Changes in Corporate Effective Tax Rates Over the Past 25 Years. Scott D. Dyreng Duke University

Transcription

1 Changes in Corporate Effective Tax Rates Over the Past 25 Years Scott D. Dyreng Duke University Michelle Hanlon Massachusetts Institute of Technology Edward L. Maydew University of North Carolina Jacob R. Thornock Brigham Young University June 2016 Abstract: We investigate systematic changes in corporate effective tax rates over the past 25 years and find that effective tax rates have decreased significantly. Contrary to conventional wisdom, we find that the decline in effective tax rates is not concentrated in multinational firms; effective tax rates have declined at approximately the same rate for both multinational and domestic firms. Moreover, we find that within multinational firms, both foreign and domestic effective rates have decreased. Finally, we find that changes in firm characteristics and declining foreign statutory tax rates explain little of the overall decrease in effective rates. The findings have broad implications for research, as well as for current policy debates about reforming the corporate income tax. We appreciate comments from Jeff Hoopes, Tom Neubig, Terry Shevlin, Nemit Shroff, Jaron Wilde, and seminar participants at the 2013 EIASM Workshop on Current Research on Taxation at the University of Münster, the 2015 MIT Asia Conference in Accounting, Cornell University, the Duke Summer Brownbag series, Michigan State University, Renmin University, the European Accounting Association 2015 Annual Congress, and the International Tax Policy Forum. We thank Brady Williams for foreign statutory tax rate data. We thank Madeline Hartman, Tyson Mayers, Meredith Mayer-Salman, Theresa Richards, and Tommy Saintsing for research assistance. Dyreng acknowledges funding from the Fuqua School of Business at Duke University, Hanlon from the Howard W. Johnson Chair at the MIT Sloan School of Management, Maydew from the Accounting Alumni Faculty Research Fund at the University of North Carolina, and Thornock from the University of Washington and Brigham Young University.

2 1. Introduction In this paper, we examine the trend in corporate effective tax rates over the past 25 years. While U.S. statutory tax rates have remained relatively constant during this period, there are indications that some firms have been able to reduce their effective tax rates through tax planning strategies and by taking advantage of favorable provisions in the tax code (Gravelle, 2013; Drucker, 2010; U.S. Senate, 2014; Dyreng, Hanlon and Maydew, 2008). Many have suggested that the tax system is broken because of the disparity between the high statutory rate and low effective rates that some companies are able to achieve, resulting in an inequitable system of taxation, perverse incentives, and undesirable consequences. 1 Despite recent attention to these issues by the U.S. government, international tax bodies such as the Organization for Economic Cooperation and Development (OECD), the academic literature, and the popular press, surprisingly little evidence exists on how effective tax rates have evolved over time, even though such data would be informative to academics and policy makers when evaluating the current system relative to a set of alternatives. There is a great deal of research on tax avoidance and effective tax rates, but almost all of this research is on the cross-section of firms (e.g., Chen, Chen, Cheng, and Shevlin, 2010; Desai and Dharmapala, 2006; Kim, Li, and Zhang, 2011). In order to investigate the existence of, and potential explanations for, a trend in effective tax rates over time, we examine a sample of 54,028 U.S. firm-years over the 25-year period from We primarily study the cash effective tax rate (worldwide cash taxes paid/pretax accounting earnings), supplemented by domestic and foreign current effective tax rates, which we use to test foreign versus domestic rates within U.S. corporations that have income from 1 For discussions and references see: 1) H.R. 1 Tax Reform Act of 2014 (113 th Congress), 2) U.S. Permanent Subcommittee on Investigations hearing report entitled Impact of the U.S. Tax Code on the Market for Corporate Control and Jobs (hearing held July 30, 2015), 3) Senate Finance Committee report entitled Comprehensive Tax Reform for 2015 and Beyond and 4) Citizens for Tax Justice letter to the Senate entitled Memo to Senate Permanent Subcommittee on Special Investigations: U.S. Corporations already Pay a Low Tax Rate. 1

3 foreign sources (multinational firms). 2 To be clear, we do not examine taxes paid to the IRS relative taxable income reported to the IRS. Broadly speaking, taxes paid to the IRS will be equal to taxable income reported to the IRS multiplied by the statutory tax rate, reduced by tax credits. Examining the ratio of these two numbers will not capture tax avoidance strategies that reduce or delay the recognition of taxable income (e.g., shifting income to a low-tax foreign country, investment in tax favored activities such as municipal bonds, investments that are subjected to accelerated depreciation for tax purposes) because those types of strategies reduce both taxes paid and taxable income reported to the taxing authority. In contrast, cash effective tax rates capture all reductions in taxes paid relative to pretax financial accounting income. Thus, our measures are intentionally broad, so that they capture any form of tax reduction relative to pretax accounting income, whether through tax sheltering, location decisions, income shifting, tax preferences within the tax code, or rule changes. 3 We begin by testing whether or not cash effective tax rates have declined over the sample period. Overall, we find a significant downward trend in effective tax rates using our broad sample of publicly traded U.S. corporations. On average, cash effective tax rates of our sample firms have decreased by about 0.4 percentage points per year over the past 25 years. The trend is economically large, representing a cumulative decline of approximately 10 percentage points. This translates into about $109 billion less in taxes paid in 2012 compared to what would have been paid had the effective rate remained constant at the 1988 level. 4 2 We explain these measures more fully and we test multiple alternate measures, such as the GAAP effective tax rate and a cash effective tax rate based on cash flows from operations, in the sections that follow. 3 Our measure does not capture tax avoidance that reduces financial accounting income, known as conforming tax avoidance. We discuss this further below. 4 $109 billion is calculated by aggregating pretax income of our sample firms in 2012 of $1,088 billion and multiplying by

4 Much attention has been focused on multinational firms because they are able shift income from high-tax jurisdictions to low-tax jurisdictions, including tax havens (Dharmapala and Riedel, 2013; Dyreng and Markle, 2016; Dharmapala, 2014), especially when cross-border transactions are based on intangible assets which are more difficult to value and easier to move across borders (Grubert and Slemrod, 1998; Kleinbard, 2011; De Simone, Mills, and Stomberg, 2014). Recent international tax reform guidance has focused on these issues for multinationals, including the Base Erosion and Profit Shifting (BEPS) initiative at the OECD (OECD 2013). If effective tax rates are declining more rapidly for multinational firms than for purely domestic firms, then multinationals may increasingly have a cost advantage over domestic firms. However, there is little evidence in the literature about changes in effective tax rates of multinational firms compared to purely domestic firms over time. To test this conjecture, we examine the time trend in effective tax rates separately for multinational and domestic firms. Surprisingly, we find essentially the same decrease in effective tax rates over time for purely domestic firms as we find for multinationals. We probe this finding further by comparing the trend in current effective tax rates on domestic income versus foreign income within our subsample of multinational firms. As expected, we find that the decline in foreign effective tax rates of multinationals is largely explained by the well-documented decline in statutory tax rates of foreign countries. However, we also find a downward trend in the domestic effective tax rates of multinationals, which cannot be explained by declining foreign statutory tax rates. These primary results are informative for tax policy. While multinationals have access to international tax planning opportunities, our results show that domestic corporations have declining effective tax rates as well. This suggests that there are significant and increasing opportunities to reduce (via planning or from provisions in the tax laws) effective tax rates on 3

5 domestic income. Thus, perhaps policymakers should not focus exclusively on tax avoidance by multinationals and should consider the extent of current provisions for domestic-only firms. We perform a number of additional tests to explain the trend in effective tax rates. First, we examine whether changing firm characteristics can explain the trend. Extant research shows that the characteristics of publicly traded firms today are different than they were 25 years ago (e.g., Fama and French, 2001; Graham, Leary and Roberts 2014; DeAngelo and Roll, 2015). If effective tax rates vary with those characteristics, then they might help explain the time trend. Consistent with prior cross-sectional research, we find that a multitude of firm-level characteristics are associated with effective tax rates, such as intangible assets and the existence of losses. However, we find that these characteristics are unable to account for much of the decrease in tax rates over time. Second, we test whether the mapping of certain firm characteristics into tax rates (i.e., the time-varying effect of those characteristics on effective tax rates) has changed over time. For example, a common assertion is that globalization has made it easier to shift the income from intangible assets into tax havens, or to escape taxation entirely by creating so-called stateless income (Kleinbard, 2011). If so, tax rates will decrease with respect to intangible assets over time, even if the level of intangible assets stays constant. We find some evidence of changes in the mapping of certain firm characteristics into tax rates over time, but again the changes explain little of the overall decrease in effective tax rates. Finally, we examine whether the decline in effective tax rates can be explained by legislative and regulatory changes that took place during the sample period, including the checkthe-box regulations, the American Jobs Creation Act of 2004 (AJCA), and the two recent bonus 4

6 depreciation regimes. Overall, we find little evidence that the trends in effective tax rates coincide with these legislative and regulatory changes. Our paper contributes to the literature by executing a thorough examination of time trends in corporate effective tax rates. 5 Our finding that effective tax rates have declined for purely domestic firms at essentially the same rate as multinationals and that multinational firms do not on average have significantly lower cash effective tax rates suggests that the current, almost singular focus on multinational corporations for tax avoidance and low tax rates may be misplaced. In addition, our study is also important for researchers interested in studying the determinants of effective tax rates. It is common to pool observations across years when studying effective tax rates (and tax avoidance), but the changing nature of effective tax rates over time suggests that researchers should be careful to examine whether results are time period specific, and they should interpret their results accordingly. The paper proceeds as follows. Section 2 develops the hypotheses, Section 3 discusses the sample, the variables, and presents descriptive statistics. The main results are presented in Section 4. Section 5 presents an examination of potential determinants of declining rates. Section 6 presents additional tests and robustness. Section 7 briefly examines foreign incorporated firms relative to U.S. firms and Section 8 concludes. 2. Hypotheses Development Corporate tax avoidance has recently received substantial attention in the academic literature, the popular press, and from policymakers. It is well known in these circles that there is substantial cross-sectional variation in corporate effective tax rates (e.g., Dyreng et al. 2008), and 5 There are a few studies that examine changes in tax avoidance over time, but not as the main focus of study (e.g., Desai, 2003; Desai and Dharmapala, 2009; Graham, Raedy and Shackelford, 2012; Klassen and LaPlante, 2012). There are also a few studies that examine trends in aggregate tax collections (e.g., Auerbach and Poterba, 1987; Auerbach, 2007). 5

7 that this variation can be partially explained by a multitude of firm-specific factors, including ownership structure (Chen et al., 2010; Badertscher, Katz and Rego, 2013), compensation structure (Rego and Wilson, 2012; Armstrong, Blouin and Larcker, 2012), corporate governance (Desai, Dyck and Zingales, 2007), labor organization (Chyz, Leung, Li and Rui, 2013), subsidiary locations (Dyreng and Lindsey, 2009), business model characteristics (Higgins, Omer and Phillips, 2015), and management characteristics (Chyz, 2013). Despite the progress made in understanding the cross-sectional determinants of effective tax rates, surprisingly little is known about how effective tax rates have evolved over time. Though little is known about time trends in effective tax rates of U.S. corporations at the microeconomic level, some research has examined trends in macroeconomic tax outcomes, such as over-time changes in corporate tax revenues. For example, Auerbach (2007) extends Auerbach and Poterba (1987) to examine the question of whether corporate tax revenues have declined over time, and finds that they declined through 1982, but leveled off thereafter. Furthermore, he finds offsetting trends in the ratio of profits to GDP (which is declining) and in the average tax on profits (which is increasing over time). He interprets the latter as evidence that tax planning may not be as prevalent as some suggest. In contrast, Slemrod (2004) suggests misreporting of corporate income taxes has increased over time from $61 billion in 1988 to $146 billion in 2000, while Desai (2003) argues that the corporate tax receipts as percentage of the budget have decreased over time. Overall, the research examining the macroeconomic trends in tax outcomes has produced somewhat mixed results. There are several reasons to suspect that, at the microeconomic level, the effective tax rates of U.S. firms are decreasing over time. First, many developed countries have reduced their statutory tax rates, even while the U.S. has kept its statutory tax rate essentially constant since 6

8 1986. For a multinational firm, a decline in foreign tax statutory tax rates can lead to a lower effective tax rate by reducing its tax on foreign income, at least until it repatriates the income to the U.S. Moreover, a declining foreign statutory tax rate could also increase the incentives to locate more real operations in that country and/or to source or shift income to that country to take advantage of the lower tax rate. Second, firms are becoming more global in nature. Figure 1 shows that the fraction of sample firms that are multinational has steadily increased over time, suggesting a clear and steady upward trend in multinational status over the sample period. In 1988, approximately 40% of the sample firms were multinationals; by 2012, nearly 70% of sample firms were multinationals. 6 Combining these two insights suggests that to the extent that U.S. multinational firms are increasingly able to source income into countries outside the U.S., their effective tax rates will decline over time. Third, recent evidence suggests that domestic firms also have tax planning opportunities available to them including tax shelters (Wilson, 2009; Lisowsky, 2010; Brown, 2011) and provisions enacted to benefit domestic firms (e.g., domestic production activities deduction, bonus depreciation). If domestic firms are increasingly utilizing these strategies over time, it could cause their effective tax rates to decline over time. In sum, changes to the global economic, competitive, and tax landscape over the sample period suggest a decline in effective tax rates. This discussion leads to our first hypothesis (all hypotheses are stated in the alternative): H1: The effective tax rates of U.S. corporations have decreased over time. Anecdotes in the popular press, as well as evidence in academic research, suggest that multinationals use transfer pricing, intracompany debt, cost-sharing agreements, and other tactics to shift income from high tax jurisdictions to low tax jurisdictions (e.g., Hines and Rice, 1994; 6 Regressing the average annual percentage of firms that are classified as multinationals on a linear time trend variable, TIME, reveals an upward trend that is highly statistically significant (t-statistic = 21.17). 7

9 Huizinga and Laeven, 2008; Dharmapala and Riedel, 2013). As U.S. firms have become more multinational, and as the disparity in statutory tax rates between the U.S. and foreign countries has increased, the opportunities and incentives to implement cross-border tax saving strategies has also likely increased. Indeed, multinational companies likely have tax strategies at their disposal that are simply unavailable to purely domestic firms because purely domestic firms cannot take advantage of the large statutory tax rate disparities that exist across countries. This leads to our second hypothesis: H2: The effective tax rates of multinational firms are declining more over time than those of purely domestic firms. To probe our second hypothesis more deeply, we consider the tax rates on domestic income of multinationals relative to the tax rates on foreign income of multinationals. If the declines in effective tax rates are in fact driven by strategies involving income shifting across international borders, we should see more stark declines in foreign taxes paid on foreign income than in domestic taxes paid on domestic income. Accordingly, our third hypothesis is: H3: For multinationals the effective tax rate on foreign income declines more over time than does the effective tax rate on domestic income. Beyond testing our three hypotheses, we examine several other potential explanations for the changes in tax rates over time, including changes in foreign tax rates, changing firm characteristics, changing tax and accounting rules, and other factors. We conduct several sets of additional analyses and robustness tests that we discuss in more detail below. 3. Sample and descriptive statistics Table 1 describes our sample selection criteria. We begin with all non-financial, nonutility firm-years listed in Compustat during the period that have total assets of at least $10 million. We start the sample in 1988 because it follows the last major overhaul of U.S. 8

10 tax system, the Tax Reform Act of 1986, which took effect in mid There are 102,925 firmyears that meet these initial requirements. We require non-missing values of cash taxes paid, pretax income and other control variables, which reduces the sample to 89,114 firm-year observations. 7 We also exclude firm-years with negative pretax income because effective tax rates are difficult to interpret when the denominator is negative this criterion further reduces the sample to 62,564 observations. 8 Finally, we require each firm to have at least five years of data. These criteria result in a sample of 54,028 firm-year observations from 4,643 unique firms, for an average of 11.6 years of data per firm. Our primary variable of interest is the firm s cash effective tax rate, CASH ETR, which is computed as the ratio of cash taxes paid to pretax accounting income. 9 CASH ETR is widely used in the literature because it captures a broad range of tax avoidance activities, including income shifting from high-tax to low-tax jurisdictions (e.g., strategic transfer pricing arrangements, cost sharing agreements, income stripping using intra-company debt), investment in tax favored assets (e.g., municipal bonds), accelerated depreciation deductions, tax credits (e.g., research and experimentation credits), and so on. This measure is advantageous because it allows us to speak to changes in tax avoidance generally, without specifying ex-ante precise tax avoidance strategies or rule changes, which could have evolved over time. To be clear, CASH ETR uses financial accounting income in the denominator, not taxable income. While this has the advantage of allowing the measure to capture tax avoidance strategies that cause reductions to taxable income (e.g., income shifting), it has the disadvantage that it is affected by regulatory 7 Compustat variable pneumonics are included in the footnote to Table 2. Unless otherwise specified, all variables are acquired from the annual fundamentals database maintained by Compustat. 8 Thus, our results do not apply to loss years. In Section 5, we conduct tests to examine whether losses are driving the results that we observe, and we find that they do not. 9 We do not adjust CASH ETR for special items, but instead include special items as a control variable in our multivariate regressions. We winsorize all effective tax rates at 0 and 1, following, among others, Dyreng et al. (2008). 9

11 changes to financial accounting rules, and is also affected by managers incentives to report high income to shareholders. CASH ETR also does not capture conforming tax avoidance, where firms engage in tax avoidance strategies that reduce both taxable income and financial accounting income (e.g., using external debt capital instead of equity capital to take advantage of interest deductions). 10 Table 2, Panel A presents descriptive statistics for the sample. CASH ETR averages 29.1% during the sample period, with a median of 27.5%, both of which are below the statutory tax rate of 35% and are consistent with rates reported in prior research (e.g., Dyreng et al., 2008). The data demonstrate considerable variation in CASH ETR the 25 th percentile of CASH ETR is only 12.6% and the 75 th percentile is 38.6%, which is also consistent with quantiles of CASH ETR reported in prior research. The remainder of Panel A of Table 2 presents descriptive statistics for firm characteristics that potentially explain variation in effective tax rates. 11 The variable MNE, which indicates whether the firm is a multinational (MNE=1) or a purely domestic firm (MNE=0), has a mean of We classify a firm as multinational (MNE) in a given year if either its pretax foreign income is greater than zero or if its absolute value of foreign tax expense is greater than zero. 12 As noted in Donohoe, McGill and Outslay (2012), the classification of firms as being strictly multinational or purely domestic is more complex than it might seem at first, and we therefore consider six alternative measures of MNE in Section We know of no established tax burden measure that captures all types of tax avoidance, and hence we focus on explicit tax avoidance activities. Our measure incorporates implicit taxes to the extent the implicit taxes reduce financial accounting income. However, our measure will not fully capture these taxes. We leave an investigation of implicit taxes to future research. 11 All variables are defined in the footnote to Table Because these Compustat data items are gathered directly from the firm s audited financial statements filed with the SEC, we expect them to be generally accurate. We hand-checked some of the largest domestic firms in our sample, such as CVS, Macy s, Time Warner, and Lowes, and find that the numbers reported by Compustat are generally accurate. In some cases, we find evidence of small amounts of foreign activity in other parts of the financial statements (generally less than 2% of total income), but the amounts were relatively small. 10

12 LOG ASSETS is the natural log of total assets. The median LOG ASSETS of 5.803, corresponding to approximately $331 million of total assets. The average firm-year reports 2.6 dollars of research and development spending for every 100 dollars in sales (mean R&D EXPENSE = 0.026); however, the median firm-year reports no research and development spending. For the mean (median) firm-year, PP&E constitute 29.4% (23.5%) of total assets. The mean (median) firm-year has INTANGIBLE ASSETS equal to 11.8% (4%) of total assets. The mean firm-year in the sample has LEVERAGE of 21.8% of total assets, with a median of 19.3%. The average (median) firm-year in the sample has CAPITAL EXPENDITURES of 25.7% (21.2%) of net property, plant and equipment. The mean (median) firm-year has ADVERTISING EXPENSE of 1% (0%) of total assets. Special items, when they exist, are usually negative (i.e., income-decreasing), which is reflected in the mean SPECIAL ITEMS of -0.3% of average total assets. The median firm-year does not report any special items. LAGGED SPECIAL ITEMS has an average value in the sample of -0.9%, while the median value is zero. NOL indicates that the firm reports a tax-loss carryforward as reported by Compustat, on average, 27.4% of the years in the sample. NOL reflects the change in net operating losses from the prior year to the current year and has a mean (median) of (0.000) in the sample, suggesting that tax loss carryforwards equal to about 0.1% of assets were either utilized to offset the tax bill, or expired in the current year. 13 Table 2, Panel B presents the Pearson (above the diagonal) and Spearman (below the diagonal) correlations for these variables. Of note, using Pearson correlations, CASH ETR is negatively and significantly correlated with R&D EXPENSE, PP&E, LEVERAGE, CAPITAL 13 Note that these NOLs are as reported in Compustat, which are subject to data concerns (e.g., Mills, Newberry, and Novack 2003). We cannot determine whether the NOLs are foreign, domestic, or state NOLs. In addition, we cannot determine whether the losses are usable or limited under certain tax code provisions, such as Internal Revenue Code Section 382 that limits the use of NOLs acquired as part of a merger or acquisition of a company and/or via a technical ownership change outside of an acquisition. 11

13 EXPENDITURES, SPECIAL ITEMS and NOL. CASH ETR is positively and significantly correlated with MNE, ADVERTISING EXPENSE, and NOL. LOG ASSETS and INTANGIBLE ASSETS are either insignificantly correlated or not consistently correlated with CASH ETR. 4. Main results of hypotheses tests 4.1 Changes in cash effective tax rates over time- Test of H1 We begin by testing our first hypothesis, which is that CASH ETRs of U.S. corporations have decreased over time. In Figure 2, we visually test this hypothesis by plotting the sample mean of CASH ETR for each year from 1988 to There is a clear downward trend in CASH ETR during the past 25 years. The average CASH ETR was just above 32% in 1988 but declined to about 27% by The figure reveals some periods of increases and decreases in average CASH ETR the average CASH ETR goes as low as 22% by 2004 before increasing to nearly 31% by Moreover, the figure shows that in some partitions of the sample period, the trend in cash effective tax rates varies in steepness. For example, the trend in effective tax rates is quite steep before 2001 and somewhat flat after However, even with the year-to-year variation in the sample average CASH ETR, the trend over the full sample period is clearly negative. To test the trend more formally, we estimate the following regression: CASH ETR it = α 0 + α 1 TIME t + ε it. (1) where CASH ETR is as defined above and TIME is calculated as the fiscal year for a given firmyear observation less the number 1988, which is the first year in the dataset. Thus, the coefficient on TIME captures the linear time trend in CASH ETR over the sample period. We present the results from estimating equation (1) for the full sample in Table 3, Panel A, Model We see that the slope on the time trend, TIME, is (t-statistic = 6.99), 14 To simplify presentation of the regression coefficients, we multiply the dependent variable in all regressions by

14 confirming the statistical significance of the visual evidence presented in Figure 2. The decrease in cash effective tax rates is economically large about 0.4 percentage points per year and likely resulted in positive cash flows for the average firm. 15 For example, a decline in the CASH ETR from 32% to 27% corresponds to a 7.4% increase in average after-tax cash flows, all else constant. 16 This finding suggests that, for the average firm, lower cash effective tax rates have become an increasingly important source of cash flow in the past 25 years. 4.2 Cash effective tax rate trends for multinational versus domestic firms Test of H2 Having documented a downward trend in effective tax rates, we now turn to testing our second hypothesis, which is that the effective tax rates of multinationals are declining more rapidly than the effective tax rates of purely domestic firms. In Figure 3, we examine our second hypothesis visually by plotting the average CASH ETR over time separately for multinationals and purely domestic firms. The dark (light) line in Figure 3 plots the average CASH ETR over time for multinationals (purely domestic firms). There are two striking results evident in Figure 3. First, multinationals on average have higher CASH ETRs than purely domestic firms in almost every year during the 25-year sample period. Thus, while multinationals have tax planning opportunities not available to purely domestic firms, the results suggest the average multinational firm has a higher cash effective tax rate than a purely domestic firm. 17 Second, in stark contrast to conventional wisdom, both multinationals and purely domestic firms exhibit declining CASH ETRs over time at approximately the same rate. At the beginning of the period in 1988, the average multinational firm had a CASH ETR of approximately 34%, while the average purely domestic firm had a CASH ETR of approximately 15 This is consistent with Hanlon, Maydew, and Saavedra (2014), who find a significantly negative correlation between after-tax cash flows and a firm s 5-year cash effective tax rate. 16 (1-0.27)/(1-0.32) = For simplicity, we assume accruals are zero in expectation. 17 This is consistent with Markle and Shackelford (2012), who show that the CASH ETR for multinationals is greater than that for domestic firms for the period and with Rego (2003) over an earlier period. 13

15 32%. By the end of the sample period, both multinationals and purely domestic firms were able to lower their average CASH ETR to about 28% and 24%, respectively. This result contradicts a commonly held belief that a decline in CASH ETRs over time is due to the rising prevalence of multinationals and/or increased tax avoidance by multinationals. We observe essentially the same decline in effective tax rates in purely domestic firms as we do in multinationals. In Table 3, Panel A we present the results from estimating equation (1) for only multinational firms (Model 2), and only domestic firms (Model 3). In Model 2, the slope on TIME is for multinationals; and in Model 3, it is for purely domestic firms. To test the difference in the TIME coefficients across Models 2 and 3, we employ a stacked regression approach, which reveals that the time trends are not significantly different from one another (difference = 0.070, t-statistic = 1.17). The intercept from estimating equation (1) represents the conditional mean CASH ETR at the beginning of the sample period. The intercept for multinationals, presented in Model 2, is , while the intercept for domestic firms, presented in Model 3 is The difference of is marginally statistically significant (t-statistic = 1.98). Thus, the average multinational has a slightly higher CASH ETR than the average purely domestic firm. As striking as the results in Figure 3 and Table 3 are in suggesting that both multinational and domestic firms have declining CASH ETRs, multinational and domestic firms likely differ on a number of dimensions. For example, multinationals are on average much larger than purely domestic firms. To address this potential confound, in Panel B of Table 3, we retain only firmyears in the upper quartile of total assets for the sample and examine whether the trend in effective tax rates for large multinationals differs from that of large domestic firms. In the first column, we see that large firms are similar to other firms in that they exhibit a statistically 14

16 significant declining time trend in their CASH ETR, which is similar in magnitude to the trend reported for the full sample (that in the first column of Panel A). Models 2 and 3 show that both sets of firms (large multinationals and large purely domestic firms) experience a decrease in CASH ETR over time, with the slope on the time trend being for large multinationals and for large domestic firms (t-statistics of and -6.19, respectively). The slopes are significantly different from one another (t-statistic = 2.26) as shown in the rightmost column, suggesting with large domestics having a faster rate of decline than large multinationals. Overall, these results support H1, that effective tax rates are declining over time, but fail to support H2, as effective tax rates of domestic firms are declining at least as much as for multinationals firms. 4.3 ETRs on foreign income versus ETRs on domestic income Test of H3 To test our third hypothesis, we focus exclusively on multinationals and examine changes in their effective tax rates on two different types of income: domestic versus foreign. 18 Therefore, in this subsection we use the domestic or foreign portion of current tax expense in place of cash taxes paid as the numerator in our effective tax rate measure. 19 Specifically, we examine three measures of current effective tax rates for multinationals: 1) the worldwide current effective tax rate, ETR ww, which is the ratio of current worldwide tax expense to total pretax income; 2) the domestic current effective tax rate, ETR dom, which is current federal tax expense divided by pretax domestic income; and 3) the foreign current effective tax rate, ETR for, which is current foreign tax expense divided by pretax foreign income. 18 Specifically, in this section, we restrict the sample to the 15,339 multinational firm-years with non-missing federal tax expense, foreign tax expense, and positive values for both pretax domestic income and pretax foreign income. 19 Although cash taxes paid are not disclosed by jurisdiction, an analysis of the differential trend on foreign versus domestic income is possible because Generally Accepted Accounting Principles (GAAP) requires firms to break down their pretax income and current tax expense into domestic and foreign components in their financial reports. Current tax expense can differ from cash taxes paid because of accruals for uncertain tax positions and other current accrued taxes, but in general current tax expense is highly correlated with cash tax paid. In our sample, for example, the correlation between current tax expense and cash tax paid is about

17 Figure 4 plots the average ETR dom and ETR for for our sample each year over the sample period. Three results immediately stand out. First, the evidence suggests that multinationals on average pay tax at a higher rate on their foreign income than their domestic income. At the beginning of the sample period in 1988, the average ETR dom was approximately 30%, whereas the average ETR for was approximately 38%. Second, multinationals experienced declines in both their domestic and foreign current effective tax rates over time, with both ETR dom and ETR for declining to approximately 28% by the end of the sample period. Third, the decline in current effective tax rates of multinationals is greater in magnitude with respect to foreign income than domestic income. We report more formal empirical tests of H3 in Table 4, Panel A. In Model 1, we show the results of a regression of ETR ww on TIME, which reveals that ETR ww has a similar downward sloping trend to that shown in Table 3 for the full sample of firms using CASH ETR. In Model 2 the dependent variable is ETR dom, and in Model 3 the dependent variable is ETR for. Comparing the coefficient on TIME between the two models, the results show that the downward trend is steeper for ETR for than for ETR dom. In Panel B, we repeat this analysis but using only large multinationals (i.e., those in the upper quartile of total assets for the sample). 20 We observe a similar pattern in the time trends for ETR for and ETR dom as we do for the full sample of multinationals. Overall, the results in Figure 4 and Table 4 support H3, suggesting that the slope in current effective tax rates is negative with respect to both foreign and domestic income over time, and is decreasing at a faster rate with regard to foreign income. Thus, when we compare multinationals to domestic firms (in Table 3), the evidence suggests that domestic firms have a 20 Because the quartiles are based on the full sample, the number of observations may not be equal in particular tables. 16

18 similar or a slightly steeper decline over time in cash effective tax rates than do multinationals. However, when we look within multinationals (in Table 4), the evidence suggests that the current effective tax rate on foreign-sourced income has declined more over the last 25 years than has the current effective tax rate on domestic-sourced income. 21 There are some important caveats to keep in mind regarding the data on foreign effective tax rates, however. First, the effective tax rate on foreign income may obscure significant heterogeneity in taxes across jurisdictions. For example, a firm may face a 10% tax rate on income earned in one country, while facing a 40% tax rate in another, but these are averaged away in computing ETR for. Thus, even when a firm exhibits an effective tax rate on its foreign income that is greater than its effective tax rate on domestic income, the firm may still have incentives to shift income (from the U.S. or high tax rate foreign countries) into low-tax foreign jurisdictions. Second, the effective tax rate on foreign income, ETR for, may obscure variation in profitability across countries. For example, a firm may report losses in one country and income in another, but will typically not be able to use the loss in the first country to offset the income in the second. 22 The result of this friction can result in an increased ETR on the firms total income Examination of potential determinants of declining rates 5.1 The effect of declining foreign statutory tax rates 21 Note that income shifting by multinational firms would not generally change the effective rate on domestic income. Domestic income would be lower or higher (if income shifted out of or into the U.S.), and overall U.S. taxes would be lower or higher, but the tax per dollar of domestic income reported would not be lower or higher. In addition, the shifting of income to low-tax foreign jurisdictions yields lower cash taxes and lower current tax expense, but if the firm later repatriates those foreign earnings, U.S. taxes (less a foreign tax credit) would likely need to be paid. Thus, there is a potential future U.S. tax liability that is unrecorded in our data. See Foley, Hartzell, Titman and Twite (2007) and Graham, Hanlon, and Shevlin (2011) for further explanation. Our objective is to examine trends in taxes paid (as well as we can measure it); thus this measurement issue related to future potential repatriation taxes is not a concern. 22 While the incomes cannot be directly offset, see De Simone, Klassen, and Seidman (2014) for a discussion of how firms could shift income via transfer pricing to utilize losses cross-country. 23 Country-by-country data might offer better data for use in tests such as ours but such data are not available now. 17

19 One explanation for declining CASH ETRs is the decline in foreign statutory tax rates. To examine the overall effect of foreign statutory tax rates, we first plot the weighted-average statutory tax rates of OECD countries (excluding the U.S.) over the sample period in Figure Countries are weighted each year by their GDP. The figure shows a clear downward trend in weighted-average foreign statutory tax rates over time among other developed countries. At the beginning of the sample period, the weighted-average statutory rate in the OECD was about 45%, declining steadily to just below 30% by We test this proposition directly by constructing a firm-year-specific foreign statutory tax rate, FORSTAT, as the simple average of the foreign statutory tax rates for the countries in which the firm has significant subsidiaries. The subsidiary data are gathered from Exhibit 21 of Form 10-K following Dyreng and Lindsey (2009) and Dyreng, Lindsey and Thornock (2013), and include subsidiaries not only in developed countries but all countries, including tax havens. 26 In Table 5 Panel A, we present the results of a regression of effective tax rates on TIME and FORSTAT. In Model 1, the dependent variable is CASH ETR, which includes both domestic and foreign taxes paid relative to the firm s worldwide pretax income. The results indicate that the coefficient on FORSTAT is positive and statistically significant, with a value of (tstatistic = 2.93). This finding, together with the inference from Figure 5 of a decrease in foreign statutory tax rate, supports the explanation that the decrease in foreign statutory tax rates is at least partially responsible for the decrease in effective tax rates over time. In Model 2, we repeat the exercise using ETR ww as the dependent variable and reach similar conclusions. In Models 3 24 Statutory tax rate and GDP data come from the OECD Tax Database ( accessed October 18, 2014) as well as from KPMG, as in Williams (2015). 25 Note that the trend is not due to the weighting schema. Re-examining the trend using an unweighted average results in a trend in foreign statutory tax rates that begins at 44% in 1988 and declines to 25% in Exhibit 21 is required by Item 601 of SEC Regulation S-K to accompany the form 10-K filed with the SEC. The sample is restricted to those that have foreign subsidiary data from the Exhibit 21, which reduces the sample size to 11,175 firm-year observations. 18

20 and 4, we present the results separately for multinationals foreign and domestic current effective tax rates, ETR for and ETR dom, respectively. In Model 3, we find that when ETR dom is the dependent variable, the coefficient on FORSTAT is (t-statistic = -1.43), suggesting that foreign statutory tax rates do not affect domestic effective tax rates for multinationals. In contrast, in Model 4, when ETR for is the dependent variable, the coefficient on FORSTAT is positive and highly significant, with a value of (t-statistic = 7.85). It is important to note, however, that the TIME coefficient is uniformly negative across all models in Panel A of Table 5, indicating that changes in statutory foreign tax rates are only a partial explanation for the decrease in current effective tax rates over time. The TIME coefficient is smaller in magnitude in the ETR for regression; nevertheless, it is still negative and significant (coefficient = , t-statistic = -2.08). In Panel B, we repeat the analysis for large firms, and reach similar conclusions. Together, the results in Figure 5 and Table 5 suggest that the decline in effective tax rates over time is associated with the decline in foreign statutory tax rates over time. However, the decline in foreign statutory tax rates is at best a partial explanation for the declining effective rates observed in our data. It does not explain the declining rates in purely domestic U.S. firms and even among U.S. multinationals there is still evidence of declining effective rates over time after controlling for variation in foreign statutory rates, both with respect to U.S. multinational firms domestic effective tax rates and their foreign effective tax rates. 5.2 The effect of changes in firm characteristics We next examine the possibility that firms themselves have changed over time (e.g., Graham, Leary and Roberts, 2014), changing their mix of characteristics and abilities towards 19

21 those that yield lower effective tax rates. 27 As one example, U.S. firms receive a tax credit for certain expenditures related to research and experimentation. If firms have shifted investment away from fixed assets and into research and experimentation, then it is possible that more firms qualify for and receive tax credits, causing effective tax rates to drop on average. Another example could arise if rules or abilities of firms change. For example, if firms are required to depreciate capital assets over specified period of time, and that period of time is reduced by the government, then the effect of capital expenditures on cash effective tax rates will also change over time. Another method may be because of improvements in tax avoidance technology over time. For example, if firms are better able to shift income related to intangible assets to low tax jurisdictions now than they were in the past, then even holding the intangibility of firms constant, we could observe a decrease in effective tax rates over time. Thus, changes in firm characteristics could account for a decline in effective tax rates over time, simply because firms today look and act differently than firms did 25 years ago. Moreover, even holding the effect of firm characteristics constant, effective tax rates could change if the mapping of firm characteristics into taxes has changed over time. Figure 6 plots changes in these firm characteristics over time for our sample firms. 28 The results clearly show that many of these characteristics are changing over time among sample firms. For example, the figure suggests an increase in intangible assets over time, as measured by R&D and INTANGIBLE ASSETS, which is important because prior research suggests that firms are increasingly R&D intensive (Bates, Kahle and Stulz, 2009) and that firms with intangible 27 These variables are, to some extent, a result of choices by the firm. As an example, as discussed above, an increase in the opportunity for tax avoidance behavior because of an exogenous change, such as a decline in foreign statutory tax rates, could prompt the firm to take tax avoidance actions, such as shifting operations overseas, that in turn affect the firm s characteristics. 28 To minimize the impact of extreme outliers, all continuous variables are winsorized at the top and bottom 1% of their respective distributions. 20

22 assets have greater capacity for tax avoidance (e.g., Grubert and Slemrod, 1998; Koester, Shevlin, and Wangerin, 2013). The figure also shows an upward trend in firm size, as measured by LOG ASSETS, and a downward trend in capital assets, as measured by PP&E. Firms in our sample show decreasing LEVERAGE over the sample, and fairly steady CAPITAL EXPENDITURES. ADVERTISING EXPENSE at first declines and then increases over time. SPECIAL ITEMS become negative and larger in magnitude over the sample period, similar to the finding in Bradshaw and Sloan (2002) that special items have become more frequent over time. The percentage of firms reporting a non-zero NOL carryfoward increases greatly during the sample period, from approximately 18% of the sample in 1988 to approximately 51% in 2012, consistent with the findings in Collins, Pincus, and Xie (1999) and Givoly and Hayn (2000), who document an increase in the frequency of losses over the last few decades. Overall, we employ a set of controls that prior research has identified to be important drivers of tax avoidance and tax sheltering (e.g., Graham and Tucker, 2006; Chen et al., 2010; Kim, Li and Zhang, 2011; Hasan, Hoi, Wu and Zhang, 2014; Desai and Dharmapala, 2009). We examine the influence of firm characteristics on the trend in cash effective tax rates by estimating a regression of CASH ETR on TIME while allowing for variation in firm characteristics. The approach follows that of prior time-trend studies (e.g., Collins, Maydew and Weiss, 1997; Francis and Schipper, 1999; Collins, Li, and Xie, 2009), which examine the continued significance of TIME with the inclusion of additional explanatory variables. Thus, the model is: CASH ETR it = β industry + β 1 TIME t + β 2 LOG ASSETS it + β 3 R&D EXPENSE it + β 4 PP&E it + β 5 INTANGIBLE ASSETS it + β 6 LEVERAGE it + β 7 CAPITAL EXPENDITURES it + β 8 ADVERTISING EXPENSE it + β 9 SPECIAL ITEMS it + β 10 LAGGED SPECIAL ITEMS it 1 + β 11 NOL it + β 12 NOL it + ε it. (2) 21

23 The model includes industry fixed effects based on the classifications in Barth, Beaver, Hand and Landsman (2005) to account for any changes in the composition of firms across industries over time. We also employ standard errors clustered by firm and year. All variables are defined in the footnote to Table 2. Table 6 presents the results of the pooled OLS estimation of equation (2). Model 1 presents the results for the full sample of firms, while Models 2 and 3 present the results for multinationals and purely domestic firms, respectively. Similar to the models we estimated earlier without firm-level controls (i.e., in reduced form), we find that the coefficient on the time trend is negative and significant, equal to in Model 1, in Model 2, and in Model 3, with t-statistics ranging from to This suggests that even after controlling for firm characteristics, we continue to see a significant decrease in effective tax rates over time, both in the sample as a whole and in the subsamples of multinationals and purely domestic firms. Moreover, the decrease in effective tax rates is economically significant. For example, a coefficient of on the time trend corresponds to an approximately 10% decrease in the CASH ETR of a firm over the 25-year sample period, holding other variables constant. The coefficients on the control variables provide some important insights. In Model 1, we find that the coefficient on MNE is positive and significant, suggesting that multinationals during our sample period exhibited higher cash effective tax rates than did otherwise similar domesticonly firms. LOG ASSETS is positive in Model 3, suggesting that larger domestic firms have higher cash effective tax rates. In all three models, we find that the coefficient on R&D EXPENSE is negative, consistent with our expectation that greater expenditures on research and development are associated with reduced taxes paid on income. The coefficient on PP&E is 22