnormal production costs (R-PROD) and abnormal discretionary expenses (R-DISX) are computed as the difference between the actual
values and the normal levels predicted from Equations (4), (7), and (9). We use these three
variables as proxies for real earnings management.
Given sales levels, firms that manage earnings upward are likely to have one or all of
these: unusually low cash flow from operations, and/or unusually low discretionary expenses, and/or unusually high production costs. In order to capture the effects of real
earnings management through all these three variables in a comprehensive measure, we
compute a single variable by combining the three individual real earnings management
variables. Specifically, we compute RM-PROXY as the sum of the standardized variables,
R_CFO, R-PROD, and R-DISX. However, we acknowledge that the three individual variables have different implications for earnings that may dilute any results using RM-PROXY
alone. We thus report results corresponding to the single real earnings management proxy
(RM-PROXY) as well as the three individual real earnings management proxies (R-CFO,
R-PROD, and R-DISX).
IV. TESTS AND RESULTS
Descriptive Statistics: Earnings Management
We begin with an exploratory analysis of the trends over time in the various earnings
management metrics. Table 1, Panel A provides summary statistics of the full sample, while
16 As long as SG&A is available, advertising expenses and R&D are set to zero if they are missing.
The Accounting Review, May 2008
767
Real and Accrual-Based Earnings Management
TABLE 1
Descriptive Statistics
25th
Percentile
Mean
Median
75th
Percentile
Standard
Deviation
89.02
76.71
94.70
0.08
124.05
0.21
1.92
-0.06
0.00
0.06
-0.02
0.06
0.01
0.04
0.01
-0.01
438.68
441.13
492.98
0.25
186.01
0.39
3.48
-0.001
0.05
0.11
-0.01
0.12
0.08
0.06
0.18
0.06
8815
11855.55
849.40
978.67
895.94
0.09
111.27
0.20
2.48
-0.06
-0.01
0.00
-0.01
0.04
0.05
-0.04
0.00
0.00
0.14
219.32
0.432
0.102
0.049
1.067
2611.80
3280.46
2678.90
0.22
160.10
0.33
4.04
-0.02
0.03
0.03
-0.00
0.08
0.11
0.05
0.12
0.06
0.25
526.80
0.874
0.347
0.128
3.057
19038.10
20283.20
13039.01
0.76
73.61
0.20
6.45
0.12
0.11
0.06
0.08
0.09
0.15
0.28
0.15
0.16
0.15
821.30
0.857
0.431
0.304
1.067
Panel A: Full Sample, 1987-2005 (n = 87,217)
Total Assets
Market Capitalization
Sales
Growth of Sales
OC (Days)
Leverage
Market-to-Book
Total Accruals
DA
Positive-DA
Negative-DA
ABS-DA
R_CFO
R-PROD
R_DISX
RM-PROXY
17.54
15.08
17.05
-0.04
79.15
0.04
1.11
-0.12
-0.05
0.02
-0.03
0.03
-0.06
-0.17
-0.09
-0.08
1401.43
1616.11
1396.32
0.13
142.56
0.41
4.94
-0.10
0.00
0.09
-0.05
0.11
-0.02
-0.06
0.08
0.01
7555.65
0.49
90.89
0.26
6.17
0.25
0.20
0.09
0.18
0.17
0.35
0.31
0.58
0.35
Panel B: ExecuComp Sample, 1992-2005 (n = 31,668)
Total Assets
Market Capitalization
Sales
Growth of Sales
OC (Days)
Leverage
Market-to-Book
Total Accruals
DA
Positive-DA
Negative-DA
ABS-DA
R_CFO
R-PROD
R_DISX
RM-PROXY
BONUS (%)
BONUS ($)
EX-OPTIONS (%)
UN-OPTIONS (%)
GRNT-OPTION (%)
OWNER (%)
334.88
388.27
331.89
0.01
72.49
0.05
1.63
-0.10
-0.05
0.00
-0.05
0.01
-0.00
-0.17
-0.10
-0.06
0.04
60.00
0.312
0.072
0.032
0.764
4232.66
5491.55
4025.76
0.18
124.69
0.22
4.70
-0.07
-0.01
0.03
-0.04
0.07
0.05
-0.05
0.01
0.00
0.16
454.04
0.745
0.212
0.129
5.191
(continued on next page)
The Accounting Review, May 2008
768
Cohen, Dey, and Lys
TABLE 1 (continued)
Variable Definitions:
Total Assets = annual Compustat data item 6;
Market Capitalization = the price per share (annual Compustat data item 199) times the number of shares
outstanding (annual Compustat data item 25);
Sales = annual Compustat data item 12;
Growth of Sales = the change in sales divided by lagged sales;
Total Accruals = the difference between operating cash flows (annual Compustat data item 308),
adjusted for extraordinary items and discontinued operations (annual Compustat data
item 124) and income before extraordinary items (annual Compustat data item 123)
divided by lagged total assets;
DA = discretionary accruals computed using the Modified Jones Model;
ABS-DA = the absolute value of discretionary accruals computed using the Modified Jones model;
(AR, + A,_,)/2
(INVt + INV,_,)/2
+
OC = the operating cycle (in days), calculated as
(Sales/360)
(COGS/360)
Leverage = total liabilities (annual Compustat data item 9 plus data item 34) divided by total
assets;
Book-to-Market = the market capitalization divided by the book value of common equity (annual
Compustat data item 60);
Positive-DA = the value of positive discretionary accruals computed using the Modified Jones Model;
Negative-DA = the value of negative discretionary accruals computed using the Modified Jones
Model;
R_CFO = the level of abnormal cash flows from operations;
R-PROD = the level of abnormal production costs, where production costs are defined as the sum
of cost of goods sold and the change in inventories;
R-DISX = the level of abnormal discretionary expenses, where discretionary expenses are the
sum of advertising expenses (annual Compustat data item 45), R&D expenses (annual
Compustat data item 46) and SG&A expenses (annual Compustat data 189);
RM-PROXY = the sum of the standardized three real earnings management proxies, i.e., R_CFO,
R-PROD and R_DISX;
BONUS (%) = the average bonus compensation as a proportion of total compensation received by the
CEO and CFO of the firm;
BONUS ($)= the BONUS variable in ExecuComp in thousands of dollars;
EX-OPTIONS (%)= exercisable options, which is the number of unexercised options that the executive held
at year-end that were vested scaled by total outstanding shares of the firm;
UN_OPTIONS (%)= unexercisable options defined as the number of unexercised options (excluding options
grants in the current period) that the executive held at year-end that had not vested
scaled by total outstanding shares of the firm;
GRNT-OPTION (%) = new option grants made during the current period scaled by total outstanding shares of
the firm; and
OWNER = the sum of restricted stock grants in the current period and the aggregate number of
shares held by the executive at year-end (excluding stock options) scaled by total
outstanding shares of the firm.
Table 1, Panel B provides summary statistics of the ExecuComp sample.’ 7 Both samples
are significantly larger (at the 0.001 level) in terms of total assets and market capitalization
when compared to the “average” firm listed on Compustat. Sample firms have a 13 percent
annual growth in sales, a market-to-book ratio of 4.94, and a leverage ratio of 0.41 (18
percent, 4.70, and 0.22, respectively, for the ExecuComp subsample).
The operating cycles of our sample firms is, on average, approximately 142.56 days
(124.69 days for the ExecuComp subsample), suggesting that accruals are likely to reverse
in the subsequent year. Consistent with prior research we find a positive and significant
correlation (0.019, Pearson; 0.053 Spearman; both significant at the 1 percent level, results
not tabulated) between the operating cycle and discretionary accruals (Dechow and Dichev
2002). Finally, requiring the availability of ExecuComp data while considerably increasing
‘7
Whenever possible, we perform the tests on both the full and the ExecuComp samples to assess the impact of
the ExecuComp selection on our results.
The Accounting Review, May 2008
769
Real and Accrual-Based EarningsManagement
firm size does not seem to have a significant impact on fundamental measures such as
leverage, growth of sales, or market-to-book ratios.
As expected, TA (total accruals deflated by prior-year total assets) is negative at -0.10
(-0.07 for the ExecuComp subsample) with a standard deviation of 0.25 (0.12 for the for
the ExecuComp subsample). In contrast the average DA (discretionary accruals) is 0.00
(standard deviation of 0.20) for the full sample and -0.01 (0.11) for the ExecuComp
8
subsample. 1
While the average DA is zero, we find that positive discretionary accruals (PositiveDA) are, on average, larger in magnitude than negative discretionary accruals (NegativeDA). This is not only true for the mean, but also for the median and the 75th percentiles.
This is however not true for the ExecuComp sample (except at the 75th percentile). Thus,
for the full sample, it appears that larger earnings increasing DAs are followed by smaller
but more frequent reversals.
The main variable of interest is the absolute value of discretionary accruals (ABS-DA).
We use the absolute value because our hypotheses do not predict any specific direction for
earnings management. Moreover, the absolute value also captures accrual reversals following earnings management. The average for ABS-DA is 0.11 (0.07 for the ExecuComp
sample). At first, this may seem a large value as percentage of total assets. However, recall
that while DA has a mean of zero, the mean is shifted to the right by taking absolute
values. 19,20
Nevertheless, we conduct some additional analyses to get more reassurance on the
magnitudes of the discretionary accrual measure. First, we selected a few firms that had
very high absolute values of DA as percentage of total assets (higher than 15 percent) and
examined their financial statements. 2′ For the 12 firms that meet this criterion, we find that
more than 50 percent of them had large asset impairment and restructuring charges and
write-offs related to goodwill, while the rest had merger-related charges and significant
growth in accounts receivables and inventory, which (at least partly) explains the large
magnitudes in DAs that we observe. To check whether our results are affected by outliers,
we winsorize the top and bottom 1 percent of the distribution. On winsorizing, the mean
(median) of ABS-DA, is 0.06 (0.05) for the full sample and 0.04 (0.04) for the ExecuComp
sample. These magnitudes of discretionary accruals appear more realistic. Moreover, our
cross-sectional results and inferences are not affected by repeating all our analyses using
the winsorized sample (results not tabulated).
Finally, we also check the discretionary accruals of the firms that were involved in
accounting scandals. Untabulated results indicate that all but one of the firms involved
18 Note that
the mean Jones model residual is zero by construction (residual of a regression), whereas the mean
modified Jones model residual is not constrained to be zero by construction.
If X follows a normal distribution with mean [Land variance a’ then the expected absolute value of X is given
by:
wee E[IX]] =- u2/e
±+[~()
1
where (1)(.) is the standard normal cumulative distribution function.
We also perform two “sanity checks.” First, we generated 1,000 random variables from a normal distribution
with mean 0.00 and a standard deviation of 0.20. The mean of the absolute values of those random variables is
0.173. Second, we computed the ROA and the absolute value of the ROA for our sample firms. While
the average ROA for our sample firms is -0.06 with a standard deviation of 0.191 (0.046 and 0.207 for the
ExecuComp subsample) the absolute value of ROA is 0.22 (0.118 for the ExecuComp subsample).
21 We thank one of the referees for this suggestion.
21
The Accounting Review, May 2008
770
Cohen, Dey, and Lys
in accounting scandals fall in the 90th percentile of the ABS-DA distribution in the SCA
period.
The last three rows of Panels A and B of Table 1 report our proxies for real earnings
management. Comparing the 25th and 75th percentiles of the real earnings management
proxies to DA suggests that accrual-based earnings management takes larger values. This
observation is consistent with Graham et al.’s (2005) survey, which suggests that real earnings management is more costly than accrual based earnings management.
The last five rows of Panel B of Table 1 report both the dollar amount obtained from
bonuses and characteristics of equity-based compensation, i.e., exercisable and unexercisable options, new option grants, and stock ownership. For the entire sample period, the
average ExecuComp CEO received annual bonuses of $454,040, representing 16 percent
of their total compensation. New option grants and other unexercisable options are on
average 0.13 percent and 0.21 percent of outstanding shares respectively, while exercisable
options are on average 0.75 percent of outstanding shares. The sum of restricted stock
grants and aggregate shares held by the CEO averages 5.19 percent.
Table 2, Panel A summarizes time-trends of the accrual and real earnings management
proxies. To summarize the data, we regress each of the variables on a time-trend variable,
Time, defined as the difference between the year and 1987, and two dummy variables, SCA,
which takes the value of 1 in the scandal period (years 2000 and 2001), and 0 otherwise,
and SOX, which takes the value of 1 in the post-SOX period (years 2002, 2003, and 2005), and
0 otherwise. We choose this procedure to describe the variables because many of our
variables exhibit significant time trends (nonstationarity), rendering traditional summary
statistics uninformative.
The coefficient for the time trend in Table 2, Panel A, row 1 (b) indicates that the
magnitude of discretionary accruals (ABS-DA) has been increasing significantly (at the 1
percent level) over the sample period with that increase being nearly symmetric for both
Positive-DA and Negative-DA (rows 2 and 3).
The magnitude of discretionary accruals increased significantly in the SCA period (0
in Row 1) with positive (i.e., income-increasing, Row 2) discretionary accruals contributing
twice as much to that increase than income-decreasing DAs (Row 3, both significant at the
1 percent level).
Finally, the magnitude of discretionary accruals declined significantly in the postSOX period (a in Row 1). Moreover, the magnitude (absolute value) of the coefficient
for Positive-DA (a in Row 2) is approximately three times larger than the coefficient for
Negative-DA (d in Row 3) suggesting that most of that decline in accrual-based management results from the reduction of Positive-DAs.
Figures 2 and 3, Panels A and B provide graphical illustrations of these results. Figure
2 indicates that the SCA period was, indeed, associated with a high level of earnings
management. Figure 3, Panels A and B plot the trends in positive and negative discretionary
accruals. Positive discretionary accruals peaked in the SCA period and negative discretionary accruals were the lowest in that period. These trends reversed in the post-SOX period.
Among the real earnings management variable, except for R-DISX, which increased
over the period, the other real earnings management variables do not show an increasing
trend over the sample period (the time trend coefficients are either not significant or negative). Abnormal production costs and abnormal cash flows both increased significantly in
the post-SOX period, but were not significantly higher in the SCA period. On the other
hand, abnormal discretionary expenses were significantly higher in the SCA period and
significantly lower in the post-SOX period. The combined variable RM-PROXY shows a
The Accounting Review, May 2008
Real and Accrual-Based Earnings Management
771
TABLE 2
Earnings Management Metrics: Time Trends and Correlation Matrix
Panel A: Time Trend Analysis of Earnings Management Metrics over Time, 1987-2005
Dep,, = a + b x Time + c X SCA + d x SOX
Dependent
Variables
ABS-DA
Positive-DA
Negative-DA
R-PROD
R_CFO
R_DISX
RM-PROXY
b
0.053***
0.028***
-0.040***
-0.040***
0.022***
0.028***
-0.018***
0.003***
0.002***
-0.003***
-0.002
-0.003**
0.005***
-0.005**
0.027***
0.026***
-0.012**
0.008
-0.007***
0.008***
0.033**
a
Adjusted R2
-0.015***
-0.019**
0.006***
0.032***
0.007**
-0.011***
0.041***
0.024
0.029
0.015
0.009
0.007
0.010
0.011
Panel B: Correlation between the Earnings Management Proxies, 1987-2005
DA
ABS-DA
R_CFO
R-PROD
R_DISX
RM-PROXY
DA
ABS-DA
R_CFO
R-PROD
R-DISX
RM-PROXY
1
0.298***
-0.214***
-0.024***
-0.143***
-0.237***
0.398***
1
-0.114***
-0.314***
-0.164***
-0.149***
-0.237***
-0.184***
1
-0.183***
-0.214***
0.381***
-0.029***
-0.047***
-0.274***
1
-0.187***
0.427**
-0.163***
-0.174***
-0.142***
-0.243***
1
0.422***
-0.331***
-0.153***
0.447***
0.579***
0.487***
1
*, **, *** Significant at 10 percent, 5 percent, and 1 percent levels, respectively.
Variable Definitions:
DA = discretionary accruals computed using the Modified Jones Model;
ABS-DA = the absolute value of discretionary accruals computed using the Modified Jones Model;
Positive-DA = the value of positive discretionary accruals computed using the Modified Jones Model;
Negative-DA = the value of negative discretionary accruals computed using the Modified Jones Model;
TIME = a trend variable equal to the difference between the current year and 1987;
SCA = a dummy variable equal to 1 if the year is either 2000 or 2001; and
SOX = a dummy variable for years equal to 2002, 2003, 2004, and 2005.
decreasing trend over time, but is significantly higher both in the SCA period and in the
post-SOX periods.
Figure 4 graphically illustrates these trends. A comparison of the post-SOX coefficients
for accrual-based and the real earnings management suggests that there may have been a
substitution effect: while accrual-based earnings management decreased, overall firms increased the use of real earnings management methods.
The correlation among accrual-based and real earnings management is reported in Table
2, Panel B. We find a significant negative relation between discretionary accruals and the
real earnings management metrics, suggesting that firms appear to use these two earnings
management methods as substitutes. Further, the three real earnings management variables
are also negatively correlated, indicating that firms switch between real earnings management methods.
In summary, the above analysis indicates that the overall level of accrual-based earnings
management decreased from the SCA period to the post-SOX period, while overall the level
The Accounting Review, May 2008
772
Cohen, Dey, and Lys
FIGURE 2
Absolute Value of Discretionary Accruals over Time, 1987-2005
0.18
01
ch~
1–ABS_
AE
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Year
This figure plots the absolute value of discretionary accruals computed using the Modified Jones Model over the
1987-2005 sample period.
of real earnings management increased in the post-SOX period. However, there was significantly higher earnings management, particularly income-increasing earnings management, during the SCA period as compared to the pre-SCA period. One interpretation of
this result is that the SCA period was characterized by higher earnings management; and
the scandal firms were not just a “few bad apples,” but evidence of the generally high level
of accrual management. Another observation is that, although earnings management using
accrual-based means increased from the pre-SCA to the SCA period, it declined significantly from the SCA period to the post-SOX period. It should be noted that most of the
decline in accrual-based earnings management seems to have been due to a reduction in
positive discretionary accruals.
Whether this decline is caused by the passage of SOX or other concurrent events (such
as the negative publicity of the most egregious governance failures and the highly visible
enforcement actions directed at the offending corporate officers) cannot be inferred from
this analysis, however. Moreover, firms appeared to have switched to managing earnings
using real management techniques after SOX, probably because such methods of manipulation are harder to detect. How this substitution between the different types of earnings
management affected the overall level of earnings management post-SOX is unclear. We
formally examine the determinants of earnings management in our multivariate analysis
next.
The Accounting Review, May 2008
773
Real and Accrual-Based Earnings Management
FIGURE 3
Discretionary Accruals over Time, 1987-2005
Panel A: Positive Discretionary Accruals over Time, 1987-2005
u. 1.
0.14
0.12
“0.1
0
I’-Pos-DAI
0 .08
0.06
0.04
SCANDAL_ SCANDAPOST SOX
0.02
0
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Year
Panel B: Negative Discretionary Accruals over Time, 1987-2005
0
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1 99 2000 2001 2002 2003 2004 2005
-0.02
-0.04
LPRE SOX
CAND
[POST SOX
-0.06-0.08a)
Z -0.1D
-0.12
-0.14
-0.16
-0.18-0.2
Year
This figure plots positive and negative discretionary accruals computed using the Modified Jones Model over the
1987-2005 sample period.
The Accounting Review, May 2008
774
Cohen, Dey, and Lys
FIGURE 4
Real Earnings Management Proxies over Time, 1987-2005
0.16
0.150.14-
POSTSX
0.130.120.11
x
0
c-
0.1
0.090.080.070.060.050.04
0.03
0.02
SCNAL
-R-CFO
/
SR-PROD
-4–R DISX
RM PROXY
0.01
0
-0.01
-0.02
-0.03
-0.04
-0.05
-0.06
-0.07
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
-u.ub 1-
Year
Figure 4 plots abnormal cash from operations, abnormal production costs, abnormal discretionary expenses, and
the sum of the standardized three real earnings management proxies, RM-PROXY over the 1987-2005 sample
period.
Trends In and Determinants of Earnings Management
We examine the trends in and determinants of the level of earnings management over
time by estimating the following regression:
DEPJ = a0 + a 1 X BIGj + 0t2 X AGDPJ + a 3 X MKTVALJ + a4 X Time
“+ oL5 X SCA + ot6 X SOX + a-7 X RM PROXY + u58 X BONUS.
“+ t9 x BONUS x SCA + a-,0 x BONUS. x SOX
“+ o, x UN OPTIONJ + t1 2 X UN OPTION X SCA
“+ a 1 3 X UN OPTIONJ x SOX + a1 14 X GRNT_OPTION.
“+ o 15 x GRNT OPTIONJ X SCA + at 16 X GRNT OPTION, X SOX
“+ a 17 x EX OPTIONj + t,8 x EX_OPTIONJ x SCA
“+ a19 X EX_OPTION, X SOX + a 2o X OWNERJ
“+ aL21 X OWNER, X SCA + a 22 X OWNER. X SOX
The Accounting Review, May 2008
(10)
Real and Accrual-Based Earnings Management
775
where:
DEPj = the various earnings management metrics, including discretionary
accruals, positive discretionary accruals, and negative discretionary
accruals;
BIG = a dummy variable equal to 1 if the auditor is a Big 5 audit firm (or
their successors);
AGDP = the change in the GDP;
MKTVAL = market value of equity;
Time = the calendar year minus 1987;
SCA = a dummy variable that is equal to 1 for the years 2000 and 2001, and
0 otherwise (represents the SCA period);
SOX = a dummy variable that is equal to 1 for the years 2002 onward, and 0
otherwise (represents the post-SOX period);
BONUS = the average bonus compensation as a proportion of total
compensation received by the CEO and the CFO of a firm;
EX-OPTION = exercisable options and is the number of unexercised options that the
executives held at year-end that were vested scaled by total
outstanding shares of the firm;
UN-OPTION = unexercisable options defined as the number of unexercised options
(excluding option grants in the current period) that the executives
held at year-end that have not vested scaled by total outstanding
shares of the firm;
GRNT-OPTION = new option grants made during the current period scaled by total
outstanding shares of the firm; and
=
the sum of restricted stock grants in the current period and the
OWNER
aggregate number of shares held by the executives at year-end
(excluding stock options) scaled by total outstanding shares of the
firm.22
The compensation variables proxy for performance-based compensation and are defined in
accordance with prior studies (Cheng and Warfield 2005). In addition, we separate out new
options granted in the current period from other unexercisable options because current
grants are likely to provide incentives to reduce earnings to benefit from lower exercise
prices, which are opposite to the incentives provided by other unexercisable options. We
include the variable GRNT-OPTION in order to capture the differential incentive effects
of new option grants awarded during the current period. We include these variables to test
the conjecture that the bonus and equity components of executive compensation are likely
to induce opportunistic behavior by managers.2 3
We include the variable AGDP as a proxy for real economic activity. We include this
to control for the effect of economic activity on earnings management, since what might
be classified as opportunistic earnings management may, in fact, be a consequence of changing economic conditions. Discretionary accruals may also reflect firms’ responses to and
We also repeat the analyses by measuring the compensation related variables for the top five executives of a
firm, and the results are qualitatively unchanged.
23 The timing of variable measurement is consistent with Cheng and Warfield (2005, 448). The equity incentive
variables are measured during or at the end of fiscal year t, whereas the earnings management variable is
measured based on information disclosed after the end of fiscal year t.
22
The Accounting Review, May 2008
776
Cohen, Dey, and Lys
representations of changes in economic conditions. If this were true, then changes in earnings management metrics would coincide with changes in measures of economic activity
such as operating cash flows, revenues, prior stock returns, industry performance, changes
in gross domestic product, etc. Further, after controlling for changes in economic activities,
there should be no relation between increases in earnings management and the compensation variables.24
We include control variables for the auditors in the above regression to examine whether
the earnings management activity of firms audited by the large audit firms were different
from the rest of the sample firms over the three subperiods analyzed. Note that we make
no claim that differences in the earnings management activities (if any) of these firms were
the result of the monitoring activities of the audit firms, since there could be a self-selection
by certain types of firms in their selection of big audit firms. In addition, to the extent that
audit firms specialize in specific industries and levels of earnings management are likely
to vary across industries, the audit firm dummies may also control for industry characteristics. Finally, we include the market value of equity as a proxy for firm size.
Next, we examine the trends in and determinants of real earnings management activities
of firms by estimating the following regression:
DEPJ = ot + a- X BIGJ + %2 X AGDPJ + ot3 x MKTVALJ + et4 X Time
“+ ot X SCA + ot6 X SOX + 0,7 X ABS-DA. + t 8 X BONUSJ
“±a X BONUSj X SCA + oto x BONUSj X SOX
“+ oal x UN OPTION, + t 12 X UN OPTIONA x SCA
“+ L1 3 X UN OPTIONJ X SOX + (x14 X GRNT OPTIONJ
“+ .15 X GRNT_OPTION, X SCA + 0X16 X GRNT OPTIONJ X SOX
“+o 1 7 x EX_OPTION, + oY8 X EX_OPTION, X SCA
“- ot9 x EX OPTION, x SOX + ’20 X OWNERJ
“+ o2l X OWNERJ X SCA + a 22 X OWNER, X SOX
(11)
where DEPj represents the three real earnings management metrics, R_CFO, R_PROD, and
R-DISX, and the combined variable RM-PROXY. All other variables are as defined above.
Firms may follow an overall earnings management strategy and use a mix of real and
accrual-based earnings management tools. Alternatively, they can choose between the two
management techniques, using the technique that is less costly for them. To control for this
possibility, we also include a variable representing accrual-based earnings management (we
include ABS-DA in the regression). If it were more costly for firms to manage earnings
using accrual-based techniques after the passage of SOX and they substituted this by using
real management techniques instead, then we should observe a significant increase in the
latter after SOX. On the other hand, firms could have decreased earnings management
activities as a whole-in which case we would observe a decrease in real earnings management after SOX as well. The results of these regressions are discussed next.
24 In the
“SUSPECT Firms Analysis” section we discuss some robustness tests we conduct to further test whether
we are indeed capturing earnings management activities.
The Accounting Review, May 2008
Real and Accrual-Based Earnings Management
777
Results
Tables 3 and 4 present the results of the determinants of the level of earnings management by firms. Table 3 reports the results when we use the absolute value of discretionary
accruals, ABS-DA, and when we split discretionary accruals into positive and negative
discretionary accruals. Table 4 reports the results when we use the measures for real earnings management activities. We discuss the results for the accrual-based earnings management variables first.
Consistent with the preliminary analysis reported in Table 2, we find a positive trend
in the level of earnings management, including income-increasing earnings management,
and a negative trend in income-decreasing earnings management. This indicates that overall
earnings management increased over the sample period. The dummy variable SOX is negative and significant for ABS-DA as well as for positive discretionary accruals and positive
and significant for negative discretionary accruals. This suggests that, controlling for the
other independent variables, the period after SOX was characterized by lower accrual-based
earnings management and this decrease primarily resulted from a decrease in incomeincreasing earnings management. Several simultaneous occurrences could have contributed
to a decrease in earnings management activities after passage of SOX, including the increased vigilance of investors, auditors and regulators, and greater care taken by managers
in financial reporting after the adverse publicity caused by the scandals. Thus, we are
cautious in attributing the decrease in the level of earnings management solely to the passage of SOX from this analysis.
Unlike our analysis in Table 2, the dummy variable representing the scandal period is
not significant for any of the variables. This implies that the entire increase in the accrualbased earnings management in the SCA period is related to the increase in equity-based
compensation. Further, we find that the percentage of bonus compensation is not correlated
with earnings management for the entire period, and this association did not significantly
change in the SCA or the post-SOX periods. Positive discretionary accruals, however, were
significant and positively associated with the percentage of compensation received from
bonuses in the scandal period.
Consistent with our conjecture, the percentage of compensation derived from unexercised options and stock ownership is significantly positively associated with discretionary
accruals as well as with positive discretionary accruals. For unexercised options this effect
increased significantly in the SCA period for positive discretionary accruals and decreased
significantly in the post-SOX period. This suggests that option compensation increased
managers’ incentives to manipulate earnings upward (e.g., Fuller and Jensen 2002;
Greenspan 2002; Coffee 2003), and this effect was significantly higher during the period
surrounding the corporate scandals. One possible explanation for the decline after SOX
could be the penalties on incentive compensation introduced by SOX. Unexercised options
and ownership are significantly negatively correlated with negative discretionary accruals
overall and after SOX. This suggests that the presence of stock and options in the compensation structure reduces the incentive to manage earnings downward. Unexercised options are also significantly negatively related to negative discretionary accruals in the scandal period indicating an even lesser inclination toward managing earnings downward during
this period.
New option grants (GRNT-OPTION) are positively associated with ABS-DA overall,
not significantly related to ABS-DA during the scandal period and significantly negatively
related to ABS-DA in the post-SOX period. The results for the positive and negative discretionary accruals are more interesting and are consistent with the notion that managers
The Accounting Review, May 2008
778
Cohen, Dey, and Lys
TABLE 3
Determinants of Accrual Based Earnings Management Activities 1987-2005
DEPj
ot, + c X BIGj + ot 2 X AGDPj + ot3 X MKTVALj + oo4 X Time
“+ X5 X SCA + ot X SOX + (X X RM_PROXY + a 8 X BONUSJ
“- ot X BONUSj x SCA + a-o x BONUS, x SOX + ax, X UN_OPTION,
“+ a12 X UN OPTION. X SCA + a 13 x UN OPTION, x SOX
“+al 4 X GRNT-OPTION, + a 1 5 X GRNT-OPTIONj x SCA
“+ Ot1 6 X GRNT_OPTIONj X SOX + a1 7 X EX_OPTIONj
“+ ao8 x EX_OPTIONJ X SCA + ot 9 X EX OPTION, x SOX
“±a 2 o X OWNER, + a 21 x OWNERJ X SCA + a 22 X OWNER. X SOX
6
Intercept
BIG
AGDP
MKTVAL
Time
SCA
SOX
RM-PROXY
BONUS
BONUS x SCA
BONUS x SOX
UN-OPTION
UN-OPTION x SCA
UN-OPTION x SOX
GRNT-OPTION
GRNT-OPTION X SCA
GRNT-OPTION x SOX
EX-OPTION
EX OPTION x SCA
EX-OPTION x SOX
OWNER
OWNER x SCA
OWNER x SOX
2
Adjusted R
F-value (Pr > F)
7
ABS-DA
Coefficient (t-stat)
0.118
(18.87)
-0.027
-0.114
-0.007
0.003
-0.002
-0.019
-0.427
0.002
-0.005
0.014
0.423
-0.124
-0.237
0.119
-0.087
-0.074
0.121
-0.006
-0.009
0.092
0.004
-0.033
(-5.87)
(-6.21)
(-17.36)
(8.21)
(-0.42)
(-4.32)
(-5.81)
(0.48)
(-0.43)
(1.27)
(10.97)
(-1.21)
(-3.94)
(2.93)
(-0.91)
(-3.63)
(1.12)
(-0.74)
(-1.09)
(5.58)
(0.49)
(-3.84)
0.084
169.16 (