Earnings Management: New Evidence. Based on Deferred Tax Expense

Transcription

1 Earnings Management: New Evidence Based on Deferred Tax Expense John Phillips Universy of Connecticut Morton Pincus * Universy of Iowa Sonja Olhoft Rego Universy of Iowa July 2001 * Corresponding author: Morton Pincus, Tippie College of Business, The Universy of Iowa, 108 PBB, Iowa Cy, Iowa , (319) , morton-pincus@uiowa.edu. The authors appreciate the comments of Dan Collins, Bruce Johnson, Bin Ke, Lil Mills, Mary Margaret Myers, Kathy Petroni, Bill Schwartz, workshop participants at Michigan State Universy, the Universy of Iowa, and the Universy of Waterloo, the programming assistance of Paul Hribar and Hong Xie, and the research assistance of Guojin Gong.

2 Earnings Management: New Evidence Based on Deferred Tax Expense ABSTRACT: We examine the usefulness of deferred tax expense relative to various accrual measures in detecting earnings management to avoid reporting an earnings decrease (Burgstahler and Dichev 1997). The motivation for using deferred tax expense to detect earnings management is that managers typically have more discretion under generally accepted accounting principles than under tax rules. We expect that managers are likely to manage income upwards by exploing the greater discretion they have under GAAP in ways that do not affect current taxable income. If so, their accrual choices will generate book-tax differences that increase deferred tax expense. Our results support the dominance of deferred tax expense over accrual measures in detecting earnings management to avoid reporting an earnings decline. We also consider the earnings management context of avoiding a loss (Burgstahler and Dichev 1997). We do not expect deferred tax expense to be useful here since likely will be immaterial. Rather, accrual metrics should dominate, and the results are consistent wh this expectation. Overall, the results suggest deferred tax expense likely is a more reliable metric for detecting earnings management than accrual-based variables in settings where earnings changes are of interest and in settings where earnings levels, on average, are not likely to be near zero or negative. Keywords: earnings management; discretionary or abnormal accruals; deferred tax expense. Data Availabily: All data used in this research are from publicly available sources. 1

3 Earnings Management: New Evidence Based on Deferred Tax Expense I. INTRODUCTION In this paper we propose and evaluate the use of deferred tax expense as a metric for detecting earnings management. Building on Burgstahler and Dichev s (1997) earnings management evidence and Mills and Newberry s (2001) research concerning earnings management and book-tax differences, we ask whether deferred tax expense better explains earnings management behavior to avoid reporting an earnings decline than do various accrual measures. The abily to detect earnings management reliably is important in assessing the qualy of reported earnings. Prior research has used a number of accounting accrual measures to detect earnings management. These measures include total accruals (Healy 1985), changes in total accruals (DeAngelo 1986), and abnormal accruals derived from the Jones (1991) model, or some variation of (e.g., DeFond and Jiambalvo 1994; Dechow et al. 1995). However, Guay et al. (1996) demonstrate that accruals derived from five alternative models reflect considerable imprecision. Only the Jones and modified-jones models (Dechow et al. 1995) yield abnormal accruals that are significantly different from a random assignment of accruals into normal and abnormal components, and thus have characteristics consistent wh accruals that reflect managerial opportunism. Moreover, Bernard and Skinner (1996) argue that abnormal accruals estimated using Jones-type models reflect measurement error due to the systematic misclassification of normal accruals as abnormal accruals and also treat accruals for nonrecurring events as abnormal accruals. Researchers have attempted to improve the abily of Jones-type models to generate abnormal accruals that reflect earnings management behavior by adding 1

4 addional explanatory variables to the model to control for factors affecting normal accrual behavior (Dechow et al. 1995; Dechow et al. 2001). We take a different tact and argue that measurement error in accrual-based metrics used to detect earnings management can be reduced by focusing on deferred tax expense instead of attempting to decompose total accruals into normal and abnormal components. Current and deferred tax expense make up a firm s total income tax expense. Deferred tax expense can be characterized as the firm s statutory income tax rate multiplied by a subset of total accruals, plus the increase in the deferred tax valuation allowance. 1 The temporary differences between book income (i.e., income reported to shareholders and other external users) and taxable income (i.e., income reported to the tax authories) that give rise to deferred tax expense result primarily from accounting accruals relating to income and expense ems that affect book and taxable income, but in different periods. In contrast, the current portion of income tax expense is generally based on income resulting from cash from operations and the subset of accruals that affect both book and taxable income simultaneously. The tax law generally allows less discretion concerning accruals relative to the discretion managers have for financial reporting purposes (Joos et al. 1999; Mills and Newberry 2001). Accordingly, we expect that managers seeking to manage earnings to avoid reporting an earnings decline are likely to do so by exploing the greater discretion they have under generally accepted accounting principles in ways that do not affect taxable income. If so, their accrual choices will generate book-tax differences that increase deferred tax expense. Thus, deferred tax expense is a potentially useful measure for detecting earnings management activy because the book-tax 1 Generally accepted accounting principles require a deferred tax asset valuation allowance to reduce deferred tax assets to an amount that will likely be realized. Increasing the valuation allowance reduces a firm s net deferred assets and increases s deferred tax expense. 2

5 differences reflects result from accruals over which managers can exercise relatively more discretion under GAAP than for tax purposes. We note, however, that firms can manage book income whout generating temporary book-tax differences. Managers can take actions that affect current operating cash flows and thus affect both book and taxable income. They can also use accruals that affect both book and taxable income, and they can engage in transactions that create permanent book-tax differences. Should managers take such actions or decisions, we would not detect earnings management using deferred tax expense. Thus, is an empirical question whether deferred tax expense is a better metric than various accrual measures for detecting earnings management to avoid reporting an earnings decline. 2 Our results indicate that deferred tax expense (and estimated abnormal deferred tax expense) consistently outperforms total accruals and also abnormal accruals estimated using two versions of the Jones model in detecting earnings management to avoid reporting an earnings decline. In fact, the accrual proxies generally are not significant in explaining such earnings management when the models also include deferred tax expense. We also consider earnings management to avoid reporting a loss (Burgstahler and Dichev 1997). This is a context in which we do not expect deferred tax expense to be useful in detecting earnings management since deferred tax expense is likely an immaterial amount when firms report zero or slightly posive earnings levels. Rather, we expect the accrual measures to be more useful in detecting earnings management in this setting, and we find results consistent wh 2 Jones-type models do not control for nonrecurring ems (e.g., restructurings); accruals for such ems end up in the residual of an estimated Jones-type model and thus are classified as abnormal accruals. While the timing and amount of accruals for nonrecurring ems undoubtedly reflect some degree of managerial discretion, researchers tend to view them as different from abnormal accruals of a more recurring nature. Because accruals for nonrecurring ems typically give rise to temporary book-tax differences, they will also be reflected in deferred tax expense. Hence, both deferred tax expense and accrual metrics reflect accruals for nonrecurring ems. 3

6 that expectation. Thus, deferred tax expense dominates total accruals and estimated Jones-type model abnormal accruals in detecting earnings management to avoid reporting an earnings decline, but is not useful in detecting earnings management to avoid reporting a loss. This suggests that deferred tax expense likely is a more reliable metric than accrual measures for detecting earnings management in settings where earnings changes are of interest and where earnings levels are likely to average above zero. In the next section we provide instutional background about the accounting for book-tax differences and summarize relevant prior research on book-tax conformy and on detecting earnings management. Section III describes the empirical design and data, and section IV presents the results. We conclude the paper in section V. II. BACKGROUND Instutional Details Statement of Financial Accounting Standards No. 109 (SFAS 109) governs the accounting for income taxes and takes a balance-sheet approach. Temporary differences between the financial accounting and tax bases of assets and liabilies are expected to reverse in the future, whereas permanent differences will not. Temporary differences include future taxable amounts, which create deferred tax liabilies, and future deductible amounts, which create deferred tax assets. An increase in deferred tax liabilies is consistent wh a firm currently recognizing revenue and/or deferring expense for book purposes relative to s tax reporting, resulting in a future taxable amount. An increase in deferred tax assets is consistent wh a firm currently recognizing expense and/or deferring revenue for book vis-à-vis tax purposes, producing a future deductible amount. All else equal, firms report higher (lower) pre-tax book income than taxable income when they have increases (decreases) in their net deferred tax 4

7 liabilies (defined as the net change in deferred tax liabilies less the net change in deferred tax assets). Furthermore, unlike prior accounting rules, SFAS 109 gives full recognion to deferred tax assets. However, if a financial statement preparer determines that is more likely than not that a deferred tax asset will not be realized, then the preparer must provide a valuation allowance that offsets the deferred tax asset. Under SFAS 109, the change in a firm s net deferred tax liabily for a period can equal s deferred tax expense for the period, but differences are common. Such differences are most likely to occur when firms engage in mergers and acquisions and divestures, or when they report other comprehensive income ems. These events can affect the balance sheet deferred tax accounts whout affecting deferred tax expense on the income statement. We focus on deferred tax expense because only reflects temporary book-tax differences associated wh the income statement. Book-Tax Conformy Some earnings management research implicly assumes that book income is a valid surrogate for taxable income and finds that in certain circumstances firms manage earnings downward to achieve lower tax costs (see Guenther 1994). This book-tax conformy assumption is valid for earnings management involving real cash flow choices such as delaying or accelerating sales or research and development expendures (Dechow and Skinner 2000) and for some accounting choices such as revenue recognion and the recording of most accounts payable. Based on this assumption, Scholes et al. (1992) show that firms shifted book income to take advantage of decreasing statutory tax rates after the Tax Reform Act of Similarly, Maydew (1997) documents that firms shifted revenue and expense recognion in response to the Tax Reform Act of 1986 to increase the amount of net operating loss carrybacks they could 5

8 apply to pre-1987 high tax rate years, thereby creating larger tax refunds. 3 Based on the evidence from these and other tax-motivated earnings management studies, book-tax conformy implies that for certain firms the expected tax savings from reporting lower taxable income exceeds the non-tax costs of reporting lower income for financial reporting purposes. 4 In general, the tax law provides relatively rigid accounting rules as compared to the flexibily in GAAP. For instance, GAAP allows managers flexibily in estimating the provision for bad debts while tax rules allow a deduction only for accounts receivable actually wrten off. Similarly, there is more discretion in choosing the estimated useful lives over which to record depreciation for book purposes as compared to the limed flexibily for determining depreciable assets cost recovery periods for tax purposes. These differences between GAAP and tax rules are inconsistent wh the book-tax conformy assumption. Accordingly, Joos et al. (1999) argue that book income is not a good surrogate for taxable income and that book-tax differences reflect earnings management activies. They report evidence of a weaker relation between secury prices and earnings for firms wh greater book-tax differences, consistent wh the notion that book-tax conformy is an indicator of the value relevance of earnings. We also hypothesize that book-tax differences result from earnings management activy. However, unlike Joos et al., we focus on settings in which earnings management is likely to have occurred. Mills and Newberry (2001) use a sample of large manufacturing firms included in the confidential Coordinated Examination Program of the IRS during , and present 3 Other studies investigate the hypothesis that firms conform book income to taxable income to lower the costs of dealing wh tax authories. Based upon experimental data, Cloyd et al. (1996) demonstrate that certain firms choose financial accounting methods that conform to tax accounting methods to avoid higher tax costs expected from IRS auds. Using confidential tax return data, Mills (1998) documents that greater book-tax differences are posively associated wh IRS aud adjustments. 4 Non-tax (or financial reporting) costs of reporting lower book income can include costs associated wh executive compensation contracts and debt contracts, and the perceptions of investors in financial markets. 6

9 evidence that firms wh earnings management incentives have greater differences between book and taxable income. In particular, they show that public (versus private) firms, highly leveraged privately-held firms, and financially distressed privately-held firms all have greater book-tax differences. In untabulated results, Mills and Newberry also note that firms reporting slightly posive earnings changes have larger book-tax differences than firms wh slightly negative earnings changes. We extend Mills and Newberry's (2001) research in several ways. First, we expand the investigation of whether firm/years wh slightly posive earnings changes have greater book-tax differences than other firm/years to the population of publicly traded firms for the period Second, we use deferred tax expense, which is publicly available, to proxy for the level of book-tax differences. The Mills and Newberry measure, actual book-tax differences, can only be calculated from confidential tax return data. Finally, and perhaps most important, we compare the relative abily of deferred tax expense and accrual-based metrics used in prior research in detecting earnings management activy. Accrual Studies Another stream of lerature that directly relates to our study concerns accrual-based measures used to detect earnings management. Healy (1985) considers compensation incentives to manage income, and uses total accruals to proxy for discretionary (i.e., abnormal) accruals. DeAngelo (1986) examines going private decisions and uses the change in total accruals as her proxy for income manipulation. DeAngelo s approach implicly assumes that normal accruals are constant over time so that a change in total accruals reflects abnormal accruals. Dechow (1994), however, shows that total accruals are mean-reverting; hence, part of the change in total accruals is expected. Jones (1991), in a study to detect earnings management by firms seeking import relief from an 7

10 international trade commission investigation, relaxes the assumption of constant normal accruals by using an expectations model to capture changes in a firm s economic environment. More specifically, Jones estimates firm-specific time-series regressions of total accruals on the change in a firm s revenues and the level of s property, plant, and equipment. She uses the estimated model coefficients to predict accruals for the next year, and compares predicted to actual accruals to derive a prediction error, which becomes her measure of the firm s abnormal accruals for that year. DeFond and Jiambalvo (1994) note that the need for a sufficiently long time-series of data and the implic assumption of intertemporally constant model coefficients lim the usefulness of the time-series version of the Jones model. Researchers can avoid these limations by estimating a cross-sectional version of the model that uses firms from the same industry and year as the treatment firm. The residuals from the estimated industry/year models are the abnormal accruals used in the analysis of earnings management. DeFond and Jiambalvo report consistent results using both time-series and cross-sectional versions of the Jones model in their investigation of the relation between debt covenant restrictions and accrual choices. Dechow et al. (1995) modify the Jones model to allow for the possibily that managers use discretion to accrue revenues when is questionable whether the creria for revenue recognion have been met. Dechow at al. assess the abily of five accrual models (total accruals, changes in total accruals, an industry model, the Jones model, and the modified Jones model) to detect earnings management. They find the modified Jones model is the most powerful in detecting earnings management in a sample of firms identified by the SEC for overstating earnings. 8

11 Guay et al. (1996) also evaluate the five accruals models considered by Dechow et al. (1995). Their results suggest that only the Jones and modified Jones models produce abnormal accruals that (1) are distinguishable from a random decomposion of earnings into normal and abnormal components, and (2) are consistent wh abnormal accruals resulting from managerial decisions to increase and/or smooth income. Dechow et al. (2001) adapt the modified Jones model to explain accrual behavior better, and thus further reduce the misclassification of normal accruals as abnormal. They drop the implic assumption that all cred sales are discretionary, and also include proxies to capture the predictable component in a firm s accruals and s future sales growth. Dechow et al. (2001) show that the revised (forward-looking) model has greater explanatory power than the modified Jones model, although explains only 20-30% of accrual behavior. 5 We extend the lerature on detecting earnings management by comparing the usefulness of deferred tax expense to various accrual-based metrics used in previous studies. The motivation for this investigation is that deferred tax expense reflects managerial decisions regarding a set of accruals that likely afford managers considerable discretion. Earnings Management to Avoid Earnings Decreases Burgstahler and Dichev (1997) hypothesize that managers have strong incentives to avoid reporting an earnings decrease, and document a significantly higher frequency of small increases in earnings relative to small decreases in cross-sectional distributions of annual scaled earnings changes. They interpret their results as evidence of earnings management (also see DeGeorge et 5 Kang and Sivaramakrishnan (1995) develop a model similar in spir to the Jones model in which their abnormal accrual proxy is the residual from a regression model having non-cash current assets less liabilies as the dependent variable. McNichols and Wilson (1988) adopt a different approach by focusing on a single accrual, the provision for bad debts, and model to derive an abnormal accrual proxy. See McNichols (2000) for an in-depth review and discussion of research design issues in earnings management research. 9

12 al. 1999). They also find similar results for zero and slightly posive earnings levels relative to slightly negative earnings levels. Barth et al. (1999) identify a capal market incentive to avoid earnings declines. They find that firms wh sustained periods of zero or posive earnings changes have higher priceearnings ratios than firms that are unable to sustain earnings growth. Further, the price-earnings multiple increases monotonically for each consecutive year a firm reports nondecreasing earnings but disappears after just two years of earnings declines. 6 Figure 1 shows our replication of Burgstahler and Dichev s (1997) results regarding earnings changes. The unusually high number of observations in the zero and slightly posive scaled earnings change intervals and the unusually low frequency of observations in the slightly negative earnings change intervals are consistent wh Burgstahler and Dichev s findings of a higher frequency of small increases in annual earnings relative to small decreases. [Insert Figure 1 here] Note that Burgstahler and Dichev (1997) document earnings management whout estimating abnormal accruals. We use their results as a benchmark to assess the relative abily of deferred tax expense and accrual metrics to detect earnings management to avoid reporting an earnings decline (or a loss). 6 This capal market-based incentive to avoid reporting an earnings decline has been supported in other studies (Bartov et al. 2000; Beatty et al. 2000; Myers and Skinner 2001), and consistent wh Barth et al. (1999) the incentive is increasing in a firm s growth opportunies (Skinner and Sloan 2000). Myers and Skinner (2001) link the duration of consecutive earnings changes to management stock ownership and unexercised stock options, and Ke (2001) finds that growth opportunies and CEO equy and accounting compensation-based incentives are posively associated wh the duration of a firm s consecutive posive earnings changes. 10

13 III. EMPIRICAL DESIGN Research Design Our empirical analysis compares the abily of deferred tax expense and various accrual measures to detect earnings management. We consider two suations in which earnings management likely is present: firm/years wh zero or slightly posive reported earnings changes and firm/years wh zero or slightly posive reported earnings levels. We focus inially on earnings changes and estimate the following cross-sectional prob model: EM = α + β1 DTE + β 2 AC + β3 CFO + ε (1) where EM = 1 if the change in firm i s net income (annual Compustat data em #172) from year t-1 to t divided by the market value of equy at the end of year t-2 (annual Compustat data ems #25 #199) is 0 and < 0.005, and 0 otherwise; DTE = firm i s deferred tax expense (annual Compustat data em #50) in year t, scaled by total assets at the end of year t-1; AC = a measure of firm i s accruals in year t; CFO = the change in firm i s cash flows from operations (annual Compustat data ems #308 - #124) from year t-1 to t, scaled by total assets at the end of year t-1; and ε = error term. The dependent variable in equation (1), EM, is an indicator variable that, following Burgstahler and Dichev (1997), equals 1 (0) if firm i reports (does not report) a scaled earnings change greater than or equal to zero and less than in year t. DTE is the amount of deferred tax expense included in firm i s total income tax expense reported in s year t income statement, scaled by s beginning-of-year total assets. 7 The basic idea of using deferred tax 7 Numerous missing observations preclude using annual Compustat data em #126, which is deferred taxes on the funds statement. 11

14 expense to detect earnings management is that managers typically have more discretion under GAAP than under tax rules. If managers are more likely to manage earnings in areas where they have more discretion, then earnings management behavior will give rise to temporary book-tax differences that will be reflected in deferred tax expense. Hence, we predict the coefficient on DTE in equation (1) will be posive, indicating that the probabily of earnings management to avoid reporting an earnings decline increases wh deferred tax expense. AC represents one of several accrual variables (discussed below) that we use to detect earnings management, and we expect to have a posive coefficient in the presence of earnings management to avoid an earnings decline. Including both DTE and AC in the model allows us to determine if deferred tax expense is a more significant explanatory factor than a given accrual measure in detecting earnings management. We also include CFO to control for the effect that changes in cash flows from operations have on a firm s status as an earnings management firm. We expect that larger increases in operating cash flows reduce the need to manage earnings to achieve a zero or posive earnings change. Figure 2 illustrates a firm wh a negative change in pre-managed earnings in excess of Y, a threshold that varies wh managers abily and incentives to manage earnings to avoid reporting an earnings decline. This firm manages s earnings upward in such a way as to increase s temporary book-tax differences. We assume that a firm having a pre-managed earnings change of less than Y cannot manage s earnings upward to avoid an earnings decline. Figure 2 also illustrates that a firm wh a zero or slightly posive pre-managed scaled earnings change that does not engage in earnings management activies would nevertheless be classified as an earnings management firm. Our inclusion of CFO in equation (1) should control for 12

15 cases where increases in operating cash flows cause this firm/year to be classified as EM = 1. If other, unspecified factors explain a classification as an earnings management firm/year, then the resulting misclassification would affect the abily of both deferred tax expense and the various accrual variables to detect earnings management. Finally, Figure 2 also shows that some firms may manage earnings downward; e.g., some firms may lower earnings to smooth reported income and some firms, particularly those wh a negative pre-managed earnings change lower than Y, may engage in big bath behavior. If a significant number of firms manage earnings downward in our setting, then neher deferred tax expense nor accrual-based measures will detect income increasing earnings management to avoid reporting an earnings decline. Accrual Models [Insert Figure 2 here] We use total accruals (Healy 1985), abnormal accruals derived from the modified Jones model (Dechow et al. 1995), and abnormal accruals derived from the forward-looking Jones model (Dechow et al. 2001) as proxies for accruals that reflect earnings management. We compute total accruals, which we scale by beginning-of-year total assets (annual Compustat data em #6), as follows: TAcc = EBEI CFO (2) where TAcc = firm i s total accruals in year t; EBEI = firm i s income before extraordinary ems (annual Compustat data em #123) in year t; and CFO = firm i s cash flows from operations in year t. 13

16 We estimate two different cross-sectional models to derive abnormal accruals. The first model is the modified Jones model. Following Dechow et al. (2001), we estimate this model as follows: TAcc = + β1 Sales Rec ) + β 2 α ( PPE + ξ (3) where Sales = the change in firm i s sales (annual Compustat data em #12) from year t-1 to t; Rec = the change in firm i s accounts receivable from operating activies from year t-1 to t (annual Compustat data em #302); PPE = firm i s year t gross property, plant, and equipment (annual Compustat data em #7); and ξ = error term. The subtraction of Rec modifies the Jones (1991) model so that only cred sales are assumed to be discretionary, rather than all sales. We scale all variables by beginning-of-year total assets, and estimate equation (3) separately for each two-dig SIC group/year. The error term represents abnormal accruals computed using this model and is hereinafter referred to as AbAccMJ. et al. (2001): The second normal accruals model we estimate is the forward-looking model of Dechow TAcc = + β1 Sales (1 k) Re c ) + β 2PPE + β3tacc 1 + β 4GR _ Salest+ 1 α ( + ξ (4) where k = the slope coefficient from a regression of Rec on Sales ; TAcc 1 = firm i s total accruals from the prior year, scaled by year t-2 total assets ; GR _ Sales +1 = the change in firm i s sales from year t to t+1, scaled by year t sales; and ξ = error term. 14

17 The forward-looking model includes three adjustments to the modified Jones model. The first adjustment is to treat part of the increase in cred sales as expected, as opposed to assuming that all cred sales are discretionary. This is accomplished by estimating the parameter k, which is allowed to range from 0 to 1 and reflects the expected change in accounts receivables from a change in sales. Hence, the expected portion of the increase in receivables that results from an increase in sales is treated as a normal accrual so that the change in sales in equation (4) is reduced by less than 100% of the increase in receivables. The second adjustment to the modified Jones model is for lagged total accruals. If we assume that a portion of total accruals is predictable, then we can capture the predictable component by including last year s accruals in the model. The last adjustment is for future sales growth. The modified Jones model treats increases in inventory made in anticipation of higher sales as an abnormal accrual reflecting earnings manipulation rather than as a rational increase in inventory. Including a proxy for future sales growth corrects for such misclassifications. Finally, the error term represents abnormal accruals computed using this model and is denoted AbAccFL. Collins and Hribar (2000) demonstrate that the lack of financial statement articulation caused primarily by mergers, acquisions, and divestures induces measurement error in Jonestype model abnormal accruals. They document a reduction in measurement error by estimating the models using data from the cash flow statement rather than data from successive balance sheets. Accordingly, we adopt their approach and use cash flow data where appropriate. Sample SFAS 109 became effective in 1993 and substantially altered GAAP for income tax reporting. To assure consistent financial reporting, we only include firm/years for the period 15

18 We require sample firms to be incorporated in the U.S. because foreign firms face different financial accounting and tax rules and incentives than U.S. firms do. We exclude utilies (SIC codes ) because as regulated businesses they may not have the same incentives to manage earnings as other competive businesses. For the same reason we exclude financial instutions (SIC codes ), for which is also the case that Compustat does not report their deferred tax accounts. We also exclude mutual funds (SIC code 6726), trusts (SIC code 6792), REITs (SIC code 6798), limed partnerships (SIC code 6799), and other flowthrough enties (SIC code 6795) because these firms do not account for income tax expense. In addion, firm/years must have non-missing data for the variables needed in the analysis, and we delete firm/years having deferred tax expense or accrual variable values below the 1 st percentile and above the 99 th percentile to control for extreme observations. Our selection procedures generate samples that range between 11,866 and 16,737 firm/year observations, depending primarily on the accrual model being considered. Of these observations 681 firm/years (262 firms) have a scaled earnings change that is zero or slightly posive (i.e., greater than or equal to zero and less than 0.005), and these comprise our earnings management (i.e., EM = 1) sample. The remaining firm/years form the control (i.e., EM = 0) sample. IV. RESULTS Descriptive Statistics and Univariate Analysis Panel A of Table 1 presents summary statistics for firm/years wh zero or slightly posive earnings changes and for all other firm/years. For the EM = 1 sample, the mean is the first year we can compute change variables. We lose 1998 observations when estimating the forwardlooking model since we need to use one-year ahead sales. 16

19 deferred tax expense, DTE, is , or 0.22% of beginning-of-year total assets (median = ), wh values ranging from -6.77% to 6.5% of total assets. Not surprisingly, total accruals are substantially larger in magnude; mean TAcc is or -4.96% of beginning-of-year total assets (median = ), and the range is from % to 54.01% of total assets. In the control sample, mean DTE is (median = ) and mean TAcc is (median = ). In Panel B, we statistically compare the two samples (p-values are two-tailed). The results indicate that both mean and median DTE are significantly larger in the sample of firm/years wh zero or slightly posive earnings changes than in the control sample. Similarly, the mean TAcc is reliably larger (i.e., less negative) for the EM = 1 sample, and the median is marginally statistically larger. We would expect that if firms manage earnings upward to avoid reporting an earnings decline, then earnings management metrics should reflect that activy. In particular, there should be greater deferred tax expense and greater accruals in those firm/years than in other (i.e., control) firm/years. [Insert Table 1 here] Somewhat surprisingly, we do not observe consistently larger abnormal accruals for the EM = 1 firm/years. Panel B reveals that estimated abnormal accruals from the modified Jones model, AbAccMJ, do not differ on average between the earnings management and control samples, although mean (but not median) abnormal accruals from the forward-looking Jones model, AbAccFL, are larger for the EM = 1 sample. We report correlation results in Table 2. TAcc is posively correlated wh abnormal accruals from both the modified Jones model and the forward-looking model, and all three accrual metrics are generally negatively correlated wh 17 CFO. As expected, EM and DTE are posively correlated, while correlations between EM and the various accrual metrics are not

20 always significant. Consistent wh Dechow (1994), we find (in untabulated results) that the Pearson correlation between the change in net income and the change in operating cash flows is , which is significant (Spearman correlation = ). We also find that the correlation between total accruals and cash flows from operations is significantly negative (Pearson = and Spearman = ), which is consistent wh Sloan (1996). [Insert Table 2 here] Primary Results Our primary results are a set of horse races that compare the importance of deferred tax expense relative to each of the three alternative accrual measures in detecting earnings management to avoid reporting an earnings decline, after controlling for changes in operating cash flows. Table 3, Panel A, displays the results of estimating equation (1) using DTE and, alternatively, TAcc, AbAccMJ, and AbAccFL. Indicated p-values are based on one-tailed tests. [Insert Table 3 here] In the first set of results, shown in the left-hand pair of columns in Table 3, the coefficient on DTE is , which is significant (p < ). The coefficient on TAcc is 0.123, which is significant only at the 0.10 level. Hence, deferred tax expense is more significant than total accruals in detecting earnings management to avoid an earnings decline. The middle two columns present the second set of results, which compare DTE and AbAccMJ. The coefficient on deferred tax expense is again posive (5.5186) and significant (p < ), while the coefficient on abnormal accruals derived from the modified Jones model is insignificant from zero (p = 0.94). The final set of columns compare DTE and abnormal accruals from the forward-looking model. The coefficient on DTE is posive (4.6351) and significant at the level, and the 18

21 coefficient on AbAccFL is also posive and significant (p = ). Abnormal accruals derived from the forward-looking model are thus successful in detecting the avoidance of an earnings decline, but deferred tax expense is still the more significant explanatory variable. Hence, the results indicate that deferred income tax expense reliably detects earnings management to avoid reporting an earnings decline, while each of the alternative accrual measures is less significant than deferred tax expense, or not significant at all. We provide addional analysis by exploring a decomposion of deferred tax expense into normal and abnormal components. Because deferred tax expense is related to accruals by virtue of the fact that is derived from the subset of accruals that give rise to temporary book-tax differences, we assume that the same models that are used to estimate normal accruals can be used to estimate the normal component of deferred tax expense. Hence, we estimate abnormal deferred tax expense using the modified Jones model (AbDTE1) and the forward-looking model (AbDTE2). 9 The univariate results in Panel B of Table 1 show that both AbDTE1 and AbDTE2 are significantly larger for firm/years wh zero or slightly posive earnings changes than for all other firm/years. The results in Table 2 document posive correlations for both of the abnormal deferred tax expense metrics wh DTE as well as wh EM. Panel B of Table 3 reports the prob results in which we compare (1) AbDTE1 wh abnormal accruals from the modified Jones model, and (2) AbDTE2 wh abnormal accruals from the forward-looking model. In the first comparison, the coefficient on AbDTE1 is posive (4.459) and significant (p = ) while the coefficient on AbAccMJ is insignificant (p = 0.83). In the second comparison, the coefficient on AbDTE2 is and significant (p = 0.006). The 9 In the forward-looking model, we substute lagged DTE for lagged total accruals. 19

22 coefficient on AbAccFL is and significant (p = ) but not as significant as the coefficient on abnormal DTE derived from the forward-looking model. Hence, the results support the conclusion that deferred tax expense and the estimates of abnormal DTE are more reliable indicators of earnings management to avoid reporting an earnings decline than are the various accrual measures that we consider. Note that we obtain stronger results when using DTE than when using estimates of abnormal DTE. We have not attempted to model normal deferred tax expense; thus, may be possible to identify better DTE expectations models than those based on Jones-type models. However, recall that deferred tax expense excludes the effects of accruals over which managers likely have ltle discretion. Hence, DTE may primarily reflect managerial behavior to avoid reporting an earnings decline, which would suggest that decomposing DTE into normal and abnormal components might induce greater measurement error. We leave this issue for future research. Results for Earnings Levels As previously discussed, Burgstahler and Dichev (1997) also provide evidence of earnings management to avoid reporting a loss. Hence, we investigate the performance of the deferred tax expense and accrual metrics in detecting earnings management in this context. It seems reasonable to conjecture that firms wh zero or slightly posive earnings levels in a given year will be more likely to have had poor operating performance in recent prior years and thus will have greater net operating loss (NOL) carryforwards than other firms. If so, that can weaken the power of deferred tax expense to detect accrual manipulation to avoid reporting a 20

23 loss. 10 It is also possible that managers seeking to avoid reporting a loss will be relatively unconcerned wh the tax costs of earnings management decisions since their firms have ltle, if any, income tax expense. If so, they will be more likely to manage earnings upwards in a tax inefficient manner and thereby generate a smaller deferred tax expense. Both of these conjectures suggest that deferred tax expense will not be useful in detecting earnings management to avoid reporting a loss, in contrast to s importance in the context of earnings management to avoid an earnings decline. Hence, we predict a zero coefficient on the deferred tax variables, and expect that the accrual metrics will be more useful in detecting earnings management in this setting. To examine the usefulness of deferred tax expense and accrual measures in the context of avoiding a loss, we use the same total sample of observations that we used for the earnings change analysis. Following Burgstahler and Dichev (1997), we define firm/years wh scaled earnings levels of greater than or equal to zero and less than 0.01 as EM = 1 firm/years; i.e., firm/years reflecting earnings management behavior to avoid reporting a loss. Descriptive statistics for our analysis of earnings levels appear in Table 4. Mean DTE for the EM = 1 firm/years is The negative mean indicates an average deferred tax benef, which implies that in the year a firm reports zero or slightly posive book income, reports higher taxable income. Recall that for firms reporting zero or slightly posive earnings 10 To illustrate this, suppose at t-1 a firm has a $1,000 NOL. At a 35% tax rate, the NOL will generate a $350 gross deferred tax asset (DTA). If the deferred tax asset valuation allowance is (say) $200, then the net DTA is $150. Assuming the firm has other deferred tax liabilies (DTLs) of $200, s net DTL at t-1 will be $50. In year t, the firm has taxable income of -$100 and manages book earnings up to zero by using accruals that generate temporary book-tax differences. Its NOL grows to $1,100, and the associated gross DTA becomes $385. Assuming the firm does not adjust the valuation allowance, s net DTA will be $185. Due to the earnings management, other DTLs increase by $35 ($100 x 35%) to $235, and the firm s net DTL at t will remain at $50. Thus, in this stylized example, net DTL stays the same and there is no deferred tax expense despe the fact that the managers increase book income relative to taxable income by $100. Thus, DTE does not pick up this earnings management activy. 21

24 changes, mean DTE was posive, implying lower income for tax purposes. Panel B of Table 4 reveals that the mean (and median) DTE for the sample of zero and slightly posive earnings levels is significantly smaller than that for all other firm/years, which again is oppose of the result in Panel B of Table 1 where the focus was on avoiding an earnings decline. Similarly, the mean and median for both abnormal DTE measures are significantly smaller than their control sample counterparts. On the other hand, mean and median values of the three alternative accrual metrics are almost always significantly larger for the sample of firm/years wh zero or slightly posive earnings than for all other firm/years. The corresponding results for the earnings change case revealed the accrual metrics were not consistently larger (Panel B of Table 1). Table 5 presents the prob results. The coefficient on each of the deferred tax expense measures is significantly negative. We expected that deferred tax expense (and abnormal DTE ) would not be useful in detecting earnings management to avoid reporting a loss, and predicted a zero coefficient. Hence, while we do not observe posive coefficients on the deferred tax expense variables, the negative coefficients are unexpected. On the other hand, the coefficients on the various accrual measures are posive and significant, as expected. Hence, in the context of detecting earnings management to avoid reporting a loss, accrual measures are more useful. This result is in contrast to the dominance of the deferred tax variables in detecting earnings management to avoid a negative earnings change, as reported in Table 3. As noted, the negative coefficients on the deferred tax variables are surprising, and may reflect the negative mean DTE (i.e., average deferred tax benef) for the firm/years wh zero or slightly posive earnings levels. We had expected that deferred tax expense would, on average, be close to zero since there is ltle or no tax expense and these firm/years could have non-trivial 22

25 NOL carryforwards. However, untabulated results reveal that the mean NOL carryforward for the EM = 1 firm/years is in fact significantly smaller than that for all other firm/years. One possible explanation for why firms/years wh zero or slightly posive earnings levels have negative average deferred tax expense is that these firm/years have significant accruals for restructurings and asset impairments. Accruals for such special ems give rise to deferred tax assets and thus reduce deferred tax expense. The occurrence of restructuring accruals in a year when a firm reports essentially zero income therefore could result in a net deferred tax benef. In untabulated results we find that the mean and median special ems (annual Compustat data em #17) for the EM = 1 firm/years are more negative than for all other firm/years, and the median is significantly more negative. However, when we re-estimate the prob regressions in Table 5 after purging deferred tax expense (and abnormal DTE ) of the tax effects of special ems and after purging the accrual metrics of the after-tax effects of special ems the basic results remain although the deferred tax expense metrics tend to be less significantly negative. V. CONCLUSIONS AND LIMITATIONS We investigate the relation between earnings management activies and deferred tax expense. Because managers generally have more discretion under generally accepted accounting principles than under tax rules, we expect managers will be more likely to manage earnings by exploing their discretion under GAAP, which in turn gives rise to temporary book-tax differences that are reflected in deferred tax expense. We rely on Burgstahler and Dichev s (1997) evidence that firm/years wh zero or slightly posive scaled earnings changes likely reflect earnings management activies to avoid reporting an earnings decline, and we compare 23

26 the abily of deferred tax expense and several accruals metrics to detect such earnings management. Our primary results support the dominance of deferred tax expense over alternative accrual measures to detect earnings management to avoid reporting a decrease in earnings. These results also add to recent findings that indicate a relation between book and tax reporting and firms incentives to engage in earnings management activies (Mills and Newberry 2001). We also explore the usefulness of deferred tax expense versus accrual measures in detecting earnings management to avoid reporting a loss (Burgstahler and Dichev 1997). Unlike the case of avoiding an earnings decline, we do not expect deferred tax expense to be useful in this context. Rather, we expect that accrual metrics will dominate in this setting, and the results are consistent wh this expectation. Overall, the results suggest that deferred tax expense likely is a more reliable metric for detecting earnings management than accrual-based variables in settings where earnings changes are of interest and in settings where earnings levels, on average, are not likely to be near zero or negative. 24

27 REFERENCES Barth, M., J. Elliott, and M. Finn Market rewards associated wh patterns of increasing earnings. Journal of Accounting Research 37 (2), Bartov, E., D. Givoly, and C. Hayn The rewards to meeting or beating earnings expectations. Working paper, New York Universy. Beatty, A., B. Ke, and K. Petroni Earnings management to avoid earnings declines across publicly and privately-held banks. Working paper, Michigan State Universy Bernard, V. and D. Skinner What motivates managers choice of discretionary accruals? Journal of Accounting and Economics 22 (1-3): Burgstahler, D. and I. Dichev Earnings management to avoid earnings decreases and losses. Journal of Accounting and Economics 24 (1): Cloyd, B., J. Pratt, and T. Stock The use of financial accounting choice to support aggressive tax posions: Public and private firms. Journal of Accounting Research 34 (1): Collins, D. and P. Hribar Errors in estimating accruals: Implications for empirical research. Working paper, Universy of Iowa. DeAngelo, L Management buyouts of public stockholders. The Accounting Review 61 (3): Dechow, P Accounting earnings and cash flows as measures of firm performance: The role of accounting accruals. Journal of Accounting and Economics 18 (1): Dechow, P., R. Sloan, and A. Sweeney Detecting earnings management. The Accounting Review 70 (2): Dechow, P. and D. Skinner Earnings management: Reconciling the views of accounting academics, practioners, and regulators. Working paper, Universy of Michigan. Dechow, P., S. Richardson, and A.I. Tuna Earnings management and costs to investors from firms meeting or slightly exceeding benchmarks. Working paper, Universy of Michigan. DeFond, M. and J. Jiambalvo Debt covenant violations and manipulation of accruals. Journal of Accounting and Economics 17 (1-2): DeGeorge, F., J. Patel, and R. Zeckhauser Earnings management to exceed thresholds. Journal of Business 72 (1):