Did FIN 48 Limit the Use of Tax Reserves as a Tool for Earnings Management? Richard Cazier Texas Christian University. Sonja Rego * Indiana University

Transcription

1 Did FIN 48 Limit the Use of Tax Reserves as a Tool for Earnings Management? Richard Cazier Texas Christian University Sonja Rego * Indiana University Xiaoli Tian University of Iowa Ryan Wilson University of Iowa August 8, 2011 * Corresponding author. Kelley School of Business, Indiana University, 1309 E. 10 th St., Bloomington, IN Phone: (812) sorego@indiana.edu. We are grateful for helpful comments received from Cristi Gleason, Max Hewitt, Paul Hribar, Bruce Johnson, Inder Kurana, Phil Quinn, Teri Yohn and workshop participants at the University of Iowa, the University of Missouri, and Indiana University. We also appreciate the summer research funding provided by the University of Iowa.

2 Did FIN 48 Limit the Use of Tax Reserves as a Tool for Earnings Management? Abstract We utilize new income tax reserve disclosures required under FIN 48 to examine whether managers use discretion over this accrual to manage earnings to meet the consensus analyst forecast. We find that firms with pre-managed earnings (i.e., earnings before the change in the tax reserve) that are below the consensus analyst forecast are far more likely to reduce their tax reserves and thus report higher net income. In fact, we find that 37 percent of firm-years with pre-managed earnings below the consensus forecast meet the forecast when the change in the tax reserve is included in earnings. In contrast, only 9.8 percent of firm-years with premanaged earnings above the consensus forecast increased their tax reserves to the extent that it caused them to miss the consensus forecast. This asymmetric result is consistent with managers using their discretion over tax reserves to meet consensus analyst forecasts. Using a proxy for changes in tax reserves developed by Blouin and Tuna (2007), we also document a decline in the use of tax reserves to meet the consensus analyst forecast following the adoption of FIN 48. Nonetheless, our results using both estimated and actual changes in tax reserves clearly suggest that managers continue to use their discretion over this account to meet the consensus analyst forecast, although at a lower rate of recurrence than during the pre-fin 48 time period. Keywords Tax reserves; tax cushion; FIN 48; earnings management; analyst forecasts JEL Classification H25, M41, M48

3 Did FIN 48 Limit the Use of Tax Reserves as a Tool for Earnings Management? 1 Introduction Accounting for income taxes involves significant complexity and managerial discretion. The combination of these factors makes it difficult for auditors and financial statement users to evaluate tax-related information disclosed in financial statements. Prior research provides evidence that managers take advantage of information asymmetry in the tax accounts to manage earnings to meet key earnings targets (e.g. Dhaliwal, Gleason, and Mills 2004; Krull 2004; Frank and Rego 2006). However, the landscape related to accounting for income taxes has changed significantly in recent years as a result of the Sarbanes Oxley Act (SOX) and Financial Interpretation No. 48 (FIN 48). New rules have resulted in both increased auditor scrutiny of tax accounts and increased disclosure of tax-related information. Consequently, it is no longer clear the tax accounts provide managers with an attractive vehicle for earnings management. We take advantage of the new disclosures of the liabilities for unrecognized tax benefits required under FIN 48, which we refer to as tax contingencies or tax reserves, 1 to examine whether managers continue to use discretion over this account to meet earnings targets in the new regulatory environment. According to the Financial Accounting Standards Board (FASB), FIN 48 was issued to compensate for SFAS No. 109 s lack of guidance regarding the accounting for uncertainty in income taxes. This lack of guidance led to diverse accounting practices and inconsistency in the criteria used to recognize, derecognize, and measure benefits related to income taxes (FIN 48). In addition, FASB Chairman Robert Herz stated that FIN 48 was issued because SEC staff had concerns that firms were able to use their tax reserves to manage earnings. There was a lot 1 Prior to FIN 48, the liability for unrecognized tax benefits was often referred to as the tax cushion or the tax contingency. 1

4 of concern that this area was extremely opaque and a little mystical (Leone and Shaw, CFO.com, January 17, 2007). Thus, the FASB issued FIN 48 to increase the relevance and comparability of tax footnote disclosures and to provide more information on tax contingencies (FIN 48). Prior studies have investigated the income tax provision as a potential earnings management tool. Dhaliwal et al. (2004) argue that because income tax expense is one of the last accounts closed prior to an earnings announcement, it provides managers a final opportunity to manage earnings. Consistent with this prediction, they find evidence managers use discretion over the tax provision to meet or beat analysts forecasts. In preparing the tax provision there are three specific tax accruals that allow managers sufficient discretion to manage earnings: tax reserves, the valuation allowance, and the amount of foreign earnings designated as permanently reinvested abroad. Our analyses focus on managers use of discretion over tax reserves to meet or beat analysts forecasts because we believe this specific tax accrual is the most likely to be used to manage earnings. It is also the account directly affected by FIN 48. Tax reserves represent management s assessment of potential future tax payments that could arise when tax authorities examine particular tax positions. Due to the large number of transactions that must be evaluated, combined with the complexity and uncertainty associated with many tax positions, tax reserves involve greater managerial discretion than the valuation allowance or the designation of foreign earnings as permanently reinvested. Gleason and Mills (2002) document that prior to FIN 48 firms rarely disclosed information regarding their tax contingencies. 2 However, for fiscal years starting after December 15, 2006, FIN 48 requires firms to disclose the beginning and ending balances of their tax contingencies, along with a 2 Gleason and Mills (2002) is one of the first studies to examine tax contingencies and they document significant diversity in practice among firms in accounting for tax contingencies during the time these accounts were governed under SFAS No. 5. They do not examine whether the tax contingency is used to manage earnings. 2

5 schedule of changes in the account. Tax footnote disclosures in the post-fin 48 time period reveal that many firms maintain large tax reserves, and thus, tax reserves have the potential to be effective earnings management tools. For example, Merck & Co., General Electric, and AT&T disclosed initial tax contingencies under FIN 48 of $7.4, $6.8, and $6.3 billion, respectively. Consistent with studies of earnings management in general (e.g., DeGeorge, Patel and Zeckhauser 1999; Brown and Caylor 2005; Burgstahler and Eames 2006) and tax expense management in particular (e.g., Dhaliwal et al. 2004; Krull 2004; Frank and Rego 2006), our empirical tests focus on firms incentives to manage earnings around the consensus analyst forecast. We construct variables intended to capture whether a firm has incentives to manage earnings up to meet or beat analysts forecasts (i.e., when pre-managed earnings are below the consensus forecast) or to smooth earnings down to create cookie jar reserves for the future (i.e., when pre-managed earnings are sufficiently above the consensus forecast). 3 We examine both the total change in tax reserves and the discretionary change in tax reserves as instruments for earnings management. We also investigate whether the implementation of FIN 48 reduced tax reserve management in the post-fin 48 time period relative to the pre-fin 48 time period. We perform a battery of univariate and multivariate tests, which consistently indicate that firms have continued to use their tax reserves to manage earnings up toward the consensus analyst forecast and to build cookie jar reserves in the post-fin 48 time period. We first construct figures that show the distribution of analyst forecast errors based on both pre-managed earnings (i.e., earnings before the change in tax reserves) and actual earnings. The figure based on pre-managed analyst forecast errors demonstrates firms that would meet the consensus analyst forecast absent a change in tax reserves are far more likely to increase their 3 For purposes of this study we define pre-managed earnings as earnings before the change in tax reserves, or alternatively earnings before the change in the discretionary component of tax reserves. We discuss our calculation of pre-managed earnings in more detail in Section 3. 3

6 tax reserves, while firms that would miss the consensus forecast absent a change in tax reserves are far more likely to decrease their tax reserves. In stark contrast to the pre-managed analyst forecast error figure, the second figure reveals a tight distribution of actual forecast errors that is centered on a one-cent earnings surprise. Taken together, these univariate results are consistent with firms using tax reserves to manage earnings toward the consensus analyst forecast. We then partition our sample based on firms with pre-managed earnings that are far below, just below, just above, and far above the consensus forecast. These partitions reveal asymmetric proportions of tax reserve increases and decreases, depending on whether premanaged earnings are far below, just below, just above, or far above the consensus forecast. For instance, 76.4 percent of firms categorized as having pre-managed earnings far above the consensus forecast increase their tax reserves, compared to the 21.7 percent of firms with premanaged earnings far below the consensus forecast that increase their tax reserves. We also demonstrate that 60.3 (37.4) percent of firms categorized as having pre-managed earnings just (far) below the consensus forecast are able to meet or beat the forecast by decreasing their tax reserves. Results from a battery of multivariate tests uniformly support the conclusions from our univariate analyses. Altogether, our empirical results consistently indicate that FIN 48 did not stop last chance earnings management. In a concurrent study, Gupta, Laux, and Lynch (2011) also examine whether managers use discretion in accounting for tax contingencies to meet quarterly consensus forecasts. They hand-collect tax footnote data for a random sample of 100 Fortune 500 firms during the pre-fin 48 time period ( ). They then examine the tax footnote disclosures for the same 100 firms in the post-fin 48 time period (2007 and 2008). They find evidence that these 100 firms used their tax reserves to meet or beat quarterly consensus forecasts in the pre-fin 48 time 4

7 period, but not in the post-fin 48 time period. Our study differs from Gupta et al. (2011) in several important dimensions. We hand-collect tax reserve data for S&P 500 and S&P 400 firms for fiscal years , and thus we utilize a larger sample of firms that are not subject to self-selection bias. 4 We also examine annual rather than quarterly earnings targets; we consider both upward and downward earnings management (i.e., income smoothing); we classify tax reserve components as discretionary and non-discretionary; and we implement a battery of empirical tests. As a result, our analyses provide a more complete understanding of the extent to which managers use tax reserves in the pre- and post-fin 48 time periods to manage annual earnings. Our study makes several important contributions to the accounting literature. It directly responds to comments by Graham, Ready and Shackelford (2010, p.43) that we know little about the use of the tax contingency to manage earnings on an ongoing basis and, in particular, whether this behavior occurs since the effective date of FIN 48. It thus informs standard setters and regulators regarding the effectiveness (or lack thereof) of FIN 48. In addition, while prior research provides indirect evidence (Dhaliwal et al. 2004) that firms use their tax contingencies to manage earnings up to meet or beat the consensus forecast, we provide robust evidence that managers use tax reserves to manage earnings up and down to meet or beat the consensus analyst forecast in the post-fin 48 time period. 5 The remainder of the study proceeds as follows. Section 4 The sample utilized by Gupta et al. (2011) is subject to self-selection bias because in practice, firms voluntarily chose whether to disclose tax reserve changes and the content of those disclosures during the pre-fin 48 time period. Gupta et al. (2011) report that only 10.2 percent of their sample firm-quarters in the pre-fin 48 time period made disclosures about tax cushion reversals. Because Gupta et al. utilize quarterly tax reserve changes, they continue to rely on voluntary disclosures from either the tax footnote or the management, discussion, and analysis section in the post-fin 48 era. Thus, their sample from the post-fin 48 time period is also subject to self-selection bias. Furthermore, the change in disclosure regime likely changed managers incentives to voluntarily disclose quarterly tax reserves in the post-fin 48 era. As a result, the self-selection bias could be affected by different sets of managerial incentives in the pre- versus post-fin 48 era. 5 Dhaliwal et al. (2004) examine changes in third- to fourth-quarter effective tax rates and whether firms meet or beat the consensus analyst forecast to make inferences about whether firms use the tax provision to manage 5

8 2 provides institutional details and develops hypotheses. Section 3 explains the research design and sample selection, while Section 4 discusses the empirical results. Section 5 concludes. 2 Background and Hypothesis Development Prior to the issuance of FIN 48, firms generally followed the liability model of accounting for uncertainty in income taxes, consistent with SFAS No. 5, Accounting for Contingencies. Under this model, firms accrued a contingent tax liability when the liability was probable to occur and the amount could be reasonably estimated. In this financial reporting regime, recognition and measurement practices varied greatly across firms and few firms disclosed any tax contingency information (Gleason and Mills 2002). Issued in June 2006, FIN 48 applies a benefit recognition model that requires a twostep process to determine the amount of tax benefits recognized in financial statements. Step one determines whether the firm can recognize any tax benefit associated with a specific tax position, 6 while step two measures the amount of tax benefit that can be recognized. FIN 48 imposes two stringent requirements in the benefit recognition model : 1) Firms must assume that all positions taken on a tax return will be audited by relevant tax authorities, and 2) Firms cannot incorporate the potential for tax audit settlements with tax authorities in step one. These requirements are significant departures from SFAS No. 5 and caused many to predict that the prevalence and size of uncertain tax benefits would increase under FIN 48. earnings. As previously discussed, we believe the tax reserve is the account managers are most likely to use to manage the tax expense. However, it is possible the results documented by Dhaliwal et al. (2004) could be, at least in-part, the result of changes in the valuation allowance or changes in the amount of earnings reported to be permanently reinvested abroad, consistent with findings in Krull (2004) and Frank and Rego (2006). 6 There must be a greater than 50 percent likelihood that a tax position will be sustained upon audit based on its technical merits, in order for any tax benefit to be recognized in the financial statements. According to FIN 48, The term tax position refers to a position in a previously filed tax return or a position expected to be taken in a future tax return that is reflected in measuring current or deferred income tax assets and liabilities for interim or annual periods. 6

9 Under both reporting regimes, increases in uncertain tax benefits typically increase total tax expense and also generate contingent liabilities (i.e., reserves) that are reversed in the future when the uncertain tax position is resolved. 7 Upon the resolution of a particular uncertain tax position, the firm may be under-reserved or over-reserved relative to the actual outcome. If the firm is under-reserved, it will reverse the accrued tax liability, report additional tax expense (which reduces current period net income), and pay the relevant tax authority. If the firm is overreserved, it will reverse the accrued tax liability, report a tax benefit (which increases current period net income), and pay the relevant tax authority, if applicable. In the latter case, firms with ample uncertain tax benefits can potentially reduce those liabilities to manage earnings upward. 2.1 Evidence of Opportunism in Accounting for Income Taxes Dhaliwal et al. (2004) provide evidence that firms use discretion in accounting for their income tax expense to manage earnings upward to meet or beat analysts forecasts. Managers have substantial discretion with respect to multiple components of the tax provision. These components include discretion over the valuation allowance account, the designation of earnings as permanently reinvested abroad, and the focus of our analysis, the tax contingency. In addition, income tax expense is one of the last accounts closed prior to earnings announcements. Thus, the tax provision has been characterized as ripe for earnings management, which likely prompted the FASB to issue FIN 48 and clarify the accounting for uncertainty in income taxes. While Dhaliwal et al. (2004) rely on third- to fourth-quarter effective tax rate changes, Blouin and Tuna (2007) utilize Compustat data to estimate changes in tax cushion. They find that firms with larger implicit claims and equity financing use the cushion to smooth earnings. Gupta et al. (2011) utilize pre-fin 48 quarterly financial statement disclosures to specifically examine 7 In some cases uncertain tax benefits may relate to deferred taxes and in those cases increases in the contingency for uncertain tax benefits would not affect the reported tax expense. See Dunbar, Kolbasovsky, and Phillips (2007) for detailed examples of the journal entries necessary to record contingencies for unrecognized tax benefits. 7

10 whether firms reduce their tax reserves to opportunistically manage earnings. Based on voluntarily disclosed tax contingency data for 100 firms from , Gupta et al. find evidence consistent with firms reducing their tax contingencies to meet or beat analysts quarterly earnings forecasts. However, Gupta et al. find no evidence that these same firms reduced their tax contingencies to meet or beat analysts quarterly earnings forecasts in the years immediately following FIN 48 (2007 and 2008). Blouin, Gleason, Mills, and Sikes (2010) investigate whether firms were more likely to settle disputes with the IRS and/or decrease their tax reserves in the quarters leading up to the adoption of FIN 48. While reversals prior to FIN 48 adoption increased net income, reversals upon FIN 48 adoption increased retained earnings. 8 Thus, if firms determined they were overreserved under the new requirements of FIN 48 they had an incentive to reduce their tax reserves prior to FIN 48 adoption to report higher net income. Blouin et al. s (2010) results indicate that firms with larger outstanding IRS proposed deficiencies during the third and fourth quarters of 2006 were more likely to settle with the IRS. Their results also suggest that on average firms reduced their tax reserves during the third and fourth quarters of 2006, and firms with the incentive to manage earnings up to meet or beat the consensus forecast experienced greater reductions in tax reserves throughout their entire sample period (i.e., fiscal years 2005 and 2006 and first quarter 2007). 2.2 Hypothesis Development Despite evidence firms used their tax reserves to manage earnings prior to FIN 48, there are several reasons to expect the new accounting rules for tax contingencies would reduce if not eliminate such earnings manipulations. SOX Section 404 had already put accounting for income 8 Reversals after FIN 48 adoption also increase net income; thus, the impact of tax reserve reversals on retained earnings upon FIN 48 adoption was a one-time event. 8

11 taxes in the public spotlight, as a significant fraction of material internal control weaknesses disclosed between 2004 and 2006 involved accounting for income taxes (Gleason, Pincus, and Rego 2010). Thus, FIN 48 further heightened the increased auditor and regulatory scrutiny of corporate tax contingencies, possibly reducing managers willingness to use this account to manipulate earnings. In addition, concurrent research demonstrates that tax contingencies reported during the post-fin 48 time period are associated with aggressive tax sheltering. (e.g., Lisowsky 2010; Lisowsky, Robinson, and Schmidt 2010). These findings suggest the tax reserves do reflect at least to some extent uncertain tax positions. Despite these factors pointing toward a reduction in managerial opportunism related to tax reserves, the complex rules of FIN 48 still allow substantial discretion in determining whether and how large a tax contingency the firm is required to accrue. In fact, De Simone, Robinson, and Stomberg (2011) examine one specific uncertain tax position across a sample of 19 firms: the taxation of excise tax credits for alternative fuel mixtures. Despite the underlying tax position being virtually identical for each firm, DeSimone et al. find substantial variation in the amounts of unrecognized tax benefits reported by the 19 firms. While some firms fully reserved against the tax credits in their financial statements, others either partially or fully recognized the tax benefits associated with the uncertain tax positions. The authors even find that firms using the same auditors take different approaches to recording reserves against the tax benefits for virtually identical underlying tax positions. This lack of consistency in applying FIN 48 to one specific uncertain tax position suggests the financial reporting for the numerous, complex tax arrangements in which large firms engage is still subject to significant managerial discretion and therefore continues to be a likely tool for earnings management. 9

12 Although firms are required to separately evaluate each position taken on their tax returns, the disclosures required by FIN 48 reflect aggregate tax reserves across all tax jurisdictions for the entire firm. This aggregation makes it difficult for outside stakeholders to evaluate the accuracy of tax contingency disclosures, further making the account attractive for continued earnings manipulations. Thus, our first hypothesis predicts that firms continue to use their tax reserves to manage earnings up, despite the FASB s intention of imposing consistent recognition and measurement requirements on firms through FIN 48: H1 Firms with pre-managed earnings below the consensus analyst forecast decrease their tax reserves to meet or beat the consensus forecast. In order for firms to have tax reserves available to manage earnings upward, they must have accrued those reserves in a prior year. This type of earnings manipulation is sometimes referred to as building cookie jar reserves. Managers wishing to consistently meet or beat analysts forecasts can engage in income-decreasing earnings management in periods of higher than expected performance, thereby creating or replenishing cookie jar reserves that can be drawn upon to increase earnings in subsequent periods when unmanaged earnings are lower than expected (Levitt 1998; Badertscher, Phillips, Pincus, and Rego 2010). Tax reserves are one form of cookie jar reserves. 9 Thus, our second hypothesis predicts that firms with sufficiently high pre-managed earnings build cookie jar reserves through their tax contingencies: H2 Firms with pre-managed earnings above the consensus analyst forecast increase their tax reserves in anticipation of future earnings management. 9 Koester (2011) finds investors positively value tax reserves, consistent with investors taking a very different view of tax reserves than other liabilities. One possible reason for the positive valuation is that investors believe a significant portion of the unrecognized tax benefits will be retained upon audit. This result is consistent with investors believing that managers create cookie jar reserves with their contingent tax liabilities. However, it is also possible the positive valuation results, in-part, because investors interpret large current tax reserves as a signal about future cash flows resulting from tax planning activity. 10

13 FIN 48 requires firms to disclose changes in tax reserves that are related to current year tax returns, prior year tax returns, settlements with tax authorities, and expirations of the statute of limitations for tax returns. While changes in reserves related to current and prior year tax returns involve substantial managerial discretion, we characterize changes related to settlements with tax authorities and statute of limitations expirations as largely non-discretionary. 10 Consistent with H2, we conjecture that certain firms prefer to maintain their tax contingencies as cookie jar reserves for future earnings manipulations. In particular, we expect firms with premanaged earnings sufficiently above the consensus forecast to offset decreases in tax reserves caused by settlements and statute of limitations expirations with net increases in other components of the tax reserve (e.g., increases related to current and prior year tax returns). Accordingly, our third and fourth hypotheses are as follows: H3 Firms with pre-managed earnings sufficiently above the consensus forecast offset decreases in tax reserves caused by settlements with tax authorities with net increases in other components of their tax reserve. H4 Firms with pre-managed earnings sufficiently above the consensus forecast offset decreases in tax reserves caused by statute of limitation expirations with net increases in other components of their tax reserve. Despite the continued, substantial discretion surrounding the recognition and measurement of uncertain tax positions under FIN 48, it is possible that the combination of increased regulatory and auditor scrutiny of the tax provision significantly reduced but did not eliminate tax reserve management in the post-fin 48 time period. Recall that Gupta et al. (2011) find no evidence that their sample of 100 Fortune 500 firms used tax reserves to manage 10 Blouin et al. (2010) find evidence consistent with firms having discretion over the timing of their settlements with tax authorities, but the amounts of such settlements are largely non-discretionary. 11

14 earnings upward in the post-fin 48 time period (i.e., 2007 and 2008). Nonetheless, given the smaller sample size and data limitations in Gupta et al. (2011), we investigate whether tax reserve management declined in the post-fin 48 time period relative to the pre-fin 48 time period. Our final two hypotheses are as follows: H5 Firms with pre-managed earnings below the consensus analyst forecast are less likely to decrease their tax reserves to meet or beat the consensus forecast in the post-fin 48 time period, compared to the pre-fin 48 time period. H6 Firms with pre-managed earnings above the consensus analyst forecast are less likely to increase their tax reserves in anticipation of future earnings management in the post-fin 48 time period, compared to the pre-fin 48 time period. 3 Research Design and Data 3.1 Empirical Models Hypothesis 1 (2) predicts that firms with pre-managed earnings below (above) the consensus analyst forecast reduce (increase) their tax reserves to meet or beat the consensus forecast (to create cookie jar reserves for the future). We test these hypotheses by regressing the total change in tax reserves on proxies for incentives to manage earnings up and down (i.e., FE_MISS and FE_MEET), and variables identified in prior research as determinants of tax avoidance and/or tax reserves. Our empirical model is shown in equation (1) below: TAX_RES it = FE_MISS it + 2 FE_MEET it + 3 PT_ROA it + 4 FOR_SALES it + 5 R&D it + 6 LEV it + 7 MTB it + 8 SG&A it + 9 SALES_GR it + 10 CASH_ETR it + it (1), 12

15 where TAX_RES is the total change in firm i s tax reserve; FE_MISS is firm i s forecast error (i.e., consensus analyst forecast less pre-managed earnings) if pre-managed earnings are below the consensus forecast (and zero otherwise); 11 FE_MEET is firm i s forecast error (i.e., consensus analyst forecast less pre-managed earnings) if pre-managed earnings are above the consensus forecast (and zero otherwise); 12 all other variables in equation (1) are control variables (discussed below). We compute pre-managed earnings as reported earnings plus the change in tax reserves from year t-1 to year t (see Appendix A for complete variable definitions). If premanaged earnings are below the consensus forecast (if FE_MISS > 0), then we predict the firm will reduce its tax reserves, which would cause the coefficient on FE_MISS ( 1 ) to be negative. If pre-managed earnings are above the consensus forecast (i.e., if FE_MEET < 0), then we predict the firm will increase its tax reserves, which would cause the coefficient on FE_MEET ( 2 ) to also be negative. In our analysis we initially assume that TAX_RES = DS_ TAXRES true +. In other words, we assume the total reported change in the tax reserve ( TAX_RES) is equal to the true discretionary change in the tax reserve (DS_ TAXRES true ) plus any error in measuring the discretionary change in the tax reserve ( ). Under the null hypothesis of no earnings management via the tax reserve we expect DS_ TAXRES true = 0. We assume that any error is unbiased (that is, E[ but we recognize that measurement error in our estimate of the discretionary change in the tax reserve will be non-zero for individual firm-years. In subsequent tests we alter this assumption and use an adjusted measure of the change in tax reserves (DS_ TAXRES) in an effort to more accurately measure the portion of the change in tax reserves that is discretionary 11 Thus, FE_MISS can only be zero or positive. 12 Thus, FE_MEET can only be zero or negative. 13

16 and would directly affect earnings. In these subsequent tests we assume DS_ TAX_RES = DS_ TAXRES true +. We discuss the calculation of DS_ TAXRES in detail below. To control for non-discretionary changes in tax reserves which are reflected in measurement error above we include in equation (1) control variables that are based on prior studies of tax avoidance and tax reserves (e.g., Rego 2003; Cazier, Rego, Tian, and Wilson 2009; Wilson 2009; Lisowsky 2010;). We expect profitable firms to have the greatest ability and incentive to engage in tax planning that involves significant uncertainty, so we include pre-tax return on assets (PT_ROA) in equation (1). Prior research suggests that firms with foreign operations and greater research and development expenditures engage in more tax avoidance than other firms (e.g., Rego 2003; Wilson 2009), so we also include the ratio of foreign sales to total sales (FOR_SALES) and R&D. Prior research demonstrates an inverse relation between leverage and aggressive tax planning (e.g., Graham and Tucker 2006; Wilson 2009). Thus, we include leverage (LEV) in equation (1). Growth firms are subject to greater capital market pressure to meet market expectations (e.g., Skinner and Sloan 2002), so we control for growth by including the market-to-book ratio (MTB) and sales growth (SALES_GR). We include selling, general, and administrative expenses (SG&A) in equation (1) because prior research has found these expenses to be significantly associated with tax contingencies (Song and Tucker 2008). Lastly, firms that avoid more income taxes are more likely to require a tax contingency under FIN 48. Thus, we include cash effective tax rates (CASH_ETR) in our model, since that measure of tax avoidance is not directly affected by changes in tax reserves (Cazier et al. 2009). We also perform several additional tests of H1 and H2. First, we refine our proxies for incentives to manage earnings by considering how far pre-managed earnings are from the consensus analyst forecast. Consistent with Frank and Rego (2006), we partition our sample into 14

17 firms with pre-managed earnings that are far below (BIG_MISS = 1 firms), just below (SMALL_MISS = 1 firms), just above (SMALL_MEET = 1 firms), and far above the consensus analyst forecast (BIG_MEET = 1 firms). Firms with pre-managed earnings that are just below the consensus forecast (i.e., SMALL_MISS = 1 firms) are the most likely to reduce their tax reserves to manage earnings up to meet or beat the forecast, while firms with pre-managed earnings far below (i.e., BIG_MISS = 1 firms) are less likely to do so unless their tax reserves are sufficiently large to allow them to meet or beat the consensus forecast. Firms with pre-managed earnings that are far above the consensus forecast (i.e., BIG_MEET = 1 firms) are the most likely to increase their tax contingencies to create cookie jar reserves that can be used in future earnings manipulations. We do not expect firms with pre-managed earnings that just meet or beat the forecast (i.e., SMALL_MEET = 1 firms) to significantly change their tax reserves, as they have already beat the consensus forecast but not excessively. We test these predictions based on equation (2) below: TAX_RES it = 1 BIG_MISS it + 2 SMALL_MISS it + 3 BIG_MEET it + 4 SMALL_MEET it + 5 PT_ROA it + 6 FOR_SALES it + 7 R&D it + 8 LEV it + 9 MTB it + 10 SG&A it + 11 SALES_GR it + 12 CASH_ETR it + it (2), The variables in equation (2) above are as defined for equation (1), except we now replace FE_MISS and FE_MEET with BIG_MISS, SMALL_MISS, BIG_MEET, and SMALL_MEET. BIG_MISS (SMALL_MISS) equals 1 for firm-years where pre-managed earnings are less than the consensus forecast by at least (less than) 2 cents; and zero otherwise. BIG_MEET (SMALL_MEET) equals 1 for firm-years where pre-managed earnings are greater than the consensus forecast by at least (less than) 2 cents; and zero otherwise. (See Appendix 1 for complete variable definitions.) We predict negative coefficients on BIG_MISS and 15

18 SMALL_MISS ( 1 and 2 ) and a positive coefficient on BIG_MEET ( 3 ). We predict that the coefficient on SMALL_MEET ( 4 ) will not be significantly different than zero. We suppress the intercept when estimating equation (2) so that we can include the four MISS and MEET indicator variables. Second, we modify our proxy for tax reserve management. While equations (1) and (2) use the total change in tax reserves (i.e., TAX_RES) as the dependent variables, we also reestimate those models with the portion of the change in tax reserves that affects a firm s effective tax rate (ETR) ( TAX_RES_ETR). It is only this portion of the total change in tax reserves that affects after-tax net income. In contrast, the portion that does not affect a firm s ETR does not affect after-tax net income, and thus is not a useful earnings management tool if the earnings target involves after-tax income. Because firms are only required to disclose the portion of the ending balance of tax reserves that affects a firm s ETR (rather than the portion of the change in total tax reserves that affects a firm s ETR), we have one less year of data for TAX_RES_ETR compared to TAX_RES, and thus these tests are less powerful. Third, we estimate the portion of the total change in tax reserves that is discretionary and re-estimate equation (1) with the discretionary change in tax reserves (DS_ TAXRES) as the dependent variable. We calculate DS_ TAXRES as the total change in tax reserves less changes related to settlements with tax authorities and statute of limitations expirations. This alternative measure is intended to isolate the changes in tax reserves over which managers have the most discretion and thus are most likely to reflect earnings manipulations. Hypotheses 3 and 4 predict that firms with pre-managed earnings sufficiently above the consensus analyst forecast offset decreases in tax reserves caused by settlements with tax authorities and statute of limitations expirations with net increases in other components of the tax 16

19 reserve, including those related to current and prior year tax returns. To test these hypotheses, we first include an indicator variable (BIG_MEET) in equation (1) that equals 1 if the firm has premanaged earnings that are far above the consensus analyst forecast (i.e., greater than the consensus forecast by at least 2 cents), and 0 otherwise. We also separately include the decrease in tax reserves caused by settlements with tax authorities ( TAXR_SETTLE) and statute of limitations expirations ( TAXR_STATLIM), and then interact those continuous variables with the BIG_MEET indicator variable. Because H3 and H4 predict that firms offset nondiscretionary decreases in tax reserves with net increases in other (discretionary) components of the tax reserve, we replace the dependent variable in equation (1) with the discretionary change in tax reserves (DS_ TAXRES). Hence, our tests of H3 and H4 are based on equation (3) below: DS_ TAXRES it = FE_MISS it + 2 FE_MEET it + 3 PT_ROA it + 4 FOR_SALES it + 5 R&D it + 6 LEV it + 7 MTB it + 8 SG&A it + 9 SALES_GR it + 10 CASH_ETR it + 11 BIG_MEET it + 12 TAXR_SETTLE it + 13 BIG_MEET TAXR_SETTLE it + 14 TAXR_STATLIM it + 15 BIG_MEET TAXR_STATLIM it + it (3), We expect the coefficients on BIG_MEET TAXR_SETTLE ( 13 ) and BIG_MEET TAXR_STATLIM ( 15 ) to be negative and significant, consistent with firms with pre-managed earnings far above the consensus forecast offsetting nondiscretionary decreases in tax reserves with net increases in other components of the tax reserve, to maintain cookie jar reserves for the future. We also continue to expect the coefficients on FE_MISS and FE_MEET to be negative and significant, consistent with H1 and H2. Hypothesis 5 (6) predicts that firms with pre-managed earnings below (above) the consensus forecast are less likely to decrease (increase) their tax reserves to manage earnings 17

20 upward (to create cookie jar reserves for the future) in the post-fin 48 time period, compared to the pre-fin 48 time period. To test these hypotheses, we require a proxy for the change in tax reserves in the pre-fin 48 time period. Because firms were not required to disclose changes in tax reserves prior to FIN 48, we follow Blouin and Tuna (2007) and estimate the change in tax reserve (aka, change in tax cushion) as the current portion of tax expense less cash paid for income taxes, less an estimate of the tax benefit from stock options, less the change in income taxes payable from the balance sheet, scaled by beginning of the year total assets. 13 We refer to the Blouin and Tuna measure as ΔCUSHION. We then calculate FE_MISS and FE_MEET in the pre- and post-fin 48 time periods based on CUSHION. We estimate equation (4) below to test H5 and H6, i.e., whether upward and downward earnings management through tax reserves decreased in the post-fin 48 time period: CUSHION it = FE_MISS it + 2 FE_MEET it + 3 POST + 4 POST FE_MISS it + 5 POST FE_MEET it + 6 PT_ROA it + 7 FOR_SALES it + 8 R&D it + 9 LEV it + 10 MTB it + 11 SG&A it + 12 SALES_GR it + 13 CASH_ETR it + it (4), where POST equals 1 for fiscal years 2007, 2008, and 2009, and is zero otherwise. The coefficients on FE_MISS and FE_MEET should be negative and significant if firms managed earnings through their tax reserves in the pre-fin 48 time period. However, H5 and H6 predict positive and significant coefficients on POST FE_MISS and POST FE_MEET if firms engaged in less tax reserve management in the post-fin 48 time period, compared to the pre-fin 48 time period. 3.2 Sample Selection 13 See Appendix A for a description of our estimate of the tax benefit from stock options based on Blouin and Tuna (2007) using ExecuComp data. 18

21 Table 1 summarizes our sample selection procedures. We begin with all firms listed on either the S&P 500 or the S&P 400 indices for fiscal year 2007, the first year for which FIN 48 disclosures were required. We collect tax reserve data from the tax footnotes of these 900 firms for the fiscal years (2,700 firm-years). FIN 48 disclosure requirements became mandatory for fiscal years beginning after December 15, Consequently, firms with 2007 fiscal years ending prior to December 15, 2007 were not required to comply with FIN 48 until the following year. As a result, we eliminate 147 firm-year observations that did not implement FIN 48 during the 2007 fiscal year. We eliminate 151 firm-years due to missing 10-K filings (largely due to mergers and acquisitions) and another 280 observations with insufficient FIN 48 data reported in tax footnotes. We remove 21 real estate investment trust observations, as these firms do not pay Federal income taxes and thus do not require tax reserves. Lastly, we eliminate 177 firm-years with insufficient financial statement data available on Compustat to calculate measures required by our empirical tests, including data necessary to calculate five-year cash effective tax rates. Our final sample consists of 1,925 firm-year observations. [INSERT TABLE 1 HERE] 3.3 Descriptive Statistics Table 2, Panel A provides descriptive statistics for the variables included in our initial set of tests. The mean and median change in total tax reserves ( TAX_RES) is 0.000, indicating that on average sample firms did not significantly increase or decrease their tax reserves during the first three years following the implementation of FIN 48. The mean (median) change in the portion of the tax reserve that affects the ETR ( TAX_RES_ETR) is (0.000), while the mean and median discretionary change in tax reserves (DS_ TAXRES) is The mean and median forecast error for firms with pre-managed earnings below the consensus analyst forecast 19

22 (FE_MISS) is and 0.000, respectively. 14 The mean and median forecast error for firms with pre-managed earnings equal to or above the consensus analyst forecast (FE_MEET) is and , respectively. Recall that FE_MEET (FE_MISS) equals zero if a firm s pre-managed earnings are below (equal or exceed) the consensus forecast. Our sample firms are generally profitable, reporting mean and median pre-tax return-on-assets (PT_ROA) of and 0.072, respectively. Sample firms also report positive mean (0.076) and median (0.063) sales growth (SALES_GR) during the sample period, and relatively low mean (median) cash effective tax rates of 22.8 (24.0) percent. Interestingly, we also find that 71.6 percent of sample firm-years report a reduction in tax reserves due to a settlement with tax authorities (SETTLE_DUMM) and 67.3 percent of firm-years report a reduction in their tax reserves due to statute of limitations expirations (STATLIM_DUMM). [INSERT TABLE 2 HERE] Table 2, Panel B presents descriptive statistics for the FIN 48 data collected from the tax footnotes of our sample firms. We note the mean and median change in tax reserves upon FIN 48 adoption was The mean beginning (BEG_TAXRES) and ending tax reserves (END_TAXRES) for sample firms is 1.5 percent of lagged total assets, consistent with the average firm maintaining its tax reserves throughout the sample period. We collect data on each of the individual components of the reported change in total tax reserves. The largest mean (median) change of any component is associated with current period tax positions ( TAXR _CURR), which is equal to 0.2 (0.1) percent of total assets. Table 2 also shows that the mean (median) value of tax reserves at year-end that would affect a firm s ETR (TAXRES_ETR) is (0.006) compared to the mean (median) ending total tax reserve (END_TAXRES) of Throughout this paper, we measure the consensus analyst forecast as the median of all analyst forecasts available just prior to a firm s earnings announcement. 20

23 (0.010). This data suggests that most changes in tax reserves affect net income through total tax expense, and thus, most tax reserves can be used to manipulate after-tax earnings. In Table 3 we present the Pearson and Spearman correlation coefficients for the variables of interest in our tests of H1-H4. Consistent with expectations, we observe a significant and negative Pearson and Spearman correlation between the total change in tax reserves ( TAX_RES) and both FE_MISS and FE_MEET. We find similar results for the change in tax reserves that affect the ETR ( TAX_RES_ETR) and for the discretionary change in tax reserves (DS_ TAXRES). We observe significant and negative Pearson and Spearman correlations between the total change in tax reserves and the current year CASH_ETR, consistent with firms that increase their tax reserves also reporting lower cash ETRs. Related to H3 and H4, we find that firms experiencing a decrease in tax reserves related to either a settlement with tax authorities ( TAXR_SETTLE) or the expiration of statutes of limitations ( TAXR_STATLIM) are more likely to experience an increase in other (discretionary) components of the change in tax reserves (DS_ TAXRES). By definition, decreases in total tax reserves due to settlements with tax authorities or statute of limitations expirations reduce total tax reserves ( TAX_RES) and generate positive Pearson and Spearman correlations. [INSERT TABLE 3 HERE] 4 Discussion of Results We begin our tests of H1 and H2 by examining some descriptive statistics for groups of firms partitioned based on the difference between our measure of pre-managed earnings and the consensus analyst forecast, where we calculate pre-managed earnings by reversing the total change in tax reserves. Inferences are substantially similar if we calculate pre-managed earnings 21

24 by reversing only the change in tax reserves that affect the ETR (results untabulated). Panel A of Table 4 presents descriptive statistics for our full sample of firm-year observations. We note that 51.9 percent of all firm-years increase their tax reserve in a given year and 44.5 percent decrease their reserves. This result is consistent with our descriptive statistics in Table 2, which indicate that tax reserves exhibit little change over our sample period. Panel A also shows that for 16.9 percent of our firm-year observations, a net decrease in the firm s total tax reserve allowed the firm to meet the consensus forecast. In contrast, we find that for only 6.0 percent of sample observations, a net increase in total tax reserves caused the firm to fall short of the consensus forecast (results untabulated). [INSERT TABLE 4 HERE] In Panel B we partition the firms into two groups based on whether pre-managed earnings are above or below the consensus analyst forecast. We note that 74.6 percent of firm-years with pre-managed earnings below the consensus forecast decrease their tax reserves, which increases net income. In contrast, only 23.3 percent of the firms with pre-managed earnings above the consensus forecast decrease their tax reserves. Moreover, while 40.9 percent of firm-years with pre-managed earnings below the consensus forecast are able to meet the forecast by reducing their tax reserve (i.e., CROSSMEET = 1), just 10.2 percent of firm-years with pre-managed earnings above the consensus forecast increase their tax reserves to the extent that it causes them to miss the consensus forecast (i.e., CROSSMISS = 1). This asymmetric result is consistent with managers using discretion over the tax reserve to meet the consensus forecast. In Panel C of Table 4 we further partition firm-years into four groups (BIG_MISS =1, SMALL_MISS =1, SMALL_MEET =1, and BIG_MEET =1), where the SMALL_MISS and SMALL_MEET groups include observations with pre-managed earnings within 2 cents of the 22

25 consensus forecast, while the BIG_MISS and BIG_MEET groups have pre-managed earnings that are more than 2 cents from the consensus forecast. This additional partitioning yields several interesting results. First, we note that 60.3 percent of the SMALL_MISS = 1 firm-years were able to meet the consensus forecast by decreasing their tax reserves (i.e., CROSSMEET = 1). 15 Second, a relatively high 37.4 percent of the BIG_MISS = 1 firm-years were also able to meet the consensus forecast by decreasing their tax reserves (i.e., CROSSMEET = 1). Moving to the SMALL_MEET = 1 group, where firms have no clear incentive to either increase or decrease their tax reserves, we find a relatively even split with 45.1 percent of firm-years increasing their reserves and 47.9 percent decreasing their reserves. Finally, for the BIG_MEET = 1 group, where pre-managed earnings significantly exceed the consensus forecast, we note that a large majority (76.4 percent) of the firm-years increase their tax reserves. This result is consistent with managers at those firms smoothing earning downward in high earnings years and building reserves for the future. Taken together, the results in Table 4 are at least suggestive of opportunistic behavior by managers in accounting for their tax reserves. Figure 1 depicts the distribution of sample observations across bins of earnings surprises, which are calculated as the difference between pre-managed earnings and the consensus analyst forecast. The distribution appears to approximate a normal distribution that is centered on a 2 cent earnings surprise. Each bin is shaded to depict the proportion of firms in that particular bin that increased vs. decreased their tax reserves, where the darker shading represents firms that decreased their tax reserves. The results in Figure 1 support our analysis in Table 4 by illustrating that a much greater proportion of firm-years with pre-managed earnings below the consensus forecast decrease their tax reserves. In contrast, most firm-years with pre-managed 15 Thus, amongst the 68.6 percent of the SMALL_MISS = 1 firm-years that decreased their tax reserves (see bottom row of Panel C), approximately 88 percent (i.e., ) of these firm-years were able to meet the consensus forecast due to a decrease in tax reserves. 23