How much of tax earnings management is really earnings management? Kathleen Powers University of Tennessee, Knoxville

Transcription

1 How much of tax earnings management is really earnings management? Kathleen Powers University of Tennessee, Knoxville

2 I. Introduction Using the Blinder-Oaxaca decomposition (Blinder 1973; Oaxaca 1973), I examine how much of the decrease between the third quarter and annual GAAP effective tax rate (ETR) is due to intentional earnings management. Findings in prior literature suggest that when firms are close to missing an earnings target, they manage their GAAP ETR downward to increase income and meet or beat the target (Cazier, Rego, Tian and Wilson 2014; Comprix, Mills, and Schmidt 2012; Dhaliwal, Gleason and Mills 2004; Frank and Rego 2006; Gleason and Mills 2008; Krull 2004; Mauler 2016; Schrand and Wong 2003). These papers argue that because the Q3 ETR proxies for management s best estimate of the firm s annual ETR, decreases between the Q3 and annual ETR are suspicious and likely indicative of intentional earnings management, especially when these decreases occur at firms also in danger of missing an earnings benchmark. Several of these studies also identify the specific component of tax expense that firms can manage when in danger of missing the earnings target and find results consistent with this behavior (Cazier et al. 2014; Frank and Rego 2006; Krull 2004; Schrand and Wong 2003). Thus, at least some of the decrease between the Q3 and Q4 ETR is consistent with earnings management. However, it is unlikely that the entire decrease in the Q3 to Q4 ETR is attributable to earnings management. First, the complexities inherent in applying the accounting for income taxes in interim periods guidance (ASC ) may result in a Q3 ETR that is not the best estimate of the annual ETR. To calculate the estimated annual ETR, firms are required to forecast their expected annual ordinary income by jurisdiction at each quarter, calculate the expected tax on that income in each jurisdiction and then aggregate the expected income and tax expense to arrive at the global, consolidated estimated annual ETR. Thus, differences in firms income projections or income allocations among jurisdictions at Q3 and their actual realization at year-end will result in movement in the Q3 to Q4 ETR. Additionally, discrete events, which must be accounted for in the Page 1

3 quarter in which they occur, can have a lasting effect on the annual expected ETR over the subsequent quarters. Although these events are accounted for discretely in one quarter, they will affect the annual estimated ETR as well. Thus, the proper implementation of ASC can result in volatility in and potentially uninformative ETR (pwc 2015). For example, Bratten, Gleason, Larocque, and Mills (2015) and Kim, Schmidt and Wentland (2015) find evidence consistent with analysts and investors struggling to understand and forecast changes in tax expense when firms have transitory or discrete items. Taken together, it is unclear how useful the Q3 ETR is as a best estimate of the firm s annual expected tax rate and whether even sophisticated financial statement users can anticipate expected movement between the Q3 and annual ETR. I propose that the observed decrease in the ETR during Q4 at firms with a high incentive to manage their earnings is composed of both an expected and managed component. I use a broad sample of firms from and identify suspect firms with a high incentive to decrease their ETR during Q4 as those that beat their consensus EPS target by less than $0.02. To determine how much of the decrease is attributable to each component, I use the Blinder-Oaxaca decomposition, a popular methodology used to study the wage gap in labor economics literature. The decomposition involves first estimating two group-specific regression models. In this case, I model the likelihood the firm decreases its ETR during Q4 on firm and ETR characteristics shown to be associated with movement in the ETR. I then estimate this model separately for suspect firms and non-suspect firms. These regressions yield the mean predicted likelihood of a decrease in the ETR during Q4 for each group. The difference in the likelihood of decreasing the ETR between the groups is then decomposed into an explained and unexplained component. The explained component accounts for differences due to underlying group differences in the predictors and is the expected difference. For example, when examining the gap in wages between Page 2

4 men and women, the explained portion of the gap will capture differences in education, work experience, job tenure, etc. (Jann 2008). If women have less education than men, it is expected that they will earn less. To determine the unexplained portion, the Blinder-Oaxaca decomposition utilizes the coefficients from the first model and imposes them on the characteristics of the observations from the second model. Referring again to the wage gap literature, if women s characteristics have the same association with wages as men s characteristics, the wage gap should be fully attributed to differences in the underlying characteristics (explained portion) and the discriminatory portion (unexplained portion) should be zero. Similarly, in my setting, if differences in the likelihood of decreasing the ETR during Q4 between suspect and non-suspect firms are fully attributable to differences in the underlying characteristics of these firms, then the explained portion should fully account for the difference, and the earnings management portion (unexplained portion) should be zero. Using a sample of 27,962 firm-year observations, I find that the mean likelihood of decreasing the ETR between Q3 and Q4 for non-suspect firms is 52.4% and 57.5% for suspect firms. Thus, suspect firms are 5.09% more likely to decrease the Q3 ETR than non-suspect firms. When decomposing this difference, I find that 3.94% of the difference is explained whereas 1.15% is unexplained. These findings suggest that 77% (3.94/5.09) of the difference in likelihood of decreasing the ETR in Q4 between suspect and non-suspect firms is expected, whereas only 23% (1.15/5.09) of the difference is due to earnings management. This finding is robust to various definitions of suspect and non-suspect firms, as well as differences in the specification of the regression models. Therefore, I conclude that although management of the ETR in Q4 to beat an earnings target is occurring, over three-quarters of the difference in likelihoods of decreasing the ETR in Q4 is expected, which has significant implications for researchers and investors. Page 3

5 First, understanding whether management intentionally walks down the ETR during Q4 or whether movement in the ETR during Q4 is a byproduct of the firms operations and ASC is an important unanswered question. Gleason and Mills (1998) and Mauler (2015) both find that firms that simultaneously just beat an earnings benchmark and decrease the ETR in Q4 are punished by the market. My finding that three quarters of the difference in behavior between suspect and non-suspect firms is expected suggests that decreases in the ETR during Q4 by suspect firms may not be entirely attributable to earnings management in Q4. In this way, my findings corroborate Comprix et al. (2012). Suspect firms may still be managing earnings through the ETR, but instead of waiting until Q4, they are managing earlier in the year. However, I cannot rule out the possibility that suspect firms simply anticipate the decrease in the ETR during Q4 due to a better understanding of the implications of income tax accruals earlier in the year. Either way, my findings and those of Comprix et al. (2012) suggest a holistic approach when examining firms quarterly ETR. Rather than only examining the change in Q4, it is important to understand movements earlier in the year as well. Second, because market participants have traditionally struggled to understand and price changes in tax expense (Amberger, Eberhartinger and Kasper 2016; Chen and Schoderbek 2000; Plumlee 2003; Powers, Seidman, Stomberg and Schmidt 2016; Weber 2009), disentangling intentional management of the ETR from an expected change should allow participants to better anticipate and classify expected versus unexpected changes in the ETR during Q4. My finding that three-quarters of the difference in behavior between suspect and non-suspect firms is expected suggests that it may be possible for market participants to anticipate a portion of the movement in the ETR during Q4 at suspect firms, and thus refine their expectations of the annual ETR. Page 4

6 The remainder of the paper proceeds as follows. Section II reviews related literature and develops my hypotheses. Section III specifies my research design. Section IV presents results, and Section V concludes. II. Prior Literature and Hypothesis Development It is well-established that firms manage earnings (Burgstahler and Dichev 1997; Degeorge, Patel, and Zeckhauser 1999; Donelson, McInnis, and Mergenthaler 2013) and that tax expense is one of the accounts they use (Dhaliwal et al. 2004). The judgment, discretion and complexity involved in estimating a company s tax expense makes this account attractive for opportunistic financial reporting. Although the majority of a firm s tax burden is a function of its pre-tax income and tax credits, the calculation of tax expense for financial statement purposes requires accruals for valuation allowances, unrecognized tax benefits, and taxes on permanently reinvested foreign unremitted earnings (PRE). ASC 740 provides guidance as to how each of these accruals should be calculated, but each requires a significant amount of subjective judgement. Indeed, prior literature has found that firms use both the valuation allowance (Frank and Rego 2006; Schrand and Wong 2003) and PRE assertions (Krull 2004) to successfully manage earnings. Studies have also found that investors and even sophisticated financial statement users, such as analysts, struggle to understand firms tax expense (Chen and Schoderbek 2000; Plumlee 2003; Powers et al. 2015; Weber 2009). Although recent work suggests that perhaps analysts have improved their understanding of the drivers of firms income tax expense (Bratten et al. 2015; Kim et al. 2015), it is unclear whether investors sufficiently understand the interim reporting guidance (ASC ) to allow them to disentangle expected movements in the ETR from earnings management. Studies examining earnings management through the tax expense generally identify suspect firms as those who just beat an earnings benchmark and use decreases in the ETR between Q3 and Q4 as the proxy for tax earnings management (Comprix et al. 2012; Dhaliwal et al. 2004; Page 5

7 Gleason and Mills 2008; Mauler 2016). However, a lesser-explored area of the literature on firms use of tax expense to manage earnings is how the guidance on accounting for income taxes in interim periods (codified in ASC ) potentially affects the movement in the ETR throughout the year. ASC requires companies to calculate an estimated annual ETR applicable to the entire year s ordinary income. Because the guidance specifically indicates that the firm should calculate its best estimate of its annual ETR for the entire year, it is intuitive to assume that companies calculate their best estimate at Q1 and that deviations from this rate represent poor quality, withheld information or earnings management. Prior research also argues that by Q3, when the firm is three-quarters through the year, it should have an accurate estimate of the annual ETR. However, because of the nuances in the guidance and the complexity in calculating and updating a best estimate of the annual ETR, the Q3 ETR may not be the best estimate of the firm s annual ETR. There are three main drivers of significant changes to the quarterly ETR and estimated annual ETR. 1 The first two drivers of changes in the estimated annual ETR relate to changes in expected ordinary income. To calculate tax expense in each interim period, the firm must estimate the annual ordinary income in every jurisdiction in which it operates, multiply the income by the relevant jurisdiction s appropriate tax rate and divide the aggregated consolidated tax expense by the aggregated consolidated pre-tax income to arrive at the estimated ETR that will be presented in the income statements. Thus, changes in either the expected consolidated annual income or the allocation of income among jurisdictions can have a substantial effect on the firm s estimated annual ETR. 1 I provide illustrative examples of how each of the three drivers can affect the quarterly and annual ETR in Appendix A. Page 6

8 Second, firms must identify which income and expenses are ordinary, which are included in the estimated annual ETR calculation, and account for other items discretely in the quarter in which they occur. The tax effect of these other events is not prorated over the entire year and assimilated into the estimated annual ETR. Rather, it is accounted for discretely in the period in which it occurs. Some examples of common discrete items are movements in firms valuation allowances, changes in the reserve for uncertain tax benefits (UTB) relating to prior-year positions, the cumulative effect on firms tax positions from the enactment of a tax law or rate change, and a change in the estimate of the prior-year tax provision (pwc 2015). Because by definition, discrete items are significant unusual or infrequently occurring events and they must be recognized in the quarter they occur, firms have limited ability to directly affect the timing of these events or to smooth their impact. For example, the tax authority determines when to finalize an audit, and thus dictates the timing of discrete items relating to movement in the UTB reserve. The statute also requires firms to wait to record the impact of a change in tax law or statutory tax rate until the law s effective date, so even if the firm knows the impact, it cannot record it until the quarter in which the effective date falls. Therefore, the presence, timing and magnitude of a discrete item can have a significant impact on not only the quarter s tax expense, but also the estimated annual ETR. Comprix et al. (2012) and Schmidt (2006) also examine how ASC affects the patterns and persistence of the quarterly ETR. Comprix et al. (2012) examine changes in quarterly ETR and find that ETR during the first three quarters are biased upward and are more likely to reverse over subsequent quarters when firms would have missed their analyst earning s target. Schmidt (2006) studies the persistence in changes in the quarterly ETR and market valuations of these changes. He decomposes changes in the tax change component of the ETR into the initial change and subsequent changes and finds that the initial change is more persistent than subsequent Page 7

9 changes, suggesting that the initial estimate has less noise than subsequent quarters due to either a lack of earnings management or events muddling the best estimate of the annual ETR. My study builds on these findings and attempts to quantify how much of the decrease in ETR during Q4 at suspect firms is attributable to earnings management and how much is expected. Because the literature has demonstrated both that earnings management through tax expense occurs (Frank and Rego 2006; Krull 2004) and that ASC creates distortions in the quarterly ETR (Schmidt 2006) which suspect firms can opportunistically use (Comprix et al. 2012), I make no predictions regarding the portion of the decrease that is attributable to earnings management and how much is expected. Instead, I examine this question empirically. III. Research Design 3.1 Sample Selection Table 1, Panel A details my sample selection. I begin by selecting all observations in Compustat that have financial data available to calculate all regression variables for years I begin my sample in 1993 so that all of the firms in my sample were required to account for income taxes according to FAS 109 (ASC 740). I identify firms as suspicious if they beat their analyst consensus earnings target by $0.02 or less. Firms that miss their analyst consensus EPS target or beat the target by more than $0.02 are considered non-suspicious. 2 Thus, I eliminate firms without a consensus EPS target in IBES. There are 41,851 firm-years that meet these criteria. I eliminate observations where the annual GAAP effective tax rate (ETR) is less than zero or greater than one, as it is unclear how to interpret these ETRs (Henry and Sansing 2014). Finally, I eliminate 2 Inferences are similar if I use different cutoffs for suspect and non-suspect firms. I focus on firms that beat or miss the consensus EPS rather than a different earnings target, because this target has arguably become the most salient to managers (Graham, Harvey, and Rajgopal 2005). Page 8

10 firms with SIC codes in either the utility industry ( ) or the finance sector ( ). This process yields 27,962 firm-year observations comprised of 4,451 unique firms. [Insert Table 1 here.] 3.2 Blinder-Oaxaca Decomposition I utilize the Blinder-Oaxaca decomposition technique to differentiate between expected and unexpected changes in the ETR during Q4. This technique was popularized in the labor economics literature and is often used to determine how much of the wage gap between two groups is due to discrimination. Implementing this technique in my settling will allow me to disentangle differences in decreasing the ETR during Q4 between suspect and non-suspect firms. The decomposition first requires the estimation of the same regression model on two different groups and then decomposes the difference in the mean outcomes into an explained and unexplained portion. 3 In this way, the decomposition is advantageous over using a single regression and including a binary variable for the suspicious firms. The coefficient on the binary variable would provide the estimated change in the ETR during Q4 for suspect firms relative to non-suspect firms but would not provide any information regarding the amount of the change that was expected versus the amount attributable to earnings management. Similar to Dhaliwal et al. (2004), my dependent variable is a binary variable equal to one if the firm s ETR decreases during Q4 (DEC_Q4) and zero otherwise. Thus, I am examining the likelihood the firm decreases its ETR during Q4, rather than the magnitude of the decrease. I make this decision for several reasons. First, I only eliminate firms with an annual ETR less than zero or greater than one. Therefore, I still have loss firms in my sample and may have firms with a Q3 ETR less than zero or greater than one in my sample. Firms with these characteristics can distort 3 See Blinder 1973; Jann 2008; or Oaxaca 1973 for mathematical proofs. Page 9

11 the magnitude of changes in the ETR during Q4. 4 Thus, I estimate the following pooled, OLS regression model separately on the group of suspect and non-suspect firms: DEC_Q4 = β0 + β1q3_etrit + β2non_recurrit + β3qtr_volit + β4dec_q3it + β5bignit + β6chg_liquidit + β7roait + β8chg_intanit + β9chg_rdit + β10chg_capxit + β11chg_bmit + β12sizeit + β13levit + β14complexit +β15growit + FE_Ind + FE_Year + eit (1) I include firm and ETR characteristics shown in prior research to affect changes in the ETR during Q4. Q3_ETR captures the level of the firm s year-to-date ETR at Q3. Thus, this variable proxies for the magnitude by which the ETR can change during Q4 (Comprix et al. 2012; Dhaliwal et al. 2004). I include two other variables related to movements in the quarterly ETR during the year. NON_RECURR is a binary variable equal to one if the firm records a discrete item (NRTXT) during the year (Donelson, Koutney and Mills 2016), and QTR_VOL is the standard deviation of the quarterly ETR over the year. I include DEC_Q3, a binary variable equal to one if the firm decreased its ETR in Q3 to see if the associations in Comprix et al. (2012) occur in my sample as well. BIGN is a binary variable equal to one if the firm was audited by a Big 8 (4) auditor during the year. Several variables capture the economic determinants of changes in ETR. These variables include changes in liquidity (CHG_LIQUID), profitability (ROA), intangible assets (CHG_INTAN), research and development expense (CHG_RD), capital intensity (CHG_CAPX) and the book-to-market ratio (CHG_BM). I also include the level of certain variables that are associated with changes in the ETR but which I expect to be stickier. These variables firm size (SIZE), leverage (LEV), firm complexity (COMPLEX) as defined by the number of segments in which the firm operates, growth (GROW). All changes are calculated as the change in value from year-to-date Q3 to Q4. I winsorize all continuous variables at the one percent level. I include industry and year fixed effects and cluster standard errors by firm (Cameron and Miller 2013). 4 In future work, I plan to incorporate a continuous variable to capture the decrease of the ETR during Q4. Page 10

12 After separately estimating the model in equation (1) on the group of suspicious and nonsuspicious firms, the Blinder-Oaxaca decomposition will separate the difference in the expected likelihood of decreasing the ETR in Q4 between the two groups into an explained and unexplained portion. The explained portion represents the difference in behavior between the groups due solely to differences in the underlying characteristics included in the regression (e.g., Q3_ETR, SIZE, FOR, COMPLEX, etc.) For example, it may be that suspect firms have higher Q3 ETR on average, which would allow them to potentially decrease the ETR more during Q4. This natural difference in group characteristics then explains the difference in behavior. The unexplained portion of the difference is the portion attributable to earnings management. To calculate the unexplained portion, the decomposition imposes the coefficients from the regression of the non-suspect firms onto the suspect firms. If the difference in behavior between the two groups is due solely to differences in characteristics, then I expect the explained portion of the decomposition to account for the entire difference in results. If the underlying characteristics between the two groups are the same and it is differences in the regression coefficients between the two groups, then I expect the difference in underlying behavior to be attributable to the unexplained (managed portion). It is important to note that for this technique to properly decompose the difference in likelihood of decreasing the ETR during Q4, the model in equation (1) must be properly specified. If important observable firm characteristics are omitted from this equation, then inferences as to the difference attributable to the explained or unexplained portions may be flawed. Page 11

13 IV. Results 4.1 Descriptive Statistics Table 2 presents descriptive statistics for the sample. Panel A displays the statistics for the entire sample, whereas Panel B displays the statistics for suspect firm-years (firms that beat their analyst consensus EPS forecast by $0.02 or less) and non-suspect firm-years. (Insert Table 2 here.) Focusing on Panel A, the average likelihood of the ETR decreasing during Q4 is 53.1% in the entire sample, suggesting that whether or not firms are in danger of missing their earnings threshold, the likelihood of a decrease in the ETR during Q4 is about 50%. When separating suspect and non-suspect firms, I find that the likelihood of the ETR decreasing during Q4 is 57.5% for suspect firms and 52.4%, and this difference is significant at p < The average Q3 ETR for the entire sample is 29.5%, and I find that the Q3 ETR is higher for suspect firms (31.8%) than non-suspect firms (29.1%), consistent with suspect firms potentially being able to manage their ETR more than non-suspect firms (Comprix et al. 2012; Dhaliwal et al. 2004). Interestingly, the suspect firms in my sample have higher growth and ROA but are less complex than the non-suspect firms, which suggests that perhaps these firms are younger or in a different stage in their lifecycle. 4.2 Regression Results Table 3 presents the results from the Blinder-Oaxaca decomposition. Panel A shows the results from the regression of the determinants on the likelihood of experiencing an ETR decrease during Q4. Column (1) presents the results of the regression on the non-suspect firms and Column (2) presents the results of the regression on the suspect firms. Both models fit the data reasonably well with Adjusted R-squared of over 10%. (Insert Table 3 here.) Page 12

14 Consistent with the statistics presented in Table 2, Panel B shows that 52.4% of nonsuspect firms experience a decrease in the ETR during Q4, whereas 57.5% of suspect firms experience a decrease in the ETR during Q4. Thus suspect firms are 5.05% more likely to experience a decrease in the ETR during Q4. The decomposition further reveals that 3.9% of this difference is explained due to differences in underlying firm characteristics. Thus, 77% (3.94/5.09) of the difference in likelihood of decreasing the ETR in Q4 between suspect and non-suspect firms is expected, whereas only 23% (1.15/5.09) of the difference is due to earnings management. This finding is robust to various definitions of suspect and non-suspect firms, as well as differences in the specification of the regression models. Therefore, I conclude that although management of the ETR in Q4 to beat an earnings target is occurring, over three-quarters of the difference in likelihoods of decreasing the ETR in Q4 is expected. V. Conclusion I utilize the Blinder-Oaxaca decomposition technique to disentangle how much of the movement in the ETR during Q4 at suspect firms is due to earnings management as the literature suggests (Cazier et al. 2014; Comprix et al. 2012; Dhaliwal et al. 2004; Frank and Rego 2006; Gleason and Mills 2008; Krull 2004; Mauler 2016; Schrand and Wong 2003) versus expected decreases in the ETR during Q4. I find that the majority of the movement in the ETR during Q4 is attributable to expected movements rather than earnings management. This finding suggests that differences in firm characteristics and accounting for income taxes in interim periods is at least partially responsible for observed decreases in the ETR during Q4. Page 13

15 Appendix A For simplicity purposes, I assume that a firm operates in two jurisdictions: Jurisdiction A has a 30% tax rate and Jurisdiction B has a 10% tax rate. At the beginning of the year, the firm expects to report pre-tax income of $100 in Jurisdiction A and $100 in Jurisdiction B, yielding a total tax expense of $40 and an estimated ETR of 20% ($40/$200). Jurisdiction A Jurisdiction B Total Pre-tax income Tax expense Net income Jurisdiction tax rate 30% 10% Annual Estimated ETR 20% Thus, the firm s quarterly and year-to-date ETR are as follows: Q1 Q2 Q3 Q4 Annual Pre-tax income Tax expense Net income Quarterly ETR 20% 20% 20% 20% Year-to-date ETR 20% 20% 20% 20% 20% Change in total amount of expected annual income: In the second quarter, the tax department of the firm receives updated profit projections. I first assume the firm now expects to only report $70 of pre-tax income in Jurisdiction A for the year. Thus, the firm now expects to incur income tax expense of only $21 in Jurisdiction A, yielding a total tax expense of $31 for the year ($21 + $10) and an estimated ETR of 18.2% ($31 / $170). Jurisdiction A Jurisdiction B Total Pre-tax income Tax expense Net income Jurisdiction tax rate 30% 10% Annual Estimated ETR 18% Page 14

16 Thus, the firm s quarterly and year-to-date ETR are as follows: Q1 Q2 Q3 Q4 Annual Pre-tax income Tax expense Net income Quarterly ETR 20% 17% 18% 18% Year-to-date ETR 20% 18% 18% 18% 18% Change in the expected allocation of annual income: Alternatively, I assume that during the second quarter, the firm expects to make the same pre-tax income for the year ($200) but has shifted its expectations about the allocation of income between Jurisdiction A and Jurisdiction B. The firm now expects to report income of $60 in Jurisdiction A and $140 of income in Jurisdiction B. Thus, the total amount of income expected at the firm has not changed ($200), but the firm now expects to incur only $18 of tax expense in Jurisdiction A and $14 of tax expense in Jurisdiction B, yielding a total tax expense of $32 and an estimated annual ETR of 16% ($32/$200). Jurisdiction A Jurisdiction B Total Pre-tax income Tax expense Net income Jurisdiction tax rate 30% 10% Annual Estimated ETR 16% Thus, the firm s quarterly and year-to-date ETR are as follows: Q1 Q2 Q3 Q4 Annual Pre-tax income Tax expense Net income Quarterly ETR 20% 13% 16% 16% Year-to-date ETR 20% 16% 16% 16% 16% Page 15

17 Discrete item in Q2: Finally, I assume that during the third quarter, the firm filed the tax return for Jurisdiction A related to the prior year and realized that its actual taxable income was $40 more than it had estimated, resulting in $12 more in tax expense ($40*30%). The firm is required to treat this event as a discrete item and recognize the benefit fully in the third quarter. Q1 Q2 Q3 Q4 Annual Pre-tax income Tax expense Net income Quarterly ETR 20% 20% 35% 20% Year-to-date ETR 20% 20% 27% 26% 26% Page 16

18 References Amberger, H., E. Eberhartinger, and M. Kasper Tax rate biases in tax planning decisions: Experimental evidence. Working paper, Vienna University of Economics and Business. Blinder, A Wage discrimination: Reduced form and structural estimates. The Journal of Human Resources 8, Bratten, B., C. Gleason, S. Larocque, and L. Mills Forecasting taxes: New evidence from analysts. Working paper, the University of Kentucky, the University of Iowa, University of Notre Dame and the University of Texas at Austin. Burgstahler, D. and I. Dichev Earnings management to avoid earnings decreases and losses. Journal of Accounting and Economics 24, Cameron, A. C. and D. Miller A practitioner s guide to cluster-robust inference. Working paper, University of California Davis. Cazier, R., S. Rego, X. Tian, and R. Wilson Did FIN 48 stop last chance earnings management through tax reserves? The Journal of the American Taxation Association 34, Chen, K. and M. Schoderbek The 1993 tax rates increase and deferred tax adjustments: A test of functional fixation. Journal of Accounting Research 38, Comprix, J., L. Mills, and A. Schmidt Bias in quarterly estimates of annual effective tax rates and earnings management. Journal of the American Tax Association 34, Degeorge, F., J. Patel, and R. Zeckhauser Earnings management to exceed thresholds. Journal of Business 72, Dhaliwal, D., C. Gleason, and L. Mills Last-chance earnings management: Using the tax expense to meet analysts forecasts. Contemporary Accounting Research 21, Donelson, D., C. Koutney, and L. Mills Nonrecurring income taxes: Do analysts and investors identify and adjust for transitory tax expense items? Working paper, the University of Texas at Austin. Donelson, D., J. McInnis, and R. Mergenthaler Discontinuities and earnings management: Evidence from restatements related to securities litigation. Contemporary Accounting Research 30, Page 17

19 Frank, M. M. and S. Rego Do managers use the valuation allowance account to manage earnings around certain earnings targets? Journal of American Taxation Association 28, Gleason, C. and L. Mills Evidence of differing market responses to beating analysts targets through tax expense decreases. Review of Accounting Studies 13, Graham, J., C. Harvey, and S. Rajgopal The economic implications of corporate financial reporting. Journal of Accounting and Economics 40, Henry, E. and R. Sansing Data truncation bias and the mismeasurement of corporate tax avoidance. Working paper, the University of Tennessee, Knoxville and Dartmouth University. Kim, S., A. Schmidt, and K. Wentland Analysts and taxes. Working paper, University of Massachusetts Boston, North Carolina State University, George Mason University. Jann, B The Blinder-Oaxaca decomposition for liner regression models. The Stata Journal 8, Krull, L Permanently reinvested foreign earnings, taxes, and earnings management. The Accounting Review 79, Mauler, L The role of additional non-eps forecasts: Evidence using pre-tax forecasts. Working paper, Florida State University. Oaxaca, R Male-female wage differentials in urban labor markets. International Economic Review 14, Plumlee, M The effect of information complexity on analysts use of that information. The Accounting Review 78, Powers, K., A. Schmidt, J. Seidman, and B. Stomberg Examining which tax rates investors use for equity valuation. Working paper, the University of Tennessee, Knoxville, North Carolina State University, the University of Virginia, and the University of Georgia. pwc Guide to accounting for income taxes. Second edition. Schmidt, A. The persistence, forecasting, and valuation implications of the tax change component of earnings. The Accounting Review 81, Schrand, C. and M. H. F. Wong Earnings management using the valuation allowance for deferred tax assets under SFAS 109. Contemporary Accounting Research 20, Page 18

20 Weber, D Do analysts and investors fully appreciate the implications of book-tax differences for future earnings? Contemporary Accounting Research 26, Zang, A Evidence on the trade-off between real activities manipulation and accrual-based earnings management. The Accounting Review 87, Page 19

21 Table 1: Sample Selection Firm-year observations with all financial data 135,528 Less: Observations missing an analyst consensus EPS forecast in IBES (93,677) Less: Observations where ETR is less than 0 or greater than 1 (5,259) Less: Observations in either the utility or finance industry (8,630) General Sample 27,962 Table 1 details my sample selection. I begin with all firm-year observations in Compustat between 1993 and 2015 that have sufficient data to calculate all of my control variables. EPS is actual earnings-per-share from IBES at the end of year t. I compare the consensus EPS estimate, calculated as the median of all EPS forecasts within a year of the earnings announcement for year t, to the EPS actual reported in IBES to determine if the firm is within two cents of its target. ETR is the annual GAAP effective tax rate, calculated as tax expense (TXT) divided by pre-tax income (PI) for year t. Page 20

22 Table 2: Descriptive Statistics Panel A: Full Sample Variable N Mean Std Dev Q1 Median Q3 DEC_Q Q3_ETR NON_RECURR QTR_VOL BIGN CHG_LIQUID ROA CHG_INTAN CHG_RD CHG_CAPX CHG_BM SIZE LEV COMPLEX GROW Page 21

23 Panel B: Samples separated into suspect and non-suspect firms Table 2: Descriptive Statistics (cont.) Non-suspect firm-years Suspect firm-years Variable N Mean Std Dev Q1 Median Q3 N Mean Std Dev Q1 Median Q3 t-test DEC_Q *** Q3_ETR *** NON_RECURR *** QTR_VOL *** BIGN *** CHG_LIQUID *** ROA *** CHG_INTAN *** CHG_RD ** CHG_CAPX *** CHG_BM *** SIZE ** LEV *** COMPLEX *** GROW *** Page 22

24 Table 3: Results from the Blinder-Oaxaca Decomposition Panel A: Reuslts from the regression of determinants on DEC_Q4 (1) (2) Q3_ETR *** *** NON_RECURR *** QTR_VOL DEC_Q *** *** BIGN CHG_LIQUID ROA *** * CHG_INTAN * CHG_RD CHG_CAPX CHG_BM *** SIZE *** ** LEV COMPLEX *** GROW * * Adj R N 24,102 3,682 Panel B: Results from the decomposition Non-suspect firms Suspect firms Difference Explained Unexplained Page 23