The Reputational Costs of Tax Avoidance

Authors


  • Accepted by Steve Salterio. The authors would like to thank Kathleen Andries, Darren Bernard, Jenny Brown, Nicole Cade, Charles Christian, Lisa De Simone, Katherine Drake, Don Goldman, Susan Gyeszly, Michelle Hanlon, Bradford Hepfer, Jeff Hoopes, Becky Lester, Kevin Markle, Zoe-Vonna Palmrose, Phil Quinn, Steven Savoy, Terry Shevlin, Michelle Shimek, Stephanie Sikes (Oxford discussant), Bridget Stomberg, Laura Wellman, Brady Williams, Ryan Wilson, participants at the 2012 Oxford University Centre for Business Taxation annual conference, Vienna University of Economics and Business, and reading groups at Arizona State, Iowa, MIT, and University of Texas-Austin for helpful comments. They especially thank Michelle Hanlon, Joel Slemrod, and Ryan Wilson for sharing tax shelter data and Bob Bowen, Andy Call, and Shiva Rajgopal for sharing Fortune Magazine reputation data. John Gallemore gratefully acknowledges the financial support of the Deloitte Foundation.

1 Introduction

This study investigates whether firms and their top executives bear reputational costs from engaging in aggressive tax avoidance activities. At least two decades of empirical tax research has shown that firms engage in a wide range of strategies for tax avoidance purposes.1 Recent studies suggest that for many firms, tax avoidance appears to be highly effective at reducing the firms' tax payments and increasing their after-tax earnings. For example, Dyreng, Hanlon, and Maydew (2008) find that more than a quarter of publicly traded U.S. firms are able to reduce their taxes to less than 20 percent of their pre-tax earnings, and are able to sustain such low rates of taxation over periods as long as ten years. Tax avoidance strategies are abundant and include a wide variety of activities such as shifting income into tax havens (Dyreng and Lindsey 2009), using complex hybrid securities (Engel, Erickson, and Maydew 1999), and engaging in other tax shelters (Wilson 2009).

While the evidence indicates there is wide variation in tax avoidance across firms, the extant literature has a difficult time explaining this variation. What is puzzling is not that some firms engage in tax avoidance, but rather why some firms engage in it enthusiastically while others appear to shun it. For example, while showing that some firms engage in substantial tax avoidance, Dyreng et al. (2008) also find that approximately one-fourth of firms pay taxes in excess of 35 percent of their pre-tax income over a ten-year period. Given a U.S. federal corporate tax rate of 35 percent, these firms appear to be engaging in little or no sustainable tax avoidance. The question of why so many firms do not avail themselves of tax avoidance opportunities has been coined the “under-sheltering puzzle” (Desai and Dharmapala 2006; Hanlon and Heitzman 2010; Weisbach 2002).

Reputational costs are often posited as an important factor that limits tax avoidance activities, particularly the most aggressive tax strategies.2 For example, the Commissioner of the Internal Revenue Service (IRS) asserts that aggressive tax strategies can pose “a significant risk to corporate reputations” and “the general public has little tolerance for overly aggressive tax planning” (Shulman 2009). However, empirical evidence on the reputational costs of tax avoidance is scarce. The most compelling evidence to date is Hanlon and Slemrod (2009) and Graham, Hanlon, Shevlin, and Shroff (2012). Hanlon and Slemrod (2009) examine the stock price responses of firms accused of engaging in tax shelters. They find evidence that firms suffer stock price declines following public revelation of tax shelter behavior. They are careful to acknowledge that there are many possible determinants of the negative returns, of which reputational costs are only one. With the exception of some tests on retail firms, they leave extensive testing of reputational costs for future research. Graham et al. (2012) survey tax executives and find that more than half agree that potential harm to their firm's reputation is an important factor in deciding whether or not to implement a tax planning strategy. This evidence is consistent with managers perceiving that aggressive tax avoidance will subject them or their firms to reputational costs.

Beyond this important initial evidence, we know very little about the reputational costs of tax shelters. In particular, we do not know if firms that are publicly scrutinized for having engaged in tax shelters actually bear reputational costs, as might have been feared ex ante. In their review of tax research, Hanlon and Heitzman (2010) call for research on the under-sheltering puzzle, specifically posing the following questions: “Why do some corporations avoid more tax than others? How do investors, creditors, and consumers perceive corporate tax avoidance?… These are interesting questions worthy of study” (137, 146). This study answers the call for research, focusing on the extent to which tax sheltering results in reputational costs.

We analyze a sample of firms identified in prior studies as engaging in aggressive tax shelters. Our study combines the samples of several prior studies of tax shelter behavior; namely, Graham and Tucker (2006), Hanlon and Slemrod (2009), and Wilson (2009). After imposing data requirements, our sample constitutes 118 firms revealed during the period 1995 to 2005 as having engaged in tax shelters. To our knowledge, this is the largest sample of publicly identified corporate tax shelters analyzed to date.

We maintain the assumption that managers act rationally in considering costs and benefits of a tax strategy for the firm. When the managers decide to engage in tax avoidance, they weigh the expected benefits of tax avoidance against the expected costs and will not engage in tax avoidance unless the net benefits are positive in an expected value sense. Hence, for our sample of tax shelter users, our assumption is that managers expected the tax shelters to yield positive net benefits. However, we distinguish between the ex post cost of getting caught and the ex ante decision to engage in tax avoidance. While the probability of bearing reputational costs may be low ex ante, those costs will be realized ex post for firms subject to scrutiny. Thus, a manager may rationally engage in tax sheltering even though doing so places the firm at risk of bearing costs if it is later challenged; our objective is to assess those potential ex post reputational costs.

Reputation is a multifaceted construct associated with several parties, including the firm and its management, as well as shareholders, customers, and tax authorities. We begin by examining the reputational effects of tax sheltering exerted by shareholders. To provide some assurance that the sample and research design have sufficient power, we replicate Hanlon and Slemrod (2009) using our sample and design to test for capital market reactions to tax shelter revelations. In the short window surrounding the revelation date, we find that the stocks publicly revealed to have tax shelters exhibit significantly negative abnormal returns, consistent with the findings in Hanlon and Slemrod (2009). We then expand the event period to 30 days past the revelation date to test whether the short-window effects on stock price are permanent or temporary. We find evidence that, although the immediate stock price response around the tax shelter is negative, in the days that follow, the stock price systematically reverses back to its pre-event levels. Thus, we confirm that the Hanlon and Slemrod (2009) short-window effect on stock price is a robust result and at the same time, we also find that it is a temporary effect that reverses within 30 days.3

Next, we examine the potential reputational effects on the firms' managers, specifically those related to their employment. We do not find evidence of increased chief executive officer (CEO), chief financial officer (CFO), or auditor turnover in the three years following tax shelter revelation relative to the turnover rates for matched control firms. Next, we assess whether customers exert a reputational cost on shelter firms. We find no differential change in sales, sales growth, or advertising expense for revealed shelter firms relative to that of control firms. We also examine how the revelation of a tax shelter influences the public media reputation of a firm, as measured by the Fortune magazine list for “Most Admired Companies,” following Bowen, Call, and Rajgopal (2010). We find no evidence that being caught in a tax shelter lowers the likelihood of making the Fortune list relative to the matched control sample.

Firms also face potential reputational consequences with the tax authorities. When the IRS learns that a firm has engaged in what it considers a tax shelter, its policy allows it to expand the scope of information that it requests from the firm, which can lead to the discovery of other aggressive tax avoidance. Accordingly, we examine whether firms accused of engaging in tax shelters become more conservative in the future regarding their tax avoidance, perhaps as a result of increased IRS scrutiny. If they do, we would expect to observe the effective tax rate (ETR) of tax shelter firms to increase following scrutiny of their activities. The data, however, show that the ETR of tax shelter firms remains approximately the same before and after discovery, and this is true regardless of whether ETRs are measured on a generally accepted accounting principles (GAAP) basis or on a cash basis. These results suggest that firms identified as tax shelter users may not suffer significant reputational costs even at the hands of the tax authorities.

In summary, across a battery of tests, we do not observe a reputational effect of tax sheltering. We are careful to note, however, that such an effect may indeed exist; but we are simply unable to find it empirically in our tests, perhaps because shelter firms are peculiar or because we have a small sample and/or low power. For example, it is possible that firms expecting nonzero reputational costs ex ante avoid tax shelter use completely and that only firms that are immune to reputational concerns engage in tax shelters. While we cannot rule out this possibility completely, it appears unlikely given the wide variety of firms we observe engaging in tax shelters.4 Moreover, in logistic regressions of tax shelter usage on proxies for reputation and other factors known to be associated with tax shelter usage (Wilson 2009), we find no evidence that reputation significantly influences the likelihood of tax shelter participation. In summary, the absence of ex post reputational costs and the insignificance of reputation as a determinant of tax shelter use suggest that reputational costs are not likely to be responsible for the significant variation in tax sheltering. Thus, it appears that under-sheltering is more of a puzzle than ever.5

2 Prior literature and research question

Prior literature on tax avoidance and tax shelters

There is a vast and growing empirical literature on tax avoidance, dating back at least as far as Scholes, Wilson, and Wolfson (1990) and continuing to the present day, that provides ample evidence that tax avoidance is pervasive and adaptable. Research has shown that firms engage in all manner of tax avoidance strategies, ranging from simple strategies like holding tax-exempt municipal bonds, to complex strategies such as debt–equity hybrid securities (Engel et al. 1999), cross-border avoidance strategies (Dyreng and Lindsey 2009), intangible holding companies (Dyreng, Lindsey, and Thornock 2013), and corporate-owned life insurance (COLI) tax shelters (Brown 2011).

While prior research indicates that some firms do not actively engage in tax avoidance (Dyreng et al. 2008), the forces that are curtailing more widespread tax avoidance are not well understood. On average, it is reasonable to assume that firms are engaging in the optimal amount of tax avoidance. A maintained assumption in the literature, and in this paper, is that the presence of costs of tax avoidance at some point outweigh the benefits. However, such costs have been difficult to identify.6 The lack of tax avoidance more generally, especially in nonregulated settings and settings with no financial reporting trade-off, has been dubbed the “under-sheltering puzzle.” Essentially, the question is: what is holding those firms back from taking advantage of known tax avoidance opportunities being used by other firms?

Reputational costs are often conjectured to be an important factor constraining tax avoidance, particularly the more aggressive forms of tax avoidance. COLI shelters are a useful example of a tax avoidance strategy that many viewed as particularly aggressive and that resulted in adverse scrutiny for firms that engaged in them. The COLI shelter involved firms taking out life insurance policies on their rank-and-file employees and then receiving the death benefits if the employee died (see Brown [2010] for a more detailed description). The COLI shelter was the subject of unflattering coverage in the media, including the Wall Street Journal, which identified the companies that engaged in COLI alongside pictures of their actual deceased employees (Schultz and Francis 2002).

As noted in Hanlon and Slemrod (2009), there is anecdotal evidence that some firms apply a “Wall Street Journal” test to their tax avoidance activities, whereby they forgo activities that would look unsavory if they were linked to the firm on the front page of the Wall Street Journal. There is also anecdotal evidence that companies want to be perceived as being good corporate citizens that pay their “fair share” of taxes, although evidence on this is mixed (Davis, Guenther, Krull, and Williams 2013; Watson 2011). General Electric, for example, has been criticized by the New York Times for paying no taxes to the U.S. government in 2011 despite being one of the largest companies in the world by earnings and by market capitalization.7 GE immediately responded by pointing out its other contributions to society, such as being a major employer and exporter, its prior tax payments, as well as tallying up a broader measure of its tax burden including payroll, property, and sales taxes paid.8

These examples are consistent with the conjecture that firms perceive reputational costs from aggressive tax avoidance, especially when subjected to media scrutiny. To date, however, there is little in the way of empirical evidence on the validity of that claim. The closest evidence on the reputational effects of tax shelters is provided by Hanlon and Slemrod (2009), who examine the change in firms' market values following public revelation that the firms engaged in tax sheltering, and by Graham et al. (2012), who survey tax executives in regard to their tax planning activities. Hanlon and Slemrod (2009) find that when tax shelter participation is revealed in the news media, the tax shelter firms suffer a decline in market value. Graham et al. (2012) survey tax executives and find that nearly 70 percent respond that potential harm to their firm's reputation is “very important” or “important” when deciding what tax planning strategies to implement. Moreover, responding to a question about tax disclosures, approximately half of executives surveyed responded that risk of adverse media attention was very important or important in reducing their firm's willingness to be tax aggressive.

Other research examines the role of political costs in determining effective tax rates, where political costs can be viewed as a type of reputational cost or at least related to reputational costs. Zimmerman (1983) hypothesizes that accounting choices are often driven by political costs, such as taxation, and finds that large firms have higher effective taxes, which he asserts are a function of the higher political costs faced by these firms. Mills et al. (2013) examine the influence of political costs on the tax avoidance in a sample of federal contractors. They find that federal contractors that are highly sensitive to political costs have higher effective tax rates, consistent with political costs driving the tax strategy of federal contractors.

Prior literature on corporate reputation

Most of the prior research on corporate reputation focuses on nontax settings. In accounting and finance, the extant literature examines the reputational costs of malfeasance on both firms and their managers. For example, Karpoff, Lee, and Martin (2008a) track the careers of over 2,000 managers who allegedly engaged in financial misrepresentation and find that 93 percent of these managers are fired and suffer a significant loss in personal wealth. Desai, Hogan, and Wilkins (2006) examine whether managers of firms suffer reputational costs following earnings restatements. They find that top managers of firms that restate their earnings experience more turnover than managers at a control sample of firms. Bowen, Call, and Rajgopal (2010) find evidence of negative effects from whistle-blowing events, including significant negative abnormal returns, restatements, shareholder lawsuits, and negative future operating performance. Prior research has shown that firms accused of accounting improprieties (Karpoff, Lee, and Martin 2008b; Dechow, Sloan, and Sweeney 1996), defective products (Jarell and Peltzman 1985), and violating environmental regulations (Karpoff, Lott, and Wehrly 2005) are subject to adverse effects following revelation of the alleged misdeeds.

However, there is also evidence that managers and firms are not always severely punished for improprieties. Beneish (1999) finds that managers of firms with extreme earnings overstatements do not have higher employment losses than managers of a matched sample of firms. Agrawal, Jaffe, and Karpoff (1999) find no evidence that managers and directors of firms charged with fraud have higher rates of turnover. Finally, Srinivasan (2005) finds that outside directors face labor market penalties following accounting restatements but not penalties through regulatory actions or litigation.

In sum, there is evidence of significant reputational costs accruing to both firms and managers in a range of nontax settings. This evidence suggests that it is reasonable to expect that tax avoidance can be accompanied by adverse reputational costs if it is discovered and subject to outside scrutiny. However, some evidence suggests the opposite is the case, in which managers go unpunished following instances of inappropriate behavior. Moreover, tax avoidance may be viewed in a completely different light than other corporate misdeeds because of the positive cash flow effects of such activity.

Research question: Does tax sheltering affect corporate reputation?

The reputation of the firm is a multifaceted construct that results from the impressions and perceptions of numerous interested stakeholders. However, as Walker (2010) notes, the concept of corporate reputation is somewhat difficult to define because it means different things in different contexts. In this paper, we take a broad view of reputation, in which stakeholders form a perception of the corporation that influences the interactions between the firm and stakeholders. The stakeholders of the firm include any party with which the firm has a vested interest in maintaining a relationship, including equity-holders, debt-holders, customers, suppliers, and governments, among others. We follow Fombrun and Shanley (1990) in defining reputation as something that is both created and consumed as part of a corporate strategy. Hence, a firm's reputation is an integral part of its comparative advantage, and as such, the firm and its management will act to protect and grow the reputation of the firm. To do so, it will strategically consider reputation in undergoing strategic initiatives and will make decisions, including tax decisions, based in part on the costs/benefits to its reputation.

Accordingly, we are interested in those aspects of the firm's reputation that have real consequences for the firm. This includes aspects of reputation that affect people's perceptions of the firm and its managers for honesty and fair dealing, which could affect the firm's brand values and sales, as well as scrutiny by parties the firm deals with such as customers, suppliers, the IRS, and outside auditors. These costs may manifest as increased advertising costs to counter reputational damage, increased effective tax rates from heightened IRS scrutiny, and increased auditor turnover.

We are also interested in the reputation for competence and the effects on top executives. Tax shelters that ultimately backfire and become the subject of intense IRS and media scrutiny may call into question the competence of the firm's managers. An important way that managers may internalize the reputational consequences of adverse scrutiny is through increased manager turnover following the tax shelter scrutiny, consistent with prior literature on nontax reputation-damaging events.

However, as Hanlon and Slemrod (2009) discuss, some stakeholders may actually react positively to news of tax shelter involvement, or at least they will not react negatively. As a result, involvement in a tax shelter may affect a firm's reputation differently than the corporate misdeeds listed above, for the following reasons. First, unlike accounting fraud, corporate tax sheltering is usually legal or at least in an area of the tax law with substantial gray area. Second, tax sheltering can improve the firm's after-tax cash flow if the IRS ultimately agrees with the firm's interpretation of the tax law. Third, the risk involved in tax sheltering may not be the same (or even rank anywhere near the gravity) as other risk factors that the firm faces, such as liquidity risk, competition, and going concern. Thus, the reputational risk of a tax shelter may not be of major importance to the firm or the parties with whom it does business.

3 Data and Sample

To assess firms' reputational costs of aggressive tax avoidance, we concentrate on firms that were subjected to media scrutiny for having participated in a tax shelter. Tax avoidance activities range from mundane strategies that are unlikely to attract adverse attention to aggressive strategies that have little economic purpose other than to reduce taxes. Tax shelters tend to be at the extreme end of the tax avoidance spectrum. By focusing on tax shelters that attracted media scrutiny, we give the best chance for the public (including shareholders and governments) to form a negative opinion about the firm. However, if we observe reputational costs in the sample of tax shelter firms, we need to be careful about generalizing the results to less aggressive forms of tax avoidance. Conversely, if we do not observe reputational costs in this extreme sample, then it is unlikely that more mundane forms of tax avoidance result in significant reputational costs.

Tax shelters are secretive by nature. As a result, it is difficult to identify a comprehensive sample of firms that are engaged in sheltering. Prior research has identified small samples of public cases of tax sheltering. This research includes Graham and Tucker (2006), 44 firm-year observations; Hanlon and Slemrod (2009), 107 firm-year observations; and Wilson (2009), 33 firm-year observations. To create a broad sample of tax shelter observations, we combine the samples in these prior studies of tax sheltering with 61 additional observations of the COLI shelter (Sheppard 1995), resulting in 245 tax shelter observations. To the best of our knowledge, we have the largest public tax shelter database employed in current and past research.9

To maximize power, we require the data to meet only minimal conditions to remain in the sample. Panel A of Table 1 presents the details of our sample composition. From the combined 245 observations, we remove 69 observations that are duplicated across the databases.10 We then remove foreign firms (3 observations), those for which the initial date of public scrutiny is unclear or missing (15), pre-1993 tax shelters (10), those with insufficient data (17), and those that upon closer inspection are not clearly tax shelters (2).11,12 Finally, we remove observations for which we cannot obtain a matched control firm (1 observation), according to the matching process described below. In the end, we are left with 118 shelter observations from 1995–2005. Panel B of Table 1 shows the number of tax shelter revelations in each year, from which we see that our shelter observations tend to cluster more in some time periods than others (e.g., tax shelters made public in 1995 comprise about 40 percent of our sample).13 The distribution of tax shelter revelations across time largely follows prior research that gave rise to the data (Graham and Tucker 2006; Hanlon and Slemrod 2009; and Wilson 2009). We perform sensitivity tests for the effects of temporal patterns in tax sheltering in section 5 below.

Table 1. Tax shelter sample
Panel A: Sample selection
 Shelter firm-years from prior researcha184
+Shelter firm-years from Tax Notesb61
Duplicate shelter firm-years69
 Total unique shelter firm-years176
Foreign firms3
Revelation date unclear or missing15
Revelation occurred before 199310
Insufficient data for matching17
Observations not classified as tax shelteringc2
 Shelter firm-years eligible for matching129
 Shelter firms eligible for matching119
Firms with no matched control1
 Final sample118
Panel B: Tax shelter revelations by year
YearRevelations

Notes

  1. a

    From Graham and Tucker (2006), Hanlon and Slemrod (2009), and Wilson (2009).

  2. b

    From September 25, 1995, Tax Notes article (Sheppard 1995).

  3. c

    Both incidents involved transfer pricing rather than tax sheltering.

199548
19967
19978
19982
19994
20004
20012
200218
200315
20049
20051

For the main tests, we match each revealed shelter firm-years (i.e., the treatment group) to a control firm from the same industry that is the closest in firm size (i.e., assets) in the year before tax shelter revelation.14,15 We employ financial statement data from COMPUSTAT and executive turnover data from ExecuComp from 1994 to 2011, although the sample using the Fortune reputation lists concludes in 2010.16 Table 2 contains the descriptive statistics (panel A) and correlations (panel B) for revealed shelter firms and their matched control firms. The variables in this table are measured in the first year in which the tax shelter was subject to public scrutiny. The statistics in panel A suggest that shelter firms are large, with mean (median) total assets of $10.3 billion ($10.0 billion). The variation in firm attributes in panel A is consistent with the notion that tax shelters are used by a wide variety of firms.17

Table 2. Descriptive statistics
Panel A: Descriptive statistics
Reputational variables N Treatment firmsControl firms
MeanStd. dev.MedianMeanStd. dev.Median
AD EXPENSE 1180.020.030.000.010.030.00
ΔAD EXPENSE 1180.000.000.000.000.010.00
SALES 1180.900.680.811.160.921.00
ΔSALES 1180.060.150.040.120.160.08
CEO TURNOVER 1070.140.350.000.050.210.00
ADMIRED 1180.040.200.000.020.130.00
Control variables N Treatment firmsControl firms
MeanStd. dev.MedianMeanStd. dev.Median
ABNORMAL RETURN 106−0.010.250.010.100.250.07
BTD 1060.000.050.000.010.030.01
CEO RETIRE 970.110.320.000.200.400.00
DISCRETIONARY ACCRUALS 98−0.010.07−0.01−0.010.060.00
EXTRAORDINARY ITEMS 1180.000.020.000.000.010.00
FOREIGN INCOME 1180.010.030.000.020.030.00
FOREIGN INCOME DUMMY 1180.470.500.000.490.500.00
INTANGIBLE ASSETS 1180.090.130.040.110.150.04
LEVERAGE 1180.190.150.170.190.130.17
MTB 1062.853.072.533.543.722.57
ΔNET OPERATING LOSSES 1180.000.030.000.000.010.00
NOL DUMMY 1180.220.420.000.160.370.00
PPE 1170.330.230.310.300.240.25
ΔPPE 1170.010.050.000.020.050.01
R&D EXPENSE 1180.020.040.000.020.030.00
ROA 1180.030.080.040.060.050.06
SENSITIVE INDUSTRIES 1180.420.500.000.420.500.00
SIZE 1189.241.769.218.831.418.74
SPECIAL 118−0.010.040.000.000.020.00
Panel B: Variable correlations (Pearson above/Spearman below)
Variable12345678910111213
1 AD EXPENSE 0.430.210.060.030.090.05−0.190.00−0.160.030.270.19
2 Δ AD EXPENSE 0.49 0.100.100.210.01−0.01−0.090.03−0.120.010.130.00
3 SALES0.280.11 0.560.08−0.01−0.03−0.11−0.11−0.080.020.080.09
4 Δ SALES0.060.200.61 −0.040.030.10−0.060.05−0.140.100.090.03
5 CEO TURNOVER0.060.160.100.01 −0.050.01−0.03−0.090.05−0.06−0.090.00
6 ADMIRED0.000.030.030.04−0.05 −0.010.060.240.010.020.170.13
7 ABNORMAL RETURN 0.060.00−0.040.070.02−0.01 0.050.06−0.110.070.04−0.04
8 BTD−0.160.00−0.15−0.040.010.090.07 0.050250.050.070.00
9 CEO RETIRE−0.120.01−0.080.07−0.090.240.080.01 0.04−0.010.200.12
10 DISCRETIONARY ACCRUALS−0.07−0.06−0.04−0.030.020.00−0.160.170.06 −0.090.010.02
11 EXTRAORDINARY ITEMS0.040.03−0.09−0.03−0.110.000.000.100.01−0.04 0.060.11
12 FOREIGN INCOME0.120.100.220.20−0.020.09−0.030.130.15−0.050.15 0.55
13 FOREIGN INCOME DUMMY0.110.070.230.140.000.13−0.060.030.12−0.040.140.82 
14 INTANGIBLE ASSETS0.140.050.060.04−0.030.020.000.04−0.01−0.050.060.250.27
15 LEVERAGE−0.010.050.02−0.030.14−0.07−0.020.10−0.090.13−0.15−0.16−0.14
16 MTB0.170.100.060.07−0.030.120.150.10−0.02−0.12−0.020.360.32
17 Δ NET OPERATING LOSSES0.080.11−0.040.020.28−0.170.040.09−0.02−0.05−0.05−0.11−0.04
18 NOL DUMMY0.050.060.100.060.130.11−0.07−0.070.06−0.020.100.080.18
19 PPE0.040.010.350.140.080.02−0.140.12−0.090.06−0.190.040.04
20 Δ PPE−0.060.140.220.49−0.010.11−0.120.180.100.040.010.100.06
21 R&D EXPENSE0.050.060.190.140.020.02−0.020.080.03−0.140.040.520.51
22 ROA 0.180.060.430.35−0.140.08−0.020.280.060.070.150.440.29
23 SENSITIVE INDUSTRIES0.150.15−0.13−0.080.05−0.050.05−0.01−0.060.090.01−0.29−0.31
24 SIZE−0.040.07−0.46−0.280.000.160.010.040.070.030.09−0.07−0.11
25 SPECIAl−0.060.070.030.070.020.09−0.030.190.050.310.030.06−0.03

Notes

  1. This table presents descriptive statistics for the variables used in our analyses, all of which are measured in the tax shelter revelation year. Panel A presents descriptive statistics for the sample and panel B presents the Pearson and Spearman correlations. The sample is composed of all treatment firms and their matched control firms with nonmissing data in the initial tax shelter revelation year. All variables are defined in the Appendix. All continuous variables are winsorized at the 1st and 99th percentiles. Differences in means between the treatment firms and matched control firms are indicated in panel A, with † denoting a significant mean difference at the 5 percent level.

Variable141516171819202122232425
1 AD EXPENSE0.130.040.200.110.010.02−0.070.100.120.06−0.13−0.08
2 ΔAD EXPENSE −0.050.070.100.180.010.050.070.15−0.020.10−0.03−0.11
3 SALES−0.05−0.040.02−0.050.060.180.15−0.040.220.02−0.400.01
4 ΔSALES−0.05−0.01−0.130.150.050.020.390.160.170.01−0.240.04
5 CEO TURNOVER−0.040.16−0.040.250.130.06−0.03−0.04−0.160.05−0.02−0.12
6 ADMIRED−0.02−0.070.21−0.050.110.010.080.040.08−0.050.190.05
7 ABNORMAL RETURN 0.050.010.110.04−0.04−0.15−0.08−0.010.100.010.000.06
8 BTD0.05−0.070.20−0.04−0.140.070.12−0.100.44−0.020.040.49
9 CEO RETIRE−0.01−0.090.100.080.06−0.100.09−0.030.03−0.060.080.08
10 DISCRETIONARY ACCRUALS0.020.080.03−0.19−0.100.070.02−0.230.140.060.050.40
11 EXTRAORDINARY ITEMS−0.03−0.08−0.01−0.02−0.02−0.140.280.050.370.08−0.030.08
12 FOREIGN INCOME0.06−0.210.42−0.070.090.020.050.270.39−0.24−0.030.08
13 FOREIGN INCOME DUMMY0.15−0.170.25−0.090.18−0.030.010.230.25−0.31−0.090.00
14 INTANGIBLE ASSETS 0.270.070.020.20−0.21−0.050.070.00−0.07−0.04−0.03
15 LEVERAGE0.24 −0.130.300.170.390.04−0.15−0.28−0.100.03−0.10
16 MTB0.20−0.07 −0.40−0.05−0.03−0.090.020.48−0.070.170.15
17 Δ NET OPERATING LOSSES−0.020.00−0.13 0.18−0.060.030.24−0.410.10−0.14−0.39
18 NOL DUMMY0.280.13−0.08−0.06 0.000.010.13−0.16−0.15−0.02−0.07
19 PPE−0.170.46−0.04−0.050.02 0.17−0.140.06−0.36−0.170.01
20 ΔPPE−0.060.000.050.08−0.020.20 0.100.210.00−0.100.06
21 R&D EXPENSE0.25−0.140.350.060.12−0.020.09 0.03−0.10−0.12−0.38
22 ROA 0.08−0.120.42−0.16−0.070.200.270.28 −0.09−0.170.49
23 SENSITIVE INDUSTRIES−0.06−0.15−0.070.16−0.15−0.370.01−0.33−0.11 0.22−0.02
24 SIZE0.010.060.14−0.060.00−0.18−0.06−0.15−0.290.21 0.08
25 SPECIAL−0.100.06−0.01−0.230.000.080.05−0.140.24−0.010.03 

4 The effects of tax shelter scrutiny on corporate reputation

In this section, we describe the empirical tests and results of possible reputational costs from being revealed as a tax shelter user. We call this the ex post analysis because it focuses on the ex post costs of being revealed as a tax shelter user.

The basic research design follows a differences-in-differences methodology, estimated in logistic form for binary dependent variables and in ordinary least squares (OLS) for continuous dependent variables:

display math(1)

In (1), REPUTATIONAL COST is one of several proxies for firm and manager reputational costs, measured in year t for a given firm i. CAUGHTFIRM is an indicator variable equal to one for firms that were revealed to have been in a tax shelter, and zero for the control firms. CAUGHTYEAR is an indicator variable set equal to one, for both treatment and control firms, in the year it was revealed the treatment firm had a tax shelter and zero otherwise. The variable of interest is the interaction of CAUGHTFIRM and CAUGHTYEAR, which reflects the reputational cost in the year the tax shelter is revealed, relative to the firm's matched control in the same year.

The difference-in-differences approach offers several advantages. First, it helps us to isolate the effect of being caught engaging in a tax shelter, separate from the effects of having the characteristics of a tax shelter user. Second, by evaluating differences between the revealed shelter firms and the matched control firms, we account for unobserved changes over time, such as changes in competitive and macroeconomic forces, which can confound interrupted time-series tests. The set of control variables differs depending on the specific reputational cost being examined. All variables are defined in the Appendix. In all analyses that follow (unless otherwise noted), we account for residual correlation by clustering the standard errors at the firm level, and all continuous variables are winsorized at the 1 percent and 99 percent levels within each group (treatment and control).18

Does tax shelter revelation affect equity pricing?

As a starting point, we replicate the result found in Hanlon and Slemrod (2009) that tax shelter revelations are associated with a short-window negative capital market response. Our objectives in doing so are twofold. The first objective is to replicate a previously documented finding from the tax shelter literature using our sample and research design to provide assurance that they have sufficient power. The second objective is to extend the event window to examine whether the negative abnormal returns associated with revealation of tax shelter involvement are permanent or temporary.

To replicate Hanlon and Slemrod (2009), we compare the abnormal returns for our sample of tax shelter firms to those for the matched sample during a three-day window surrounding the date of tax shelter revelation (CAUGHTPERIOD). Table 3 presents the results from regressing the daily abnormal return on CAUGHTFIRM, CAUGHTPERIOD, and their interaction. We find that for our sample of tax shelter firms on the revelation date, the average three-day abnormal return is negative and significant compared to the control group on the same dates, as indicated by the significant and negative coefficient on CAUGHTFIRM * CAUGHTPERIOD, with magnitudes similar to those found by Hanlon and Slemrod (2009). Figure 1 presents the average cumulative abnormal returns for the tax shelter firms, which shows the same effect visually.

Table 3. Capital market reaction to tax shelter revelation—replication of Hanlon and Slemrod (2009)
Variables(1)
ABNORMAL RETURN

Notes

  1. This table presents the results of OLS regression of abnormal daily returns on an indicator for the date on which a firm is revealed to have a tax shelter, an indicator for the firm, and the interaction of these two variables. The sample consists of treatment firms and control firms matched on industry and size, and includes the three-day period centered around the tax shelter revelation and the sixty days before and after this period. The dependent variable is the abnormal daily return, calculated as the firm's raw daily return minus the daily return for the CRSP value-weighted index. CAUGHTFIRM is an indicator variable equal to one if the firm has ever been revealed as having a tax shelter (i.e., a treatment firm), and zero otherwise. CAUGHTPERIOD is an indicator variable equal to one, for both the treatment and matched control firm, in the three-day period surrounding the first trading day subsequent to the day in which it was revealed that the treatment firm had engaged in a tax shelter, and zero for all other days. The estimated three-day CAR is calculated by multiplying the coefficient on the interaction term CAUGHTYEAR * CAUGHTPERIOD by 3. Daily abnormal returns higher than 25 percent or lower than −25 percent are winsorized. Coefficients are presented with t-statistics based on standard errors clustered by firm and trading day in parentheses. * represents significance levels of 5 percent all for two-tailed tests.

CAUGHTFIRM −0.0003
(−1.351)
CAUGHTPERIOD 0.0001
(0.083)
CAUGHTFIRM  *  CAUGHTPERIOD −0.0025*
(−2.030)
CONSTANT 0.0002
(0.959)
Observations25,810
Adjusted R20.000
Estimated three-day CAR 0.75%
Figure 1.

Mean cumulative abnormal return around shelter revelation

  • Notes: This figure plots the mean cumulative abnormal return for shelter firms in event time around the first revelation that the firm has engaged in a tax shelter. Cumulative abnormal returns are calculated by taking the difference between the firm's stock return and the value-weighted CRSP return each day and summing these differences starting on day −5. The y-axis denotes the cumulative abnormal return. The x-axis denotes the day through which the cumulative abnormal return was calculated, with day 0 being the day the shelter was revealed.

Under the notion that reputation is a multiperiod construct, we next evaluate whether short-window effects on stock price are temporary or permanent. Prior research has shown permanent stock price declines following discoveries of corporate malfeasance (e.g., Hennes et al. 2008). To test the permanence of the stock price decline, we plot in Figure 1 the cumulative abnormal return over a 36-day window, with the accumulation starting five days before tax shelter revelation and ending 30 days after revelation. The figure shows that when the event window is extended to 30 days following the tax shelter revelation, the abnormal returns revert to effectively zero. This evidence is consistent with a temporary response by investors.

Does tax shelter scrutiny lead to top executive turnover?

To examine whether firms that are revealed as having participated in a tax shelter experience higher CEO or CFO turnover subsequent to the revelation, we estimate (1) separately for CEO and CFO turnover.19 TURNOVER equals one if the CEO or CFO of firm i changed in year t, and zero otherwise. The coefficient of interest in this estimation of (1) is β3, which, if positive, indicates a greater incidence of CEO or CFO turnover following being accused of being in a tax shelter, relative to control firms over the same time period. In the implementation of (1) for executive turnover, the set of control variables are those that previous research has shown to influence CEO or CFO turnover (e.g., Engel, Hayes, and Wang 2003; Gilson 1989; Hennes, Leone, and Miller 2008; Menon and Williams 2008). The control set includes SIZE, ABNORMAL RETURN, ROA, LEV, and CEO RETIRE, as defined in the Appendix.20 All control variables in this test are measured in the year prior to the tax shelter revelation.

Figure 2 plots the frequency of CEO and CFO turnover in event time around the year in which the tax shelter was revealed (year 0). For each event year, we present the number of CEOs (panel A) or CFOs (panel B) that left the company. The darkly shaded columns measure the turnover frequency for revealed shelter firms and the lightly shaded columns measure the turnover frequency for matched control firms. For both CEOs and CFOs, visual inspection shows very little difference in turnover frequency between the revealed tax shelter firms and the control firms before or after the tax shelter revelation (i.e., in years 0, 1, and 2). In fact, Figure 2 reveals that only one CFO in our sample turned over in year 0, which is the smallest amount of CFO turnover in any year examined, and far less than the amount of CFO turnover for the control firms in the same year. With only a single instance of CFO turnover in the year of tax shelter revalation, multivariate tests on CFO turnover would not be very meaningful. Therefore, our multivariate analyses of executive turnover focuses on CEOs.

Figure 2.

CEO, CFO, and auditor turnover before and after tax shelter revelation

  • Notes: This figure plots the frequency of turnover for CEOs, CFOs, and auditors for treatment and control firms in event time around the revelation that the treatment firm has engaged in a tax shelter. Treatment firms are firms that are publicly revealed to have been engaging tax sheltering, while control firms are matched to treatment firms on total assets within the same industry in the year before shelter revelation. Panel A presents CEO turnover, panel B presents CFO turnover, and panel C presents auditor turnover. The horizontal axis represents the event year of tax shelter revelation, where year 0 is the year in which the tax shelter became public. The vertical axis represents the frequency of turnover for CEOs, CFOs, and auditors, respectively.

Table 4 presents the results of logistic regressions estimating (1), in which the dependent variable is an indicator variable equal to one if the firm experienced CEO turnover that year and zero otherwise.21 In column (1), there is no evidence that revealed shelter firms experience a significantly higher likelihood of CEO turnover than a matched control sample. In column (2), we find the same lack of significant increase in CEO turnover after including control variables that prior research has found to be associated with executive turnover. Overall, neither the univariate tests (Figure 2) nor the multivariate tests (Table 4) are consistent with reputational cost of tax shelter revelation manifesting in increased CEO or CFO turnover.22

Table 4. Tax shelter revelation and top management turnover
Variables(1)(2)
CEO TURNOVERCEO TURNOVER

Notes

  1. This table presents the results of a logistic regression of an indicator for CEO turnover on tax shelter variables and predicted determinants of executive turnover. CEO TURNOVER is an indicator variable equal to one if the firm's CEO changed that year, and zero otherwise. CAUGHTFIRM is an indicator variable equal to one if the firm has ever been revealed as having a tax shelter (i.e., a treatment firm), and zero otherwise. CAUGHTYEAR is an indicator variable equal to one, for both the treatment and matched control firm, in the year it was revealed that the treatment firm had engaged in a tax shelter, and zero otherwise. All control variables are measured with a one-year lag and are as defined in the Appendix. Coefficients are presented with Z-statistics based on firm-clustered standard errors in parentheses. ***, **, * represent significance levels of 1 percent, 5 percent, and 10 percent, respectively, all for two-tailed tests.

CAUGHTFIRM −0.0200.114
(−0.157)(0.736)
CAUGHTYEAR −0.599−0.777*
(−1.278)(−1.667)
CAUGHTFIRM  *  CAUGHTYEAR 0.7710.908
(1.310)(1.504)
SIZE t−1  0.045
 (1.064)
ABNORMAL RETURN t−1  −0.391
 (−1.544)
ROA t−1  −1.736
 (−1.507)
LEVERAGE t−1  0.830
 (1.599)
CEO RETIRE t−1  1.852***
 (9.913)
Observations2,8342,834
Pseudo R20.0010.080

Does tax shelter scrutiny lead to auditor turnover?

The implementation of tax shelters often involves the assistance or at least acquiescence of the firm's external auditor, particularly during the early portions of our sample period (Maydew and Shackelford 2007). Thus, if the revealed shelter firms face reputational costs of tax avoidance, they may hold their auditor responsible when the shelter goes bad. Figure 2, panel C presents the frequency of auditor turnover for revealed shelter firms relative to the matched control sample in event time surrounding the year of revelation. The figure shows that across the sample period as a whole, the frequency of turnover is exactly the same for the revealed shelter firms and their controls (ten cases of auditor turnover in each group). Moreover, in the year of revelation, none of the revealed shelter firms experienced an auditor turnover.

Does tax shelter scrutiny influence sales revenue and advertising expense?

We next examine whether firms accused of engaging in tax shelters suffer lost sales from customers and whether they had to increase advertising as a result. In this estimation of (1), SALES and AD EXPENSE are the dependent variables (both in levels and in changes form), and are defined in the Appendix. When SALES is the dependent variable, β3 <  0 would indicate a reduction of sales revenue for firms following adverse media coverage accusing them of being in a tax shelter. If firms respond to the negative publicity from the media coverage by increasing their advertising expenditures, then we expect β3 > 0 when AD EXPENSE is the dependent variable. Both findings would be consistent with firms suffering reputational costs from aggressive tax avoidance. We include a number of control variables in this implementation of (1) including SIZE, PPE, math formulaPPE, LEVERAGE, INTANGIBLE ASSETS, R&D EXPENSE, AD EXPENSE, NOL DUMMY, math formulaNET OPERATING LOSSES, SPECIAL ITEMS, EXTRAORDINARY ITEMS, FOREIGN INCOME DUMMY, and FOREIGN INCOME, all of which are defined in the Appendix.

Table 5 presents the results of the sales and advertising regressions. In columns (1) and (2), in which the dependent variables are SALES and ΔSALES, we see that neither sales nor sales growth decrease for revealed shelter firms following the tax shelter revelation.23 In columns (3) and (4), in which the dependent variables are AD EXPENSE and ΔAD EXPENSE, we see that advertising expense is no different for treatment firms than for control firms, nor do treatment firms increase their advertising relative to control firms following revelation of tax sheltering. We note that in each of these models, the coefficients on many of the control variables are significant, indicating that the estimation had sufficient power to yield significance. For example, in column (1), 10 out of 13 coefficients on control variables are significant. However, β3 is insignificant across all four estimations. Thus, across all models, we find no evidence of a significant reputational effect of tax shelter revelation that manifests in the form of reduced sales or increased advertising expense.

Table 5. Tax shelter revelation and sales revenue and advertising expense
 (1)(2)(3)(4)
Variables SALES Δ SALES AD EXPENSE Δ AD EXPENSE

Notes

  1. This table presents the results of OLS regression of sales and advertising expense on tax shelter variables and control variables. SALES (AD EXPENSE) is sales (advertising expense), divided by average total assets. ΔSALES (ΔAD EXPENSE) is the change in sales (advertising expense), divided by average total assets. CAUGHTFIRM is an indicator variable equal to one if the firm has ever been revealed as having a tax shelter (i.e., a treatment firm), and zero otherwise. CAUGHTYEAR is an indicator variable equal to one, for both the treatment and matched control firm, in the year it was revealed that the treatment firm had a engaged in a tax shelter, and zero otherwise. All other variables are as defined in the Appendix. Coefficients are presented with t-statistics based on firm-clustered standard errors in parentheses. ***, **, and * represent significance levels of 1 percent, 5 percent, and 10 percent, respectively, all for two-tailed tests.

CAUGHTFIRM −0.190**−0.035***0.003−0.000
(−2.147)(−3.288)(0.909)(−1.032)
CAUGHTYEAR −0.0240.0110.0010.001*
(−0.738)(0.870)(0.388)(1.666)
CAUGHTFIRM  *  CAUGHTYEAR −0.0140.010−0.000−0.000
(−0.365)(0.562)(−0.154)(−0.550)
SIZE −0.168***−0.018***−0.003***−0.000*
(−7.975)(−5.267)(−3.962)(−1.815)
PPE 0.368*−0.096***0.005−0.001
(1.658)(−3.176)(0.882)(−1.319)
ΔPPE 2.082***1.811***−0.0090.020***
(4.385)(13.260)(−0.524)(5.668)
LEVERAGE −0.824***−0.066*0.001−0.002**
(−3.091)(−1.945)(0.110)(−2.133)
INTANGIBLE ASSETS −0.371*−0.0450.0130.001*
(−1.807)(−1.587)(1.433)(1.680)
R&D EXPENSE −3.132***−0.132−0.057−0.010*
(−3.323)(−0.842)(−1.246)(−1.683)
AD EXPENSE 5.590***0.069  
(4.603)(0.472)  
NOL DUMMY 0.245***0.022**−0.003−0.000
(3.054)(2.010)(−1.122)(−0.397)
ΔNET OPERATING LOSSES 0.105−0.1250.050*−0.004
(0.169)(−0.570)(1.726)(−0.483)
SPECIAL ITEMS −0.887**0.194**−0.020−0.005
(−2.219)(2.018)(−0.894)(−1.180)
EXTRAORDINARY ITEMS −3.081**−0.277−0.002−0.012
(−2.354)(−0.787)(−0.040)(−1.432)
FOREIGN INCOME DUMMY 0.108−0.0170.001−0.000
(1.067)(−1.351)(0.269)(−0.751)
FOREIGN INCOME −0.3110.390**0.211***0.014**
(−0.273)(2.256)(3.169)(2.187)
Observations4,0694,0684,0694,069
Adjusted R20.2710.2580.0770.031

Does tax shelter scrutiny negatively influence firm reputation in the media?

We next examine the impact of tax shelter scrutiny on direct measures of the firm's overall reputation in the media. Following Bowen et al. (2010), we use a firm's presence on the Fortune “Most Admired Companies” list as a proxy for a high overall reputation.24 Specifically, we estimate (1) in logistic form with ADMIRED as the dependent variable, where ADMIRED is an indicator set equal to one if the firm makes the Fortune list in a given year, and zero otherwise.25 If firms suffer significant costs to their overall reputation in the media from tax shelter behavior, then we expect β3 < 0. We note that all of the firms in the sample are the subject of media scrutiny for their tax shelter use since that was a requirement to be in the sample. Thus, this test is an assessment of whether the scrutiny over one particular activity (i.e., tax shelter use) has adverse effects on the firm's overall reputation in the media. In Table 6, we find no evidence that, relative to the control sample, firms with tax shelters experience a significant change in their reputation once the tax shelter is made public. Many of the control variables are statistically significant in the direction one would expect.

Table 6. Tax shelter revelation and firm reputation in the media
 (1)(2)
Variables ADMIRED t+1 ADMIRED t+1

Notes

  1. This table presents the results of a logistic regression of firm reputation in the media on tax shelter variables and control variables. Our proxy for firm reputation in the media is ADMIRED, which is an indicator variable equal to one if the firm-year is included on Fortune Magazine's “Most Admired Companies” list, and zero otherwise. CAUGHTFIRM is an indicator variable equal to one if the firm has ever been revealed as having a tax shelter (i.e., a treatment firm), and zero otherwise. CAUGHTYEAR is an indicator variable equal to one, for both the treatment and matched control firm, in the year it was revealed that the treatment firm had engaged in a tax shelter, and zero otherwise. All other variables are as defined in the Appendix. Coefficients are presented with Z-statistics based on firm-clustered standard errors in parentheses. ***, **, and * represent significance levels of 1 percent, 5 percent, and 10 percent, respectively, all for two-tailed tests.

CAUGHTFIRM 0.087−0.228
(0.256)(−0.622)
CAUGHTYEAR −0.535−0.362
(−0.971)(−0.619)
CAUGHTFIRM  *  CAUGHTYEAR −0.069−0.104
(−0.099)(−0.132)
SIZE  0.775***
 (6.455)
PPE  1.872**
 (2.376)
ΔPPE  1.940
 (0.842)
LEVERAGE  −2.264
 (−1.447)
INTANGIBLE ASSETS  1.770**
 (2.065)
R&D EXPENSE  0.389
 (0.086)
AD EXPENSE  5.002
 (0.986)
NOL DUMMY  0.383
 (1.267)
ΔNET OPERATING LOSSES  −2.624
 (−0.779)
SPECIAL ITEMS  18.429***
 (3.712)
EXTRAORDINARY ITEMS  −6.137
 (−1.093)
FOREIGN INCOME DUMMY  0.574
 (1.284)
FOREIGN INCOME  7.735*
 (1.899)
Observations3,8363,836
Pseudo R20.0020.193

5 Additional tests

Reputation and tax shelter engagement

The prior tests have examined the ex post consequences to the firm reputation from public scrutiny of tax shelter involvement. In this subsection, we examine whether the firm's reputation is associated with the ex ante probability of engaging in a tax shelter. Panel A of Table 7 reports the frequency with which firms on the Fortune “Best Companies To Work For” or “Most Admired Companies” lists are identified as having engaged in a tax shelter, compared to publicly traded firms that do not make either of the Fortune lists. The data do not indicate that high reputation firms avoid engaging in tax shelters. Firm-years that make Fortune's lists have an 18 percent chance of being in our tax shelter sample, whereas firm-years not on the Fortune lists have about a 1.5 percent chance of being in our tax shelter sample. Finding that high reputation firms engage in tax sheltering is consistent with the tax sheltering resulting in little or no reputational cost. However, another possibility is that high reputation firms are immune to reputational costs. Perhaps some firms have innately high reputations without incurring costs to build those reputations. Accordingly, in panel B we examine actual advertising expenditures to reflect reputations that are costly to build. The results in panel B also do not suggest that high reputation firms avoid tax shelters. We are careful to emphasize, however, that these are merely univariate tests.

Table 7. The effect of reputation on tax shelter participation
Panel A: Percentage of firm-years included on Fortune “Most Admired Companies” or “Best Companies to Work For” list by tax shelter status
Reputation variableFirm-yearsTax shelter firm%
ADMIRED & BEST = 11,20917.78%
ADMIRED & BEST = 085,4991.49%
Difference 16.29%***
p-value 0.000
Panel B: Percentage of firm-years with above median advertising expense by tax shelter status
Reputation variableFirm-yearsTax shelter firm%

Notes

  1. This table reports differences in the proportion of sample firms with a tax shelter based on measures for high and low reputation. The proxy for reputation is the firm's inclusion on Fortune “Most Admired Companies” or “Best Companies to Work For” list (ADMIRED & BEST) or by the level of advertising expense (AD EXPENSE). Panel A measures high reputation as being included on one of the two Fortune lists. Panel B measures high reputation as having advertising expense above the median in a given year. For panels A and B, firm-years are classified as either being shelter firm-years (i.e., firms in our treatment sample) or control firm-years (all nontreatment firm-years on COMPUSTAT with total assets greater than $10 million). Panel A covers the period 1998–2010 and panel B covers the period 1983–2010. Proportional differences are tested using a two-sample t-test, and ***, **, * represent significance levels of 1 percent, 5 percent, and 10 percent, respectively, all for two-tailed tests.

High AD EXPENSE26,8453.43%
Low AD EXPENSE26,8281.81%
Difference 1.63%***
p-value 0.000

In Table 8, we extend these univariate tests to the multivariate setting to control for other determinants of tax shelter usage, following Wilson (2009) and Lisowsky (2010). For this test, we only use shelter firms for which we have data on when the firm was actively participating in the shelter. We collapse all years in which the firm was actively participating in a shelter into one observation, and we match each shelter-firm to a firm in the same industry with closest total assets in the first shelter participation year. We estimate a logistic regression of tax shelter usage (SHELTER) on variables from prior research associated with tax shelter usage and a number of proxies for reputational sensitivity: research and development (R&D EXPENSE), AD EXPENSE, the firm's market-to-book ratio (MTB), intangible assets (INTANGIBLE ASSETS), an indicator variable (SENSITIVE INDUSTRIES) for firms in the industries that are likely to be especially sensitive to reputational concerns, and ADMIRED as follows:26

display math(2)
Table 8. The likelihood of tax shelter participation and reputation
Variables(1)(2)
Shelter dummyShelter dummy

Notes

  1. This table presents the results of a logistic regression of tax shelter participation on tax shelter predictor variables and reputational variables, as in Wilson (2009) and Lisowsky (2010). For the purposes of this test, all years in which the firm was actively engaged in a tax shelter are collapsed into one observation, and all independent variables are calculated yearly then averaged over the shelter period. Treatment firms are matched to a firm in the same industry with the closest total assets in the first tax shelter year. The dependent variable is an indicator variable equal to one if the observation is a tax shelter, and zero otherwise. Column (1) contains all reputational proxies except for ADMIRED, and column (2) contains all reputational proxies. Reputational proxies include R&D EXPENSE, AD EXPENSE, MTB, INTANGIBLE ASSETS, SENSITIVE INDUSTRIES, and ADMIRED. All variables are defined in the Appendix. Coefficients are presented with Z-statistics based on firm-clustered standard errors in parentheses. ***, **, * represent significance levels of 1 percent, 5 percent, and 10 percent, respectively, all for two-tailed tests.

BTD 9.95725.316*
(1.253)(1.783)
DISCRETIONARY ACCRUALS 13.428***31.206***
(1.964)(2.824)
LEVERAGE −0.340−2.067
(-0.150)(−0.759)
SIZE 0.164−0.088
(1.056)(−0.244)
ROA 9.772*−0.074
(1.739)(−0.008)
FOREIGN INCOME DUMMY 0.0461.152
(0.074)(1.546)
R&D EXPENSE 3.071−10.897
(0.449)(−1.081)
AD EXPENSE 0.9843.564
(0.200)(0.630)
MTB −0.0090.275
(−0.056)(1.339)
INTANGIBLE ASSETS 2.2374.825
(0.768)(1.318)
SENSITIVE INDUSTRIES 0.3570.255
(0.561)(0.338)
ADMIRED  −2.143
 (−1.388)
Observations8458
Pseudo R20.1060.188

We include the market-to-book ratio, intangible intensity, and research and development as measures of reputational sensitivity because firms with high growth opportunities, high intangibles, or high research and development operate in businesses where explicit contracts are costly to write and enforce, which places more reliance on the trust of management by outside parties.27 Since this empirical specification is quite similar to that in Wilson (2009), we include a vector of control variables similar to the one used in that study.

The results in Table 8 reveal that several factors are associated with the likelihood of shelter participation, including discretionary accruals and return on assets, consistent with the findings in prior studies.28 However, in neither column do we observe any of the reputation variables being significantly associated with tax shelter use. Overall, these results do not suggest that firm-level reputation concerns affect whether firms engage in these tax shelters.

Alternate methods for creating matched control sample

In this section, we replicate Tables 4–6 using two alternative control groups: (1) a propensity score matched control group based on the probability of tax sheltering and (2) a control group that matches on effective tax rate, industry, and size decile. The objective with both of these alternative control groups is to match the revealed tax shelter firm with a control firm that had a similar ex ante probability of having engaged in a tax shelter but was not revealed as a tax shelter user ex post. The propensity-score matching that we use follows the procedure used in McInnis and Collins (2011) in that each firm is included just once in the sample.29 For each sample firm, we calculate a predicted likelihood (i.e., propensity score) of having used a tax shelter using the variables from the reduced model in Lisowsky (2010).30 We then match treatment firms to control firms (without replacement) from the same industry that have the closest propensity score in the year prior to the tax shelter revelation. Our second alternate control sample matches on GAAP ETR (defined in the Appendix) industry and size in the year prior to the tax shelter revelation. The idea behind this control sample is that while market participants may use effective tax rates as a simple heuristic to assess the likelihood that a firm has engaged in aggressive tax planning strategies, it is unlikely that the average investor assesses a firm's tax planning strategies using complex models such as those in Lisowsky (2010) and Wilson (2009).

Table S9 presents the propensity score regression (panel A) and the replication of Tables 4-6 using each of these alternate control samples (panels B and C).31 The propensity score regression shows sufficient ability to predict variation in tax sheltering with a psuedo R2 of 33 percent. Across panels B and C, the coefficient testing for reputation effects is significant in the predicted direction in only one regression out of 12.

Different time horizons

In the results presented above, a maintained assumption is that the reputational effect of tax sheltering manifests itself in a one-year time horizon. However, it is quite possible that the reputational effect of tax sheltering takes more time to affect the firm. To account for this possibility, we expand the time window for each of the tests presented above by changing CAUGHTYEAR in (1) to extend two years (CAUGHTYEAR2) or three years (CAUGHTYEAR3) past the revelation of the tax shelter.

We present the results of these tests in Table S10, with the two-year and three-year variables presented in panels A and B, respectively.32 Across the 12 tests in panels A and B, we do not find that firms caught engaging in tax shelters experience significant reputational damage during the two- to three-year period after the news of the tax shelter becomes public. The lack of evidence holds regardless of the reputational measure we use (executive turnover, sales, advertising, or media reputation).

Cross-sectional and time period sample partitions

It is reasonable to expect that the reputational effects of tax avoidance vary in the cross-section and over time. In this subsection, we repeat our analyses on several different cross-sections and time periods of the sample where reputation effects might be more likely to manifest. The subsamples are as follows: (1) firms with a low likelihood of being in a tax shelter (i.e., a low tax shelter propensity score, as described above); (2) firms with above-the-sample median ETRs; (3) firms with below-the-sample median ETRs; (4) firms in sensitive industries, as defined above; (5) shelter revelations that occurred in or after the year 2000; (6) firms specifically involved in the COLI tax shelter; and (7) firms not engaged in the COLI tax shelter. We expect that firms in subsamples 1 and 2 may suffer a larger reputational effect because the tax shelter revelation is more likely to be a surprise to outsiders. Subsamples 3 and 4 are designed to consider the possibility that high-reputation firms may have “reputational goodwill” to expend and thus not suffer as much when bad news comes forth (Decker 2010). However, Hanlon, and Slemrod (2009) find that low ETR firms suffer declines in market value around tax shelter revelations whereas high ETR firms suffer no loss of market value, consistent with the market imposing a penalty on only on firms perceived as too aggressive.

In addition, there may be changing perceptions about tax shelter use during our sample period. The period of the 1990s has been described as a period of high tax shelter use among corporations (Weisbach 2002). However, the financial scandals that came to light in the early 2000s (e.g., Enron and WorldCom) and the associated Sarbanes-Oxley Act may have changed firm and stakeholder perceptions of the tax shelter involvement. With that in mind, it is possible that a tax shelter revelation in the post-2000 period (subsample 5) is more likely to produce adverse reputation effects. Finally, as briefly mentioned above, the COLI shelters are an example of what some observers view as a particularly egregious shelter, and thus may be systematically different from non-COLI shelters (subsamples 6 and 7).33 We note that for each of these subsamples, we face a trade-off in that we focus even more sharply on those firms likely to suffer reputational costs of tax avoidance, but at the cost of a decrease in the number of observations.

We estimate the regressions from Tables 4, 5, and 6 on these seven different sample partitions and, for parsimony, present only the coefficients on the interaction term (CAUGHTFIRM CAUGHTYEAR) in Table S11.34 Hence, with six different dependent variables and seven different sample partitions, Table S11 presents the interaction coefficients from 42 estimations of equation ((1)). The results show that all 42 test coefficients are insignificant.

The effects of tax shelter revelation on subsequent tax avoidance

Being identified as a firm that engages in tax aggressive behavior could also have reputational consequences with the tax authorities. When faced with resource constraints, tax authorities are likely to allocate more audit resources toward firms known to engage in aggressive tax behavior, and to cast a more skeptical eye on those firms' activities when conducting the audit. If tax authorities increase their scrutiny of such firms, we expect the firms to react by reducing their level of tax avoidance in the post-revelation period.

In Figure 3, we plot the mean effective tax rates in event time around the year in which the tax shelter revelation occurred (event year 0) for both the revealed shelter firms and control firms. Panel A (panel B) presents the results for the annual measures of CASH ETR (GAAP ETR), as defined in the Appendix. The figures do not reveal a stark change in the ETRs of firms following revelation of tax shelter involvment. For the CASH ETR, the average hovers around 28 percent for revealed shelter firms and exhibits no distinguishable difference from the control firms. In summary, this analysis does not indicate that tax shelter revelation leads to decreased tax avoidance following the revelation of tax shelter use.35

Figure 3.

Effective tax rates before and after tax shelter revelation

  • Notes: This figure plots the effective tax rates (ETRs) for treatment and control firms in event time around the first revelation that the treatment firm has engaged in a tax shelter. Treatment firms are firms that are publicly revealed to have been engaging tax sheltering, while control firms are matched to treatment firms on total assets within the same industry in the year before shelter revelation. Panel A presents the CASH ETR, which is measured as cash taxes paid divided by adjusted pre-tax income (pre-tax income minus special items). Panel B presents the GAAP ETR, which is measured as current tax expense divided by pre-tax income. The y-axis shows the mean ETR, and the x-axis shows the year the ETR mean is calculated in relation to the year in which the shelter was revealed (year 0). Both ETR measures are winsorized at 0 and 1.

6 Conclusion

Across a multititude of tests, we do not find evidence that firms or their top executives face significant reputational costs from tax shelter involvment. The only exception is a temporary decline in stock price around tax shelter revelations that fully reverses within 30 days. When weighed against the large benefits to after-tax earnings from tax avoidance documented in Dyreng et al. (2008), the evidence in this paper casts doubt on reputational costs as a primary factor in explaining the “under-sheltering puzzle,” in which some firms take advantage of tax sheltering opportunities while others do not.

This paper also contributes to the literature on corporate misconduct, which has a long history in finance and accounting, including studies of managerial and corporate fraud, accounting restatements, and environmental violations. Across these studies, the evidence is mostly consistent that capital markets and labor markets exert heavy reputational penalties for corporate misconduct. However, our results suggest that firms and executives engaged in tax avoidance do not face significant reputational consequences, which is consistent with tax sheltering not being perceived as in the same category as misconduct. There are several possible explanations for the lack of an empirical association between tax sheltering and reputation penalties. First, if some stakeholders prefer tax avoidance while others do not, the net effect on a firm's reputation may be zero. Second, some stakeholders prefer that managers take appropriate risks, one of which may be tax sheltering (Rego and Wilson 2012). Finally, stakeholders may view tax sheltering as different from corporate misdeeds and thus not react in the same way to tax shelter news as they would to news of corporate misconduct.

Moreover, the empirical results are subject to some interrelated caveats. First, although we use tax shelters to focus on aggressive tax strategies, by definition our sample consists of tax shelters that were actually implemented. It is entirely possible that there are tax shelters so egregious that no firm implements them due to reputational concerns, and we cannot generalize to tax strategies that are never implemented. Second, firms and managers self-select into tax shelters, which means that it is possible that some firms forgo the tax benefits of sheltering to avoid jeopardizing their good name. While there is no way fully to rule out this possibility, we test this directly and find no association between the decision to engage in a tax shelter and several measures of reputation. Third, we can only speak to the reputational costs that we examine. Reputation is a broad concept that means different things to different people. We have endeavored to consider reputational costs imposed by a wide variety of parties (e.g., shareholders, customers, and tax authorities) that could manifest in various ways (e.g., managerial turnover, lost sales, and increased taxes). Despite our best efforts, however, it is possible that there are reputational costs of some form that we have not examined. For example, it is possible that lower-level executives, such as those in the tax department, do suffer turnover or other reputational costs. This is worthy of future inquiry. Fourth, it is possible that using other methodologies, including survey, case, and clinical research methodologies, may yield different inferences from those we find. We welcome more research on the subject.

In terms of future research, the results suggest that the “under-sheltering puzzle” is more of a puzzle than ever. If we assume that firms are acting optimally, there must be costs of tax avoidance that prevent more firms from engaging in it, but so far they have proved largely elusive. The evidence here suggests that reputational consequences, at least as they manifest at the firm level, are unlikely to be a major factor in explaining under-sheltering. Other costs can explain only small pieces of the puzzle. For example, financial reporting costs exist for certain tax avoidance strategies, but there are whole classes of tax strategies for which there is no financial reporting trade-off (Badertscher et al. 2009). The question of why some firms forgo tax avoidance, while others enthusastically engage in it, is very much an open question in the literature. We look forward to future research on the question.

Appendix

Variable measurements and descriptions

VariableDescriptionSource
AD EXPENSE Advertising expense, divided by average total assetsCOMPUSTAT
math formulaAD EXPENSECurrent advertising expense minus last year's advertising expense, divided by average total assetsCOMPUSTAT
SALES Sales, divided by average total assetsCOMPUSTAT
math formula SALES GROWTH Current sales minus last year's sales, divided by average total assetsCOMPUSTAT
CEO TURNOVER Indicator variable equal to one if the firm's CEO changes, and zero otherwiseExecuComp
ADMIRED Indicator variable equal to one if the firm is included on Fortune's “Most Admired Companies” list in a given year, and zero otherwise (variable is nonmissing from 1983 through 2011)Bowen et al. (2010)
ADMIRED & BEST Indicator variable equal to one if the firm is included on either Fortune's “Most Admired Companies” or “Best Companies to Work For” lists in a given year, and zero otherwise (variable is nonmissing from 1998 through 2010)Bowen et al. (2010)
CASH ETR Cash taxes paid, divided by adjusted pre-tax income (pre-tax income minus special items)COMPUSTAT
GAAP ETR Current tax expense, divided by pre-tax incomeCOMPUSTAT
ABNORMAL RETURN Annual stock return in excess of Center for Research in Security Prices (CRSP) value-weighted indexCRSP
BTD Book-tax difference; pre-tax income minus estimated taxable income (current tax expense grossed up by 35 percent, minus change in the tax loss carryforward), divided by total assetsCOMPUSTAT
CEO RETIRE Indicator variable equal to one if the CEO is 64 or older, and zero otherwiseExecuComp
DISCRETIONARY ACCRUALS Discretionary accruals as calculated using the cross-sectional modified Jones model with lagged ROA per Kothari et al. (2005)COMPUSTAT
EXTRAORDINARY ITEMS Extraordinary items, divided by average total assetsCOMPUSTAT
FOREIGN INCOME Pre-tax foreign income, divided by average total assetsCOMPUSTAT
FOREIGN INCOME DUMMY Indicator variable equal to one if pre-tax foreign income is nonzero, and zero otherwiseCOMPUSTAT
INTANGIBLE ASSETS Average intangible assets, divided by average total assetsCOMPUSTAT
LEVERAGE Average long-term debt, divided by average total assetsCOMPUSTAT
MTB Ratio of market value of equity to book value of equityCRSP
ΔNET OPERATING LOSSES Current net operating loss carryforward minus last year's net operating loss carryforward, divided by average total assetsCOMPUSTAT
NOL DUMMY Indicator variable equal to one if net operating loss carryforward is positive, and zero otherwiseCOMPUSTAT
PPE Average property, plant and equipment, divided by average total assetsCOMPUSTAT
ΔPPE PPE minus last year's property, plant and equipment, divided by average total assetsCOMPUSTAT
R&D EXPENSE Research and development expense, divided by average total assetsCOMPUSTAT
ROA Net income, divided by average total assetsCOMPUSTAT
SENSITIVE INDUSTRIES Indicator variable equal to one if firm is in the food, healthcare, retail, or financial industries (using the Fama-French 30 classification), and zero otherwiseCOMPUSTAT
SIZE Natural log of average total assetsCOMPUSTAT
SPECIAL ITEMS Special items, divided by average total assetsCOMPUSTAT

Notes

  1. 1

    For reviews of the literature on tax avoidance, see Hanlon, and Heitzman (2010), Maydew (2001), and Shackelford and Shevlin (2001).

  2. 2

    We define reputation as a general perception of the firm by all interested stakeholders and further discuss this concept in section 2.

  3. 3

    By comparison, prior research shows that the stock price declines associated with corporate malfeasance are permanent. For example, Hennes et al. (2008) show that the stock prices decline by nearly 30 percent following discovery of financial irregularities and stay at reduced levels.

  4. 4

    For example, our sample of shelter users contains firms from 15 of the 17 industries in the Fama-French 17 industry classification and has total assets ranging from $13 million to $1.3 billion.

  5. 5

    We test the robustness of our results using two different matching techniques (i.e., a propensity score matched control sample and an industry, size, and effective tax rate matched control sample), multiyear periods (i.e., one, two, and three years following tax shelter revelation), and subsamples of firms for which the potential reputational effect of tax shelter involvement is likely to be higher. Throughout all of these different specifications, we do not find significant reputational costs from tax avoidance.

  6. 6

    The costs identified are financial reporting costs and regulatory costs, which apply to a subset of tax avoidance strategies and a subset of firms, respectively. See Frank, Lynch, and Rego (2009) and Badertscher, Philips, Pincus, and Rego (2009) for examples of financial reporting costs of tax avoidance and Mills, Nutter, and Schwab (2013) for an example of the regulatory costs of tax avoidance .

  7. 7

    http://www.nytimes.com/2011/03/25/business/economy/25tax.html?_r=1, accessed September 30, 2011.

  8. 8

    http://www.gereports.com/setting-the-record-straight-ge-and-taxes/, accessed September 30, 2011.

  9. 9

    Other researchers (e.g., Lisowsky, Robinson, and Schmidt 2012) have obtained larger samples of tax shelter transactions using confidential IRS data. Although that sample is useful for many research questions, the tax shelters in those studies are unobservable to the public and thus would not be suitable for tests of reputational penalties exerted by outside stakeholders.

  10. 10

    Since this study is conducted at the firm-year level, if there is more than one tax shelter revelation in a year, we count that as one observation.

  11. 11

    We begin our sample period in 1993 because SFAS 109 changed the rules regarding the financial reporting of taxes, and we want to maintain similar accounting treatment of income taxes for all our observations.

  12. 12

    Both incidents involved transfer pricing, which can be considered an aggressive form of tax planning. However, these observations are likely to be quite different from the other tax shelters, and hence we remove them from the sample.

  13. 13

    This time clustering results because the Tax Notes article identifying COLI users was published in 1995.

  14. 14

    We require that potential control firms have data for each analysis. We require that control firms have a minimum of five years of data for the one-year regressions (CEO and CFO turnover, sales and ad expense, Fortune reputation lists). We make no such requirement of treatment firms. Thus in the one-year regressions, there is not necessarily a one-to-one mapping of firm-years between the treatment and control groups.

  15. 15

    For matching purposes in our ex post tests, we employ the Fama-French 17 industry classification, which can be found on Ken French's website: http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/

  16. 16

    To maximize the sample, we follow Dyreng and Lindsey (2009, 1296) and set the following variables to zero if missing: advertising expense, research and development expense, tax loss carryforwards, intangible assets, special items, and long-term debt. We also use their methodology to correct for errors in foreign tax expense, foreign pre-tax income, pre-tax domestic income, total pre-tax income, federal current tax expense, and worldwide current tax expense.

  17. 17

    Three other points regarding the sample are noteworthy. First, tests of differences between the treatment and matched control firms suggest that they are similar for many of the variables we consider. Second, although we have data for tax shelter revelations for 118 firms, we were able to obtain the actual years the firm was engaged in the shelter for only 24 of these firms. Most shelter revelations do not contain the years for which the firm was actively engaged in the tax shelter. Thus, we rely on the tax shelter usage data from Graham and Tucker (2006) and Wilson (2009), which constitutes only a small portion of our sample. From these limited observations, we find that the average time between the firm's engagement in the tax shelter and its revelation is approximately three to four years, which roughly corresponds to the IRS audit cycle. Finally, there are ten firms that are “repeat offenders” and for these firms we include only the first instance of tax shelter revelation in the sample.

  18. 18

    We do not include year fixed effects because CAUGHTYEAR will absorb time-specific differences in the year the tax shelter is revealed. In untabulated tests, we repeat the analysis including year fixed effects and find similar inferences as in the main tests.

  19. 19

    We are interested in CEO and CFO turnover because these are the most important positions in the firm. Moreover, executives set the tone at the top and are commonly held accountable for the decisions made under their watch, regardless of whether they directly influenced the decision. Ideally, we would also examine turnover in tax directors, who also play a critical role in the tax decisions of the firm (Armstrong et al. 2012). The tax director is the individual most likely to suffer reputational costs from aggressive tax avoidance. However, data on tax director turnover is not publicly available.

  20. 20

    We do not require a retirement variable when examining CFO turnover because the data are available for only about one-third of our sample. When we do require a retirement variable for CFO observations, inferences are the same (untabulated).

  21. 21

    Ai and Norton (2003) and Greene (2010) show that several mainstream statistical packages incorrectly calculate interaction effects in nonlinear models such logit regressions. However, Kolasinski and Siegel (2010) argue that the interaction effects presented by these statistical packages are economically meaningful. Our interaction effects throughout the paper are calculated as usual, but inferences are robust to using the “inteff” procedure in Stata developed by Ai and Norton.

  22. 22

    One possibility for this finding is that the legal proceedings for a tax shelter are sufficiently longer than the one-year window we evaluate. In subsequent tests we examine longer windows (up to three years) and find results consistent with the one-year results.

  23. 23

    Some research suggests that aggressive tax avoiders are more likely to inflate their sales or earnings (e.g., Frank et al. 2009). If we had found a decline in sales following tax shelter scrutiny that would have made interpretation of the decline more difficult, since it could have been due to actual lost sales or due to the firm curtailing its sales-inflating behavior. In an untabulated analysis, we find that inferences are unchanged if we examine operating cash flows instead of sales.

  24. 24

    The Fortune lists can be found at: http://money.cnn.com/magazines/fortune/most-admired/ and http://money.cnn.com/magazines/fortune/best-companies/, accessed May 13, 2013.

  25. 25

    We also use the firm's presence on the Fortune “Best Companies to Work For” list. Inferences are unchanged if we use an alternate dependent variable, ADMIRED & BEST, equal to one if a firm makes either list in that year, and zero otherwise (results untabulated).

  26. 26

    Specifically, SENSITIVE INDUSTRIES is equal to one for firms in the food products, healthcare, retail, and financial industries, according to the Fama-French 30 industry classification.

  27. 27

    We thank a referee for this suggestion. We also acknowledge that intangible intensity and research and development can proxy for other factors as well, including tax avoidance incentives and activities, which could complicate interpretation of those variables.

  28. 28

    In column (1) of Table 8, we present the results with all explanatory variables except for ADMIRED, because this variable is not available for part of the sample period and reduces the sample size. In column (2) we present the results with the full set of explanatory variables, including ADMIRED.

  29. 29

    We follow the design of Collins and McInnis (2011). The dependent variable in the propensity score model is an indicator variable equal to one if the firm is a shelter firm and zero otherwise. The sample for this test spans the period for which we have shelter revelations. Each firm is represented once in the regression; if control firms have more than one eligible firm-year in the period, we randomly select one of those firm-years to be included. All variables are measured the year before the revelation.

  30. 30

    Because not all of the variables examined by Lisowsky (2010) are readily available, we use the reduced prediction model from that paper, which is similar to the model in Wilson (2009).

  31. 31

    Please see supporting information, Table S9: “Alternate control groups” as an addition to the online article.

  32. 32

    Please see supporting information, Table S10: “Different time horizons” as an addition to the online article.

  33. 33

    We note that COLI firms make up nearly half of our sample but do not appear to be strongly different from non-COLI firms. We find significant differences between COLI and non-COLI firms in only 3 of 25 descriptive variables (SIZE, PPE, and SENSITIVE INDUSTRIES; tests unreported).

  34. 34

    Please see supporting information, Table S11: “Subsample analyses” as an addition to the online article.

  35. 35

    Our inferences remain the same if we use multivariate tests to control for cross-sectional and time-series determinants of ETRs (tests unreported).

Ancillary