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 signi? cant decline after the passage of SOX. Conversely, the level of real earnings management activities declined prior to SOX and increased signi? antly after the passage of SOX, suggesting that ? rms 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 ? nd that ? rms that just achieved important earnings benchmarks used less accruals and more real earnings management after SOX when compared to similar ? rms 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 ? ancial 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 raised concerns about the integrity of the accounting information provided to investors and resulted in a drop in investor con? dence (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. ’’1 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. Speci? cally, it is unclear whether there really was a widespread breakdown of the reliability of ? ancial reporting prior to the passage of SOX or whether the highly publicized scandals were isolated instances of individuals engaging in blatant ? nancial manipulations. If it were the former, then how did the passage of SOX affect ? rms’ ? nancial 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
T 1 2 3 Bumiller (2002). 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 Earnings Management 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 ? d 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 War? eld (2005), we do not ? nd a signi? cant 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 ? st to document this evidence empirically. Following the passage of SOX accrual-based earnings management declined signi? cantly, while real earnings management increased signi? cantly. At the same time, optionbased compensation decreased. Consistent with the results of a recent survey by Graham et al. (2005), this suggests that ? rms 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 ? ms that are more likely to have managed earnings (we refer to these ? rms as the SUSPECT ? rms). Speci? cally, 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 speci? c motives are among the most important reasons for earnings management behavior. Consistent with our full sample results, we ? nd that both before and after SOX, SUSPECT ? rms had signi? antly higher discretionary accruals when compared to ? rms that either just missed those benchmarks or ? rms that were not close to either making or missing those benchmarks. However, these SUSPECT ? rms used signi? cantly less incomeincreasing accrual-management after SOX when compared to ? rms in similar circumstances before SOX. An analysis of the real earnings management behavior of these ? rms indicates that the SUSPECT ? rms had signi? cantly higher real earnings management activities after SOX when compared to the ? rms 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 ? nd an increase in accrual-based earnings management prior to SOX, our evidence suggests that ? rms 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 bene? ts 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 ? nancial information presumed to be accurate and to re? ect economic reality. 4 The series of corporate scandals occurring in 2000–2001 eroded that trust in ? ancial reports. Indeed, one of the main objectives of SOX was to restore the integrity of 4 Paulson (2007). The Accounting Review, May 2008 760 Cohen, Dey, and Lys ?nancial 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 ? rms’ arnings management activities is motivated in part by the literature documenting that managerial propensity to manage earnings and to avoid negative earnings surprises has increased signi? cantly 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, ? rms 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 ? nd 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. %) state that they would delay starting a new project to meet an earnings target, even if such a delay entailed a small sacri? ce 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 in? uenced 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 dif? culty detecting earnings management (see Fields et al.  for a discussion on the pricing of earnings management. )7 5 6 7 In a related study, Lobo and Zhou (2006) investigate whether the SOX mandate that ? nancial statements be certi? ed by ? rms’ CEOs and CFOs resulted in an increase in the conservatism in ? nancial reporting, and ? nd evidence of greater conservatism in reported earnings.
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. Greenp (2002) opines that ‘‘the highly desirable spread of shareholding and options among business managers perversely created incentives to arti? ially in? ate 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 War? eld 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) ? d 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 signi? cant 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. Speci? cally, 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 ? rms who were accused of questionable or even fraudulent reporting practices. Our ? nal 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 ? m (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 ? rms 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 certi? ation 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 ? rms 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 ? les for the period 1987–2005. We restrict our sample to all non? nancial ? rms with available data, and require at least eight observations in each two-digit SIC grouping per year. Further, we require that each ? rm-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 ? rms.
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 ? ows from operations obtained from the Statement of Cash Flows reported under the Statement of Financial Accounting Standards No. 95 (SFAS No. 95, FASB 1987). 9 The sample period of 1987–2005 permits us to use SFAS No. 95 statement of cash ? ow data to estimate accruals, rather than a balance sheet approach. The sample obtained from Compustat consists of 8,157 ? rms representing 87,217 ? rm-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 ? rms and 31,668 ? rm-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 classi? ed into the pre-SCA and the SCA periods), and the postSOX 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). 10 Figure 1 depicts these different time periods analyzed. 11 9 10 11 SFAS No. 95 requires ? rms to present a statement of cash ? ows for ? scal years ending after July 15, 1988. Some ? rms early-adopted SFAS No. 95, so our sample begins in 1987. 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. The 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). Speci? cally, we de? ne 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 con? dence in the results obtained using annual data.
Finally, as suggested by the referee we repeat all analyses by de? ning 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 Real and Accrual-Based Earnings Management 763 FIGURE 1 Time Periods Analyzed 1987 2000 2002 2005 Pre-SCA Period 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 classi? ed 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 coef? cients to vary across time (e. g. , Kasznik 1999; DeFond and Jiambalvo 1994). 12 Our primary model is the modi? ed cross-sectional Jones model (Jones 1991) as described in Dechow et al. (1995). 13 The modi? ed Jones model is estimated for each twodigit SIC-year grouping as follows: TAit Assetsi,t k1t 1 1 Assetsi,t k2 1 REVit Assetsi,t 1 k3 PPEit Assetsi,t ?it 1 (1) where, for ? scal year t and ? rm i, TA represents total accruals de? ed as: TAit EBXIit CFOit, where EBXI is the earnings before extraordinary items and discontinued operations (annual Compustat data item 123) and CFO is the operating cash ? ows (from continuing operations) taken from the statement of cash ? ows (annual Compustat data item 308 annual Compustat data item 124); total assets (annual Compustat data item 6); change in revenues (annual Compustat data item 12) from the preceding year; and Assetsit 1 REVit 12 13 We obtain qualitatively the same results when we use a time-series approach that assumes temporal stationarity of the parameters for each ? rm.
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. . , we will obtain lower R2s. 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 PPEit Cohen, Dey, and Lys gross value of property, plant, and equipment (annual Compustat data item 7). The coef? cient estimates from Equation (1) are used to estimate the ? rm-speci? c normal accruals (NAit) for our sample ? rms: NAit ? k1t 1 Assetsi,t ? k2 1 ARit) ( REVit Assetsi,t 1 ? k3 PPEit Assetsi,t (2) 1 where ARit 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 industryspeci? c 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 ? rms 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 ? tted normal accruals, de? ned as DAit (TAit /Assetsit 1) NAit. In contrast to studies that focus on a speci? 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. , in? ate 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. 14 In contrast, our test of the SUSPECT ? rms (that is, ? rms 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 ? rst stage: TAit Assetsi,t k1t 1 1 Assetsi,t k2 1 ARit) ( REVit Assetsi,t 1 k3 PPEit Assetsi,t ?it. 1 (3) Using the coef? cient estimates obtained from Equation (3), we calculated the level of normal accruals (NAit) 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 ? m-year observation with another from the same two-digit SIC code and year with the closest return on assets in the current year, ROAit (net income divided by total assets). 15 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 ? ow 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 14 15 We 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. 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
Real and Accrual-Based Earnings Management 765 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 ? rm 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 ? ws 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 ? xed overhead costs over a larger number of units, thus lowering ? xed costs per unit. As long as the reduction in ? xed 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 ? rm can report higher operating margins. However, the ? m will still incur other production and holding costs that will lead to higher annual production costs relative to sales, and lower cash ? ows 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 ? ows (at the risk of lower future cash ? ows) if the ? rm generally paid for such expenses in cash. We ? rst 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: CFOit Assetsi,t 1 Assetsi,t Salesit Assetsi,t Salesit Assetsi,t 1 k1t 1 k2 1 k3 1 ?it. (4) Abnormal CFO is actual CFO minus the normal level of CFO calculated using the estimated coef? cient from Equation (4). Production costs are de? ned as the sum of COGS and change in inventory during the year. We model COGS as a linear function of contemporaneous sales: COGSit Assetsi,t 1 1 Assetsi,t Salesit Assetsi,t k1t k2 1 it. 1 (5) Next, we model inventory growth by the following: INVit Assetsi,t 1 1 Assetsi,t Salesit Assetsi,t 1 Salesi,t Assetsi,t k1t k2 1 k3 1 1 ?it. (6) The Accounting Review, May 2008 766 Cohen, Dey, and Lys Using Equations (5) and (6), we estimate the normal level of production costs as: Prodit Assetsi,t k1t 1 1 Assetsi,t k2 1 Salesit Assetsi,t k3 1 Salesit Assetsi,t 1 k4 Salesit Assetsi,t 1 1 ?it. (7) We model the normal level of discretionary expenses as: DiscExpit Assetsi,t 1 k1t 1 Assetsi,t k2 1 Salesit Assetsi,t ?it. 1 (8) Modeling discretionary expenses as a function of current sales creates a mechanical problem if ? ms manage sales upward to increase reported earnings in a certain year, resulting in signi? cantly 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: DiscExpit Assetsi,t 1 k1t 1 Assetsi,t k2 1 Salesi,t Assetsi,t 1 1 ?it. (9) In the above equations CFO is cash ? ow from operations in period t (Compustat data item 308 annual Compustat data item 124); Prod represents the production costs in period t, de? ed 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, de? ned 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, ? rms that manage earnings upward are likely to have one or all of these: unusually low cash ? ow 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. Speci? cally, 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 Real and Accrual-Based Earnings Management 767 TABLE 1 Descriptive Statistics 25th Percentile Panel A: Full Sample, 1987–2005 (n 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 Mean 87,217) 1401. 43 1616. 11 1396. 32 0. 13 142. 56 0. 41 4. 94 0. 0 0. 00 0. 09 0. 05 0. 11 0. 02 0. 06 0. 08 0. 01 31,668) 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 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. 1 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 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 Median 75th Percentile Standard Deviation Panel B: ExecuComp Sample, 1992–2005 (n 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 ($) EX OPTIONS (%) UN OPTIONS (%) GRNT OPTION (%) OWNER (%) 334. 88 388. 27 331. 89 0. 01 72. 49 0. 05 1. 63 0. 0 0. 05 0. 00 0. 05 0. 01 0. 00 0. 17 0. 10 0. 06 0. 04 60. 00 0. 312 0. 072 0. 032 0. 764 4232. 66 5491. 55 4025. 76 0. 18 124. 69 0. 22 4. 70 0. 07 0. 01 0. 03 0. 04 0. 07 0. 05 0. 05 0. 01 0. 00 0. 16 454. 04 0. 745 0. 212 0. 129 5. 191 (continued on next page) The Accounting Review, May 2008 768 TABLE 1 (continued) Variable De? nitions: Total Assets Market Capitalization Sales Growth of Sales Total Accruals Cohen, Dey, and Lys DA ABS DA OC Leverage Book-to-Market Positive DA Negative DA R CFO R PROD R DISX RM PROXY BONUS (%) BONUS ($) EX OPTIONS (%) UN OPTIONS (%) GRNT OPTION (%) OWNER nnual Compustat data item 6; the price per share (annual Compustat data item 199) times the number of shares outstanding (annual Compustat data item 25); annual Compustat data item 12; the change in sales divided by lagged sales; the difference between operating cash ? ows (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; discretionary accruals computed using the Modi? d Jones Model; the absolute value of discretionary accruals computed using the Modi? ed Jones model; (ARt At 1) / 2 (INVt INVt 1) / 2 the operating cycle (in days), calculated as ; (Sales / 360) (COGS / 360) total liabilities (annual Compustat data item 9 plus data item 34) divided by total assets; the market capitalization divided by the book value of common equity (annual Compustat data item 60); the value of positive discretionary accruals computed using the Modi? ed Jones Model; the value of negative discretionary accruals computed using the Modi? ed Jones Model; the level of abnormal cash ? ws from operations; the level of abnormal production costs, where production costs are de? ned as the sum of cost of goods sold and the change in inventories; 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); the sum of the standardized three real earnings management proxies, i. e. , R CFO, R PROD and R DISX; the average bonus compensation as a proportion of total compensation received by the CEO and CFO of the ? m; the BONUS variable in ExecuComp in thousands of dollars; 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 ? rm; unexercisable options de? ned 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 ? rm; new option grants made during the current period scaled by total outstanding shares of the ? m; and 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 ? rm. Table 1, Panel B provides summary statistics of the ExecuComp sample. 17 Both samples are signi? cantly larger (at the 0. 001 level) in terms of total assets and market capitalization when compared to the ‘‘average’’ ? rm listed on Compustat. Sample ? rms 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. 2, respectively, for the ExecuComp subsample). The operating cycles of our sample ? rms 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 ? nd a positive and signi? cant correlation (0. 019, Pearson; 0. 053 Spearman; both signi? cant 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 17
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 Real and Accrual-Based Earnings Management 769 ?rm size does not seem to have a signi? cant impact on fundamental measures such as leverage, growth of sales, or market-to-book ratios. As expected, TA (total accruals de? ated 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. 0 (standard deviation of 0. 20) for the full sample and 0. 01 (0. 11) for the ExecuComp subsample. 18 While the average DA is zero, we ? nd that positive discretionary accruals (Positive DA) are, on average, larger in magnitude than negative discretionary accruals (Negative DA). 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 speci? c 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 ? rst, 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. 9,20 Nevertheless, we conduct some additional analyses to get more reassurance on the magnitudes of the discretionary accrual measure. First, we selected a few ? rms that had very high absolute values of DA as percentage of total assets (higher than 15 percent) and examined their ? nancial statements. 21 For the 12 ? rms that meet this criterion, we ? nd 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 signi? ant 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 ? ms that were involved in accounting scandals. Untabulated results indicate that all but one of the ? rms involved 18 19 Note that the mean Jones model residual is zero by construction (residual of a regression), whereas the mean modi? ed Jones model residual is not constrained to be zero by construction. If X follows a normal distribution with mean and variance 2 then the expected absolute value of X is given by: 2 E[ X]] 2/ e 2 2 2 1 20 21 where ( ) 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. 0 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 ? rms. While the average ROA for our sample ? rms 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). We thank one of the referees for this suggestion. 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 ? ve 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, de? ed 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 signi? cant time trends (nonstationarity), rendering traditional summary statistics uninformative. ? The coef? cient for the time trend in Table 2, Panel A, row 1 (b) indicates that the magnitude of discretionary accruals (ABS DA) has been increasing signi? antly (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 signi? cantly in the SCA period (c ? 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 signi? cant at the 1 percent level). Finally, the magnitude of discretionary accruals declined signi? cantly in the post? SOX period (d in Row 1). Moreover, the magnitude (absolute value) of the coef? cient ? or Positive DA (d in Row 2) is approximately three times larger than the coef? cient for ? in Row 3) suggesting that most of that decline in accrual-based manageNegative DA (d ment 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 coef? cients are either not signi? cant or negative). Abnormal production costs and abnormal cash ? ows both increased signi? cantly in the post-SOX period, but were not signi? cantly higher in the SCA period. On the other hand, abnormal discretionary expenses were signi? cantly higher in the SCA period and signi? cantly 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 Dependent Variables ABS DA Positive DA Negative DA R PROD R CFO R DISX RM PROXY Depjq a ? 0. 053*** 0. 028*** 0. 040*** 0. 040*** 0. 022*** 0. 028*** 0. 018*** ? b a b Time c ? c SCA d ? d SOX Adjusted R2 0. 024 0. 029 0. 015 0. 009 0. 007 0. 010 0. 011 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. 08*** 0. 033** 0. 015*** 0. 019** 0. 006*** 0. 032*** 0. 007** 0. 011*** 0. 041*** Panel B: Correlation between the Earnings Management Proxies, 1987–2005 DA DA ABS DA R CFO R PROD R DISX RM PROXY 1 0. 298*** 0. 214*** 0. 024*** 0. 143*** 0. 237*** ABS DA 0. 398*** 1 0. 114*** 0. 314*** 0. 164*** 0. 149*** R CFO 0. 237*** 0. 184*** 1 0. 183*** 0. 214*** 0. 381*** R PROD 0. 029*** 0. 047*** 0. 274*** 1 0. 187*** 0. 427** R DISX 0. 163*** 0. 174*** 0. 142*** 0. 243*** 1 0. 422*** RM PROXY 0. 331*** 0. 153*** 0. 447*** 0. 579*** 0. 487*** 1 *, **, *** Signi? cant at 10 percent, 5 percent, and 1 percent levels, respectively.
Variable De? nitions: DA discretionary accruals computed using the Modi? ed Jones Model; ABS DA the absolute value of discretionary accruals computed using the Modi? ed Jones Model; Positive DA the value of positive discretionary accruals computed using the Modi? ed Jones Model; Negative DA the value of negative discretionary accruals computed using the Modi? ed 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. ecreasing trend over time, but is signi? cantly higher both in the SCA period and in the post-SOX periods. Figure 4 graphically illustrates these trends. A comparison of the post-SOX coef? cients for accrual-based and the real earnings management suggests that there may have been a substitution effect: while accrual-based earnings management decreased, overall ? rms 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 ? nd a signi? ant negative relation between discretionary accruals and the real earnings management metrics, suggesting that ? rms appear to use these two earnings management methods as substitutes. Further, the three real earnings management variables are also negatively correlated, indicating that ? rms 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 0. 16 0. 14 0. 12 ABS_DA 0. 1 ABS_DA 0. 08 0. 06 PRE SOX 0. 04 SCANDAL POST SOX 0. 02 0 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Year This ? gure plots the absolute value of discretionary accruals computed using the Modi? ed Jones Model over the 1987–2005 sample period. of real earnings management increased in the post-SOX period. However, there was signi? cantly 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 ? rms 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 signi? cantly 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 of? cers) cannot be inferred from this analysis, however. Moreover, ? rms 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 Real and Accrual-Based Earnings Management 773 FIGURE 3 Discretionary Accruals over Time, 1987–2005 Panel A: Positive Discretionary Accruals over Time, 1987–2005 0. 16 0. 14 0. 12 Postive DA 0. 1 0. 08 Pos_DA 0. 06 0. 04 PRE SOX 0. 02 SCANDAL POST SOX 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 1999 2000 2001 2002 2003 2004 2005 -0. 2 -0. 04 PRE SOX -0. 06 Neg_DA -0. 08 -0. 1 -0. 12 SCANDAL POST SOX Neg_DA -0. 14 -0. 16 -0. 18 -0. 2 Year This ? gure plots positive and negative discretionary accruals computed using the Modi? ed 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. 15 0. 14 0. 13 0. 12 0. 11 0. 1 0. 09 0. 08 0. 07 0. 06 0. 05 0. 04 0. 03 0. 02 0. 01 0 -0. 01 -0. 02 -0. 03 -0. 04 -0. 05 -0. 06 -0. 07 -0. 08 POST SOX PRE SOX RM_PROXY SCANDAL R_DISX R_CFO R_PROD RM_PROXY 987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 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 BIGj SCA BONUSj 6 0 5 9 11 13 15 17 19 21 1 2 GDPj SOX 7 10 12 3 MKTVALj 8 4
Time BONUSj RM PROXY BONUSj SOX UN OPTIONj 14 16 SCA UN OPTIONj UN OPTIONj GRNT OPTIONj EX OPTIONj EX OPTIONj OWNERj SCA SCA SOX SOX SCA 18 GRNT OPTIONj GRNT OPTIONj SCA SOX (10) OWNERj EX OPTIONj 20 22 SOX OWNERj The Accounting Review, May 2008 Real and Accrual-Based Earnings Management 775 where: DEPj BIG GDP MKTVAL Time SCA SOX BONUS EX OPTION UN OPTION the various earnings management metrics, including discretionary accruals, positive discretionary accruals, and negative discretionary accruals; a dummy variable equal to 1 if the auditor is a Big 5 audit ? rm (or their uccessors); the change in the GDP; market value of equity; the calendar year minus 1987; a dummy variable that is equal to 1 for the years 2000 and 2001, and 0 otherwise (represents the SCA period); a dummy variable that is equal to 1 for the years 2002 onward, and 0 otherwise (represents the post-SOX period); the average bonus compensation as a proportion of total compensation received by the CEO and the CFO of a ? rm; 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 ? rm; unexercisable options de? ed 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 ? rm; new option grants made during the current period scaled by total outstanding shares of the ? rm; and the sum of restricted stock grants in the current period and the aggregate number of shares held by the executives at year-end (excluding stock options) scaled by total outstanding shares of the ? rm. 22 GRNT OPTION OWNER The compensation variables proxy for performance-based compensation and are de? ned in accordance with prior studies (Cheng and War? ld 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 bene? t 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. 3 We include the variable GDP 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 classi? ed as opportunistic earnings management may, in fact, be a consequence of changing economic conditions. Discretionary accruals may also re? ect ? rms’ responses to and 22 23 We also repeat the analyses by measuring the compensation related variables for the top ? ve executives of a ? rm, and the results are qualitatively unchanged. The timing of variable measurement is consistent with Cheng and War? ld (2005, 448). The equity incentive variables are measured during or at the end of ? scal year t, whereas the earnings management variable is measured based on information disclosed after the end of ? scal year t. 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 ? ows, 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 ? rms audited by the large audit ? rms were different from the rest of the sample ? rms over the three subperiods analyzed. Note that we make no claim that differences in the earnings management activities (if any) of these ? rms were the result of the monitoring activities of the audit ? ms, since there could be a self-selection by certain types of ? rms in their selection of big audit ? rms. In addition, to the extent that audit ? rms specialize in speci? c industries and levels of earnings management are likely to vary across industries, the audit ? rm dummies may also control for industry characteristics. Finally, we include the market value of equity as a proxy for ? rm size. Next, we examine the trends in and determinants of real earnings management activities of ? rms by estimating the following regression: DEPj BIGj SCA BONUSj 6 0 5 9 11 13 15 17 19 21 1 2 GDPj SOX 7 10 12 3 MKTVALj 8 4 Time
ABS DAj BONUSj UN OPTIONj 14 16 BONUSj SCA SOX SCA SOX UN OPTIONj UN OPTIONj GRNT OPTIONj EX OPTIONj EX OPTIONj OWNERj SCA SOX SCA 18 GRNT OPTIONj GRNT OPTIONj SCA SOX (11) OWNERj EX OPTIONj 20 22 SOX OWNERj 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 de? ned 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 ? rms 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 signi? cant increase in the latter after SOX. On the other hand, ? rms 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 ? rms. 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 ? rst. Consistent with the preliminary analysis reported in Table 2, we ? nd 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 signi? cant for ABS DA as well as for positive discretionary accruals and positive and signi? ant 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 ? nancial 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 signi? cant 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 ? nd that the percentage of bonus compensation is not correlated with earnings management for the entire period, and this association did not signi? cantly change in the SCA or the post-SOX periods.
Positive discretionary accruals, however, were signi? cant 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 signi? cantly positively associated with discretionary accruals as well as with positive discretionary accruals. For unexercised options this effect increased signi? cantly in the SCA period for positive discretionary accruals and decreased signi? cantly in the post-SOX period. This suggests that option compensation increased managers’ incentives to anipulate earnings upward (e. g. , Fuller and Jensen 2002; Greenp 2002; Coffee 2003), and this effect was signi? cantly 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 signi? cantly 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 signi? antly 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 signi? cantly related to ABS DA during the scandal period and signi? cantly 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 0 5 9 12 14 16 18 20 1 BIGj SCA BONUSj 6 2 GDPj SOX 7 10 3 MKTVALj 8 4 Time BONUSj UN OPTIONj SOX RM PROXY BONUSj 13 SCA SOX 11 UN OPTIONj GRNT OPTIONj GRNT OPTIONj EX OPTIONj OWNERj 21 SCA 15 UN OPTIONj EX OPTIONj EX OPTIONj SCA 22 GRNT OPTIONj 17 19 SCA SOX OWNERj SOX SOX SCA OWNERj Intercept BIG GDP MKTVAL Time SCA SOX RM PROXY BONUS BONUS SCA BONUS SOX UN OPTION UN OPTION SCA UN OPTION SOX GRNT OPTION GRNT OPTION SCA GRNT OPTION SOX EX OPTION EX OPTION SCA EX OPTION SOX OWNER OWNER SCA OWNER SOX Adjusted R2 F-value (Pr F)
ABS DA Coef? cient (t-stat) 0. 118 (18. 87) 0. 027 ( 5. 87) 0. 114 ( 6. 21) 0. 007 ( 17. 36) 0. 003 (8. 21) 0. 002 ( 0. 42) 0. 019 ( 4. 32) 0. 427 ( 5. 81) 0. 002 (0. 48) 0. 005 ( 0. 43) 0. 014 (1. 27) 0. 423 (10. 97) 0. 124 ( 1. 21) 0. 237 ( 3. 94) 0. 119 (2. 93) 0. 087 ( 0. 91) 0. 074 ( 3. 63) 0. 121 (1. 12) 0. 006 ( 0. 74) 0. 009 ( 1. 09) 0. 092 (5. 58) 0. 004 (0. 49) 0. 033 ( 3. 84) 0. 084 169. 16 ( 0. 0001) Positive DA Co