Does operational efficiency spill over onto the tax return?

Transcription

1 Does operational efficiency spill over onto the tax return? Allison Koester McDonough School of Business Georgetown University Terry Shevlin Merage School of Business University of California - Irvine tshevlin@uci.edu Dan Wangerin* Broad College of Business Michigan State University wangerin@bus.msu.edu Current Draft: December 9, 2013 First Draft: May 2013 Abstract We find managers with greater ability to efficiently utilize firm resources engage in greater tax avoidance. We define managerial ability as a manager s capacity to maximize the efficiency of [firm] resources used for revenue-generating purposes (Demerjian et al. 2012, p.1) and view cash outflows to the taxing authorities as an inefficient utilization of resources. Moving from the lower to upper quartile of managerial ability is associated with a 3.16 (4.39) % reduction in a firm s one-year (five-year) cash-based effective tax rate. Cross-sectional tests reveal that higherability managers structure investments in R&D, capital expenditures, and foreign operations more tax-efficiently. JEL Classifications: G30; M41 Keywords: tax avoidance; management style; managerial ability Acknowledgements: An early draft of this paper benefited from insightful comments by Bok Baik, Bill Baber, Michelle Hanlon, Dan Lynch, Sarah McVay, Ed Outslay, Kathy Petroni, Isabel Wang, Ryan Wilson, and brownbag participants at Georgetown University. We thank Peter Demerjian, Baruch Lev, and Sarah McVay for sharing their managerial ability data and Pete Lisowsky for sharing his tax shelter likelihood data. Koester acknowledges financial support from Georgetown University s Center for Financial Markets and Policy. *Corresponding author: 632 Bogue Street, N231 Business College Complex, East Lansing, MI Phone: (517) , Fax: (517) , wangerin@bus.msu.edu

2 1. Introduction This paper examines and finds evidence that managers with greater ability to efficiently utilize firm resources engage in greater tax avoidance. Taxes are a significant source of corporate cash outflows, ranging from 20 to 40 % of pretax income for most public U.S. companies (Dyreng, Hanlon, and Maydew 2008). As every dollar of taxes paid is a dollar that cannot be reinvested within the firm, paying taxes could be viewed as an inefficient utilization of resources. 1 Prior research shows that managers have significant influence in shaping firms tax avoidance policies. Dyreng, Hanlon, and Maydew (2010), hereafter DHM, use a manager fixed effects research design to demonstrate that unobservable manager characteristics have an economically significant effect on cash-based ETRs. DHM fail to find evidence of a systematic relation between observable manager characteristics (e.g., education, tenure, age, gender, etc.) and tax avoidance, leading the authors to conclude that the executive effects on tax avoidance appear to be idiosyncratic (p.1165). Managers ability to efficiently utilize firm resources represents a potentially important factor explaining how managers affect corporate tax avoidance. All else equal, we hypothesize a positive relation between managers ability to efficiently utilize firm resources and corporate tax avoidance. Higher-ability managers have a superior understanding of the environment in which their firm operates (Demerjian et al. 2012), suggesting they are more likely to identify and/or create tax planning opportunities, better aligning business and tax strategies. If higher-ability managers themselves do not possess taxspecific expertise, they can hire individuals or engage outside consultants who are skilled at 1 While firms do benefit from government expenditures (i.e., infrastructure maintenance, well-functioning legal systems enforcing contracts, public education systems, etc.) that are in part funded by corporate income tax collections, the free rider problem suggests firms could benefit from these government expenditures without paying income taxes. 1

3 identifying and implementing tax avoidance strategies (Dyreng et al. 2010). Higher-ability managers can also create a tone at the top throughout the organization that emphasizes maximizing revenues and minimizing costs. While the same incentives to reduce income taxes also exist to reduce operating costs, reductions in operating costs are more likely to negatively impact firm operations. For example, reducing advertising expenditures can reduce future sales, and reducing product input costs by purchasing inferior materials can result in lower quality products. Reducing employee compensation can hinder firms ability to attract and retain a highquality work force. Efficiently utilizing resources by reducing cash outflows to taxing authorities could be particularly appealing to higher-ability managers especially if such cash reductions do not directly affect a firm s operations (e.g., sales, product quality, or human capital). On the other hand, there are several reasons why we could fail to find evidence consistent with our hypothesis. The skills necessary to maximize revenues and effectively deploy firm resources could be distinct from the expertise needed to plan and implement tax avoidance strategies which busy managers could find too costly to obtain. Even if it is possible to hire individuals or engage consultants possessing this expertise, the direct costs of tax planning and implementation and the indirect (or non-tax) costs of tax avoidance (e.g., political costs, reputation concerns, financial reporting effects, etc.) could exceed the benefits. Therefore, it is possible we will fail to find a relation between managerial ability and tax avoidance. If higherability managers believe their time is better spent focusing on core operating decisions as opposed to tax avoidance strategies, or fail to structure their core operating decisions taxefficiently, it is possible that higher-ability managers engage in even less tax avoidance than their lower-ability peers. 2

4 We operationalize managers ability to efficiently utilize firm resources using the MA_SCORE measure developed in Demerjian, Lev, and McVay (2012). Demerjian et al. (2012) construct this measure using data envelopment analysis (DEA) which captures how efficiently managers convert corporate resources into revenues relative to their industry peers, controlling for firm-specific characteristics affecting firm efficiency. Demerjian et al. (2012) validate the MA_SCORE by documenting that the measure is strongly associated with manager fixed effects and has superior ability to explain stock market reactions to CEO turnovers and changes in future performance relative to other measures used in the literature (e.g., historical firm performance, CEO media mentions, CEO tenure, etc.). Using the MA_SCORE allows us to examine how a specific dimension of managerial ability the efficient utilization of firm resources is associated with corporate tax avoidance. In our main analysis, tax avoidance is operationalized using the cash-based effective tax rate (ETR), with lower cash ETRs indicating greater tax avoidance. 2 After controlling for known determinants of tax avoidance, year fixed effects, and firm fixed effects, we find a negative and significant association between managerial ability and cash ETRs. Our results suggest that managers ability to efficiently utilize firm resources has an economically significant impact on corporate tax avoidance. We find that moving from the lower to upper quartile of managerial ability is associated with a 3.16 (4.39) % reduction in a firm s one-year (five-year) cash-based ETR. Our results hold after considering several alternative explanations including using an industry- and size-adjusted cash ETR as well as controlling for the effects of incentive 2 We use a cash-based ETR measure (as opposed to a GAAP-based ETR measure) because we are interested in examining whether higher-ability managers are better able to generate resources to use in revenue-generating activities. A cash-based ETR measure captures permanent and temporary tax deferral strategies, both of which retain cash resources within the firm. Because a GAAP ETR measure is accrual-based, it does not reflect cash savings from temporary tax deferral strategies and is affected by non-cash accruals like changes in the valuation allowance, tax expense accrued for foreign earnings not considered permanently reinvested, unrecognized tax benefits, etc. 3

5 compensation and corporate governance on tax avoidance. Our results also hold after controlling for manager fixed effects, highlighting the ability of managerial teams (e.g., the joint effects of individual managers working together) to efficiently deploy firm resources as a new and economically important determinant of corporate tax avoidance. 3 To draw stronger inferences, we conduct a difference-in-difference test examining changes in tax avoidance surrounding CEO turnovers. This test allows us to explore whether replacing a lower-ability CEO with a higher-ability CEO is associated with an increase in tax avoidance (and vice versa). Using a turnover event to isolate the CEO s effect on a change in MA_SCORE and a change in tax avoidance also helps to rule out the possibility of correlated omitted variables driving our main findings. Holding all else constant, moving from the lower to upper quartile of managerial ability is associated with a 3.32 % decline in cash-based ETRs during the three years following a CEO turnover relative to the three years prior to the turnover. We conduct a set of tests to shed light on the channels through which higher-ability managers avoid taxes. While we find no relation between managerial ability and the likelihood of tax sheltering, we do find that higher-ability managers record larger liabilities for unrecognized tax benefits (UTBs) related to tax positions unlikely to be sustained upon tax return audit. These findings suggest that the reduction in cash ETRs associated with managerial ability is in part attributable to transactions with greater uncertainty in tax treatment. 3 A concurrent working paper by Francis, Sun, and Wu (2013) also studies the relation between managerial ability and tax avoidance using the Demerjian et al. (2012) ability measure. The authors find a negative relation between tax avoidance in period t and managerial ability in period t-1, leading the authors to conclude that higher-ability managers engage in less tax avoidance. We model tax avoidance and managerial ability concurrently, as a dollar of firm resources saved through tax avoidance can be immediately invested within the firm for revenue-generating activities. When we include both lagged and concurrent period MA_SCORE as independent variables in our model, we find a negative coefficient on MA_SCORE t (p < 0.01) and a positive coefficient on MA_SCORE t-1 (p < 0.01). The sum of the two coefficients is negative and significant (p < 0.01), consistent with higher-ability managers engaging in greater tax avoidance. 4

6 Prior literature demonstrates that firms operating and financing characteristics explain variation in tax avoidance and these characteristics reflect managers operating and financing decisions. Therefore, in our final set of tests we explore whether higher-ability managers make more tax-efficient operating and financing decisions given their firms operating and financing characteristics. For example, firms often undertake R&D and establish foreign operations for non-tax reasons. However, we expect higher ability managers to structure these operating decisions tax-efficiently (e.g., optimize the U.S. R&D tax credit, structure transfer prices to shift profits to low tax jurisdictions). We find that R&D activities, capital expenditures, leverage, and foreign operations are important avenues through which tax avoidance is achieved by higherability managers. Transfer pricing and R&D credits are the two most commonly cited areas of tax position uncertainty firms report on Schedule UTP (IRS 2013), consistent with these channels reducing firms cash ETRs and increasing their UTBs. Identifying the avenues through which higher-ability managers are able to achieve tax avoidance is an important contribution to the literature. In addition, these cross-sectional findings help to further rule out correlated omitted variable concerns, as these correlated omitted variables would have to explain our main results as well as all significant interactions in our cross-sectional tests. We contribute to the tax literature by identifying a new and economically significant determinant of tax avoidance. With the exception of DHM, prior studies model tax avoidance as a function of firm-level characteristics and fail to take into account the influence of individual managers (Gupta and Newberry 1997; Mills 1998; Rego 2003). Our study answers the call by Hanlon and Heitzman (2010) to further explore the manager effect on tax avoidance by explicitly considering the impact of individual corporate decision-makers on corporate tax strategies. We compliment and build on the findings reported by DHM in several ways. First, 5

7 while a manager fixed effects research design captures only unobservable and time-invariant manager characteristics, we identify an observable and time-varying managerial characteristic associated with managers tax avoidance decisions. 4 Second, while directional predictions are not possible in a fixed effects research design, the Demerjian et al. (2012) MA_SCORE allows us to make directional predictions regarding the relation between this dimension of managerial ability and tax avoidance. Third, a manager fixed effects research design requires observing managers moving across multiple firms over time, limiting studies (and potentially their inferences) to a relatively small sample. The Demerjian et al. (2012) ability measure allows us to provide large-sample evidence on the relation between managerial ability and tax avoidance for a broad set of firms over a long period of time. Finally, Fee, Hadlock, and Pierce (2013) raise questions about the research design employed in management style studies by showing standard F-tests for joint significance of manager fixed effects do not provide valid statistical inferences. Our study is not subject to the econometric concerns raised by Fee et al. (2013). We also add to the literature linking managerial ability to financial reporting quality, firm characteristics, and economic outcomes. Specifically, prior research has shown managerial ability to be associated with greater earnings persistence, accruals quality, and the frequency and information content of management forecasts (Baik, Farber, and Lee 2011; Demerjian, Lev, Lewis, and McVay 2013). Other research finds managerial ability to be associated with a lower likelihood of bankruptcy (Leverty and Grace 2012), greater employment opportunities for executives (Fee and Hadlock 2003; Rajgopal, Shevlin, and Zamora 2006), and more efficient 4 In a recent study, Law and Mills (2013) find that CEO military experience is associated with a 1-2% reduction in one-year cash and GAAP ETRs and explains about 4% of the variation in manager fixed effects on corporate tax avoidance. Using stock option backdating as a proxy for personal tax aggressiveness, Chyz (2013) shows the presence of executives engaging in this type of behavior is positively associated with the likelihood of tax sheltering. While these studies examine time-invariant individual characteristics, we study a time-varying characteristic that captures management teams ability to efficiently utilize firm resources. 6

8 investments in labor (Jung, Lee, and Weber 2013). 5 Our findings provide new insights into the managerial ability literature by identifying a relation between managerial ability and corporate tax avoidance. Our findings should be of particular interest to board members when considering the costs and benefits of hiring executives, as we find that managerial ability affects not only firm operations but also the efficiency with which firms internally generated resources are utilized to obtain tax savings. The remainder of the paper is organized as follows. Section 2 discusses related literature and our hypothesis. Section 3 describes our sample and empirical method. Section 4 presents our empirical results and Section 5 concludes. 2. Related Literature and Hypothesis Development 2.1 Tax Avoidance, Management Style, and Managerial Ability Following Hanlon and Heitzman (2010), we broadly define tax avoidance as the reduction of explicit taxes (p.137). Thus, we assume tax avoidance includes the effects of tax savings from all activities in which the firm engages (e.g., real activities that are tax-advantaged, identifying and capitalizing upon tax planning opportunities, and targeted tax benefits from lobbying and political connectedness). Researchers have used a variety of proxies to capture tax avoidance. Some measures are broad in nature (e.g., cash-based ETRs, GAAP-based ETRs, and book-tax differences) while others are designed to capture certain types of transactions (e.g., permanent discretionary book-tax differences, unrecognized tax benefits, and tax sheltering). Hanlon and Heitzman (2010) note that most tax research in corporate tax avoidance focuses on firm characteristics as determinants (Gupta and Newberry 1997; Rego 2003; Wilson 2009; 5 Note that only Demerjian et al. (2013) and Jung et al. (2013) use the managerial ability measure from Demerjian et al. (2012). 7

9 Lisowsky 2010), and the authors identify research that considers the impact individual corporate decision-makers have on a firm s tax avoidance strategies as a gap in the tax literature. A separate stream of research has begun to examine the effect of individual managers on corporate decisions. Bertrand and Schoar's (2003) study of managers impact on corporate financial policy and investment decisions (e.g. dividends, capital expenditures, and mergers and acquisitions) serves as the foundational paper in what is referred to as the management style literature. The Bertrand and Schoar (2003) management style research design involves tracking individual managers who move across multiple firms over time and uses manager fixed effects to capture the influence of individual managers unobservable characteristics on corporate decisions. The management style framework has since been extended to other settings where researchers have examined the relation between management style and voluntary disclosure decisions, earnings quality, and the intersection of financial reporting and managerial decision making (Bamber, Jiang, and Wang 2010; Ge, Matsumoto, and Zhang 2011; Dejong and Ling 2013). DHM extend the management style framework to a tax setting and show a managerspecific effect on corporate ETRs using a manager fixed effects research design. A recent paper by Fee, Hadlock, and Pierce (2013) calls into question the causal role of managerial style. These authors fail to find a relation between managerial style and corporate policies in CEO turnovers likely to be exogenously determined (e.g., planned retirements due to CEO age, sudden illness, and death). Moreover, Fee et al. (2013) demonstrate that standard F- statistics for a test of joint significance of manager fixed effect coefficients used in prior studies could be econometrically invalid in detecting the presence of significant individual management style effects. Specifically, the authors highlight that standard asymptotic theory does not apply to tests where variables are highly serially correlated because the properties of standard F-tests for 8

10 joint significance are unknown (Wooldridge 2002). Highlighting these concerns, Fee et al. (2013) find F-tests reveal very significant manager style effects after randomly assigning CEOto-CEO movers to a different hiring firm than the one he actually joined. These authors suggest a more nuanced role for individual manager styles to affect corporate policies that depends on both (1) time-invariant unobservable idiosyncratic effects and (2) time-varying observable individual characteristics. By using a time-varying individual management characteristic such as MA_SCORE, we are able to produce potentially more reliable inferences relative to a manager fixed effects research design. Demerjian et al. (2013) modify the management fixed effects research design by replacing manager fixed effects with the measure of managerial ability from Demerjian et al. (2012) to assess the impact of individual managers on earnings quality. Specifically, Demerjian et al. (2013) show that managerial ability is positively (negatively) associated with accruals quality and earnings persistence (restatements), suggesting that higher-ability managers are better able to make complex accounting judgments and estimates that reflect the underlying nature of the firm s transactions. Replacing manager fixed effects with the Demerjian et al. (2012) MA_SCORE allows us to (1) examine an observable manager characteristic, (2) develop and test directional hypotheses, (3) produce valid inferences generalizable to a broader population with respect to the relation between managers ability to efficiently utilize firm resources and the variable of interest. 2.2 Hypothesis We predict that, all else equal, managers possessing greater ability to efficiently utilize firm resources will engage in greater tax avoidance. Following Demerjian et al. (2012), we 9

11 define managerial ability as how efficiently managers are able to maximize revenues using their firms limited set of resources. By reducing cash outflows to taxing authorities through tax avoidance, managers retain resources within the firm. If higher-ability managers objective is to enhance firm value, naturally they focus on maximizing profit and cash flow, and a reduction in taxes increases profit and after-tax cash flow dollar-for-dollar. While tax avoidance strategies that defer tax payments to future periods do not reduce income tax expense as reported in firms financial statements, temporary tax deferral strategies allow firms to allocate cash in different ways that can have an indirect effect on profit and cash flow. For example, managers with greater ability to efficiently utilize resources could increase profit indirectly by investing cash retained through tax deferral strategies towards revenue-generating activities such as increased advertising or production. In addition, financially constrained firms could use temporary tax deferral strategies to generate cash as a less costly substitute for debt financing (Edwards, Schwab, and Shevlin 2013). However, there are several reasons why we could fail to find evidence consistent with our prediction. First, the managerial skills necessary to maximize revenues could be very distinct from the skills required to identify tax planning opportunities. Even if higher-ability managers can identify and implement tax planning opportunities, executives could lack the specialized training and expertise required to implement the tax avoidance strategies and the effort to acquire this type of expertise could be too costly for high-ability executives to obtain. In addition, the (potentially uncertain) benefits of tax avoidance could not exceed the costs (e.g., direct costs of tax planning and implementation and non-tax costs associated with avoidance like political costs, reputation concerns, etc.), resulting in higher-ability managers choosing to engage in less tax 10

12 avoidance. Thus, the relation between managerial ability and tax avoidance is an empirical question. In light of the discussion above, one might wonder how the skills needed to maximize revenues from a given set of inputs translate to tax avoidance. One possibility is that managers themselves (e.g., CFOs or tax directors) possess the technical skills to implement successful tax avoidance strategies. Even if higher-ability managers lack these skills, they possess a superior understanding of their firm s business environment, enabling them to better identify tax avoidance opportunities that can be implemented by outside consultants or internal staff (Dyreng et al. 2010). Managers with superior ability to efficiently utilize firm resources can also create a tone at the top that emphasizes maximizing revenues and minimizing costs. The same incentives to reduce income taxes also exist to reduce operating costs, but the consequences of reducing operating costs could differ. While cutting operating expenditures such as marketing or product costs could be viewed as efficiently using limited firm resources, cuts to marketing can reduce sales and cuts to product inputs can yield lower quality products. In contrast, reducing taxes has no direct effect on a firm s sales or product quality, suggesting the reduction of taxes relative to other types of costs could be particularly appealing to higher-ability managers. 3. Research Design 3.1 Main Analysis In the management style literature, the dependent variable of interest is regressed on a set of year, firm, and manager fixed effects (Bertrand and Schoar 2003; Bamber et al. 2010; Dyreng et al. 2010; Ge et al. 2011; DeJong and Ling 2013). Year fixed effects capture the average impact of unobservable time-variant economy-wide characteristics on the dependent variable, and firm 11

13 and manager fixed effects capture the average impact of unobservable time-invariant characteristics of the firm and the manager. To further isolate the manager effect, the model can also be expanded to include additional variables that capture the impact of observable timevarying firm characteristics on the dependent variable. Demerjian et al. (2013) modify this research design to examine the relation between managerial ability and earnings quality by replacing manager fixed effects with the Demerjian et al. (2012) managerial ability score (MA_SCORE). We adopt this approach to examine the relation between managers ability to efficiently utilize firm resources and tax avoidance. Specifically, we construct our model based on DHM and substitute MA_SCORE for manager fixed effects when estimating the following equation: [1] TaxAvoid it = α 0 + β 1 MA_SCORE it + Controls it + Year fixed effects + Firm fixed effects + ε it A finding of 1 < 0 is consistent with managers with greater ability to efficiently utilize firm resources engaging in greater tax avoidance. 6 When estimating Equation 1 and in all other tests, we estimate standard errors clustering at the firm level Dependent Variables Our primary dependent variable of interest, CASHETR, is firm i s cash-based ETR in year t. There are a wide range of proxies used to capture tax avoidance, and researchers are cautioned to select the proxy most appropriate for their particular research question of interest (Hanlon and Heitzman 2010). DHM examine the impact of manager fixed effects on both cashbased and GAAP-based ETRs. Although our research question builds upon DHM, the main 6 We thank Peter Demerjian, Baruch Lev, and Sarah McVay for making their MA_SCORE publicly available at: 12

14 focus of our study is on the cash-based ETR because this measure best captures tax strategies that retain resources within the firm that can be invested in revenue-generating activities. Specifically, the cash ETR reflects both permanent and temporary tax avoidance, as well as the tax effects of uncertain tax positions unlikely to be sustained upon tax return audit. In contrast, GAAP-based ETRs do not reflect tax savings from temporary deferral strategies that reduce cash tax payments and increase firm resources available for deployment in revenue-generating activities. For example, the GAAP-based ETR also does not reflect the (1) temporary tax savings from common book/tax differences like accelerated depreciation; (2) tax benefits from taking uncertain tax positions that could not be sustained upon audit; and (3) future tax benefits a firm offsets with a valuation allowance. Following DHM, CASHETR is measured as cash taxes paid as a percentage of pre-tax book income before special items. Consistent with prior literature, we require observations to have positive cash taxes paid and positive pre-tax book income before special items, and we winsorize CASHETR values at zero and one. We specify our cash-based ETR as a 1-year measure which is appropriate for our tests because MA_SCORE is also constructed at the firmyear level. In addition, our research design closely follows DHM who also use a 1-year cashbased ETR measure. However, Dyreng et al. (2008) note that one-year cash-based ETRs are not strong predictors of long-run cash-based ETRs, suggesting a one-year measure can be a noisy proxy for long-run corporate tax avoidance. While some of this concern with respect to our study is alleviated by the Dyreng et al. (2008) finding that low one-year cash-based ETRs are more persistent than high one-year cash-based ETRs, we also employ 3-year and 5-year cash-based ETR measures in our empirical analysis. CASHETR3 (CASHETR5) is defined as the sum of cash taxes paid in years t through t+2 (t+4) divided by the sum of pre-tax income before special 13

15 items in years t through t+2 (t+4). We also measure MA_SCORE and the other control variables contemporaneously in our long-run tax avoidance analyses, averaging each variable over the same corresponding 3-year and 5-year windows. 7 Later in the paper we examine if higher-ability managers engage in tax avoidance strategies at the riskier and uncertain end of the tax avoidance spectrum. To capture the likelihood of engaging in tax-sheltering (a riskier tax strategy), we re-estimate Equation 1 using the predicted probability of tax shelter activity (PRED_SHELTER) from Lisowsky (2010) as the dependent variable. To capture the amount of tax avoidance generated by uncertain tax positions (an uncertain tax strategy), we re-estimate Equation 1 using actual and predicted unrecognized tax benefits (UTB and PRED_UTB, respectively) as the dependent variable. For brevity, we relegate the definitions of these variables to the Appendix and expand more on what these alternative measures capture in Section Test Variable We use the variable MA_SCORE, developed in Demerjian et al. (2012), to capture managers ability to efficiently deploy firm resources. The intuition underlying this measure is based on how efficiently managers can convert resources (e.g., capital, labor, and intangible assets) into revenues relative to the firm s industry competitors, with higher-ability managers being able to generate a higher rate of output from a given set of inputs. Demerjian et al. (2012) use data envelopment analysis (DEA) to estimate firm efficiency at the industry-year level by comparing the sales generated by each firm conditional on a vector of inputs (cost of goods sold, 7 We require the sum of cash taxes paid and pretax income before special items to be positive over these 3-year and 5-year windows. We also reset CASHETR3 and CASHETR5 values less than zero to zero and values greater than one to one. 14

16 SG&A expenses, net PP&E, net operating leases, net R&D, goodwill, and other intangibles). 8 Specifically, the authors conduct DEA at the industry-year level (where industry is defined using the Fama-French 48 classifications (Fama and French 1997)) by solving the following optimization problem shown below in equation 2a: [ ] The DEA optimization determines a firm-specific vector of optimal weights on the seven input variables by comparing the inputs of firm i to the inputs of all other firms within the same industry-year and computes a firm efficiency score θ which takes a value between 0 (for the least efficient firms) and 1 (for the most efficient firms). To isolate the portion of the efficiency score attributable to the managerial team, Demerjian et al. (2012) then estimate the Tobit regression shown below in equation 2b, regressing firm efficiency scores by industry on a set of firm-level characteristics. After taking into account the effect of firm-level characteristics that explain firm efficiency, MA_SCORE is constructed using the unexplained portion of θ (e.g., the information in the residuals) as a measure of managers ability to efficiently convert internal firm resources into revenues. 9 [2b] θ it = α 0 + β 1 Ln(TotalAssets it ) + β 2 MarketShare it + β 3 PositiveFreeCashFlow it + β 4 Ln(Age it ) + β 5 BusinessSegmentConcentration it + β 6 ForeignCurrencyIndicator it + Year Fixed Effects + ε it Demerjian et al. (2012) validate their measure in multiple ways and also demonstrate the measure is superior to other proxies for managerial ability used in prior literature (e.g., historical stock returns, accounting-based performance, and CEO media citations). First, they correlate MA_SCORE with CEO pay and stock returns. Second, they show a positive stock market 8 Note that inputs and outputs are measured in pre-tax dollars, mitigating the concern that MA_SCORE and tax avoidance are mechanically related and biasing against finding a significant relation between these variables. 9 See Demerjian et al. (2013) Appendix A for additional information regarding Equation 2b. 15

17 reaction to CEO turnover announcements when a higher-ability CEO replaces a lower-ability CEO (and a negative stock market reaction when the new CEO has lower ability). Finally, the authors find that hiring a CEO with greater (lesser) ability than the firm s former CEO is associated with improvements (declines) in future firm performance Control Variables We include time-varying firm-level characteristics known to be associated with tax avoidance as control variables in Equation 1 to reduce the possibility that MA_SCORE is simply capturing the effect of these characteristics on tax avoidance. Following DHM, we include the following tax avoidance control variables in Equation 1: research and development expense (R&D), advertising expense (AD), capital expenditures (CAPX), leverage (LEV), foreign operations (FOREIGN), firm size (SIZE), and intangible assets (INTANG). Prior research generally finds that cash-based ETRs are decreasing in R&D, CAPX, LEV, FOREIGN, and INTANG, and increasing in AD (Chen et al. 2010; Dyreng et al. 2010; Rego and Wilson 2012). We also include a variable that captures net operating loss utilization (NOL_DECREASE) and predict a negative relation between this variable and cash-based ETRs. 10 All variables are defined in the Appendix and all continuous variables are winsorized at the 1 st and 99 th percentiles. In addition, we also include year fixed effects and firm fixed effects in Equation 1, eliminating the possibility that MA_SCORE picks up the effects of either macro-economic characteristics that affect all firms in a particular year or firm characteristics that do not vary over time. Thus we 10 Rego and Wilson (2012) use pre-tax return on assets as a proxy for tax planning when modeling tax avoidance. In addition, accounting-based measures similar to pre-tax return on assets have been used as proxies for managerial ability (e.g., Baik et al. 2011). We argue that higher-ability managers engage in tax planning strategies that enable the more efficient deployment of firm resources. Thus, we do not include pre-tax return on assets in Equation 1 to avoid controlling for the effects of tax planning captured by MA_SCORE. Untabulated tests reveal that pre-tax return on assets and MA_SCORE are positively correlated (Spearman coefficient=0.33). When we include the variable as an additional control in Equation 1, the coefficients on pre-tax return on assets and MA_SCORE are both negative and significant (p<0.01). 16

18 identify the relation between corporate tax avoidance and managerial ability using firm-specific variation across time in the two variables. 3.2 Isolating the Effect of a Single Manager versus the Managerial Team One potential limitation of the MA_SCORE measure is that it captures both managers ability to efficiently utilize firm resources and all firm characteristics not explicitly included in Equation 2b. Demerjian et al. (2012) address this limitation by demonstrating that MA_SCORE is associated with manager fixed effects, reducing the concern that the residual simply captures firm characteristics omitted from Equation 2b. To provide stronger evidence of MA_SCORE capturing the effects of individual managers, we employ a difference-in-difference research design that exploits CEO turnovers to further rule out the concern that MA_SCORE is capturing firm and not manager characteristics. If MA_SCORE does in fact capture a manager effect and not a firm effect, we should observe a change in tax avoidance after a new CEO with differing ability joins a firm. In contrast, we should fail to observe a significant difference in the relation between changes in tax avoidance and managerial ability following CEO turnovers if MA_SCORE simply captures firm characteristics. 11 We modify Equation 1 and estimate the following regression to examine the association between changes in tax avoidance and changes in managerial ability following CEO turnovers: [3a] ΔCASHETR3 it = α 0 + β 1 ΔMA_SCORE3 it + β 2 TURNOVER it + β 3 ΔMA_SCORE3 it *TURNOVER it + ΔControls it + Year fixed effects + ε it 11 We focus on CEOs and not CFOs in this analysis for two reasons. First, CEOs generally have the most influence setting the tone at the top, the effects of which cascade down to other managers within the organization (Feng, Ge, Luo, and Shevlin 2011). Second, we rely on Execucomp data to identify managerial turnover events. Execucomp collects data relating to the five most highly compensated employees, which includes the CEO in almost every firm. In contrast, other types of executives (e.g., CFOs) are not uniformly among the five most highly compensated employees across Execucomp firms over time, resulting in the potential misclassification of a treatment firm (e.g., a firm with a CFO departure) as a control firm (e.g, a firm without a CFO departure). 17

19 The dependent variable in Equation 3a, ΔCASHETR3, is defined as the difference between the 3-year cash ETR in t+1 through t+3 and the 3-year cash ETR in t-3 though t-1 for firm i. ΔMA_SCORE3 is the difference between firm i s managerial ability score summed over t+1 through t+3 (which reflects the new CEO s ability) and firm i s managerial ability summed over t-3 though t-1 (which reflects the prior CEO s ability). We add the indicator variable TURNOVER, which is set equal to one if a CEO departs firm i in year t and zero otherwise. We omit the turnover year from our analysis to control for the fact that CEOs depart on various dates throughout year t. We also restrict the sample to exclude any observations with more than one CEO turnover occurring from t-3 through t+3. We are interested in the coefficient on the interaction of ΔMA_SCORE3 and TURNOVER, and β 3 < 0 is consistent with a higher-ability CEO engaging in greater tax avoidance relative to his lower-ability predecessor (or a lower-ability CEO engaging in less tax avoidance relative to his higher-ability predecessor). The control variables are the same as presented in Equation 1 but measured as the difference between their values summed from t+1 through t+3 and their values summed from t-3 though t-1. We denote these controls adding the prefix Δ and the suffix 3 to each variable (i.e., ΔR&D3, ΔAD3, ΔCAPX3, ΔLEV3, ΔFOREIGN3, ΔSIZE3, ΔINTANG3, and ΔNOL_DECREASE3). We also control for the effects of potentially correlated omitted variables related to CEO turnover and tax avoidance by using propensity score matching (PSM). Following prior literature (e.g., Desai et al. 2006; Skaife et al. 2013), we estimate a logistic regression modeling the probability of a CEO departure in year t as a function of firm characteristics including size (SIZE), leverage (LEV), growth (GROWTH and BTM), performance (INDROA), and the monitoring environment (ANALYST and INST_OWN) using Equation 3b shown below. See the 18

20 Appendix for detailed variable definitions. We include MA_SCORE in Equation 3b to ensure the ability of the managerial team in year t is not a correlated omitted variable. [3b] TURNOVER it = α 0 + β 1 SIZE it + β 2 INDROA it + β 3 GROWTH it + β 4 LEV it + β 5 BTM it + β 6 ANALYST it + β 7 INST_OWN it + β 8 MA_SCORE it + Year fixed effects + ε it After estimating the probability of CEO turnover, we match without replacement each treatment firm (TURNOVER=1) with a control firm (TURNOVER=0) using a caliper of three %. By using PSM, we impose no assumptions about the functional form of the relation between the selection variables with the outcome variable. However, the results from PSM can be sensitive to the variables included in the regression used to estimate propensity scores. Therefore, we also use OLS to estimate Equation 3a on the sample of Execucomp firms with data available to estimate Equation 3b. The difference-in-difference design we use in both the PSM and OLS estimations controls for the effects of any time-invariant correlated omitted variables that are not included in the models. In addition, using PSM mitigates the Fee et al. (2013) critique that managerial style is observable only after endogenous turnover decisions, as PSM controls for factors found to be associated with forced CEO dismissals. 4. Findings 4.1 Sample and Summary Statistics Table 1 shows the distribution of firm-years in our sample. We begin our analysis in 1994 to allow for the adoption of SFAS 109 (i.e., a consistent financial reporting regime) and end our analysis in 2010 because this is the latest year in which the MA_SCORE data is available for the full set of Compustat firms at the time of our study. We require firm-year observations to have non-missing values for the variables required to estimate Equation 1, yielding a sample of 44,616 firm-year observations that are approximately evenly distributed across our 17-year time period. 19

21 Table 2 reports descriptive statistics for our regression variables. The distribution of CASHETR and the control variables are comparable with descriptive statistics reported in prior studies (Dyreng et al. 2010; Edwards et al. 2013). The mean CASHETR value is 27.5 %, with an interquartile range of 8.8 to 36.9 %. Consistent with Dyreng et al. (2008), values for our long-run ETR measures (CASHETR3 and CASHETR5) are higher than the one-year cash-based ETR measure. The firm-years in our sample are smaller (SIZE) than those in DHM s sample because we use firms from the Compustat universe while DHM limit their analysis to Execucomp firms. 12 Table 3 presents the Pearson correlations between our variables of interest. We find a negative and significant correlation between MA_SCORE and CASHETR, CASHETR3, and CASHETR5 (p<0.01), consistent with our prediction. 13 We also find that MA_SCORE is positively and significantly correlated with proxies capturing the riskier and uncertain ends of the tax avoidance spectrum (e.g., PRED_SHELTER, UTB, and PRED_UTB). 14 Overall, the correlations between MA_SCORE and our tax avoidance proxies provide initial evidence consistent with our predictions. The majority of the control variables exhibit significant correlations with our tax avoidance proxies, highlighting the importance of controlling for these factors in our multivariate tests. 4.2 Main Analysis 12 While MA_SCORE values in our sample are slightly higher than those reported in Demerjian et al. (2012), this difference appears to be primarily attributable to eliminating observations from our sample with negative pre-tax income before special items to calculate the cash ETR. Our sample firms have greater profitability relative to the Demerjian et al. (2012) sample, which corresponds to higher managerial ability. The standard deviation of MA_SCORE is comparable to Demerjian et al. (2012), suggesting MA_SCORE values are distributed similarly within our sample. 13 Consistent with our multivariate tests of long-run ETRs, we sum MA_SCORE over the 3-year (5-year) period corresponding to CASHETR3 (CASHETR5) and report these pairwise correlations in Table We discuss in more detail what underlying constructs these alternative proxies for tax avoidance capture and their relation to managerial ability later in the paper. 20

22 Table 4 presents the results from estimating Equation 1. In Column 1 of Panel A we present the results from a baseline model which regresses CASHETR on the control variables from DHM and year fixed effects. Most of the control variable coefficients are significant in the predicted direction. For example, we find that CASHETR is decreasing in research and development expenses (R&D), capital expenditures (CAPX), leverage (LEV), and NOL utilization (NOL_DECREASE). In Column 2 we add firm fixed effects to the model. While most of the time-varying firm characteristics become insignificant in Column 2, the adjusted R 2 increases from 1.26 to % relative to Column 1, consistent with firm fixed effects subsuming much of the explanatory power of these firm-level characteristics. When we include MA_SCORE in Column 3 its coefficient is negative and highly significant (p < 0.01), consistent with our prediction that higher-ability managers engage in tax avoidance that increases firm resources (in this case, cash) available for revenue-generating activities. The results also suggest that this dimension of managerial ability has an economically significant effect on tax avoidance. Using the coefficient estimates presented in Column 3 and holding all control variables constant, moving from the lower to upper quartile of managerial ability is associated with a 3.16 % reduction in a firm s 1-year cash-based ETR. In Panel B we re-estimate Equation 1 using cash-based ETRs measured over longer time periods. Column 1 presents our analyses using CASHETR3 as the dependent variable. Because CASHETR3 is measured over the period t through t+2, we also average the values of MA_SCORE and each control variable over the same time period in our regression analysis so our dependent and independent variables are measured concurrently. Column 2 presents the results from repeating this analysis using a 5-year cash-based ETR calculated over the period from t through t+4 (CASHETR5) and measuring MA_SCORE and all control variables over the 21

23 same 5-year window. In both columns, we continue to find a negative and significant coefficient on MA_SCORE (p<0.01), consistent with higher-ability managers engaging in tax avoidance strategies which reduce cash tax payments over the long-run. 15 We also continue to find that managerial ability has an economically significant effect on tax avoidance. Using the coefficient estimates presented in Column 1 (2) and holding all control variables constant, moving from the lower to upper quartile of managerial ability is associated with a 3.70 (4.39) % reduction in a firm s 3-year (5-year) cash-based ETR. 4.3 Robustness Tests Having provided evidence that managerial ability is associated with greater tax avoidance, we next perform several robustness tests to help rule out potential alternative explanations for our findings. The results of these tests are shown in Table 5 where we report the coefficient and t-statistics on MA_SCORE but suppress the coefficients and t-statistics on all control variables for brevity. One potential alternative explanation for our results is that the negative association between MA_SCORE and cash-based ETRs is driven by firms with greater resources available for tax planning that also belong to industries with lower effective tax rates. Balakrishnan, Blouin, and Guay (2012) highlight the importance of industry membership and resources available for tax planning in assessing the extent of a firm s tax avoidance. To ensure our primary results are not driven by firms belonging to industries with lower ETRs and with greater resources available for tax planning, we re-estimate Equation 1 using the Balakrishnan et al. (2012) industry-size adjusted cash ETR measure. Specifically, we adjust CASHETR for firm i in 15 Untabulated results confirm our findings are robust to using a GAAP-based ETR measure as the dependent variable in Table 4. Using a one-year GAAP ETR yields a MA_SCORE coefficient of (p<0.05), while threeand five-year GAAP ETRs yield MA_SCORE coefficients of (p<0.01) and (p<0.01), respectively. 22

24 year t by subtracting the mean CASHETR corresponding to the same size quintile and industry. 16 Column 1 of Table 5 shows that we continue to find a negative and significant coefficient on MA_SCORE (p<0.01). Another possible alternative explanation for our main findings is that MA_SCORE is simply capturing better corporate governance. If better governance structures allow firms to identify, hire, and retain managers with higher ability to efficiently utilize resources, it could be that the corporate governance structure of a firm is driving our results. We conduct several robustness tests controlling for corporate governance to help rule out this possible alternative explanation. Because it is difficult to capture the construct of corporate governance with a single variable, we employ multiple proxies for corporate governance commonly used in the tax avoidance literature. Following Desai and Dharmapala (2006, 2009), we control for corporate governance using the %age of shares held by institutional owners (INST_OWN). In addition, we also control for corporate governance using the G-Index from Gompers et al. (2003) following Hanlon and Slemrod (2009) and Wilson (2009). Our finding that managers with greater ability to utilize firm resources engage in more tax avoidance is robust to the inclusion of these corporate governance controls. In Column 2 of Table 5 we find that INST_OWN is positively related to CASHETR (p<0.05) and the MA_SCORE coefficient remains negative and significant (p<0.01). In Column 3 we use G_INDEX as our governance proxy. Note that the inclusion of this variable reduces our sample by 75%, as G_INDEX is only available for S&P 1500 firms through The MA_SCORE coefficient again remains negative and significant (p<0.01) after controlling for corporate governance using these proxies, and we fail to find a relation between CASHETR and the G_INDEX (p>0.10). Replacing G_INDEX with the E-Index variable from Bebchuk et al. 16 Size quintiles are formed based on total assets and industry membership is defined using the Fama-French 48 industry classifications. 23