A Reexamination of U.S. Corporate Tax Avoidance over. the Past Twenty-Five Years: Estimating Corporate Tax. Avoidance with Accounting-Based Measures

Transcription

1 A Reexamination of U.S. Corporate Tax Avoidance over the Past Twenty-Five Years: Estimating Corporate Tax Avoidance with Accounting-Based Measures Preliminary Draft - Please Do Not Cite Noel P. Brock Eastern Michigan University Roy Clemons New Mexico State University Adam Nowak West Virginia University May 10, 2017 JEL Codes: F38, H25, H26 Eastern Michigan University, Department of Accounting & Finance, 406 Gary M. Owen Building Ypsilanti, MI 48197; noel@noelpbrock.com New Mexico State University, Accounting & Information Systems Department, 1320 E University Ave, Las Cruces, NM 88003; rclemons@nmsu.edu West Virginia University, Economics Department, 1601 University Ave, Morgantown, WV 26501; adam.d.nowak@gmail.com 1 We thank seminar participants at The American Taxation Association Midyear Meeting. We are very grateful to Terry Shevlin who provided useful comments.

2 Abstract ABSTRACT: Dyreng et al. (2017) find that the effective tax rates for both foreign and domestic corporations have steadily declined over the past quarter century. However, contrary to conventional wisdom, the authors also find that U.S. multinational corporations do not have a tax-based cost advantage relative to their domestic counterparts. We investigate this unexpected finding by reexamining corporate income taxes over the past quarter century employing an alternative tax avoidance measure developed by Henry and Sansing (2014). The authors measure addresses both sample selection bias and measurement error that exists when using income as the denominator when calculating effective tax rates. Using the Henry and Sansing (2014) measure of tax avoidance, we find that U.S. multinational corporations do have a tax-based cost advantage relative to their domestic counterparts. Thus, sample selection bias is a plausible explanation for the unexpected tax-based cost advantage of US domestic firms reported in prior research.

3 1 Introduction There is no shortage of anecdotal evidence that suggests multinational corporations (MNCs) take advantage of the low statutory tax rates in foreign countries. Despite this, Dyreng et al. (2017) examine effective tax rates (ETRs) over 25 years and find that US MNCs do not have a tax-based cost advantage relative to US domestic corporations.we explain this surprising result using a sample selection argument inherent to studies that use ETR as a measure of tax avoidance. In order to correct for the sample selection bias, we adopt the bias-free meaasure of tax avoidance developed in Henry and Sansing (2014). Using the Henry and Sansing (2014) measure of tax avoidance, we find that MNCs do have a tax-based cost advantage relative to purely domestic corporations. Broadly defined, ETR is a ratio of taxes to income and represents the total amount of taxes paid per dollar of income. Many studies use ETR as a measure of tax avoidance, but not all studies calculate ETR in the same way. Initially, researchers used financial statement tax expense and pre-tax financial accounting income (PI) as the numerator and denomoninator of ETR, respectively. More recently, Dyreng et al. (2008) put forth a Cash ETR where cash taxes paid is used in the numerator. Dyreng et al. (2008) argue that cash taxes paid is a more appropriate numerator when estimating ETRs as cash taxes paid: 1) captures tax avoidance through deferral strategies, 2) excludes the effects of financial statement accruals, and 3) includes the full tax consequences of employee stock options. Although the Cash ETR effectively addresses the shortcomings associated with the numerator in ETRs, Henry and Sansing (2014) argue that significant issues still exist in the denominator regardless of the choice of numerator. In particular, when using Cash ETR as a measure of tax avoidance, researchers must delete many observations where Cash ETR is undefined or not meaningful. To wit, Cash ETR cannot be calculated when pre-tax income is equal to zero, and Cash ETR is not easily interpreted when PI is negative but cash tax expense is positive. Hence, results from studies that exclude firm-years with non-positive PI are conditional rather than unconditional. In this study, we demonstrate that the uncondi- 1

4 tional results for all firm-years are in contrast to previously reported conditional results for firm-years with strictly positive PI. In order to address these issues related to the denominator, Henry and Sansing (2014) develop an alternative measure of corporate tax avoidance (HS). 1 In HS, the numerator is a measure of cash tax non-conformity defined as the difference between a firms cash taxes paid and the product of its pre-tax book income multiplied by the statutory rate. The authors then scale the numerator by firm size, measured as the market value of firm assets. This tax avoidance measure provides three distinct advantages over Cash ETR: 1) scaling by the market value of assets eliminates the need to discard or winsorize observations based on missing, negative, or outlying values of the dependent variable, 2) the asymmetric treatment of income and loss years is avoided, and 3) scaling by market value of assets allows for a direct comparison of large and small firms. To the extent that MNC tax advantages are more likely to result in loss years, we hypothesize that HS will avoid the sample selection bias inherent in ETR estimation. We provide evidence that restricting the sample to observations with useable Cash ETRs provides a biased estimate of the tax avoidance differential between domestic corporations and MNCs. In contrast to the results in Dyreng et al. (2017), we find that U.S. MNCs do have a tax-based cost advantage relative to U.S. domestic corporations when using the tax avoidance measure in HS. Our findings suggest that the unexpected results previously reported by Dyreng et al. (2017) may reflect empirical artifacts that remain after creating a viable data set when using Cash ETR as a measure of tax avoidance. The remainder of the paper is organized as follows. The second section provides a review of the literature that motivates our hypothesis. The third section presents our data and empirical models. The fourth section describes our univariate analyses. The fifth section describes our multivariate analyses, and the sixth section offers concluding remarks. 1 We use HS when referring to the tax avoidance measure developed in Henry and Sansing (2014) and Henry and Sansing (2014) when referring to the study, itself. 2

5 2 Literature Review 2.1 Measuring Corporate Tax Avoidance The academic literature has a long history of using ETRs to measure corporate tax avoidance going back to Surrey (1973). Typically, the numerator of the ETR was any measure of a corporations tax expense for financial reporting purposes, and the denominator was equal to a corporations PI. Although ETRs are widely used and easy to calculate, researchers argue that use of ETRs to measure tax avoidance presents significant theoretical and empirical challenges. For example, researchers often must either discard data where the ETR is undefined or winsorize a significant portion of data based on an ad hoc discomfort when an ETR falls outside of a particular range. In short, academic researchers argue there are issues with both the ETRs numerator, Dyreng et al. (2008), and denominator, Henry and Sansing (2014). 2.2 Dyreng et al. (2008): Cash ETR as a Tax Avoidance Measure Dyreng et al. (2008) develop a meaasure of the ETR based on a corporations ability to pay a low amount of cash taxes per dollar of PI. The authors use cash taxes paid instead of financial statement tax expense. The authors findings suggest that using cash taxes paid in the numerator offers the following advantages. First, cash taxes paid reflects tax avoidance through deferral strategies that are not reflected in financial statement tax expense. Second, financial statement accruals do not impact cash taxes paid. Third, cash taxes paid includes the full tax consequences of employee stock options. In addition to analyzing the numerator, Dyreng et al. (2008) analyze Cash ETR over one, five and ten year horizons. Using the longrun cash ETRs further mitigates the measurement error inherent in year-to-year fluctuations measuring taxes owed by corporations. 3

6 2.3 Changes in Cash ETRs over the past 25 years. Dyreng et al. (2017) perform an extensive study of changes in Cash ETRs over the past twenty-five years. the authors find that Cash ETRs have decreased significantly for both U.S. multinational corporations and domestic corporations. 2 On average, the findings suggest that Cash ETRs of U.S. corporations have decreased by approximately 10% over the sample period; in the final year of the sample, this 10% decrease is approximately equal to a decrease in corporate tax tax liability of $109 billion, Dyreng et al. (2017). Surprisingly, the authors also find the same percentage decrease in Cash ETRs over time for both purely domestic and MNCs. The authors examine several possible explanations for this surprising result. For example, they examine changes to 1) firm characteristics, 2) foreign statutory tax rates, and 3) regulatory and tax changes as possible explanations for the Cash ETR dynamics. The authors find that none of these possibilities can explain the initial finding that US domestic corporations have tax-based cost advantages relative to their MNC counterparts. In a follow-up study, Chyz et al. (2016) examine the relationship between implicit taxes and corporate tax avoidance. Implicit taxes are cross-sectional variations in pretax market returns that offset variations in the level of explicit taxes Scholes et al. (2015). That is, corporations invest more in tax-favored projects which drives up prices for the requisite inputs and drives down the profits. As such, implicit taxes offset tax savings from low explicit taxes, Chyz et al. (2016). The authors find that implicit taxes are higher, on average, for U.S. domestic corporations relative to U.S. MNCs. They further suggest a partial explanation for the unexpected finding: tax avoidance efforts of MNCs and domestic corporations are not different over the past 25 years. 3 Although the authors find that implicit taxes are correlated with effective tax rates, the authors do not find a significant difference between the effective 2 The authors also find the same declining trend in other measures of ETRs, including the total GAAP ETR, the current ETR, and the ratio of cash taxes paid to cash from operations. 3 Chyz et al. (2016) emphasize the difference between tax preferences and implicit taxes. Tax preferences are situations that reduce taxes (generally estimated by tax avoidance measures) while implicit taxes reduce the benefit of tax preferences Jennings et al. (2012). 4

7 tax rates of MNCs and domestic corporations. Overall, the prior literature examining the tax avoidance of U.S. corporations over the past quarter century is not consistent with the wide-held belief among policy makers that MNCs have explicit tax advantages. Although it is possible that legislators concentrated focus on MNCs tax avoidance is misguided, we suggest that the ETR proxies used in prior research may not effectively capture the difference in tax avoidance between MNCs and domestic corporations. In this paper, we extend the literature by testing for a difference in explicit taxes between MNCs and domestic corporations using the tax avoidance measure recently developed by Henry and Sansing (2014). 2.4 The Henry and Sansing (2014) Tax Avoidance Measure HS develop a tax avoidance measure to address issues that exist when the Cash ETR is used as a tax avoidance proxy. HS express the Cash ETR as the sum of the statutory tax rate and the effect of tax preferences on cash taxes paid ( ) scaled by pre-tax book income; here, is referred to as a measure of cash tax non-conformity, and is defined as the difference between a corporations cash taxes paid and the product of its pre-tax book income and the statutory tax rate. If > 0 ( < 0), the corporation is tax disfavored (favored). In order to make this difference comparable across firms, the authors then scale by firm size. Henry and Sansing (2014) use the market value of a corporations assets (MVA) as their measure of firm size. However, MVA requires the market value of equity (MVE) which presents a practical issue. Dropping firm-years with missing MVE decreases the number of firm-years in the sample by 8.7%. This decrease in the sample is non-negligible and would reflect an additional criteria for the sample not present in previous tax research. For this reason, we focus on scaling using book value of assets (AT) which does not decrease the sample size and is, in most cases, already used as a cleaning criteria in most studies. For completeness, we report results using both AT and MVA in the denominator. 5

8 Henry and Sansing (2014) argue that using of PI in the denominator of the Cash ETR creates limitations to the measurement of corporate tax avoidance. The authors argue that the Cash ETR creates statistical sampling bias and measurement error when it is used as a proxy for corporate tax avoidance. Specifically, Cash ETR is not calculated when pre-tax income is zero or negative; if the elimination of pre-tax income is systematically related to tax avoidance, studies using Cash ETR as a measure of tax avoidance are using biased samples. The HS tax avoidance measure addresses the statistical sampling bias that can exist when pre-tax book income is used in the denominator of the Cash ETR estimate. Since pre-tax income can be negative, its use in the denominator creates Cash ETRs which are difficult or impossible to interpret. As a result, researchers estimating the Cash ETR drop loss years, which results in a significant discarding of the overall population. HS argue that this discarding of the data is troublesome for two reasons. First, the literatures focus on the subsample of profitable corporations is driven not by the research question, but rather by the limitations of measuring the variable of interest. This non-random deletion of the sample observations based on realizations of the dependent variable generates a data truncation bias, Teoh and Zhang (2011). Ex post trimming of even a small fraction of a sample can lead to spurious effects, particularly in regressions where the independent variables possess low explanatory power, Henry and Sansing (2014). Second, discarding observations with negative PI induces an asymmetric treatment of income in loss years. This is important because, although loss-years comprise a significant percentage of the overall population of firms (18% of firm-years in the authors pre-recession sample), Hanlon and Heitzman (2010) note We do not have a very good understanding of loss firms, the utilization and value of tax-loss carryforwards, and how the existence of losses affects behavior (e.g., tax and accounting reporting and real decisions) of any of the involved parties. For example, consider a firm that experiences a loss in one year, but is profitable in the next. 6

9 Using the Cash ETR to measure tax avoidance for this firm is misleading because it treats the mere use of a net operating loss carryover as a tax preference by including the reduction in cash taxes paid in the year the loss carryover is used, while deleting the year in which the loss is generated from the dataset, Henry and Sansing (2014). Using HS as a measure of tax avoidance allows for the separate study of the tax avoidance incentives or activities of loss firm-years. The HS tax avoidance measure also mitigates measurement error that can exist when PI is used in the denominator of the Cash ETR estimate. Specifically, many tax preferences are related to fixed costs of the firm. Because fixed costs do not vary with output, tax preferences associated with fixed costs do not increase as PI increases. Therefore, holding fixed costs constant, firms with higher pre-tax profitability will have higher Cash ETRs, as the additional income increases cash taxes paid at the statutory rate. The reverse occurs if the firm has unfavorable book-tax differences that do not vary with PI. In each case, higher PI moves the Cash ETR toward the statutory rate. Henry and Sansing (2014) provide the following example to illustrate this point. Consider two corporations, each with $6 million of tax credits for increased research activities, and each facing a 35 percent statutory tax rate. Corporation A has PI=$60 million, and Corporation B has PI=$40 million. The corporations are the same size, face the same statutory tax rate, have the same dollar value of tax preferences, but have different levels of pretax profitability. The difference in pre-tax profitability results in Corporation A and Corporation B having different Cash ETRs of 25% and 20%, respectively. 4 The scaling bias induced by the use of PI in the denominator of a corporations ETR causes two corporations of the same size with the same amount of tax preferences to have different ETRs simply due to different levels of pre-tax profitability. In particular, higher profitability biases the ETR of Corporation A towards the statutory tax rate. In summary, by using MVA instead of PI in the denominator, Henry and Sansing (2014) 4 Corporation As Cash ETR = 25% = (.35*60-6)/60 and Corporation Bs Cash ETR = 20% = (.35*40-6)/40. 7

10 argue that their measure makes several contributions to the tax avoidance literature. Most important, their measure is defined for all firm-years regardless of PI. This feature mitigates the data truncation bias introduced when using the Cash ETR attributable to the asymmetric treatment of income and loss years. Thus, conclusions about tax avoidance using HS are generalizable to the entire Compustat population, as opposed to only strictly positive PI firm-years. In doing so, the authors argue that HS can provide an explanation for mixed findings regarding the association between tax avoidance and pre-tax profitability in prior studies. 2.5 Hypothesis Development Tax Avoidance of U.S. Multinational Corporations It is widely believed that U.S. MNCs are experiencing a growing income tax advantage over U.S. domestic corporations, Gravelle (2009); Drucker (2010); Senate (2014). Academic research indirectly supports this notion and suggest three reasons for this: 1) an increased importance of intangible assets (Grubert and Slemrod (1998); Kleinbard (2012); De Simone et al. (2014)) and 2) effective cross-border tax planning techniques are facilitating MNCs tax advantage over their domestic counterparts, Klassen and Laplante (2012); Dharmapala and Riedel (2013); Klassen et al. (2016); Dyreng and Markle (2015); Dharmapala (2014), and 3) multiple tax strategies that are simply unavailable to purely domestic firms, including strategies involving aggressive transfer pricing, hybrid entities, cost sharing agreements, intra-company debt agreements, and the deferral of offshore earnings (Dyreng et al. (2017)). Overall, the literature suggests that MNCs should experience greater income tax savings relative to their purely domestic counterparts. However, in an exhaustive empirical study using data from , (Dyreng et al., 2017) find find that U.S. MNCs do not have a tax-based cost advantage. We suggest that the sample selection bias discussed above can potentially explain the unexpected finding in (Dyreng et al., 2017) and reexamine the same data set using the HS measure of tax advantage and compare the tax avoidance of MNCs 8

11 compared to purely domestic corporations. 5 Alternatively, the HS tax avoidance measure allows us to test the following hypothesis regarding unconditional tax avoidance H1: all else equal, U.S. multinational corporations have a tax-based cost advantage relative to their domestic counterparts. 3 Estimation and Variable Description In order to test whether or not MNCs have a tax-based cost advantage, we regress measures of tax avoidance (either ETR or HS) on an indicator function for MNCs and other controls. We estimate the following for firm n, in industry i, in year t y nit = α + β 1 MNC + β 2 SIZE + β 3 RD + β 4 P P ENT + β 5 INT AN + β 6 LEV +β 7 CAP X + β 8 AD + β 9 SP I + β 10 lagsp I + β 11 NOL + β 12 lagnol + δ t + φ i + u nit (1) In 1, y nit is either ETR or HS. The parameter of interest is β 1. When β 1 < 0 (0 < β 1 ), MNCs have a tax-based cost advantage (disadvantage). The remaining variables in 1 are controls variables believed to explain variation in tax avoidance. SIZE is the natural log of assets. RD is the ratio of research and development expenses to total sales. PPENT is the ratio of property, plants, and equipment to total assets. INTAN is the ratio of intangible assets to total assets. LEV is the ratio of total debt to total assets. CAPX is the ratio of capital expenditures to property, plants, and equipment. AD is the ratio of advertising expenses to total sales. SPI is the ratio of special items to total assets, and lagspi is its lagged value. NOL is an indicator function equal to 1 if there is a tax-loss carry forward, and lagnol is its lagged value. Detailed descriptions of the construction of these variables and the Compustat data items used are presented in Table A1. The parameters δ t and φ i are annual and 2-digit SIC fixed effects, respectively, and u nit is an error term. 5 Using the same criteria as Dyreng et al. (2017), we create a comparable data set in both the number of observations and descriptive statistics. 9

12 We are interested in comparing the coefficient on MNC when using either Cash ETR or HS. We calculate ETR as the ratio of cash taxes paid to PI. In keeping with the methodology in Dyreng et al. (2017), we truncate ETR by setting ETR equal to 0 if Cash ETR < 0 and seting ETR equal to 1 if 1 <ETR. As mentioned above, when PI< 0, ETR is undefined. When calculating HS, we use the equation in Henry and Sansing (2014) HS = CashT axp aid 0.35 P retaxincome T otalasets (2) In 2, we experiment with two alternative measures of total assets. One measure of total assets is the book value of total assets (AT) and is directly available in Compustat. The second measure of total assets is the market value of total assets (MVA) that reflects the difference between market equity and book equity. Market value of equity is calculated as the product of the fiscal year closing share price and the number of common shares outstanding. Unfortunately, these variables are not recored for all firm-years in our sample, and we must delete some firm-years when either of these variables are missing. 6 4 Sample and Descriptive Statistics Table 1 describes the criteria for our sample. We use criteria similar to DHMT in order to facilitate the comparison between our study and theirs. We begin with all 284,360 firm-years listed in Compustat during the period We then remove firms not incorporated in America, financial and utility firms, and firms with less than $10 million in assets. 7 In order to calculate ET R, we remove firm-years with missing cash taxes paid or pretax income as well as missing control variables. After applying these 8 criteria, the number of firm-years in our sample (90,077) is comparable to the number of firm-years in DHTM (89,078). 6 Henry and Sansing (2014) suggest replacing MVA with AT when MVA is missing. 7 We use the following filters on the Compustat variables 1) fiscal year : 1988 FYEAR ) American companies: FIC =USA, 2) non-financial companies: SIC / {6000,..., 6799}, 4) non-utility companies: SIC / {4800,..., 4999}, 5) greater than $10m in assets: $10mil AT. 10

13 In order to create ET R, researchers must drop firm-years with non-positive P I. If we were to do so with our sample, we would lose 30% of our sample (90,077 firm-years to 63,024 firm-years). This decrease is nearly identical to the 30% decrease in sample size reported in DHMT (89,078 firm-years to 62,542 firm-years). As a final comparison of our sample to the DHMT sample, we report in the final row of Table 1 the number of firm-years for firms with 5 or more years of data.; our sample size of 53,933 firm-years is comparable to the 54,005 firm-years used in DHMT. Using criteria 1-8 in Table 1, we then drop all firms with less than 5 years of data and then bifurcate the sample into positive pretax income firm-years (0<PI) and non-positive pretax income firm-years (PI 0). The descriptive statistics are presented in Table 2. 8,9 As expected, the descriptive statistics in Panel A of Table 2 are comparable to the descriptive statistics in DHMT. The mean ETR is equal to 28.10% and is nearly 7 percentage points less than the statutory rate of 35%. However, there does appear to be a large amount of variation in ETR as the inter-quartile range is 9.72% to 38.23%. In contrast, ETR is undefined for firms where PI=0 and is omitted from Panel B. The Henry and Sansing (2014) measure of tax avoidance, HS, for positive PI firm-years has a mean value of with an inter-quartile range of to The HS for nonpositive PI firm-years has a mean value of 6.74 with a much larger and wider inter-quartile range of 1.50 to Although we cannot compare ETR for positive and non-positive P I firm-years, we can compare HS. Results in Panels A and B of Table 2 provide preliminary evidence that firms with non-positive PI have a larger tax burden than firms with positive PI. In addition to measures of tax avoidance, Table 2 also provides information about firm operations. Firms with non-positive P I are comparable to firms with positive P I. 8 We also remove 266 firm-years with the largest 0.1% of AD,RD, or CAPX. A manual inspection of the data finds that outlying observations of these 3 variables are almost always inflated a result of a denominator close to 0. Table A1 describes the denominators for these variables in more detail. We do not remove other ratio variables in Table 2 that use total assets in the denominator, as total assets are bounded above $10mil via criteria 5 in Table 1. 9 The 58,248 observations in Panel A of Table 2 differ from the 53,933 observations in line 10 of Table 1 because of the order in which the 5 year criteria is applied. 11

14 4.1 Multivariate Analysis Table 3 presents the least-squares results for regression 1. The first two columns use the 58,248 firm-years with strictly positive PI. Column 1 uses ETR as the dependent variable, and Column 2 uses HS as the dependent variable. Column 3 uses all 80,654 firm-years regardless of PI with HS as the dependent variable. All models include industry and year fixed-effects. Similar to previous studies, we cluster standard errors two ways at the firm and year level. In unreported results, the conclusions are robust to various clustering schemes for the standard errors. The results in Column 1 are comparable to results in Dyreng et al. (2017) and indicate that MNCs have ETRs that are 3.3% larger than ETRs for purely domestic firms. Using HS as the dependent variable, results in Column 2 indicate that MNCs do not have a tax based cost advantage; this result is significant at the 5% level. However, the first 2 columns of Table 3 use those firms with strictly positive PI. Results in Column 3 use all firm-years and indicate that MNCs do have a tax based cost advantage; this result is significant at the 5% level. In addition to using AT as the denominator in the HS tax measure, we also perform the same estimation using MVA as the denominator. Table 4 presents the results when using MVA. Columns 1 and 2 use the 53,537 firm-years with strictly positive PI and non-missing MVA. 10 Column 3 uses the 73,589 firm-years with non-missing MVA. The results in Table 4 are comparable to the results in Table 3. 5 Conclusion Overall our findings suggest that Cash ETR creates a sample selection bias. Using the alternative HS measure of tax avoidance mitigates this bias. In comparing MNCs to purely domestic firms, we find tax based advantages for purely domestic firms when using Cast 10 In order to calculate MVA, we require non-missing values for both the fiscal year closing price of the common stock and the number of shares outstanding. 12

15 ETR and tax based advantages for MNCs when using HS. References Chyz, J., Luna, L., and Smith, H. (2016). Implicit taxes of us domestic and multinational firms over the past quarter-century. De Simone, L., Mills, L., and Stomberg, B. (2014). Measuring income mobility. Dharmapala, D. (2014). What do we know about base erosion and profit shifting? a review of the empirical literature. Dharmapala, D. and Riedel, N. (2013). Earnings shocks and tax-motivated income-shifting: Evidence from european multinationals. Journal of Public Economics, 97: Drucker, J. (2010). Google 2.4% rate shows how $60 billion lost to tax loopholes. Bloomberg- Business & Financial News, Breaking News Headlines. Dyreng, S., Hanlon, M., Maydew, E. L., and Thornock, J. R. (2017). Changes in corporate effective tax rates over the past 25 years. Journal of Financial Economics. Dyreng, S. and Markle, K. (2015). The effect of financial constraints on tax-motivated income shifting by us multinationals. Dyreng, S. D., Hanlon, M., and Maydew, E. L. (2008). Long-run corporate tax avoidance. The Accounting Review, 83(1): Gravelle, J. G. (2009). Tax havens: International tax avoidance and evasion. National Tax Journal, pages Grubert, H. and Slemrod, J. (1998). The effect of taxes on investment and income shifting to puerto rico. Review of Economics and Statistics, 80(3):

16 Henry, E. and Sansing, R. C. (2014). Data truncation bias and the mismeasurement of corporate tax avoidance. Jennings, R., Weaver, C. D., and Mayew, W. J. (2012). The extent of implicit taxes at the corporate level and the effect of tra86. Contemporary Accounting Research, 29(4): Klassen, K. J. and Laplante, S. K. (2012). Are us multinational corporations becoming more aggressive income shifters? Journal of Accounting Research, 50(5): Klassen, K. J., Lisowsky, P., and Mescall, D. (2016). Transfer pricing: Strategies, practices, and tax minimization. Contemporary Accounting Research. Kleinbard, E. D. (2012). The lessons of stateless income. Scholes, M., Wolfson, M., Erickson, M., Hanlon, M., Maydew, E., and Shevlin, T. (2015). Taxes & Business Strategy. Pearson. Senate, U. (2014). Caterpillar s offshore tax strategy. Permanent Subcommittee on Investigations, Washington, DC. Surrey, S. S. (1973). Pathways to tax reform: the concept of tax expenditures. Harvard Univ Pr. 14

17 6 Tables and Figures Table 1: Sample Criteria and Remaining Observations Criterion Firm-Years Firms 1 Fiscal year ,360 29,158 2 American incorporated 230,632 22,802 3 Non-financial company 165,946 16,609 4 Non-utility company 150,164 15,226 5 More than 10m in assets 108,641 12,220 6 Non-missing cash taxes paid 93,716 10,552 7 Non-missing pretax income 93,713 10,551 8 Non-missing controls 90,077 10,467 9 Positive income 63,024 8, or more years per firm 53,933 4,773 Note: Criteria 1-8 are free from sample selection regarding 0 < P I. Criteria 9-10 indicate possible biases created when requiring positive income. Source: Compustat fundamentals and authors calculations. 15

18 Table 2: Descriptive Statistics Panel A: Positive Income, N = 58, 248 Variable Mean Std Dev 25 th percentile Median 75 th percentile ETR.NORM HS MNE SIZE RD PPENT INTAN LEV CAPX AD SPI NOL Panel B: Non-Positive Income, N = 22, 406 Variable Mean Std Dev 25 th percentile Median 75 th percentile ETR.NORM HS MNE SIZE RD PPENT INTAN LEV CAPX AD SPI NOL Note: Descriptive statistics for the tax avoidance measures and control variables. In addition to criteria 1-8 in Table, we also require 5 or more years of data for each firm. Source: Compustat fundamentals and authors calculations. 16

19 Table 3: Alternative Measures of Tax Avoidance (1) (2) (3) MNE (0.446) (0.047) (0.075) SIZE (0.151) (0.018) (0.058) RD (3.902) (0.451) (0.122) PPENT (1.263) (0.153) (0.314) INTAN (1.175) (0.127) (0.315) LEV (1.003) (0.106) (0.803) CAPX (0.991) (0.117) (0.332) AD (4.840) (0.711) (1.730) SPI (8.723) (1.591) (1.216) lagspi (2.365) (0.550) (0.511) NOL (0.581) (0.056) (0.072) lagnol (0.594) (0.050) (0.080) Industry and Year FE Non-positive PI Firm-Years Observations 58,248 58,248 80,654 R p < 0.001, p < 0.01, p < 0.05 Note: Model 1 uses ETR as the dependent variable. Models 2 and 3 use HS as the dependent variable. HS is calculated using AT as the denominator. All models are estimated using least-squares with 2-digit SIC industry fixed effects and annual fixed effects. All standard errors are two-way clustered at the firm and year levels. Source: Authors calculations. 17

20 Table 4: Market Value of Assets (1) (2) (3) MNE (0.430) (0.035) (0.061) SIZE (0.131) (0.010) (0.038) RD (3.896) (0.214) (0.057) PPENT (1.312) (0.098) (0.190) INTAN (1.279) (0.091) (0.297) LEV (0.950) (0.086) (0.183) CAPX (1.029) (0.084) (0.247) AD (5.417) (0.388) (1.910) SPI (9.076) (3.184) (5.829) lagspi (2.576) (0.687) (0.515) NOL (0.604) (0.042) (0.124) lagnol (0.583) (0.045) (0.090) Industry and Year FE Non-positive PI Firm-Years Observations 53,527 53,527 73,589 R p < 0.001, p < 0.01, p < 0.05 Note: Model 1 uses ETR as the dependent variable. Models 2 and 3 use HS as the dependent variable. HS is calculated using MVTA as the denominator. All models are estimated using least-squares with 2-digit SIC industry fixed effects and annual fixed effects. All standard errors are two-way clustered at the firm and year levels. Source: Authors calculations. 18

21 Figure 1: Case ETR for Positive PTI Firm Years Cash ETR Domestic Multinational Year Notes: Figure 1 presents the average Cash ETR by year for all firm years with strictly CashT axesp aid positive PI where CashET R = P retaxincome. The solid line is for firm years with foreign income equal to 0. The dashed line is for firm years with non-zero foreign income. Source: Authors calculations. 19

22 Figure 2: HS for Positive PTI Firm Years HS Domestic Multinational Year Notes: Figure 3 presents the average HS by year for all firm years with strictly positive CashT axesp aid 0.35 P T I PI where HS = MarketV alueofassets. The solid line is for firm years with foreign income equal to 0. The dashed line is for firm years with non-zero foreign income. Source: Authors calculations. 20

23 Figure 3: HS for All Firm Years HS Domestic Multinational Notes: Figure 2 presents the average HS by year for all firm years where HS= CashT axesp aid 0.35 P I AT. The solid line is for firm years with foreign income equal to 0. The dashed line is for firm years with non-zero foreign income. Source: Authors calculations. 21

24 A1 APPENDIX Table A1: Variable Descriptions Variable Description AD Ratio of advertising expense to total sales. Calculated as AD= XAD and set SALES equal to 0 if XAD is missing. CAPX Ratio of capital expenditures to plant, property, and equipment. Calculated as CAPX= CAPX and set equal to 0 if PPENT is missing. PPENT ETR Effective tax rate. Calculated as the ratio of cash taxes paid to pretax income (ETR = 100 TXPD ) and winsorized so that ETR= 0 if ETR < 0, ETR= 100 PI if 100 <ETR, and ETR=ETR if 0 ETR 100. HS Henry and Sansing (2014) measure of tax. Calculated as the ratio of the difference between cash taxes paid and 35% of pretax income to total assets (HS= 100 TXPD 0.35PI ). AT INTAN Ratio of intangible assets to total assets. Calculated as INTAN= INTAN AT LEV Ratio of total debt to total assets. Calculated as LEV= DLTT+DLC AT MNE An indicator function equal to 1 if foreign income is positive (0 < PIFO) or foreign taxes paid is positive (0 < TXFO) PI PPENT RD SIZE SPI NOL Pretax income. Equal to PI. Ratio of property, plant, and equipment to total assets, Calculated as PPENT= PPENT AT Ratio of research and development expense to total sales. Calculated as AD= XRD and set equal to 0 if XRD is missing. SALES Natural log of total assets. Calculated as SIZE= ln AT Ratio of special items to total assets. Calculated as SPI= SPI AT An indicator function equal to 1 if there is a tax loss carry forward form the previous year (0 < TLCF) Note: Variables used in the paper are CAPITALIZED letters and variables in the Compustat annual file are BOLD letters. 22