Option Compensation and Tax Avoidance: Substitution Effect or Change in Behavior?

Transcription

1 Option Compensation and Tax Avoidance: Substitution Effect or Change in Behavior? Jeri K. Seidman The University of Texas at Austin Bridget Stomberg The University of Texas at Austin January 5, 2011 Desai and Dharmapala (2006) document a negative relation between option compensation and tax sheltering. They provide a model in which tax sheltering and rent extraction are complementary activities that are both decreased when manager incentives are appropriately aligned. We test their positive feedback theory as well as an alternative theory based jointly on the accounting treatment of option compensation and the substitution between tax minimization strategies. Although our hypotheses tests provide some support for both theories, additional descriptive analyses are consistent with our alternative theory. Overall, our results suggest that the negative relation between option compensation and tax sheltering is more likely due to the joint effect of accounting treatment and the firm substituting option compensation for alternative tax minimization strategies, rather than from conservative tax planning arising from a complementarity between rent extraction and tax sheltering. Preliminary and Incomplete Please do not quote without permission of the authors. Acknowledgements: We would like to thank Lillian Mills for valuable comments and suggestions. Stomberg thanks the AICPA Foundation for financial assistance through the Accounting Doctoral Scholarship Fund as well as her summer paper committee members: Robert Freeman, Lillian Mills and Michael Williamson.

2 I. Introduction Because firm managers choose the amount of corporate income to report to taxing authorities, Desai and Dharmapala (2006) conjecture that stock-based incentive compensation should be associated with tax sheltering. 1 Agency theory suggests that stock-based compensation aligns managerial and shareholder interests, and incentivizes managers to enhance firm value. Assuming tax sheltering increases expected future cash flow, conventional wisdom suggests tax shelters should enhance firm value. Therefore, stock-based compensation and tax sheltering should be positively associated. However, Desai and Dharmapala (2006) find a negative relation between stock-based compensation and tax sheltering, and provide an alternative theory of positive feedback effects between tax sheltering and rent diversion. Their theory characterizes tax shelters as complex transactions that allow managers to divert rent. As managerial incentives become more closely aligned with shareholder interests, managers are less likely to divert rents and, therefore, their demand for the tax shelters that facilitate rent diversion decreases. Hanlon and Heitzman (2010) comment that the theory from Desai and Dharmapala (2006) that rent extraction and tax avoidance require complementary technologies is an interesting angle and is underexplored. Deriving much direction and theoretical framework from Desai and Dharmapala (2006), this paper answers the call in Hanlon and Heitzman (2010) by further exploring this somewhat counterintuitive relation between managerial stock compensation and tax avoidance. We successfully replicate the previously documented negative relation between stock option compensation and tax avoidance from and in a longer 1 Desai and Dharmapala (2006) do not explicitly define tax sheltering in their paper, but instead rely on one generally accepted view of tax shelters as complex transactions lacking valid business purpose. However, their empirical measure of tax sheltering is consistent with a broader definition that includes legal tax minimization strategies. 1

3 period extending through Next, we test the positive feedback theory outlined in Desai and Dharmapala (2006) by focusing on how option compensation affects changes in tax sheltering behavior. Finding mixed results, we then examine an alternative explanation for the documented negative relation. Finally, we examine whether the negative relation holds over different time periods, firms and industries. The positive feedback theory outlined in Desai and Dharmapala (2006) suggests a change in tax sheltering behavior. Proposition 2 in Desai and Dharmapala (2006) posits that an increase in manager-shareholder alignment will decrease rent extraction and tax sheltering, assuming certain interdependencies between the costs of sheltering and diversion. Grants of stock options increase manager-shareholder alignment, so we study the relation between grants of option compensation and changes in tax sheltering. Additionally, depending on when the stock option is granted and on the complexity of implementing or unraveling the selected tax shelter, the negative relation may persist beyond the current period. We test the effect of option compensation on changes in tax sheltering, both contemporaneously and in the subsequent period. We find a negative and significant relation only when testing the contemporaneous effect of options on tax sheltering. Stock options granted this period are insignificantly related to the change in tax sheltering in the subsequent period, even after controlling for contemporaneous grants. These results suggest that although option compensation decreases contemporaneous tax sheltering, it is not related to future tax sheltering, even though executives tend to hold granted options for several years after grant due to vesting restrictions. Overall, our results are mixed with respect to the positive feedback theory outlined in Desai and Dharmapala (2006). Next, we build on the proposal in Graham, Lang and Shackelford (2004) that tax shields are substitutes and outline an alternative explanation for the negative relation observed between 2

4 option compensation and tax avoidance. 2 Prior to SFAS 123R, firms were not required to record option compensation expense. As a result, the tax benefit received upon option exercise was not reflected in the income tax accrual. Thus, ceteris paribus, firms with large tax deductions for option compensation report a smaller book-tax gap than firms that engage in alternative, nonconforming tax minimization strategies. 3 Due to the substantial tax benefits received from option compensation deductions, the substitution effect suggests that firms with significant amounts of option deductions are less likely to engage in other tax minimization strategies because their taxable income is already low. Taken together, the conforming treatment of option compensation prior to SFAS 123R and the substitution effect between stock-based compensation deductions and alternative non-conforming tax minimization strategies could produce a negative empirical relation between option compensation and tax sheltering. We refer to the combined effect of conforming accounting treatment and the substitution between option compensation and alternative non-conforming tax minimization strategies as the joint explanation. The joint explanation outlined above suggests that the negative relation documented in Desai and Dharmapala (2006), which was documented during a period of steadily increasing stock-option usage and accounting treatment under APB 25, should weaken as the conforming accounting treatment of option compensation ends, increasing the book-tax gap for firms with significant stock-option usage. 4 We study the impact of accounting treatment by separating the 2 Desai and Dharmapala (2006) allude to the substitution effect in their discussion of tax exhaustion. However, they do not consider the substitution effect together with conforming accounting treatment as an alternative explanation, but rather test the effect of tax exhaustion on their sample selection. 3 We refer to the treatment of option compensation outlined in ABP 25 as conforming because by omitting option compensation expense from pre-tax income and the tax benefit of the deduction from tax expense, the book-tax gap appears smaller than if a non-conforming tax minimization strategy (such as a lease-in-lease-out transaction) had been used. For further discussion, see Hanlon and Shevlin (2002) who study 100 NASDAQ firms and find that these firms report relatively high effective rates and small book-tax gaps while paying very low taxes due to high stock option deductions. 4 The substitution explanation also suggests that the negative relation should weaken as option compensation usage declines and firms select alternative, potentially non-conforming tax minimization strategies. Following Desai and 3

5 sample into two subgroups: a group that expenses option compensation for book purposes and a group that does not. We find a significant negative relation for the APB 25 group and an insignificant relation for the SFAS 123R group. The difference in the relation between tax sheltering and stock-based compensation for the APB 25 and SFAS 123R sub-groups is weakly significant. 5 These results are consistent with the joint effect of conforming accounting treatment and substitution between alternative tax minimization strategies driving the negative empirical relation between stock-option compensation and tax sheltering. Given that we find some support for both the positive feedback story outlined in Desai and Dharmapala (2006) as well as the joint effect of 'conforming accounting treatment and substitution between tax minimization strategies outlined here, we next attempt to further distinguish between these hypotheses by examining when and for which types of firms the negative relation is strongest. First, we undertake a sub-period analysis broken into roughly equal periods that correspond with shifts in option usage and accounting treatment of options. Consistent with the joint effect explanation, the negative and significant relation between option compensation and tax sheltering is only documented in periods of both accounting conformity and high option usage. Next, we examine whether profitability moderates the negative relation between option compensation and tax sheltering. Desai and Dharmapala (2006) limit their sample to firms with positive estimated taxable income, but do not impose limitations based on the sign of pre-tax book income or the book-tax gap. Manzon and Plesko (2002) note that profitable firms are better Dharmapala (2006), we measure stock-based compensation as a continuous variable which partially tests this supposition. We do not test this supposition more directly because optimal levels of tax minimization differ across firms and it is therefore difficult to determine discrete breakpoints where we expect the substitution effect to change. 5 While an ideal test of the effect of the conforming accounting treatment would be to adjust both pre-tax book income and estimated taxable income from the non-conforming period to be as it would have been reported in a conforming period, machine-readable data does not fully allow this. Data limitations regarding stock-based compensation expense are discussed further in Section IV.B. 4

6 positioned than their non-profitable counterparts to benefit from tax sheltering. We find that results are sensitive to sample selection decisions regarding the levels of both book and tax profit. Although including (excluding) firms with negative estimated taxable income (negative prior year book-tax gap), does not affect the sign or significance of most results, results become insignificant when we exclude firms with negative pre-tax book income. To achieve a significant negative relation, the sample must include firms without income to shelter a sample selection requirement counter to the positive feedback theory outlined in Desai and Dharmapala (2006). Lastly, we examine the correlation between option compensation and tax sheltering by stock exchange and by industry. Graham, et al. (2004) provide evidence that the substitution effect between option deductions and interest deductions is found primarily in firms listed on the NASDAQ, likely because these firms use high percentages of stock options to compensate employees. Although we find that the correlation between option compensation and tax sheltering is most negative and significant among firms listed on the NASDAQ, we also find a negative and significant correlation for firms listed on the NYSE/AMEX. Raedy, Seidman, and Shackelford (2010) suggest tax minimization activities vary across industries. We find a significant negative correlation between option compensation and tax sheltering in only three industries computers, pharmaceutical, and transportation. Regression analysis suggests that the negative relation between option compensation and tax sheltering in the sample as a whole is significantly influenced by firms in the computer industry, where option usage is extremely high and the adjusted book-tax gap is on average negative, indicative of higher taxable income than book income. In fact, when these firms are removed, no significant relation is observed. Overall, our results suggest that the negative relation between option compensation and tax sheltering documented by Desai and Dharmapala (2006) is more likely due to the joint effect 5

7 of conforming accounting treatment and the firm substituting option compensation for alternative tax minimization strategies, rather than from conservative tax planning arising from a complementarity between rent extraction and tax sheltering. We find that the negative relation between stock option compensation and tax sheltering is not a generalizable phenomenon, but rather is driven by firms with irregular profitability and high option usage. The remainder of this paper is organized as follows. Section II provides background information on differences between book and taxable income, including a discussion of related prior literature. Section III details the sample selection and research design. Section IV presents results and Section V concludes. II. Background Publicly traded firms are required to report two separate measures of economic income. First, firms must compute and report book income in their financial statements in accordance with Generally Accepted Accounting Principles (GAAP). Second, firms must calculate and confidentially report taxable income to the IRS in accordance with the Internal Revenue Code (IRC). Mills and Plesko (2003) explain that the basic objectives of financial and tax reporting regimes create differences in income measurement. For example, GAAP aim to prevent overstatement of income to investors while the IRC seeks to prevent understatement of income to taxing authorities. Accordingly GAAP and the IRC establish distinct sets of rules to accomplish these objectives. In addition to the different underlying objectives and specific rules of GAAP and the IRC, differences between book and taxable income may also arise from earnings management 6

8 and tax sheltering. 6 Researchers use the difference between book and taxable income to infer information about these activities. For example, Desai (2003) documents a substantial increase in the book-tax gap during the late 1990s and presents evidence suggesting that the growing booktax gap is inconsistent with increased levels of earnings management but is consistent with increasing levels of tax sheltering. In his setting, tax sheltering has a negative connotation suggesting aggressive tax minimization strategies. The incentives and characteristics of managers may affect the level of tax avoidance undertaken. Armstrong, Blouin and Larcker (2010) find a negative association between tax director incentive compensation and financial effective tax rate (ETR). Robinson, Sikes and Weaver (2010) find that firms with tax departments evaluated as profit centers (versus cost centers) have lower ETRs. Dyreng, Hanlon and Maydew (2010) provide evidence that the educational and professional backgrounds of top executives are significant determinants of firm tax avoidance. Similarly, Desai and Dharmapala (2006) [DD hereafter] hypothesize that because managers are responsible for tax planning, managerial incentives should be associated with tax sheltering. It is not clear, however, whether managerial incentives should be positively or negatively associated with tax avoidance. Agency theory suggests that option compensation incentivizes managers to enhance shareholder value. Because tax sheltering can increase firm value, it follows that greater incentive compensation should result in more tax sheltering activity. Similarly, because rent diversion erodes shareholder value, greater option compensation should result in less rent diversion. 6 Tax sheltering exists on a continuum; tax minimization strategies can range from perfectly legal and even government-promoted strategies such as the use of accelerated depreciation and credits, to far more devious strategies that would be overturned if discovered. As used in this paper, tax sheltering means any activity that reduces a firm s tax liability, including both legal and illegal tax strategies. 7

9 However, DD posit that interactions between rent diversion and tax sheltering can overturn these anticipated results. Because tax shelters per se are typically complicated, they create opportunities for managers to manipulate earnings, mislead investors and divert rents. Therefore, as managers become less interested in rent diversion, they may simultaneously become less interested in tax sheltering. If rent diversion and tax sheltering are substitutes, then as the cost of one activity increases, the level of that activity will decrease and the level of the other activity will increase. However, if rent diversion and tax sheltering are complements, an increase in the cost of one activity will result in a simultaneous decrease in the level of both activities. 7 Because rent diversion and tax sheltering could be substitutes or complements, an increase in the manager s focus on firm value could be associated with increased or decreased tax sheltering. DD use various measures of managerial stock-based compensation as proxies for the importance the manager places on after-tax firm value and examine how increases in the manager s focus on after-tax firm affect the level of tax sheltering activity. DD study the relation between tax sheltering and option compensation from We extend the time period to determine whether the results hold in periods with different accounting treatment and option usage. Prior to SFAS 123R firms could account for option compensation in accordance with either SFAS 123 or APB 25. Under SFAS 123, firms expensed option compensation and included the tax benefit of the deduction in their tax accrual. Under APB 25, firms did not expense option compensation if the option had no intrinsic value to the recipient or cost to the firm at the time of issuance. This was typically true as long as the exercise price was at least equal to the current market price. Nearly all firms elected to follow APB 25, 7 DD conjecture that decreased rent diversion is direct and intended consequence of the increase in [the importance the manager places on after-tax firm value]. 8

10 resulting in accounting treatment for options that suggested book-tax conformity. 8 As Hanlon (2003) describes, because the tax benefit of options is omitted from tax expense, taxable income estimated from the financial statements is higher than taxable income reported to the IRS. During this period of conformity, firms with large option deductions and small non-conforming tax minimization strategies will report a smaller book-tax gap than firms with small option deductions and large non-conforming tax minimization strategies. Under SFAS 123R, firms must expense option compensation and record a deferred tax asset for the expected future tax benefit. Although the option compensation expense reduces pretax book income, the deferred tax benefit recorded does not affect estimated current period taxable income. Only as stock options are exercised and the tax deduction is realized is estimated taxable income affected. We are interested in whether the results in DD hold as many firms change their accounting treatment of stock options. III. Research Design A. Hypotheses The positive feedback theory outlined in Desai and Dharmapala (2006) explicitly predicts a change in behavior. Specifically, Proposition 2 states that under certain conditions, an infinitesimal increase in [the weight the manager places on firm value] leads to decreases in diversion and sheltering Additionally, if options incentivize managers to increase or decrease tax sheltering behavior, there may be a lag between the period options are granted and the period 8 In fact, whether or not they expense option compensation, firms often receive a deduction for option compensation, though the deduction is recognized in the year in which the option is exercised while any expense is recognized in the year in which the option is granted The tax deduction for option compensation usually differs from the book expense in amount as well as in the period it is recognized. Estimates of future book value, number of options exercised, etc. is required to calculate the expense for book purposes. The tax the deduction is the realization of these estimates. For firms using APB 25, the realized tax benefit reduces cash taxes paid and increases stockholders equity but has no effect on reported tax expense. 9

11 when changes in tax sheltering behavior are observed. For example, options granted later in the year may not have time to affect tax sheltering behavior in the current period. Our first hypothesis focuses on the change in observed tax sheltering; tests of this hypothesis allow for either concurrent or lagged effects. H1: Option compensation and changes in tax sheltering are negatively associated. Graham, et al. (2004) study two specific tax sheltering strategies available to firms-- deductions for stock option compensation and interest expense and find that they are substitutes for each other. In other words, as taxable income is successfully sheltered using one strategy, the value of additional shelters decreases. We extend this theory and hypothesize that option compensation (which appears as a conforming tax strategy if the company does not expense options) and other tax minimization strategies are substitutes. The substitution effect should weaken as option usage declines and the company uses other conforming and nonconforming activities to minimize tax expense. The effect should also weaken as accounting treatment of options becomes less conforming. When options are expensed in pre-tax income and the corresponding benefit is recognized in deferred tax expense, firms with large option deductions will have a larger book-tax gap relative to what it would be in the APB 25 period. Therefore, the negative relation observed by DD should be attenuated as firms begin to expense option compensation. H2: Option compensation and tax sheltering are negatively associated when option compensation is reported in accordance with APB 25. B. Sample We construct our sample using data on executive compensation from the Standard and Poor s Execucomp database for The availability of information on stock-based compensation is the largest sample constraint. Although Execucomp began collecting data in 10

12 1992, DD begin their sample in 1993 to increase the number of firms covered. We follow this approach and begin our sample in Execucomp compiles detailed information on executive compensation by firm-year, including a breakout of total compensation between salary, bonus, stock options and restricted stock awards. We use these data to calculate a measure of stockoption compensation that proxies for the degree to which manager and shareholder interests are aligned. Next, we use data from Compustat to calculate the empirical measure of tax sheltering developed by DD. Because firms tax returns are confidential, direct measures of taxable income and tax sheltering are not readily available. DD develop a measure to estimate tax sheltering activity using information from financial statements. They define tax sheltering as the residual obtained from regressing each firm s book-tax gap on total accruals. Lastly, following DD, we require estimated taxable income to be positive. Our final sample consists of 9,817 firm-year observations from 1,457 unique firms. Our sample almost doubles the years and firm-year observations included by DD. Table 1 details our sample selection criteria. [Insert Table 1 here.] During the sample period used in DD of 1993 through 2001, our sample size is 4,942, which is slightly larger than DD s sample of 4,702 observations. The difference is likely attributable to backfilling of the Execucomp database, which has made information on executive compensation available for more firm-years since the publication of DD. Appendix A provides additional information regarding our replication of DD during the period in their study. 11

13 C. Variable Definitions Stock-Option Compensation STKMIXGRANT captures the value of stock-option grants as a percentage of total compensation. Specifically, for each executive j of firm i in year t, we calculate STKMIXGRANT as follows: STKMIXGRANT i,t = ΣOPTIONS i,j,t (1) ΣOPTIONS i,j,t + ΣSalary i,j,t + ΣBonus i,j,t Salary and Bonus are Execucomp variables. Prior to the adoption of SFAS 123R, OPTIONS is defined as the Black-Scholes value of stock options granted, Execucomp variable Options_Award_Blk_Value. For fiscal years beginning after June 15, 2005, Execucomp introduced a new variable, Option_Awards, to measure the amount of compensation expense now required to be recognized for stock option grants in accordance with SFAS 123R; OPTIONS equals Option_Awards when it is populated. For all years in the sample, SEC rules required firms to disclose detailed compensation information for the chief executive officer, chief financial officer and the three other most highly compensated executives, for a total of five executives. However, the number of firm managers for whom compensation information is available in Execucomp may vary from year to year. In our sample, the number of managers for whom information is available in any given firm-year varies from two to fifteen managers, with a mean of six. Because the stock-option compensation measure is a ratio of stock-option compensation to total compensation for all executives for whom information is disclosed, the measure captures the same construct for all firms, regardless of the number of executives included. We obtain similar results including compensation data for only the top five executives of each firm. 12

14 Tax Sheltering Because firms are not required to publicly disclose information regarding taxable income, any empirical measure of taxable income obtained from publicly available financial information is imperfect. By extension, capturing true tax sheltering activity of any firm in a given year is equally difficult. We use the empirical measure developed by DD as our proxy for tax sheltering activity. This measure is the residual obtained from regressing a firm s book-tax gap in a given year on total accruals for that same year. DD define the book-tax gap (BT) as the difference between domestic pre-tax book income (PIDOM) calculated in accordance with GAAP, and an estimate of federal taxable income (TI), where TI=TXFED/35%. 9 While Hanlon (2003) discusses the difficulty in estimating taxable income from disclosed financial statements and we acknowledge the potential error in this measure, researchers without access to IRS records must contend with these inaccuracies. As calculated, TI can be positive or negative depending on whether TXFED is greater than or less than zero. Negative values of TI are assumed to represent an estimated taxable loss whereas positive amounts indicate estimated taxable income. Consistent with DD, we eliminate firms with an estimated taxable loss or with zero estimated taxable income (TXFED 0) from the sample; we remove this restriction in a later analysis. Following the calculation of the tax sheltering measure in DD, we regress the book-tax gap on total accruals to estimate the portion attributable to earnings management (as measured by total accruals) and the remaining portion which may include tax sheltering. Using an 9 This formulation of book income results in a significant number of firm-year observations being eliminated from the sample. PIDOM is missing for firms that don t separately disclose domestic and foreign income, for example if foreign income is not 5% of total pre-tax income. 13

15 approach similar to that outlined by Healy (1985), we calculate the measure of total accruals for each firm as follows: 10 TA t = -DP t -XI t + AR t + INV t - AP t - TXP t -(TXDFED t +TXDFO t ) (2) where DP t is depreciation expense in year t. XI t is is extraordinary items in year t, AR t is the accounts receivable balance in year t less the accounts receivable balance in year t-1. INV t is the inventory balance in year t less the inventory balance in year t-1. TXP t is the taxes payable balance in year t less the taxes payable balance in year t-1. TXDFED t is deferred federal income tax expense and TXDFO t is deferred foreign income tax expense, both in year t. We then regress the book-tax gap, scaled by lagged assets, on total accruals from Equation (2) to estimate the portion of book-tax gap explained by tax sheltering activity. We use ordinary least squares regression and include firm fixed effects to obtain the following specification: BT i,t = β 1 TA i,t + µ i + ε i,t. (3) By removing the effect of earnings management from the book-tax gap, we obtain an estimate of tax sheltering activity. Specifically, TS = µ i + ε i,t, (4) where µ i captures time-invariant and unmeasured firm characteristics and ε i,t. represents deviation from µ i in each year, t. IV. Results A. Descriptive Statistics Descriptive statistics are shown in Table 2. Relative to the period studied in DD, firms became more profitable on average (pre-tax domestic income (PIDOM) increases). Despite this 10 Desai and Dharmapala (2006) use various measures of accruals including total, abnormal and discretionary accruals with similar results. 14

16 increase in pre-tax income, the value of option compensation, as measured by STKMIXGRANT, is nearly identical to that reported in DD. Taxable income (TI) is significantly lower than PIDOM in both the DD sample period and the extended period we study. The effect of adjusting the book-tax gap (BT) for total accruals to calculate tax sheltering (TS) is negligible. 11 [Insert Table 2 here.] B. Tests of the Positive Feedback Theory Desai and Dharmapala (2006) outline a behavioral explanation for the observed negative relation between option compensation and tax sheltering that hinges on rent extraction and tax sheltering being complementary activities. They posit that managers use complicated tax shelters to extract rent but that as the incentives of managers and shareholders become more aligned, rent extraction will decrease, along with tax sheltering. This theory is therefore best tested by examining the change in tax sheltering induced by option compensation. Additionally, because tax sheltering strategies may take some time to implement and because stock option backdating is a concern in the early period, we allow for the option grant to affect the current year as well as the following year. Table 3 presents results of our changes analysis. Column (1) presents a negative and significant relation between concurrent option grants and changes in tax sheltering. The results in Column (1) can be interpreted as providing evidence that option compensation granted this period is associated with a decrease in tax sheltering this period. Columns (2) and (3) present results regarding the effect of option grants in the one period on changes in tax sheltering during the subsequent period. We find no significant relation between option grants and future changes in tax sheltering. The relation does not change when we control for concurrent option grants. The 11 Desai and Dharmapala (2006) do not report coefficients from their first stage regression or statistics on the unadjusted book-tax gap to allow for comparison. However, because we are able to replicate their results so closely, we are confident that TS is correctly specified. See Appendix A for further information on the replication. 15

17 positive feedback theory suggests a negative relation between both current period and prior period option grants because (a) options granted late in one period may affect behavior in the subsequent period and (b) tax sheltering often involves a great deal of planning and can take time to implement or unwind. Although Column (1) provides support for the positive feedback theory, Columns (2) and (3) do not. [Insert Table 3 here.] C. Tests of the Joint Effects Explanation Following Graham, et al. (2004), we outline an alternative explanation for the negative relation between TS and STKMIXGRANT documented in DD. We expect firms with high option usage to select fewer alternative tax minimization strategies. In periods with conforming accounting treatment for options, option compensation will not increase the estimated book-tax gap because the expense is excluded from pre-tax book income and the tax benefit is excluded from the income tax accrual. To test this joint effect, we separate the sample into firm-years where options were expensed and those where they likely were not. The joint effect should induce a negative relation in firm-years where options were accounted for under APB 25. When firms expense options, the estimated book-tax gap will increase; the negative relation should weaken. The ideal test of the joint effect would be to adjust the SFAS 123R figures to what would have been reported under APB 25. However, machine-readable data is incomplete in this regard because unless a firm discloses the stock option compensation expense separately on the face of the income statement, Compustat reports the expense as missing. Because our sample includes only firms currently using option compensation, we expect 100% of our firm-year observations subsequent to the adoption of SFAS 123R to record stock compensation expense. However, only 33.5% of the observations in our non-conforming sample period of

18 have either STKCO or TXBCOF populated in Compustat. Because of this data limitation, we are unable to adjust TS for observations with stock compensation expense to remove the effect of the expense and benefit. We instead separate the firm-years with option expense from those with no (or immaterial) option expense. We set EXP = 1 for firm-year observations subsequent to the effective date of -SFAS 123R, and for observations for which Compustat reports either stock option compensation expense (STKCO) or the excess tax benefit from stock options disclosed in the financing portion of the Statement of Cash Flows (TXBCOF). We acknowledge that the EXP = 0 group includes observations for firms expensed options prior to the effective date of SFAS 123R but did not disclose the expense on the face of the income statement. Table 4 presents results of our regression replicating DD form as well as testing H1. Column (1) documents a negative and significant relation between option compensation and tax sheltering, consistent with DD. Column (2) tests the effect of conforming accounting treatment using a binary variable EXP, which equals 1 when the treatment is deemed non-conforming. Columns (3) and (4) present separate regressions for EXP=0 and EXP=1 for ease of interpretation. Together, Columns (2) through (4) document a negative and significant relation between option compensation and tax sheltering in the sample reporting under APB 25 and a significantly less negative relation in the sample which reported under SFAS 123R. Table 4 provides support for the joint effect of conforming accounting treatment and substituting between tax minimization strategies. [Insert Table 4 here.] 17

19 D. Time, Profitability, and Industry Analysis In Sections B and C, we present some support for both the positive feedback theory of DD and the joint effect of conforming accounting treatment and substitution between tax minimization strategies outlined here. To further establish which theory is most influential in the observed negative relation between option compensation and tax sheltering, we study whether the relation varies with time or type of firm. Before we undertake these analyses, we provide a graphical analysis of the relation between tax sheltering and option compensation by partitioning the sample by decile on our two variables of interest, TS and STKMIXGRANT. For each decile, we calculate the mean and median of the other variable of interest. Panel A of Figure 1 presents the mean and median of STKMIXGRANT for each decile of TS. We find that the relation is roughly U-shaped, such that firms with the lowest values of TS have the highest option usage. Panel B presents the mean and median of TS for each decile of STKMIXGRANT. Panel B suggests an overall negative relation but one that appears significantly weaker at the median than at the mean. Overall, the graphical representations in Figure 1 suggest that the negative relation is driven by extreme observations of the tax sheltering variable among firms with highest levels of option compensation. [Insert Figure 1 here.] Time Analysis First, we conduct a sub-period analysis to determine during which periods the negative and significant results are present. We break our sample into four roughly equal sub-periods, considering both option usage and accounting treatment of options. Figure 1 plots STKMIXGRANT, the ratio of option compensation to bonus, salary and option compensation, throughout our sample period. The first ( ) and second ( ) sub-periods 18

20 approximately coincide with periods of increasing option usage and conforming accounting treatment. These two sub-periods also encompass the entire sample period in DD. The third subperiod ( ) marks increased expensing of stock option compensation, following scrutiny of option compensation created by accounting scandals such as Enron and Worldcom, as well as decreased option usage. 12 The final sub-period ( ) coincides with required expensing of stock option compensation under SFAS 123R. [Insert Figure 2 here.] Table 5 presents our fundamental regressions for the entire period and for the four separate sub-periods. Consistent with the joint explanation, the negative and significant relation between option compensation and tax sheltering is found only during the periods of increasing option usage and conforming accounting treatment. We find no significant relation in the subperiods or Results in Table 5 are consistent with the joint explanation, rather than the positive feedback theory. [Insert Table 5 here.] Profitability Analysis DD limit their sample to include observations with positive estimated taxable income without similar restrictions on the sign of pre-tax book income or the book-tax gap. However, the positive feedback theory should be most applicable to firms with both income to shelter and with a history of tax sheltering that can be mitigated by increases in option compensation. We revisit the sample selection decisions in DD and perform sub-sample analysis to determine for which firms the negative relation between option compensation and TS is strongest. Our profitability sub-sample analysis focuses on firms with (a) positive pre-tax income (suggesting profit that 12 The number of firms in the Compustat universe with non-missing and non-zero STKCO more than doubles between 2001 and 2002 and continues to increase nearly monotonically through

21 would taxable absent tax sheltering behavior) (b) a positive prior-year book-tax gap (suggesting that the firm was previously engaging in tax sheltering) or both. Table 6 presents results of these analyses. Although DD eliminate observations with negative estimated taxable income under the assumption that these firms do not have incentives to further shelter taxable income, we present results both with and without these observations. We include these observations because it is possible that these firms have negative estimated taxable income as a result of extremely successful tax sheltering 13 Column (1) presents results for the entire sample, without restrictions on pre-tax income or the book-tax gap. The positive feedback theory in DD suggests that the negative relation should be stronger in Columns (2)-(4) of Table 6 than in Column (1). Additionally, it suggests that the negative relation should be stronger in Column (4) (where both pre-tax income and the book-tax gap are positive) than in Column (2) (pre-tax income is positive) or Column (3) (the book-tax gap is positive) because observations in Column (4) have both pre-tax profit to shelter as well as demonstrated effective tax sheltering that can be mitigated with increased manager-shareholder alignment. [Insert Table 6 here.] Results in Table 6 are inconsistent with the positive feedback theory outlined by DD. The negative relation observed in Column (1) for the entire sample is not found in the sub-sample presented in Column (2) we observe no significant relation between option compensation and tax sheltering for profitable firms, regardless of the sign of estimated taxable income. Similarly, we observe no significant relation between option compensation and tax sheltering for firms that are both profitable and historically effective at sheltering taxes (Column (4)), suggesting that 13 Additionally, it is not obvious that firms with negative taxable income in any one year have no such incentive. For example, a single year taxable loss may be the result of successful tax minimization strategy resulting in net operating losses expected to be used in future years. 20

22 increased manager-shareholder alignment does not significantly affect the tax avoidance behavior of firms who are already undertaking tax avoidance. Overall, results in Table 6 suggests that the negative relation between tax avoidance activities and option compensation observed in DD is not driven by firms with high expected benefits from tax avoidance behavior and observed historical effective tax sheltering. Industry Analysis Following Graham, et al. (2004), who find that the substitution between option and interest deductions is strongest for firms listed on the NASDAQ, we divide the sample based on exchange. Panel A of Table 7 presents sample statistics for TS and STKMIXGRANT for each group, as well as the correlation between these two variables. Firms listed on the NASDAQ award a much higher percent of total compensation to their top employees in the form of stock option grants than do firms listed on other exchanges. We find that the correlation is negative and significant for firms listed on the NASDAQ and the NYSE/AMEX, though both the coefficient and the statistical significance are greater for firms listed on the NASDAQ. [Insert Table 7 here.] Next, following Raedy, et al. (2010), who find that the level and type of book-tax differences differ dramatically across industries, we divide the sample into the 14 industries outlined in Barth, Beaver, Hand and Landsman (1999). Panel B of Table 7 presents sample statistics for TS and STKMIXGRANT for each industry, as well as the correlation between these two variables. For all industries, median TS is positive, indicating that pre-tax book income exceeds taxable income for the median firm. However, mean TS is negative for two industries Computers and Insurance indicating that estimated taxable income exceeds book income for these industries, on average. The Computers industry also has the highest values of 21

23 STKMIXGRANT; for the average (median) computer firm, over 50% (55%) of executive compensation is comprised of options. The correlations shown in Panel B of Table 7 present an interesting picture of which types of firms exhibit a negative association between option compensation and tax avoidance activities. For the sample as a whole, only four of the fourteen correlations are significant; three of these are negative the Computers, Pharmaceutical and Transportation Industries. The median tax sheltering variable suggests that firms in the Computers industry report (estimated) taxable income nearly equal to pre-tax book income, which suggests an absence of tax sheltering behavior. However, given that these firms have very significant option usage and, therefore, large corresponding tax deductions, the correlations presented in Panels A and B of Table 6 are consistent with the joint effect of substitution and conforming accounting treatment. Finally, we run separate regressions for the three industries with significant, negative correlations. Table 8 presents these results. In Column (1) of Panel A, we find a negative and statistically significant relation between tax sheltering and option compensation for the Computers industry. The coefficient for the computer industry sub-sample is more than twice as large as the coefficient for the full sample regression presented in Column (1) of Table 3. We also find a negative relation for the Pharmaceutical industry and Transportation industry; these results are insignificant likely due in part to the small sample size. Excluding the observations in the Computers industry results in an insignificant relation between tax sheltering and option compensation for the remaining sample (Column (2) of Panel A), while excluding observations from either the Pharmaceutical or Transportation industries does not dramatically alter results in the remaining sample (Column (2) of Panels B and C, respectively). Results in Table 8 suggest that the large option usage in the Computers industry, coupled with the small, and sometimes 22

24 negative, book-tax gap drives the negative relation between option compensation and tax sheltering documented in Desai and Dharmpala (2006). [Insert Table 8 here.] V. Concluding Remarks This paper extends the work of Desai and Dharmapala (2006). We first confirm results in DD for the original sample period of and in a longer period extending through We first test the positive feedback theory outlined in DD using a changes analysis. DD posit that tax sheltering and rent extraction are complementary activities; managers with misaligned interests will desire to undertake rent extraction and will employ complicated tax sheltering structures to facilitate this extraction. Aligning the interests of managers should result in less rent extraction and therefore, less need for complicated tax sheltering structures. Results of the effect of option grants on changes in tax sheltering provide mixed support for the positive feedback theory. Next, we posit an alternative explanation for the negative relation that is driven by the substitution between deductions for option compensation and alternative non-conforming tax minimization strategies; option compensation appears to be a conforming strategy because it does not increase the book-tax gap. We test the joint effect of conforming accounting treatment and the substitution among tax minimization strategies by segregating firms based on whether they expense stock option compensation. Results provide support for our joint effect hypothesis. Finally, we undertake a more descriptive analysis to determine whether the relation between stock option compensation and tax sheltering differs with time or type of firm. Graphical analyses using deciles of tax sheltering and option compensation suggest that the 23

25 negative relation is driven by extreme observations of tax sheltering among firms with high option usage. The sub-period analysis shows that the negative relation is only significant during periods characterized by increasing option usage and conforming accounting treatment. Firms listed on the NASDAQ have a more negative and statistically significant correlation between option compensation and tax sheltering than to firms listed on the NYSE/AMEX or other exchanges; firms listed on the NASDAQ use significantly higher option compensation than do firms listed on other exchanges. We also find that the negative correlation is only significant in three industries computers, pharmaceutical, and transportation; firms in the computers and pharmaceutical industries report the highest percentage of option compensation of all industries. The descriptive analyses are all consistent with the joint effect explanation for the negative relation observed between option compensation and tax sheltering. Taken together, our results suggest that the negative relation between tax sheltering and option compensation found in DD is more likely due firms with high option deductions substituting away from alternative nonconforming tax minimization strategies than from conservative tax planning arising from the complementarity between rent extraction and tax sheltering. 24

26 References Armstrong, C., J. Blouin and D. Larcker. The Incentives for Tax Planning, University of Pennsylvania working paper, Barth, M., W. Beaver, J. Hand and W. Landsman, Accruals, Cash Flows, and Equity Values, Review of Accounting Studies 4(3-4): Desai, M., The Divergence between Book Income and Tax Income, In Tax Policy and the Economy 17: (ed. J. Poterba). Desai, M. and D. Dharmapala, Corporate tax avoidance and high-powered incentives, Journal of Financial Economics 79(1): Dyreng, S., M. Hanlon, and E. Maydew, The Effects of Executives on Corporate Tax Avoidance, The Accounting Review 85(4): Graham, J., M. Lang, and D. Shackelford, Employee Stock Options, Corporate Taxes, and Debt Policy, Journal of Finance 59(4): Hanlon, M., What can we infer about a firm s taxable income from its financial statements? National Tax Journal 56(4): Hanlon, M. and S. Heitzman, A Review of Tax Research, Journal of Accounting and Economics, forthcoming. Hanlon, M. and T. Shevlin, Accounting for tax benefits of employee stock options and implications for research, Accounting Horizons 16(1): Healy, P., The Effect of Bonus Schemes on Accounting Decision, Journal of Accounting and Economics 7(1-3): Manzon Jr., G., and G. Plesko, The Relation between Financial and Tax Reporting Measures of Income, New York University School of Law Tax Law Review 55(2): Mills, L. and G. Plesko, Bridging the Reporting Gap: A Proposal for More Informative Reconciling of Book and Tax Income, National Tax Journal 56(4): Raedy, J., J. Seidman. and D. Shackelford, Is There Information Content in the Tax Footnote? University of North Carolina working paper. Robinson, J., S. Sikes and C. Weaver, Performance Measurement of Corporate Tax Departments, The Accounting Review 85(3):