Is Corporate Social Responsibility (CSR) - SSRN papers

Is Corporate Social Responsibility (CSR) Associated with Tax Avoidance? Evidence from Irresponsible CSR Activities

Chun-Keung (Stan) Hoi* Saunders College of Business Rochester Institute of Technology Rochester, NY 14623, USA Tel: (585) 475-2718 Email: [email protected] Qiang Wu Lally School of Management and Technology Rensselaer Polytechnic Institute Troy, NY 12180, USA Tel: (518) 276-3338 Email: [email protected] Hao Zhang Saunders College of Business Rochester Institute of Technology Rochester, NY 14623, USA Tel: (585) 475-3981 Email: [email protected]

The authors are grateful for helpful comments from participants of research workshops at RIT and RPI. For their guidance, we are particularly indebted to two anonymous referees and John Harry Evans III (the senior editor). We thank Patrick Scanlon for editorial help and Guangjun Ding for research assistance. Hoi and Zhang thank the Saunders College of Business at RIT for summer research support. *Corresponding author: Chun-Keung (Stan) Hoi, Saunders College of Business, Rochester Institute of Technology, 105 Lomb Memorial Drive, Rochester, NY 14623. Email: [email protected]; Tel.: (585) 475-2718; Fax: (585) 475-6920.

Is Corporate Social Responsibility (CSR) Associated with Tax Avoidance? Evidence from Irresponsible CSR Activities ABSTRACT: We examine the empirical association between corporate social responsibility (CSR) and tax avoidance. Our findings suggest that firms with excessive irresponsible CSR activities have a higher likelihood of engaging in tax sheltering activities and greater discretionary/permanent book-tax differences. Moreover, at the onset of FASB Interpretation No. 48, these firms have more uncertain tax positions; also, these firms’ initial tax positions are likely supported by weaker facts and circumstances as indicated by their larger post-FIN 48 settlements with tax authorities and their higher likelihood of a net decrease in the overall level of uncertain tax positions after FIN 48. Collectively, these results suggest that firms with excessive irresponsible CSR activities are more aggressive in avoiding taxes, lending credence to the idea that corporate culture affects tax avoidance.

Keywords: corporate social responsibility; tax avoidance; tax aggressiveness; corporate culture; FIN 48 Data Availability: Data are available from public sources identified in the study.

I. INTRODUCTION Corporate tax avoidance and corporate social responsibility (CSR) are important research topics in both accounting and management literatures, 1 and recent years have witnessed steady increases in both tax avoidance and CSR activities in corporate America. To date, research efforts in these areas have remained largely independent as little attention has been focused on the linkage between them. At the same time, both academicians and commentators have called for more investigation into the relation between them (e.g., Hanlon and Heitzman 2010; Sikka 2010). In this study, we provide evidence to explore the empirical association between aggressive tax avoidance practices and irresponsible CSR activities. Following Moser and Martin (2012), we adopt a broader perspective to evaluate CSR activities. CSR activities are corporate actions affecting all of the firm’s stakeholders including shareholders, employees, communities, government, customers, etc. From this point of view, irresponsible CSR activities include corporate actions that are widely regarded as damaging to corporate governance, employee relations, communities, public health, human rights, diversity, the environment, etc. 2 On one hand, consistent with corporate culture theories (e.g., Kreps 1990), one can treat CSR as a shared belief within a firm. In this context, CSR is the belief about the “right” course of actions that takes into account not only economic but also social, environmental and other

1

See, Hanlon and Heitzman (2010) for a recent comprehensive survey of tax research and Margolis et al. (2007) for a comprehensive survey on CSR. 2 We acknowledge that some activities that have a negative impact on non-shareholder stakeholders might have a positive impact on shareholders. An example is low-cost mining practices that pollute the environment; these activities are detrimental to the environment but they might be potentially beneficial to shareholders. It is difficult to determine whether these activities actually benefit shareholders or not. For instance, Margolis et al. (2007) review an extensive management literature on CSR and conclude that there is mixed evidence on the relation between CSR activities and shareholder value. As a robustness check, when we define irresponsible CSR activities as only corporate actions that are detrimental to non-shareholder stakeholders and use it in all of the tests, we find that the estimates have the same expected signs and comparable levels of statistical significance. For brevity, we do not report or discuss these results.

1

externalized impacts of company actions; it follows that irresponsible CSR activities are inconsistent with CSR. Likewise, aggressive tax avoidance practices are costly to society (Weisbach 2002) and they are widely viewed as “unethical” and “irresponsible” by the public and the popular press; it follows that overly aggressive tax avoidance is also likely to be viewed as inconsistent with CSR. Accordingly, aggressive tax avoidance practices should be positively, or at least non-negatively, associated with irresponsible CSR activities. On the other hand, one can treat CSR activities as a risk management strategy that a firm uses to enhance its CSR reputation, which, in turn, protects the firm against the risk of adverse political, regulatory and social sanctions/penalties in the case of negative corporate events (Godfrey 2005; Minor and Morgan 2011). Aggressive tax avoidance practices may lead to severe negative sanctions such as loss of firm/executive reputation, increased political/media pressure, potential IRS fines and penalties, and even consumer boycott (e.g., Hanlon and Slemrod 2009; Wilson 2009). Accordingly, firms could manage their CSR reputation by reducing (increasing) irresponsible (responsible) CSR activities (Godfrey 2005,) so as to lessen the expected costs associated with aggressive tax avoidance practices. If participation in CSR activities is a risk management strategy, aggressive tax avoidance practices should be negatively related to irresponsible CSR activities. Hanlon and Heitzman (2010, 137) state that “if tax avoidance represents a continuum of tax planning strategies where something like municipal bond investments are at one end, then terms such as ‘noncompliance,’ ‘evasion,’ ‘aggressiveness,’ and ‘sheltering’ would be closer to the other end of the continuum.” Given our focus, we are interested in capturing more aggressive practices toward the latter end of this continuum. We use three measures to capture aggressive tax avoidance, including Wilson’s (2009) tax sheltering probability measure, the permanent

2

book-tax differences (Frank et al. 2009), and the discretionary book-tax differences (Desai and Dharmapala 2006). Among these three measures, the tax sheltering probability is likely to capture the more aggressive practices (Lisowsky et al. 2013). For comprehensiveness and to relate to prior literature (Lanis and Richardson 2012), we also use the cash effective tax rate to measure the consequences of broad corporate tax avoidance practices (Hanlon and Heitzman 2010). We use negative social ratings obtained from KLD Research & Analytics, Inc., hereafter KLD, to measure irresponsible CSR activities. In particular, we capture egregious cases of firm activities that harm shareholders and other stakeholders using dummy variables that indicate firms with at least four irresponsible CSR activities. We compare the tax avoidance practices of these firms to those of other firms with fewer irresponsible CSR activities. All else equal, according to the culture hypothesis, we expect irresponsible CSR activities to be positively related to sheltering probability and discretionary/permanent book-tax differences. We expect these relations to reverse if the risk management argument is true. Using a large sample of U.S. public firms over the period 2003-2009, we find that firms with more irresponsible CSR activities, particularly those with excessive irresponsible CSR activities in a given year, have a higher probability of engaging in tax sheltering, greater discretionary/permanent book-tax differences and a lower cash effective tax rate, after controlling for firm performance, earnings quality, corporate governance, industry effect, year effect, and other factors influencing tax avoidance. In contrast, we find little evidence of systematic associations between tax avoidance measures and responsible CSR activities. We use the implementation of FASB Interpretation No. 48 (Accounting for Uncertainty in Income Taxes), hereafter FIN 48, as a natural quasi-experiment to further explore the link

3

between irresponsible CSR activities and aggressive tax avoidance. We find that at the onset of FIN 48, firms with four or more irresponsible CSR activities have more uncertain tax positions, and more importantly, these firms’ initial tax positions are likely to be supported by weaker facts and circumstances (i.e., weak positions) as indicated by their larger post-FIN 48 settlements with tax authorities and their greater likelihood of a net decrease in the overall level of uncertain tax positions after FIN 48. Further, settlements disclosed pursuant to FIN 48 could reflect material tax deficiencies/noncompliance detected by the IRS (Blouin et al. 2010, 808). Accordingly, the settlement finding could also imply that firms with excessive irresponsible CSR activities have more prior positions at the onset of FIN 48 that resemble tax noncompliance than firms with fewer irresponsible CSR activities. Overall, our findings lend credence to the idea that CSR could be viewed a facet of corporate culture that influences aggressive tax avoidance, including tax sheltering, undertaking weak positions and even noncompliance. The findings add to a line of emerging research which suggests that corporate culture influences corporate policies (Cronqvist et al. 2007; Fleischer 2007; Frank et al. 2011). This study is related to a growing stream of accounting literature on CSR (Dhaliwal et al. 2011, 2012; Kim et al. 2012); and, it relates most closely to Lanis and Richardson (2012), hereafter LR, who find a negative association between CSR disclosure levels and effective tax rates. LR argue that their findings may indicate that CSR influences tax avoidance decisions that account for “the well-being of society as a whole (LR 87).” Nevertheless, they concede that effective tax rates do not accurately reflect aggressive avoidance practices and CSR disclosures do not necessarily reflect CSR activities (LR 105).

4

In this study, we go beyond LR to examine the association between irresponsible CSR activities and aggressive tax avoidance such as discretionary/permanent book-tax differences, tax sheltering, and uncertain tax positions. In addition, we extend LR’s findings and show that only irresponsible CSR activities have a significantly negative association with effective tax rates while responsible CSR activities are not significantly related to effective tax rates. Further, we examine the association based on a large sample of firms over the period 2003-2009 using KLD data, a comprehensive third-party source to measure CSR activities. In contrast, LR use a small sample of Australian firms during the 2008/2009 fiscal years and rely on company self-reported CSR disclosures data, which are more likely to suffer from self-selection bias. Hanlon and Heitzman (2010, 137) point out, “clearly, most interest, both for researchers and for tax policy, is in intentional actions at the aggressive end of the [tax avoidance] continuum.” In this context, our results on tax sheltering probability and uncertain tax positions are relevant to tax researchers and policy makers. Prior studies have used financial statement information to predict the likelihood of noncompliance (Mills 1998) and tax sheltering (Wilson 2009). Our findings suggest that non-financial information such as irresponsible CSR activities could be incorporated into these models to improve their precision.

II. LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT Prior studies have examined the costs and benefits of tax avoidance (e.g., Desai and Dharmapala 2006; Chen et al. 2010), but have largely ignored the important issues of how corporate culture affects the decision to avoid taxes. In this section, we lay out two arguments relating CSR activities to aggressive tax avoidance practices and develop our hypotheses. The Corporate Culture Perspective on CSR and Tax Avoidance

5

Corporate culture could be viewed as a set of shared beliefs within the firm about the “right” corporate behavior, or the “optimal” course of action, or a set of conventions of doing business (Kreps 1990; Hermalin 2001). These complementary views lead to the notion that culture evolves as a means of defining appropriate corporate behavior, which could serve the interests of shareholders by facilitating coordination within the firm, reducing transaction costs, and serving as a substitute for costly explicit communication and negotiation (Hermalin 2001). In addition, some researchers argue that corporate culture reduces agency problems because it produces shared beliefs and/or conventions of business practices that persist over time (Van den Steen 2005, 2010). Theorists have also formalized the role of corporate culture in corporate policies (Kreps 1990; Hermalin 2001). Because corporate culture is complex and latent, identifying its effects in empirical analysis is difficult. Nevertheless, Cronqvist et al. (2007) find evidence of a significant culture effect using a matched-pair sample of spinoff-parent firms. Fleischer (2007) finds that tax sheltering activities and aggressive compensation strategy are related. Frank et al. (2011) find that aggressive financial policies and aggressive operating policies are related. Overall, findings from these studies indicate that corporate culture is an important factor affecting many corporate policies. Although informative, these prior studies have ignored CSR—another potentially important dimension of corporate culture. Consistent with Kreps (1990), we view CSR as the shared belief within the organization about the “right” course of action that takes into account the economic, social, environmental, and other externalized impacts of the company’s activities. 3 As corporate culture systematically

3

Other dimensions of corporate culture could include innovation, creativity, delegation of power, kinship, etc. It is likely that each firm has its own unique culture given its history, products, customers, employees, communities, regulatory environment, visibility, etc. Nevertheless, if culture is a cost-effective means of defining appropriate

6

affects corporate decisions (Kreps 1990; Hermalin 2001; Fleischer 2007; Frank et al. 2011), we conjecture that CSR should influence corporate practices affecting the firm’s various stakeholders. CSR activities are intimately related to the well-being of many of the firm’s stakeholders; tax avoidance activities affect the government’s claim on the firm and the welfare of the society (Weisbach 2002). As such, CSR should influence both CSR activities and tax avoidance activities. Consistent with Moser and Martin (2012), we adopt a broad perspective to evaluate CSR activities. CSR activities are corporate actions widely regarded as having a significant impact on all of the firm’s stakeholders including shareholders, employees, communities, government, customers, etc. Moser and Martin (2012, 799) call for more accounting studies to go beyond the shareholder’s perspective when examining CSR on the basis that using this approach “will expand our understanding of CSR issues beyond what can be learned if we maintain the traditional (shareholder) perspective.” We specifically focus on irresponsible CSR activities because (1) these activities should be inconsistent with CSR as they are widely regarded as detrimental to shareholders and nonshareholder stakeholders; (2) companies most likely will try to avoid these activities but they might voluntarily pursue responsible CSR activities that are viewed as beneficial to stakeholders; 4 and (3) existing empirical evidence suggests that irresponsible CSR activities have explanatory power regarding the underlying CSR construct they intend to capture (Chatterji et al. 2009; Goss and Roberts 2011) while responsible CSR activities are too self-serving and tainted to provide such information (Neu et al. 1998; Cho and Paton 2007).

corporate behavior, it is likely that a firm’s culture evolves in such a way that the resulting corporate practices account for the welfare of shareholders and those of other non-shareholder stakeholders of the company. 4 We thank an anonymous referee for making this suggestion.

7

Although the payment of tax is a fundamental way corporations engage with society, it is seldom classified as a significant CSR activity. 5 But corporate tax avoidance practices are costly to society (Weisbach 2002) 6 and aggressive tax avoidance practices are often regarded by some of society’s arbiters as “irresponsible,” “unethical” and even “unpatriotic”.

7

Moreover,

aggressive tax avoidance practices can be viewed as an opportunistic behavior whereby the firm is exploiting the implicit contract between the firm and society at the expense of the latter. It follows that aggressive tax avoidance should be inconsistent with CSR. Taken together, the preceding arguments imply that if corporate culture drives company policies, then irresponsible CSR activities and aggressive tax avoidance practices are likely to be positively related, leading to the following hypothesis. H1: Aggressive tax avoidance practices are positively associated with irresponsible CSR activities. The Risk Management Perspective on CSR and Tax Avoidance

5

This phenomenon could be due to the generally accepted legal principle that asserts that tax payers may organize their activities in such a way as to pay the least tax possible under the law. Consistent with this view, Christensen and Murphy (2004) find that corporate directors and executives frequently regard tax minimization as a primary fiduciary duty. Sikka (2010, 153) characterizes the significant discrepancy between CSR and corporate tax avoidance practices as “organized hypocrisy”. 6 A significant cost is the revenues lost due to tax avoidance activities which amount to tens of billions of dollars in any given tax year. The reduction in revenues due to tax avoidance could lead to a variety of detrimental effects. For example, if government budget is constant then this gap must be plugged by other means, implying higher taxes on those who do not avoid taxes, an externalized cost. In contrast, if government budget is shrunk because of tax avoidance then at least some government programs must be cut back, leading to a welfare transfer from those who rely on these programs. Indeed, Weisbach (2002) argues that tax planning activities are non-productive in the sense that they do not generate net social benefits. As such, resources allocated to it can be regarded as a deadweight loss to the society. 7 In 2011, Senator Carl Levin introduced the “Stop Tax Haven Abuse Act” while claiming that “offshore tax abuses are not only undermining public confidence in our tax system, but increasing the tax burden on middle America.” Society’s arbiters rallying against aggressive corporate tax avoidance include the public media, legislators, regulators, non-government organizations, etc. For example, the off-shore tax avoidance practices by Microsoft and Hewlett-Packer, for instance, have recently received extensive media coverage in both the U.S. and the U.K., including popular outlets such as the NBC News, the BBC News, etc. Moreover, these corporate avoidance activities have resulted in an U.S. Senate panel investigation headed by both Senator Levin, a democrat, and signed on by Senator Tom Coburn, a republican. In a press conference, Senator Levin stated that such tax practices are “gimmicks range from egregious to dubious validity.”

8

In contrast, there is an emerging trend among academics and in the popular press focusing on risk management implications of CSR activities. This perspective is expressed as “CSR is best seen as the management of risk, as the avoidance of damages to the company’s reputation” (Financial Times 2004). According to the risk management argument, a firm could serve the interests of its shareholders by managing its positive CSR reputation which can potentially mitigate the risk associated with negative corporate events. When a negative corporate event occurs, society’s arbiters must interpret the conditions surrounding the events and may conclude that sanctions are appropriate. Godfrey (2005) theorizes that a positive CSR reputation is particularly important when negative corporate events occur because it provides some degree of insurance protection by increasing the likelihood of positive attributions from society’s arbiters “who then temper their negative judgments and sanctions toward firms because of this goodwill” (Godfrey 2005, 425). Negative corporate events can arise inadvertently and unexpectedly, such as oil spills. As a pre-emptive measure, firms could either reduce irresponsible CSR activities or increase responsible CSR activities to build up the firm’s positive CSR reputation so as to mitigate the risk of severe sanctions when negative events occur. Godfrey et al. (2009) and Minor and Morgan (2011) find evidence in support of the risk management argument. Their results suggest that (1) reducing irresponsible CSR activities and/or increasing responsible CSR activities can enhance a firm’s positive CSR reputation and (2) a positive CSR reputation provides some degree of insurance protection against the risk of market, political, regulatory and social sanctions when negative corporate events occur. Aggressive tax avoidance practices might result in significant negative sanctions and judgments toward firms because they are costly to society and likely resemble opportunistic

9

behavior contrary to societal interests. Survey evidence indicates that tax executives are sensitive to these potential negative sanctions when deciding what tax avoidance strategies to implement (Graham et al. 2012, 3). If a positive CSR reputation reduces the severity of these negative sanctions, then firms might manage CSR activities to hedge against the consequences of aggressive tax avoidance activities. More specifically, firms could either reduce irresponsible CSR activities or increase responsible CSR activities to build up their CSR reputation to lessen the severity of potential negative sanctions associated with undertaking aggressive tax avoidance activities. Accordingly, the risk management perspective suggests that irresponsible (responsible) CSR activities and aggressive tax avoidance activities should be systematically and negatively (positively) related. 8 Our focus is on the link between aggressive tax avoidance and irresponsible CSR activities. As such, we formulate the corresponding prediction from the risk management perspective as follows. H2: Aggressive tax avoidance practices are negatively associated with irresponsible CSR activities.

III. RESEARCH DESIGN Measures of Aggressive Tax Avoidance We use various measures to capture aggressive tax avoidance because each measure has its own limitations (Hanlon and Heitzman 2010; Lisowsky et al. 2013). Using multiple measures

8

However, if tax avoidance activities result in either highly uncertain or trivial negative sanctions/judgments toward firms, then responsible or irresponsible CSR activities and tax avoidance may not be systematically associated. Consistent with this conjecture, Hanlon and Slemrod (2009) find a negative but relatively small stock price reaction to news about company involvement in tax shelters. Further, they also provide some evidence that not all social arbiters view tax avoidance as a significant issue because only 67 percent, or 72 of 108, of all the sampled news stories about tax shelters are covered by major press outlets such as Associated Press, Reuters, the Wall Street Journal, etc. Gallemore et al. (2012, 1) find “no consistent evidence that firms or their top executives bear significant reputation costs as a result of being accused of engaging in tax shelter activities.”

10

to triangulate the results is advantageous because if results across various measures are consistent, one can be more confident that they are robust. We use two adjusted book-tax difference measures to capture aggressive tax avoidance practices. These measures are (1) the Desai and Dharmapala (2006) discretionary book-tax difference for firm i in year t (DD_BTit) and (2) the Frank et al. (2009) permanent book-tax difference for firm i in year t (DTAXit). For comprehensiveness and to relate our findings to those in a prior study (Lanis and Richardson 2012), we also use CETRit to capture consequences of broad tax avoidance practices. CETRit is the ratio of cash tax paid over pretax income for firm i in year t (Dyreng et al. 2010). 9 We present detailed definitions of all variables in Appendix A. Given our focus, we are most interested in empirical proxies that could capture extremely aggressive tax avoidance practices. Researchers have used incidences of a tax audit adjustment from the IRS as a proxy for noncompliance (Mills 1998; Mills and Sansing 2000). Likewise, incidences of a tax shelter position disclosed in the firm’s tax return on Form 8886 or IRS Schedule M-3 (Lisowsky 2010; Lisowsky et al. 2013) and public disclosures of large tax shelter cases (Graham and Tucker 2006) have been used as proxies for tax sheltering activities. However, empirical analyses based on these measures may be subject to selection bias and endogeneity issues (Hanlon and Heitzman 2010). Moreover, data requirements for cases of tax sheltering and IRS audit adjustment have limited prior studies to analyzing small samples that require access to confidential IRS data. Using actual tax sheltering cases, Wilson (2009) establishes several empirical models that predict the likelihood that a firm is currently engaging in tax sheltering activities. These models

9

We also use other tax rate measures to capture consequences of broad tax avoidance practices, including the effective tax rate and the ratio of cash taxes paid to pre-tax cash flows. The results are quantitatively similar to those based on CETRit in that the estimates have the same expected signs and comparable levels of statistical significance. For brevity, we do not tabulate or discuss results using these alternate tax rate measures.

11

allow researchers to use publicly available financial/accounting information to estimate concurrent sheltering probabilities. However, they require an out-of-sample estimation procedure that may generate noisy measures. Nevertheless, Kim et al. (2011) and Rego and Wilson (2012) find that Wilson’s (2009) sheltering probabilities have construct validity. Further, Wilson’s (2009) sheltering probabilities are associated with stock price crash risk (Kim et al. 2011) and the sensitivity of a manager’s wealth to stock return volatility (Rego and Wilson 2012). Following Kim et al. (2011) and others, we use the model as reported in Table 5, Column 3 in Wilson (2009) to estimate a tax sheltering probability for each firm in each year. We then rank the estimated tax sheltering probabilities annually to create a dummy variable, SHELTERit, that equals one if the firm i’s estimated tax sheltering probability belongs to the top quartile in that year and zero otherwise. We use SHELTERit as a proxy for extremely aggressive tax avoidance practices (Lisowsky et al. 2013). 10 Appendix A provides detailed information concerning this variable. Measuring Irresponsible CSR Activities We use negative social ratings from KLD to measure corporate activities that are widely recognized as having a negative impact on shareholders and non-shareholder stakeholders. The KLD database contains firm-year data on 34 binary scores for such activities in the seven categories of corporate governance, employee relations, environment, community, diversity, human rights, and product quality and safety. The variable, NEG_CSRit, is the total number of

10

The sheltering model in Table 5, Column 3 in Wilson (2009) is relatively comprehensive in that it includes the effect of discretionary accruals. Nevertheless, we also use two alternative models reported in Wilson (2009) to check the robustness of our findings. Following Rego and Wilson (2012), we use the alternative sheltering model as reported in Table 5, Column 1 in Wilson (2009) which does not include discretionary accruals. We also use the model as reported in Table 4, Column 3 in Wilson (2009) to estimate the tax shelter prediction score; this model include discretionary accruals but the estimates are based on a smaller, more precise set of control firms. The estimates from these alternative models are comparable to those obtained from the sheltering model in Table 5, Column 3 in two regards: they have the same expected signs and comparable levels of statistical significance.

12

these scores for firm i in year t. Appendix B provides detailed information concerning the KLD database. We expect NEG_CSRit to be correlated with the firm’s beliefs concerning CSR, but we recognize that it is an imperfect proxy because it is difficult to operationalize the underlying CSR construct in a continuous scale. Some irresponsible CSR activities could be accidental and unintended, such as oil spills, making it difficult to find a positive association between NEG_CSRit and tax avoidance among firms with very limited exposure to irresponsible CSR activities. Even so, if excessive irresponsible CSR activities are more likely to reflect negative behavior toward shareholders and non-shareholder stakeholders, then excessive irresponsible CSR activities and aggressive tax avoidance should be positively related. As a consequence, we use the dummy variable, HIGH_NEG_CSRit, to capture firms that are engaging in a high level of irresponsible CSR activities. The dummy variable, HIGH_NEG_CSRit, equals one for firm i in year t if firm i has four or more irresponsible CSR activities in year t, and zero otherwise. 11 We regard a positive association between HIGH_NEG_CSRit/NEG_CSRit and SHELTERit/DTAXit/DD_BTit as prima facie evidence for the culture argument. We use HIGH_NEG_CSRit as the main test variable for our analyses. Baseline Regression Models Following Manzon and Plesko (2002), Frank et al. (2009), and Chen et al. (2010), we use the following baseline regression model to test our hypotheses: AGGRESSIVEit = β0 + β1 NEG_CSRit/HIGH_ NEG_CSRit + β2 POS_CSRit + β3 ABS_DAit + β4 IOit + β5 CASHit + β6 ROAit + β7 LEVit 11

We choose four irresponsible CSR activities as the cutoff point because the data in our sample, as reported in Table 2, suggest that 7.5 percent of sample firms have four or more irresponsible CSR activities in any given year. In addition, as discussed in Section 4.2, our untabulated results also show that firms with HIGH_NEG_CSRit equal to one have a more than 85 percent probability of maintaining the same level of irresponsible CSR activities in the next three years, indicating that firms with excessive irresponsible CSR activities are likely to be persistently different from other firms with significantly fewer irresponsible CSR activities. We examine the sensitivity of our findings with respect to different cutoffs and discuss those findings in Section VII.

13

+ β8 NOLit + β9 ∆NOLit+ β10 FIit + β11 PPEit + β12 INTANGit + β13 EQINCit + β14 R&Dit + β15 EMPit+ β16 ∆SALEit + β17 SIZEit-1 + β18MBit-1 + β19 Lag(Dependent Variable) + Year Dummies + Industry Dummies + εit;

(1)

where AGGRESSIVEit represent the several empirical measures for aggressive tax avoidance as discussed earlier. NEG_CSRit and HIGH_NEG_CSRit represent the two measures of irresponsible CSR activities. We report baseline results in Table 3 using both NEG_CSRit and HIGH_NEG_CSRit. We include a number of variables to control for the effects of earnings quality, corporate governance, firm performance, and other firm characteristics. Appendix A presents the detailed definitions of these control variables. Absolute value of performance-adjusted abnormal accruals (ABS_DA) are negatively related to CSR activities (Kim et al. 2012) and positively related to aggressive tax avoidance (Frank et al. 2009). Therefore, we use ABS_DA as a control variable in the baseline regression models to ensure that the association between CSR activities and tax avoidance is not driven by earnings quality. Following Desai and Dharmapala (2009), we use institutional ownership (IO) to capture the impact of corporate governance on tax avoidance. In addition, we include a range of variables to control for profitability (ROA, NOL, ∆NOL), liquidity (CASH), leverage (LEV), foreign operations (FI), firm size (SIZE and EMP), firm growth opportunity (∆SALE and MB), and other firm attributes (i.e., PPE, INTANG, EQINC, R&D) that could potentially affect our tax avoidance measures. We include POS_CSR to explore the effects of responsible CSR activities. Lastly, we include dummy variables to control for year fixed effects and we use two-digit SIC industry dummy variables to control for industry fixed effects.

IV. SAMPLE SELECTION AND SUMMARY STATISTICS

14

Sample Selection We estimate the baseline regression model using data from three sources. We obtain financial accounting data from Standard and Poor’s Compustat database, data on CSR activities from KLD’s social ratings database, and institutional ownership (13f) data from Thomson Reuters Ownership Database. Following prior literature on tax avoidance, we exclude utilities (SIC codes 4900-4949) and finance companies (SIC codes 6000-6999). We merge firm-year observations in all three databases between 2003 and 2009. Our sampling period is restricted by limitations of the KLD database because since 2003 KLD has expanded its coverage to include up to 3,000 U.S. companies. After removing firm-year observations with incomplete Compustat, KLD or institutional ownership (13f) data, we obtain an initial sample of 2,620 unique firms with 11,006 firm-year observations covering the period 2003-2009. 12 Summary Statistics Table 1, Panel A reports sample statistics of measures that capture aggressive tax avoidance. Due to data requirements in the estimation procedures, the samples for these measures vary significantly from 6,839 and 6,393 for SHELTERit and DTAXit, respectively, to 4,191 firm-year observations for DD_BTit. 13 Our sample statistics for aggressive tax avoidance measures—SHELTERit, DTAXit, and DD_BTit—are consistent with those in the extant

12

Specifically, the initial sample consists of 11,006 firm-year observations for which data for CSR variables, control variables of our baseline regressions, and at least one tax avoidance variable are available. 13 We follow Desai and Dharmapala (2006) and include only firm-years with positive taxable income when calculating DD_BTit. This procedure and missing values in Compustat variables significantly reduce the sample size of the DD_BT variable. The implied attrition rate is about 62 percent (0.62 = 1 – 4191/11006), consistent with prior studies. Using a sample of S&P 1,500 firms in a 9-year period between 1993 and 2001, Desai and Dharmapala (2006) obtain a final sample of 4,702 firm-year observations for their DD_BT measure, implying an attrition rate of about 65 percent.

15

literature. 14 The mean value of CETRit is 0.2533, and the corresponding sample contains 9,147 firm-year observations. [Insert Table 1 here] Panel B presents descriptive statistics on the two measures of irresponsible CSR activities and control variables for the initial sample of 11,006 firm-year observations. The mean value for the earnings quality variable, ABS_DAit, is 0.145. 15 Other sample statistics are in the range of those reported in earlier studies. 16 In general, the sample statistics suggest that our sample falls between the samples of large firms examined by Desai and Dharmapala (2009) and Chen et al. (2010) versus the Compustat universe examined by Frank et al. (2009). With respect to the measures of irresponsible CSR activities, the mean value of NEG_CSRit is 1.9219, suggesting that on average the sample firm has about 2 irresponsible CSR activities per year. The mean value for HIGH_NEG_CSRit is 0.135, suggesting that only 13.5 percent of firms have four or more irresponsible CSR activities in any given year. [Insert Table 2 here] Table 2 reports mean values of NEG_CSRit and HIGH_NEG_CSRit by years as well as the frequency distribution of irresponsible CSR activities by years and across levels of irresponsible CSR activities (i.e., levels of NEG_CSRit). As evident in Panel A, the proportion of

14

The mean value of SHELTERit is 0.2677. Our sample means for DD_BTit and DTAXit are -0.0001 and 0.0080, respectively. The mean value of our DTAX measure is higher than the corresponding value of -0.05 in Frank et al. (2009). This is reasonable because Frank et al. (2009) uses all firm-years from Compustat while our sample is limited mainly to firms in the Russell index which are likely larger. On the other hand, Chen et al. (2010) report a higher mean value of DD_BTit of 0.015. This difference could be driven by firm size as our sample firms are smaller than those in Chen et al. (2010) which focuses on firms in the S&P 1500 index. 15 This value is comparable to the corresponding mean value of 0.20 in Kim et al. (2012). In untabulated results, we find positive and significant correlations between ABS_DAit and book-tax differences. For example, we find that the correlation between ABS_DAit and DTAXit is 0.0365 and it is significant at the one percent level. These empirical regularities are consistent with Frank et al. (2009). 16 The mean value of IOit is 0.7465, the mean value of CASHit is 0.2218, the average value of ROAit is 0.0496, the mean value of LEVit is 0.2005, the mean value of FIit is 0.0203, the average PPEit is 0.2826, the mean INTANGit is 0.2327, and the mean value of R&D it is 0.0443. Around 48 percent of the observations have positive loss carry forward (NOLit). The mean value of MBit is 3.30.

16

firms with none or one irresponsible CSR activity in a given year ranges from 41.7 percent in 2007 (0.417=(183+483)/1596) to 70 percent in 2003 (0.70=(582+425)/1438). In contrast, we note that relatively few firms have four or more irresponsible CSR activities in a given year, with the proportion of such firms ranging from 7.58 percent in 2003 to 16.87 percent in 2008 and increasing over the years. Moreover, we find that firms with four or more irresponsible CSR activities have an approximately 85 percent probability of maintaining the same level of irresponsible CSR activities in the next three years. 17 The lower mean values for both NEG_CSRit and HIGH_NEG_CSRit in 2003 could reflect the change in KLD coverage in that year. 18 The data presented in Panel B reveal that the most heavily represented industry is Business Services (SIC code 73), followed by Chemicals & Allied Products (SIC code 28), and Electronic & Other Electric Equipment (SIC code 36).

V. EMPIRICAL RESULTS Baseline Regression Results Table 3, Panel A presents the results for the model in equation (1) based on logistic regressions and OLS regressions with clustered standard errors at the firm level (Petersen 2009). We use SHELTERit as the dependent variable for the logistic regressions. The other aggressive tax avoidance measures, DTAXit and DD_BTit, and CETRit are the dependent variables for the

17

To examine the cross-year persistence of excessive irresponsible CSR activities, we compute the percentage of firm-year observations with four or more irresponsible CSR activities that will continue to have the same level of irresponsible CSR activities in all of the next three years. Because our sample ends at 2009, we conduct the analysis based on the observations during 2003-2006. The untabulated results indicate that approximately 85 percent of the firm-year observations with HIGH_NEG_CSRit equal to one during 2003-2006 will continue to have four or more irresponsible CSR activities during the next three years. 18 We control for the effects of these documented time trends in two ways. We include year dummies to control for year fixed effects in the baseline regressions involving firm-year data. In addition, we examine the robustness of the results by excluding the 2003 firm-year observations from the sample; we obtain estimates with the same expected signs and comparable levels of statistical significance if we exclude the 2003 firm-year observations from the sample.

17

OLS regressions. We use two alternative proxies to capture irresponsible CSR activities and we estimate two regression models per dependent variable: NEG_CSRit is the test variable in models 1, 3, 5, and 7 while HIGH_NEG_CSRit is the test variable in the other models. [Insert Table 3 here] The coefficients on both NEG_CSRit and HIGH_NEG_CSRit are significant and positive across all models when sheltering probability and discretionary/permanent book-tax differences are used to capture aggressive tax avoidance. These results show that firms with more irresponsible CSR activities in a given year are more tax aggressive than other firm-years with fewer irresponsible CSR activities in that they have a higher likelihood of engaging in concurrent tax sheltering activities and a higher discretionary/permanent book-tax difference. In particular, the results show that those firms with four or more irresponsible CSR activities in a given year, HIGH_NEG_CSR firms, are more tax aggressive when compared to other firm-years with fewer irresponsible CSR activities. Consistent with Lanis and Richardson (2012), we find that NEG_CSRit and HIGH_NEG_CSRit are significant and negative when CETRit is used to capture consequences of broad tax avoidance practices. The results are also economically significant. Based on parameter estimates in Model 2, HIGH_NEG_CSR firms are 4.25 percent more likely to engage in concurrent sheltering activity when compared to other firms in the sample.19 Given the 26.7 percent sample mean value of SHELTERit, this implies an economically meaningful difference in sheltering probabilities between HIGH_NEG_CSR firms and other firms in our sample. The implied effects on cash effective tax rates are also economically significant. After controlling for other factors that might affect tax avoidance and based on parameter estimates in Model 8 of Panel A, HIGH_NEG_CSR 19

The 4.25 percent increase in sheltering probability is the estimated marginal effect of the HIGH_NEG_CSR variable on tax sheltering probability, which is the expected increase in the sheltering probability as a function of the HIGH_NEG_CSR variable, holding all other variables at the sample mean.

18

firms have cash effective tax rates that are about 3.04 percent lower than other firms, implying a tax saving of around $14 million based on a mean pre-tax income of $463 million in our sample (not tabulated). By way of comparison, Chen et al. (2010) report a difference in cash effective tax rate between family and non-family firms of about 1.20 percent, implying an average tax saving of $6.7 million for family firms in their sample. Using the Heckman Procedure to Correct Potential Self-selection Bias The empirical association between tax avoidance and irresponsible CSR activities could reflect self-selection bias. Irresponsible CSR activities could be influenced by external factors such as political and societal pressure and internal factors such as firm performance, financial constraint, etc. Although the baseline regression model of equation (1) controls for some of these factors, it is possible that there are other underlying factors affecting irresponsible CSR activities that our model does not control for. If these other factors drive irresponsible CSR activities, then this could provide an alternative interpretation of the association between irresponsible CSR activities and tax avoidance. In order to mitigate potential self-selection bias, we reestimate the regressions using the Heckman two-stage procedure. We use HIGH_NEG_CSRit as the dependent variable in the first-stage probit regression. 20 Since the model in equation (1) already includes a range of factors internal to the firm, we use political preference (Rubin 2008) as the primary instrumental variable to capture external/environmental factors that could influence irresponsible CSR activities. Because the Democratic platform is more congruent with the underlying precepts of CSR, company involvement in CSR activities could be related to political preference as captured by election

20

In the first-stage regressions we also control for financial constraints as captured by KZ score (Kaplan and Zingales 1997), industry-adjusted advertisement expenditure and other firm characteristics such as institutional ownership, performance, leverage, size, growth, stock return volatility, and firm age; and we include year and industry dummies.

19

outcomes (Rubin 2008). According to this view, companies located in states with a Democratic majority will likely have fewer irresponsible CSR activities than will companies located in Republican states. However, aggressive tax avoidance is incompatible with either the Democratic or the Republican political platform. As such, we expect little or no significant association between political preference and aggressive tax avoidance. 21 We operationalize political preference as the relative electoral strength of the Democratic/Republican Party as captured by election outcomes in states where a firm’s headquarter is located. We consider all election outcomes for presidential, senatorial, congressional, and gubernatorial races and for members of the state’s legislature. 22 Our sample period covers four two-year election cycles; however, the data for the 2008 election cycle are not yet available. Accordingly, we use the 2002-2003 election data to capture political preference for firm-years in 2003; we use 2004-2005 election data for firm-years in 2004 and 2005; we use 2006-2007 election data for other firm-years in the sample. Given the context of our analysis, a valid instrument should be related to irresponsible CSR activities but uncorrelated with aggressive tax avoidance measures. In untabulated results, we find that the validity of the political preference variable is supported by its being significantly

21

A key Republican principle is to promote economic growth by lowering taxes. Accordingly, it is difficult to exclude political preference from the second-stage regressions when CETRit is the dependent variable and the corresponding results must be viewed with caution. Fortunately, the more aggressive end of tax avoidance practices fits our story well. With these caveats, we report estimates from all models including those with CETRit as the dependent variable. 22 A State is given a maximum of 100 points in each election cycle. The party that won a plurality of the votes in the most recent presidential election gets 20 points, as does the party that won the most recent gubernatorial election. The party that won a Senatorial seat gets 15 points, with a total of 30 points assigned to the two Senatorial elections. A maximum of ten points is assigned to each of the following three federal and state-level elections including congressional races and elections for upper and lower chambers of the State’s legislature. In these cases, the specific points assigned depend on the percentage of seats each party won in the most recent election. Finally, in those States which hold their State elections in odd-numbered years, said elections are combined with Federal elections from the immediately preceding even-numbered year to produce a Party's score. We obtained the relative electoral strength data from the Green Papers using the following hyperlink http://www.thegreenpapers.com/G08/StatewideStrength.phtml. To maximize the variations of the variable we set the Republican party’s scores as negative values in our analysis.

20

correlated with HIGH_NEG_CSRit but uncorrelated with all the tax avoidance measures, including CETRit, if we include them in the second-stage regressions. 23 For brevity, we do not report the first-stage regression results. Table 3, Panel B reports the second-stage regression results where the inverse mills ratio from the first-stage regression, INVERSE_MILLS_RATIOit, is added to the baseline regression model of equation (1) to control for potential self-selection bias. The estimates of INVERSE_MILLS_RATIOit indicate that self-selection bias is a potential issue. Nevertheless, the documented associations between HIGH_NEG_CSRit and aggressive tax avoidance measures remain unchanged—the estimates on HIGH_NEG_CSRit retain the same signs and significance in the second-stage regressions—after correcting for potential self-selection bias.

VI. CORROBORATING EVIDENCE FROM FIN 48 This section uses FIN 48 as a quasi-experiment to provide further evidence on our primary research question. An advantage of this setting is that FIN 48 could change the relative benefits of some tax avoidance strategies (Mills et al. 2010) but it is less likely to have a significant effect on CSR activities. Moreover, there is some evidence that the undertaking of tax positions supported by weaker facts and circumstances, i.e., weak positions, can be considered as a natural outcome of aggressive tax avoidance (Mills et al. 2010; Lisowksy et al. 2012); and the implementation of FIN 48 provides a unique opportunity to estimate the strength of a firm’s tax positions.

23

Firm visibility may represent another institutional factor affecting company involvement in irresponsible CSR activities because it reflects a firm’s exposure to investors, media, and the influence of other social arbiters. Following Garcia-Castro et al. (2010) we operationalize it as a dummy variable indicating whether a firm is listed in the Standard & Poors 500 index or not and use it as a second instrument in the first-stage prediction model. We find that it is also a valid instrument in the sense that it is correlated with HIGH_NEG_CSRit but uncorrelated with any aggressive tax avoidance measure if we include it in the second-stage regressions.

21

Background Information and Conceptual Framework FIN 48 was enacted in June 2006 and became effective for publicly listed companies with fiscal years beginning after December 15, 2006. Among other things, it requires firms to disclose a tabular reconciliation of the beginning and ending tax reserve balances, termed unrecognized tax benefit (UTB), with detailed information on settlements with tax authorities and other activities affecting UTB balances. FIN 48 stipulates a recognition threshold and then a measurement step for evaluating all open tax positions.24 The recognition threshold permits the recognition of a tax position if it has greater than 50 percent likelihood of being sustained under audit. If a tax position fails the test, the firm must establish a tax reserve for the entirety of the benefit. If a tax position passes the test, the company should measure its benefit as the largest portion of the position that is greater than 50 percent likely of being realized upon ultimate settlement. In the latter case, the firm must establish a tax reserve for the disallowed benefit. Tax positions could be strong or weak depending on the underlying tax uncertainty. In the context of FIN 48, the principal tax uncertainty is defined in the recognition threshold as the likelihood of being sustained upon audit. Weak (strong) positions should have a less than (better than) 50 percent chance of being sustained upon audit. Thus, weak positions are less likely to stand up to IRS challenges. According to the Financial Accounting Standards Board (FASB FIN 48 Summary) “it is not controversial to recognize the benefit of a tax position in an enterprise’s financial statements when the degree of confidence is high that the tax position will be sustained upon examination by a taxing authority.” Consistent with the FASB’s interpretation, Mills et al.

24

Firms must base their evaluation of tax positions solely on a tax position’s technical merits reflecting the facts and circumstances underlying the position, and assume that the tax authorities will audit the company’s book with knowledge of all relevant information.

22

(2010) and Lisowksy et al. (2012), we consider the undertaking of weak positions as a natural outcome of aggressive tax avoidance. UTB Level and CSR There is evidence that the UTB level is positively associated with aggressive tax avoidance. 25 Further, using confidential tax shelter data from the Office of Tax Shelter Analysis, Lisowsky et al. (2013) find that the UTB level is positively associated with the likelihood of tax sheltering, a proxy for extremely aggressive tax avoidance, while other tax avoidance measures are not. The culture argument described earlier implies that irresponsible CSR activities and aggressive tax avoidance practices are likely positively related. Accordingly, based on these prior findings concerning UTB level, an implication of the culture hypothesis is that firms with excessive irresponsible CSR activities prior to FIN 48 should have a higher UTB level at the onset of FIN 48 when compared to other firms with fewer irresponsible CSR activities. 26 CSR and Post-FIN 48 Settlements/Changes in the UTB Level A high UTB balance could be a result of undertaking weak positions. Alternatively, firms undertaking a large number of tax positions overall could also have a high UTB balance even if a majority of the underlying positions are strong positions rather than weak positions. In this way, the aggregate UTB balance could be less informative about the nature and character of the firm’s underlying tax positions; that is, the UTB level alone does not necessarily indicate whether the firm has more weak positions. In the ensuing analysis, we use settlements and

25

For instance, Frischmann et al. (2008) find a positive association between FIN 48 tax reserves affecting the effective tax rate and traditional tax avoidance measures. Cazier et al. (2009) find that the level of UTB is driven by determinants of tax avoidance. Rego and Wilson (2012) use the level of FIN 48 reserve as a proxy for aggressive tax avoidance and find that it is positively related to the sensitivity of a manager’s wealth to stock return volatility. 26 This is plausible because our earlier baseline regressions indicate that HIGH_NEG_CSR firms are more tax aggressive, implying that HIGH_NEG_CSR firms could favor weak positions, particularly during the pre-FIN 48 years when transparency of tax reserve disclosures is low (Gleason and Mills 2002) and substantial diversity in measurement exists (FASB FIN 48 summary).

23

changes in the UTB level in the three-year period 2007- 2009 after FIN 48 implementation to examine whether firms have more weak positions at the onset of FIN 48. The analytical model of Mills et al. (2010) predicts that FIN 48 will decrease the expected benefits of weak positions but increase the expected benefits of strong positions, implying that FIN 48 could have a disproportionate effect on firms with more weak positions. Consistent with this prediction, Blouin et al. (2010) find that firms are on average more likely to settle with tax authorities and more likely to reduce tax reserve balance prior to FIN 48, suggesting that negotiating settlements and reducing the overall level of UTB are probable firm responses to FIN 48. Nevertheless, firms are naturally more confident about the technical merits of strong positions and therefore they could be less likely to settle strong positions even after FIN 48. 27 In contrast, there is evidence that FIN 48 has caused firms to become less confident about weak positions. 28 As a consequence, one would expect that if weak positions comprise a larger fraction of the firm’s UTB balance at the onset of FIN 48, the firm will have larger subsequent settlements and also will be more likely to reduce the overall UTB level post-FIN 48.

27

Lisowsky et al. (2012) provides an anecdotal evidence for this conjecture as follows. Prior to 2008, IRS audited Wells Fargo (Wells) and Consolidated Edison (Con Ed) and disallowed the deductions these companies claimed in certain leveraged leases because they were considered abusive leasing transactions. In 2008, the IRS launched a settlement initiative providing the companies an opportunity to settle these disputes out of court; Wells accepted the offer while Con Ed did not. The Court of Federal Claims subsequently ruled in favor of Con Ed and upheld the tax deductions, concluding that the leases demonstrated a valid business purpose. In addition, according to Mills et al. (2010) firms would have reduced incentives to settle strong positions after FIN 48 since the law increases the expected benefits of strong positions. 28 Firms could be less confident about weak positions because there is widespread belief that FIN 48 disclosures will increase the IRS’s ability to screen firms for audit and to perform the audit. For instance, about 89 percent of the 4,000 participants who viewed the KPMG Tax Governance Institute’s webcast “FIN 48: The First Quarter Experience” indicated that they believe it is “highly likely” or “likely” that tax enforcers will increase audits after FIN 48. In particular, because FIN 48 workpapers are considered tax accrual workpapers, Jones (2008) argues that a major concern is that the IRS could enjoy broad authority to demand them, potentially exposing privileged information contained in the workpapers, such as legal theories, views on settlement viability and detailed analyses of individual positions to the IRS. As such, Jones (2008, 799) conjectures that FIN 48 could “compile a ‘roadmap’ to lead the IRS to the positions taxpayers feel least confident about.” Consistent with this view, Blouin et al. (2010) find that firms are more inclined to settle outstanding disputes with the IRS prior to FIN 48 implementation; to the extent that IRS disputes likely represent challenges of weak positions, this result implies that firms are already less confident about weak positions detected and challenged by the IRS before FIN 48 implementation.

24

All else equal, based on these prior studies, we conjecture that if firms with excessive irresponsible CSR activities have more weak positions at the onset of FIN 48, these firms should have larger subsequent settlements post-FIN 48 and should be more likely to reduce the overall UTB level after FIN 48 when compared to other firms with fewer irresponsible CSR activities. In contrast, if firms with excessive irresponsible CSR activities do not have more weak positions at the onset of FIN 48 than other firms, we should find little empirical evidence to support our conjectures. Empirical Design of FIN 48 Analyses We use cross-sectional firm-level data to conduct the FIN 48 analysis. We hand-collect information from proxy statements to complement FIN 48-related data from Compustat database when it is appropriate to do so. 29 Consistent with Lisowsky et al. (2013) and others who use the UTB level as a proxy for aggressive tax avoidance, we use the natural logarithm of the beginning UTB balance (TXTUBBEGIN) for fiscal year 2007, LN(UTB_BEGIN2007), to measure the beginning UTB balance at the onset of FIN 48. Following Blouin et al. (2010), we construct a dummy, DECREASE, to capture the UTB level changes. DECREASE takes the value of one if the ending UTB balance (TXTUBEND) for fiscal year 2009 is less than TXTUBBEGIN for fiscal year 2007, and it equals zero otherwise. To capture the settlements with the tax authorities, we construct the variable, PCT_SETTLE, which is the sum of all settlements with tax authorities (TXTUBSETTLE) over the three-year period 2007-2009 divided by TXTUBBEGIN for fiscal year 2007. We scale the variable by the beginning UTB balance at the onset of FIN 48 to mitigate the issue that firms

29

FIN 48 related variables from Compustat have a significant level of missing data. Many firms with valid FIN 48 data are coded as have missing UTB balances and/or settlement information. Moreover, there are cases where FIN 48 related Compustat data are erroneous. Given these data problems, we hand-collect data from proxy statements to verify and complement the FIN 48 related data from Compustat.

25

with more open tax positions overall naturally have more subsequent settlements. In addition, we use the scaled variable rather than the total settlements because the scaled variable reflects the proportion of uncertain tax positions at the onset of FIN 48 that is eventually settled in the threeyear period after the implementation of FIN 48, and thus it is more likely to capture the proportion of weak positions within the initial UTB level. PCT_SETTLE is undefined for firms with zero beginning UTB balance in 2007. To circumvent this issue, we set PCT_SETTLE to zero for those firms with zero beginning UTB balance in 2007 and zero total settlement during the three-year period of 2007-2009. 30 The FIN 48 regressions are based on the baseline regression model of equation (1) with several modifications as follows. First, we use LN(UTB_BEGIN2007), PCT_SETTLE and DECREASE as the dependent variables. Second, all independent variables, including the test variable, HIGH_NEG_CSRpre, and control variables, are pre-determined vis-à-vis the dependent variables. HIGH_NEG_CSRpre equals one if the firm has four or more irresponsible CSR activities prior to the implementation of FIN 48. Specifically, we use data in fiscal year 2006 to construct all independent variables, including HIGH_NEG_CSRpre. Lastly, we expand the model of equation (1) to include an additional control variable, UTB_BEGIN2007, that isolates the effect of beginning UTB balance at the onset of FIN 48, except when LN(UTB_BEGIN2007) is the dependent variable. FIN 48 Regression Results [Insert Table 4 here]

30

We also use an alternative measure, PCT_SETTLE_ALT, which is similar to PCT_SETTLE in construction except that when firms report no beginning UTB balance in 2007, we use the beginning balance in 2008 as the denominator and when that is zero we use the beginning balance in 2009 to compute PCT_SETTLE_ALT. We obtain estimates with the same expected signs and comparable levels of statistical significance when using PCT_SETTLE_ALT.

26

Table 4 presents the FIN 48 regression results. Sample size varies due to missing information in FIN 48 disclosures; the LN(UTB_BEGIN2007) regression contains 1,137 firms and the PCT_SETTLE and DECRESE regressions contain 1,121 firms. We use OLS and Tobit regression methods for the LN(UTB_BEGIN2007) and PCT_SETTLE regressions, respectively; we run logistic regression when DECREASE is the dependent variable. Across all regression models, the coefficients on HIGH_NEG_CSRpre are positive and significant. All else equal, these results indicate that firms with excessive irresponsible CSR activities prior to FIN 48 (1) have a higher beginning UTB balance at the onset of FIN 48, (2) have higher subsequent settlements in the three-year period after FIN 48, and (3) are more likely to reduce the overall UTB balance in the same period after FIN 48. We also perform the Heckman procedure described in Section V to correct for potential self-selection bias in the FIN 48 regressions. Table 4, Panel B reports the second-stage regression results of that procedure. Again, the INVERSE_MILLS_RATIO variable indicates that selection bias is a potential issue but the estimates on the HIGH_NEG_CSRpre variable remain positive and retain the same significance in the second-stage regressions. These results suggest that the FIN 48 results are also unaffected by the correction for self-selection bias. We interpret these findings as indicating that firms with excessive irresponsible CSR activities prior to FIN 48 have more uncertain tax positions; and, more importantly, these firms’ initial uncertain tax positions are likely weak positions as indicated by their larger post-FIN 48 settlements with tax authorities and their higher likelihood of a net decrease in the overall level of uncertain tax positions after FIN 48. As the undertaking of weak positions resembles aggressive tax avoidance (FASB FIN 48 Summary; Mills et al. 2010; Lisowksy et al. 2012), these results indicate that firms with excessive irresponsible CSR activities prior to FIN 48,

27

HIGH_NEG_CSRpre firms, are more tax aggressive than other firms in that they favor the undertaking of weak positions during the pre-FIN 48 years. Moreover, one could interpret the result on post-FIN 48 settlements as providing circumstantial evidence of a positive association between irresponsible CSR activities and tax noncompliance, further buttressing the culture argument. Blouin et al. (2010) find that firms are more likely to settle outstanding disputes with the IRS prior to FIN 48 implementation, suggesting

that

settlements

with

tax

authorities

are

partially

due

to

material

deficiencies/noncompliance that have already been detected and challenged. In this context, the post-FIN 48 settlements could reflect tax noncompliance arising from prior uncertain tax positions. The finding that HIGH_NEG_CSRpre firms have higher post-FIN 48 settlements could imply that HIGH_NEG_CSRpre firms have weak positions at the onset of FIN 48 that more likely resemble tax noncompliance.

VII. SENSITIVITY TESTS AND ROBUSTNESS CHECKS More Refined Proxies for Irresponsible CSR Activities This section entertains the notion that firms with moderate level of irresponsible CSR activities are more tax aggressive than firms with none or one irresponsible CSR activity. Data in Table 2 reveal that roughly half of the sample firms have either none or one irresponsible CSR activity in any given year. Accordingly, we define a moderate level of irresponsible CSR activities as a scenario in which a firm has two or three irresponsible CSR activities in a given year. We perform the analysis by adding to the baseline regression model of equation (1) a dummy variable, SOME_NEG_CSRit, which equals one if NEG_CSRit equals two or three in a given year and equals zero otherwise. Table 5, Panel A presents the results of this sensitivity

28

analysis. Likewise, for the FIN 48 regressions, we add another dummy variable, SOME_NEG_CSRpre, that parallels SOME_NEG_CSRit in construction but uses pre-FIN 48 data in 2006. Panel B reports the results of that sensitivity analysis. [Insert Table 5 here] In both panels and across all regressions, the estimates on HIGH_NEG_CSRit, SOME_NEG_CSRit, HIGH_NEG_CSRpre and SOME_NEG_CSRpre are generally significant and they remain positive when SHELTERit, DD_BTit, DTAXit, and the three FIN 48 variables are used as the dependent variable; they are negative and significant when we use CETRit as the tax avoidance

measure.

These

results

suggest

that

both

HIGH_NEG_CSR

firms

and

SOME_NEG_CSR firms are more tax aggressive than firms with none or one irresponsible CSR activity. The relative magnitudes of these estimates are informative too. For instance, based on parameter estimates of Model 1 in Panel A, HIGH_NEG_CSR firms are 6.61 percent more likely to engage in concurrent sheltering activity when compared to sample firms with none or one irresponsible

CSR

activity;

the

comparable

increase

in

sheltering

probability

for

SOME_NEG_CSR firms is around 3.13 percent. 31 These findings suggest an economically meaningful difference in sheltering probabilities between SOME_NEG_CSR firms and firms with none or one irresponsible CSR activity. Alternate Control Variables We perform a number of sensitivity analyses using different proxies for earnings management: (1) we substitute absolute values of performance-adjusted accruals, ABS_DA, with raw values of performance-adjusted accruals (Frank et al. 2009); and (2) we add real activities

31

The 6.61 percent and 3.13 percent increases in sheltering probabilities are the estimated marginal effects of the HIGH_NEG_CSR and SOME_NEG_CSR in the SHELTER regression, respectively.

29

manipulation (Roychowdhury 2006) and the incidence of Accounting and Auditing Enforcement Releases (Dechow et al. 1996) to the regression model of equation (1). In addition, we add the proportion of outside directors on the board (Shivdasani 1993), the total number of anti-takeover provisions (Gompers et al. 2003), and the total number of corporate governance mechanisms (Chung and Zhang 2011) to the regressions to examine their robustness with respect to different corporate governance measures. Lastly, because managerial and employee incentives provided by equity-based compensation could affect tax avoidance (Desai and Dharmapala 2006; Rego and Wilson 2012), we create two variables to capture the managerial and employee incentive effects and add them to the regression models. The results, not tabulated, show that all of these alternate control variables do not affect the documented associations between irresponsible CSR activities and aggressive tax avoidance. In particular, the test variables HIGH_NEG_CSRit and HIGH_NEG_CSRpre have the predicted signs and remain significant in all modified baseline and modified FIN 48 regression models. Propensity Score Matching Test Following Li and Prabhala (2007) and Lennox et al. (2012), we use the propensity score matching (PSM) procedure to cross-check the results of the Heckman procedure reported in Table 3. The PSM procedure involves a logistic regression with the same specification as the first-stage regression in the Heckman procedure as described in Section V. Using the predicted propensity score from this logistic regression, we match without replacement a firm-year observation with HIGH_NEG_CSR equals to one, a treatment observation, against another firmyear observation with HIGH_NEG_CSR equals to zero, a control observation. We use the caliper matching method and match within a caliper of three percent, where caliper refers to the difference in the predicted probabilities between the treatment observation and the control

30

observation (Dehejia and Wahba 2002). We are able to match roughly 65 to 74 percent of the treatment observations, depending on which aggressive tax avoidance variable is under investigation. We pool the treatment and matched observations into four respective test samples for each of the four aggressive avoidance measures and perform the regression analyses. The pooled test samples vary from 962 observations with 481 corresponding matched pairs for DD_BTit to 1894 observations with 947 matched pairs for CETRit. Table 6, Panel A presents the results. The estimates on HIGH_NEG_CSRit from the PSM procedure are all significant and retain their respective signs, indicating that our baseline regression findings hold after partially controlling for potential self-selection bias using the PSM procedure. [Insert Table 6 here] Firm-level Regressions We also perform an analysis at the firm level using the baseline regressions in Table 3. This sensitivity analysis is motivated by two factors. First, corporate culture is likely to remain relatively stable over time. Second, there might be potential serial dependence in our data because irresponsible CSR activities and tax avoidance practices could remain fairly stable over time. We use the average of the variables over the seven-year sampling period to calculate firmlevel measures for all the variables in the baseline regression model of equation (1). 32 We then use the firm-level average variables to run the regressions. The firm-level regression results, as presented in Panel B of Table 6, are generally consistent with the baseline regression results. Using Fama-MacBeth Estimation To mitigate statistical concerns arising from serial dependence of regression errors, we estimate the baseline models of equation (1) in Table 3 using the Fama-MacBeth (1973) method. 32

For a dummy variable, such as SHELTERit, we construct the firm-level measure as a dummy variable that equals one if SHELTERit equals one in at least half of the years during 2003-2009; it equals zero otherwise. We use the same procedure to calculate firm-level measure for other dummy variables, including HIGH_NEG_CSRit and NOLit.

31

More specifically, we drop the year dummies from the specification, estimate the revised models by year, and then test the statistical significance of the average coefficients using a t-test. 33 Overall, the Fama-MacBeth regression results, as presented in Panel C of Table 6, are consistent with the baseline regression results. Using Different Sampling Methods Finally, we use three alternate samples to assess the sensitivity of our baseline findings with respect to sampling methods. For brevity, we do not tabulate these results. We first expand the baseline regression sample to include utility and finance firm-years. We also extend the sample backward to cover firm-year observations from 1995 to 2009. Lastly, we purge the impact of the financial crisis during 2008-2009 on our results by using a sample that covers only firm-year observations during the pre-crisis period 2003-2007. Our results remain quantitatively unchanged in all these alternate samples; the estimates on the HIGH_NEG_CSR variable retain the same signs and statistical significance in all empirical models.

VIII. CONCLUSION This paper provides a comprehensive empirical examination of the association between irresponsible CSR activities and aggressive tax avoidance practices. We triangulate our results using different measures of aggressive tax avoidance and find that they are robust. We find that firms with four or more irresponsible CSR activities are more likely to undertake tax sheltering activity and have higher discretionary/permanent book-tax differences. Using FIN 48 as a natural quasi-experiment, we find that firms with excessive irresponsible CSR activities prior to the implementation of FIN 48 have more uncertain tax positions at the onset of FIN 48. More 33

Following Fama and French (2001), we estimate the logit regressions year-by-year and report the average coefficients in Column 1 of Table 6 Panel C. The t-statistics are computed based on time-series standard deviations of the regression coefficients.

32

importantly, these firms’ initial tax positions are likely supported by weaker facts and circumstances as indicated by their higher proportion of settlements of uncertain tax positions with tax authorities over the period 2008 to 2009 immediately after FIN 48 and their higher likelihood of a net decrease in the level of overall uncertain tax positions over the same period. Overall, our results suggest that firms with excessive irresponsible CSR activities are more aggressive in avoiding taxes, lending credence to the idea that CSR could be viewed as a facet of corporate culture that affects corporate tax avoidance. This study’s focus is the association between irresponsible CSR activities and tax avoidance practices. Our findings suggest a general pattern whereby in some firms the corporate culture promotes less responsible CSR activities and more aggressive tax avoidance, while in other firms a different culture promotes more responsible CSR activities and less aggressive tax avoidance. An important research question, one that is beyond the scope of this study, is whether these practices that emerge from a firm’s unique culture are benefitting their shareholders or not. In the end, how they affect shareholders will likely depend on the firm’s history, its competitive environment, the extent to which the policies are influenced by managerial self-dealing, and how the markets and the firm’s stakeholders including shareholders, employees, regulators, customers, and communities respond to the policies. 34 We leave this research question for future studies to explore.

34

It should be noted that a voluminous management literature on CSR has found mixed empirical results concerning the relation between CSR and firm performance (Margolis et al. 2007). Some researchers, such as McWilliams and Siegel (2001), question whether a relationship between CSR and firm performance should exist. Others, such as Margolis and Walsh (2003), argue that such a relation, even if it exists, may be too complex to be found.

33

REFERENCES Blouin, J., C. Gleason, L. Mills, and S. Sikes. 2010. Pre-empting disclosure? Firms’ decisions prior to FIN 48. The Accounting Review 85 (3): 791–816. Cazier, R., S. Rego, X. Tian, and R. Wilson. 2009. Early evidence on the determinants of unrecognized tax benefits. Working paper, University of Iowa. Chatterji, A. K., D. I. Levine, and M. W. Toffel. 2009. How well do social ratings actually measure corporate social responsibility? Journal of Economics and Management Strategy 18 (1): 125–169. Chen, S., X. Chen, Q. Cheng, and T. Shevlin. 2010. Are family firms more tax aggressive than nonfamily firms? Journal of Financial Economics 95 (1): 41–61. Cho, C. H., and D. M. Paton. 2007. The role of environmental disclosures as tools of legitimacy: A research note. Accounting, Organizations and Society 32 (7–8): 639–647. Christensen, J., and R. Murphy. 2004. The social irresponsibility of corporate tax avoidance: Taking CSR to the bottom line. Development 47 (3): 37–44. Chung, K. H., and H. Zhang. 2011. Corporate governance and institutional ownership. Journal of Financial and Quantitative Analysis 46 (1): 247–273. Cronqvist, H., A. Low, and M. Nilsson. 2007. Does corporate culture matter for investment and financial Policies? Working paper, Ohio State University, Nanyang Technological University, and University of Colorado at Boulder. Dechow, P. M., and R. G. Sloan. 1995. Detecting earnings management. The Accounting Review 70 (2):193–225. ———, ———, and A. Sweeney. 1996. Causes and consequences of earnings manipulation: An analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting Research 13 (1): 1–36. Dehejia, R., and S. Wahba. 2002. Propensity score-matching methods for nonexperimental causal studies. Review of Economics and Statistics 84(1): 151–161. Desai, M. A., and D. Dharmapala. 2006. Corporate tax avoidance and high-powered incentives. Journal of Financial Economics 79 (1): 145–179. ———, and ———. 2009. Corporate tax avoidance and firm value. Review of Economics and Statistics 91 (3): 537–546. Dhaliwal, D. S., O. Z, Li, A. Tsang, and Y. G. Yong. 2011. Voluntary nonfinancial disclosure and the cost of equity capital: The initiation of corporate social responsibility reporting. The Accounting Review 86 (1): 59–100. ———, S. Radhakrishnan, A. Tsang, and Y. G. Yong. 2012. Nonfinancial disclosure and analyst forecast accuracy: International evidence on corporate social responsibility disclosure. The Accounting Review 87 (3): 723–759. Dyreng, S. D., M. Hanlon, and E. L. Maydew. 2010. The effects of executives on corporate tax avoidance. The Accounting Review 85 (4): 1163–1189. Fama, E. F., and J. D. MacBeth. 1973. Risk, return, and equilibrium: Empirical tests. Journal of Political Economy 81 (3): 607–636. ———, and K. R. French. 2001. Disappearing dividends: Changing firm characteristics or lower propensity to pay? Journal of Financial Economics 60 (1): 3–43. Financial Accounting Standards Board (FASB). 2006. Accounting for Uncertainty in Income Taxes: An Interpretation of FASB Statement No. 109. FASB Interpretation No. 48. Norwalk, CT: FASB. Financial Times. 2004. Why corporate laggards should not win ethics awards (July 21): 10. Fleischer, V. 2007. Options backdating, tax shelters, and corporate culture. Virginia Tax Review 26 (4): 1031–1034. Frank, M. M., L. J. Lynch, and S. O. Rego. 2009. Tax reporting aggressiveness and its relation to aggressive financial reporting. The Accounting Review 84 (2): 467–496. ———, ———, and ———. 2011. Are aggressive reporting practices associated with other aggressive corporate policies? Working paper, University of Virginia, and Indiana University.

34

Frischmann, P., T. Shevlin, and R. Wilson. 2008. Economic consequences of increasing the conformity in accounting for uncertain tax benefits. Journal of Accounting and Economics 46(2–3): 261–278. Gallemore, J., E. L. Maydew, and J. R. Thornock. 2012. The reputation costs of tax avoidance and the under-sheltering puzzle. Working paper, University of North Carolina, and University of Washington. Garcia-Castro, R., M. Anno, and M. Canela. 2010. Does social performance really lead to financial performance? Accounting for endogeneity. Journal of Business Ethics 92 (1): 107–126. Gleason, C., and L. Mills. 2002. Materiality and contingent tax liability reporting. The Accounting Review 77 (2): 317–342 Godfrey, P. C. 2005. The relationship between corporate philanthropy and shareholder wealth: A risk management perspective. Academy of Management Review 30 (4): 777–798. ———, C. B. Merrill, and J. M. Hansen. 2009. The relationship between corporate social responsibility and shareholder value: An empirical test of the risk management hypothesis. Strategic Management Journal 30 (4): 425–445. Gompers, P., J. Ishii, and A. Metrick. 2003. Corporate governance and equity prices. Quarterly Journal of Economics 118 (1): 107–156. Goss, A., and G. S. Roberts. 2011. The impact of corporate social responsibility on the cost of bank loans. Journal of Banking and Finance 35 (7): 1794–1810. Graham, J. R., and A. Tucker. 2006. Tax shelters and corporate debt policy. Journal of Financial Economics 81 (3): 563–594. ———, M. Hanlon, T. Shevlin, and N. Shroff. 2012. Incentives for tax planning and avoidance: Evidence from the field. Working paper, Duke University, Massachusetts Institute of Technology, and University of California, Irvine. Hanlon, M., and J. Slemrod. 2009. What does tax aggressiveness signal? Evidence from stock price reactions to news about tax shelter involvement. Journal of Public Economics 93 (1–2): 126–141. ———, and S. Heitzman. 2010. A review of tax research. Journal of Accounting and Economics 50 (2–3): 127–178. Hermalin, B. E. 2001. Economics and corporate culture. In The International Handbook of Organizational Culture and Climate, edited by C. L. Cooper, S. Cartwright, and P. Christopher Earley. Chichester, UK.: John Wiley & Sons. Jones, A. W. 2008. FASB-The IRS's new best friend: How FIN 48 affects the taxpayer-IRS relationship and potential taxpayer challenges. Gerogia State University Law Review 25 (3): 767–800. Kaplan, S. N., and L. Zingales. 1997. Do investment-cash flow sensitivities provide useful measures of financing constraints. Quarterly Journal of Economics 112 (1): 169–215. Kim, J.-B., Y. Li, and L. Zhang. 2011. Corporate tax avoidance and stock price crash risk: Firm-level analysis. Journal of Financial Economics 100 (3): 639–662. Kim, Y., M. S. Park, and B. Wier. 2012. Is earnings quality associated with corporate social responsibility? The Accounting Review 87 (3): 761–796. KLD Research and Analytics, Inc. (KLD). 2008. Getting Started with KLD Stats and Ratings Definitions. Boston, MA: KLD Research & Analytics, Inc. Kreps, D. M. 1990. Corporate culture and economic theory. In Perspectives on Positive Political Economy, edited by J. E. Alt and K. A. Shepsle. Cambridge, U.K.: Cambridge University Press. Lanis, R., and G. Richardson. 2012. Corporate social responsibility and tax aggressiveness: An empirical analysis. Journal of Accounting and Public Policy 31 (1): 86–108. Lennox, C., J. Francis, and Z. Wang. 2012. Selection models in accounting research. The Accounting Review 87 (2): 589–616. Li, K., and N. Prabhala. 2007. Self-selection models in corporate finance. In Handbook of Corporate Finance: Empirical Corporate Finance, edited by B. E. Eckso. Amsterdam, The Netherlands: NorthHolland. Lisowsky, P. 2010. Seaking shelter: Empirically modeling tax shelters using financial statement information. The Accounting Review 85 (5): 1693–1720.

35

———, L. A. Robinson, and A. P. Schmidt. 2013. Do publicly disclosed tax reserves tell us about privately disclosed tax shelter activity? Journal of Accounting Research 51 (3): 583–629. Manzon, G. B., and G. A. Plesko. 2002. The relation between financial and tax reporting measures of income. Tax Law Review 55 (2): 175–214. Margolis, J. D., and Walsh, J. P. 2003. Misery loves companies: Rethinking social initiatives by business. Administrative Science Quarterly 48 (2): 268–305. ———, H. A. Elfenbein, and J. P. Walsh. 2007. Does it pay to be good? A meta-analysis and redirection of research on the relationship between corporate social and financial performance.Working paper, Harvard Business School, University of California, and University of Michigan. Mattingly, J. E., and S. L. Berman. 2006. Measurement of corporate social action: discovering taxonomy in the Kinder Lydenburg Domini ratings data. Business and Society 45 (1): 20–46. McWilliams, A. and Siegel, D. 2001. Corporate social responsibility: A theory of the firm perspective. Academy of Management Review 26 (1): 117–127. Mills, L. 1998. Book-tax differences and Internal Revenue Service adjustments. Journal of Accounting Research 36(2): 343–356. ———, L. Robinson, and R. Sansing. 2010. FIN 48 and tax compliance. The Accounting Review 85 (5): 1721–1743. Minor, D. B., and J. Morgan. 2011. CSR as reputation insurance: Primum non nocere. California Management Review 53 (3): 40–59. Moser, D. V., and P. R. Martin. 2012. A broader perspective on corporate social responsibility research in accounting. The Accounting Review 87(3): 797–806. Neu, D., H. Warsame, and K. Pedwell. 1998. Managing public impressions: Environmental disclosures in annual reports. Accounting, Organizations and Society 23 (3): 265–282. Petersen, M. A. 2009. Estimating standard errors in finance panel data sets: Comparing approaches. Review of Financial Studies 22 (1): 435–480. Rego, S. O., and R. Wilson. 2012. Executive compensation, tax reporting aggressiveness, and future firm performance. Journal of Accounting Research 50 (3): 775–810. Roychowdhury, S. 2006. Earnings management through real activities manipulation. Journal of Accounting and Economics 42 (3): 335–370. Rubin, A. 2008. Political views and corporate decision making: The case of corporate social responsibility. Financial Review 43 (3): 337–360. Shivdasani, A. 1993. Board composition, ownership structure, and hostile takeovers. Journal of Accounting and Economics 16 (1–3): 167–198. Sikka, P. 2010. Smoke and mirrors: Corporate social responsibility and tax avoidance. Accounting Forum 34: 153–168. Van den Steen, E. 2005. Organizational beliefs and managerial vision. Journal of Law, Economics, and Organization 21 (1): 256–283. ———. 2010. On the origin of shared beliefs (and corporate culture). RAND Journal of Economics 41 (4): 617–648. Waddock, S. 2003. Myths and realities of social investing. Organization and Environment 16 (3): 369– 380. Weisbach, D. A. 2002. An economic analysis of anti-tax-avoidance doctrines. American Law and Economics Review 4 (1): 88–115. Wilson, R. 2009. An examination of corporate tax shelter participants. The Accounting Review 84 (3): 969–999.

36

APPENDIX A Variable Definitions Variable

Definition

Measures of Aggressive Tax Avoidance (Dependent Variables for Baseline Regressions): SHELTERit

We use Wilson’s (2009) sheltering probability to capture incidence of the most aggressive avoidance practices as follows. We use the regression model reported in Table 5 Column 3 in Wilson (2009). The sheltering probability equation is SHELTER_PROBit=-4.86 + 5.20 × BTDit + 4.08 × DAit - 0.41 × LEVit + 0.76 × ATit + 3.51 × ROAit + 1.72 × FOREIGN INCOMEit + 2.43× R&Dit; where SHELTER_PROBit is the sheltering probability for firm i in year t, BTDit is a booktax difference measure as defined by Kim et al. (2011), DAit is discretionary accruals from the performance-adjusted modified cross-sectional Jones Model, LEVit is firm leverage, ATit is the log of total assets for firm i in year t, ROAit is return on assets, FOREIGN INCOMEit is a dummy variable set equal to one for firm-years that report foreign income and zero otherwise, and R&Dit is research and development expense ratio. Following Kim et al. (2011), we define BTD as book income less taxable income scaled by lagged assets (AT). Book income is pretax income (PI) in year t. Taxable income is calculated by summing current federal tax expense (TXFED) and current foreign tax expense (TXFO) and dividing by the statutory tax rate and then subtracting the change in NOL carryforwards (TLCF) in year t. If current federal tax expense is missing, total current tax expense is calculated by subtracting deferred taxes (TXDI), state income taxes (TXS) and other income taxes (TXO) from total income taxes (TXT) in year t. Following Rego and Wilson (2012), we rank SHELTER_PROBit each year and create dummy variable to capture those firms that have high sheltering probability. SHELTERit is an indicator variable that equals to one if the firm’s estimated sheltering probability is in the top quartile in that year, and zero otherwise.

DTAXit

Frank et al. (2009) discretionary permanent book-tax difference for firm i in year t. DTAXit is the εit from the following regression estimated by 2-digit SIC code and fiscal year: PERMDIFFit =β0 +β1 INTANGit + β2 UNCONit+ β3 MIit+ β4 CSTEit + β5 ∆NOLit+ β6 LAGPERMit + εit; Where: PERMDIFFit = BIit – [(CFTEit+ CFORit) / STRit] – (DTEit / STRit); BIit= pre-tax book income (PI) for firm i in year t; CFTEit= current federal tax expense (TXFED) for firm i in year t; CFORit = current foreign tax expense (TXFO) for firm i in year t; DTEit= deferred tax expense (TXDI) for firm i in year t; STRit = statutory tax rate in year t; INTANGit = goodwill and other intangibles (INTAN) for firm i in year t; UNCONit = income (loss) reported under the equity method (ESUB) for firm i in year t; MIit = income (loss) attributable to minority interest (MII) for firm i in year t; CSTEit = current state income tax expense (TXS) for firm i in year t; ∆NOLit = change in net operating loss carryforwards (TLCF) for firm i in year t; and LAGPERMit = one-year lagged PERMDIFF for firm i in year t. We follow the method in Frank et al. (2009) to handle the missing value problems in estimating DTAXit. If minority interest (MII), current foreign tax expense (TXFO), income from unconsolidated entities (ESUB), or current state tax expense (TXS) is missing on Compustat, then we set MI, CFOR, UNCON, or CSTE, respectively, to zero. If current federal tax expense (TXFED) is missing on Compustat, then we set the value of CFTE to: total tax expense (TXT) less current foreign tax expense (TXFO) less current state tax expense (TXS) less deferred tax expense (TXDI). If information for goodwill and other

37

intangibles (INTANG) is missing on Compustat, then we set the value for INTANG to 0. If INTANG = “C”, then we set the value of INTANG to that for goodwill (GDWL). DD_BTit

Desai and Dharmapala (2006) discretionary book-tax difference (DD_BT) for firm i, year t. DD_BT is equal to µi + εit, from the following firm fixed-effect regression: BTit = β1TAit + µi + εit, where BTit is the Manzon-Plesko (2002) book-tax difference measure (described below); TAit is Dechow et al. (1995) total accruals measure for firm i in year t, scaled by the lagged value of assets; µi is the average value of the residual for firm i over the sample period; and εit is the deviation of the residual in year t from firm i's average residual. BT is defined as (US domestic financial income – US domestic taxable income – Income taxes (State) – Income taxes (Other) – Equity in Earnings)/lagged assets = (PIDOM – TXFED/Statutory tax rate – TXS – TXO – ESUB)/ATt-1. Firms with zero or negative taxable income are assumed to have attenuated incentives, at the margin, to engage in tax sheltering activity. We follow prior literature, e.g., Desai and Dharmapala (2006), and include only firm-years with positive TXFED.

Measure of Cash Effective Tax Rate (Dependent Variable for Baseline Regressions): CETRit

Cash effective tax rate (CETR) for firm i in year t. CETR is defined as cash tax paid (TXPD) divided by pre-tax book income (PI) before special items (SPI). CETRit is set as missing when the denominator is zero or negative. We truncate CETRit to the range [0, 1].

38

CSR Measures (Test Variables for Baseline Regressions): NEG_CSRit

Sum of all observed engagement in CSR activities that negatively affect the firm’s stakeholders including shareholders, employees, customers, government, suppliers, etc. for firm i in year t reported by KLD.

HIGH_NEG_CSRit

Equals to one if NEG_CSRit ≥ 4 for firm i in year t, zero otherwise.

Control Variables: POS_CSRit

ABS_DAit

Sum of all observed engagement in CSR activities that positively affect the firm’s stakeholders including shareholders, employees, customers, government, suppliers, etc. for firm i in year t reported by KLD. Absolute value of discretionary accruals for firm i, year t, where discretionary accruals are computed using the modified Jones model including lagged ROA as an additional regressor.

IOit

Institutional ownership for firm i, year t, defined as the fraction of a firm's outstanding shares owned by institutional investors.

CASHit

Cash holding for firm i, year t, defined as cash and marketable securities (CHE) divided by lagged assets (AT).

ROAit

Return on assets for firm i, year t, measured as operating income (PI – XI) scaled by lagged assets (AT).

LEVit

Leverage for firm i, year t, measured as long-term debt (DLTT) scaled by lagged assets (AT).

NOLit

A dummy variable coded as one if loss carry forward (TLCF) for firm i is positive as of the beginning of the year t.

∆NOLit

Change in loss carry forward (TLCF) for firm i, year t, scaled by lagged assets (AT).

FIit

Foreign income (PIFO) for firm i, year t, scaled by lagged assets (AT). Missing values in PIFO are set to zero.

PPEit

Property, plant, and equipment (PPENT) for firm i, year t, scaled by lagged assets (AT).

INTANGit

Intangible assets (INTAN) for firm i, year t, scaled by lagged assets (AT).

EQINCit

Equity income in earnings (ESUB) for firm i, year t, scaled by lagged assets (AT).

R&Dit

Research and development expense ratio for firm i, year t, measured as research and development expense (XRD) scaled by lagged assets (AT). Missing values in XRD are set to zero.

EMPit

The natural logarithm of the number of employees (EMP) for firm i, year t.

∆SALEit

Changes in sales (SALE) scaled by lagged sales for firm i, year t.

SIZEit-1

Natural logarithm of the market value of equity (PRCC_F×CSHO) for firm i at the beginning of year t.

MBit-1

Market-to-book ratio for firm i, at the beginning of year t, measured as market value of equity (PRCC_F × CSHO), scaled by book value of equity (CEQ).

39

APPENDIX B CSR Activities in KLD Data KLD is an independent rating agency that specializes in assessment of corporate social actions. Mattingly and Berman (2006) find that KLD data have construct validity. Because of KLD’s independence and the comprehensiveness of its information sources, Waddock (2003, 369), a prominent scholar in CSR research, states that KLD data are “the de facto research standard at the moment” for measuring company CSR engagement in scholarly research. KLD began providing social ratings in 1991 covering roughly 650 companies in the S&P500 and Domini 400 Social indices. In 2003, KLD expanded coverage to include the largest 3,000 U.S. companies by market capitalization. A calendar-year schedule is employed in the data collection process. KLD staff members apply the same set of criteria consistently across all selected companies with a wide range of data gathered from internal sources, e.g., company disclosures, regulatory filings and interviews with company executives, and sources external to the firm, e.g., the media, KLD’s research partners around the world and information from government and nongovernment organizations. The table below lists all 75 social ratings according to KLD since 2003. Changes and new items added after 2003 are reported in parentheses. KLD Research & Analytics, Inc. (2008) provides detailed information and the definitions of the following social ratings.

Responsible CSR Activities (positive social ratings)

Irresponsible CSR Activities (negative social ratings)

Corporate Governance

Limit Compensation Strength in Ownership Transparency/Communications (moved 2005) Political Accountability Strength (added 2005) Public Policy Strength (added 2007) Other Strength

High Compensation Concern in Ownership Accounting Controversies (added 2005) Transparency Concern (added 2005) Political Accountability Concern (added 2005) Public Policy Concern (added 2007) Other Concern

Employee Relations

Strength in Union Relations Cash Profit Sharing Involvement Strong Retirement Benefits Health and Safety Strength Other Strength

Concern in Union Relations Safety Controversies Workforce Reductions Pension/Benefits Other Concern

Environment

Beneficial products and Services Pollution Prevention Recycling Clean Energy Property, Plan, and Equipment Management System Strength (added 2006) Other Strength

Hazardous Waste Regulatory Problems Ozone Depleting Chemicals Substantial Emissions Agricultural Chemicals Climate Change Other Concern

Community

Generous Giving Innovative Giving Support for Housing Support for Education Non-U.S. Charitable Giving Volunteer Programs (added 2005) Other Strength

Investment Controversies Negative Economic Impact Tax Disputes (moved 2005) Other Concern

Diversity

CEO Promotion Board of Directors Work/Life Benefits

Controversies Non-Representation Other Concern

Category

40

Women/Minority Contracting Employment of the Disabled Gay and Lesbian Policies Other Strength Human Rights

Indigenous People Relations Labor Rights Other Strength

Burma International Labor Indigenous People Relations Other Concern

Product Quality and Safety

Quality R&D/Innovation Benefit to Economically Disadvantaged Other Strength

Product Safety Marketing/Contracting Controversies Antitrust Other Concern

41

TABLE 1 Descriptive Statistics Panel A presents descriptive statistics on the aggressive tax avoidance measures and cash effective tax rate for the observations used in the baseline regression analysis. The initial sample consists of 11,006 firm-year observations for which data for CSR variables, control variables of our baseline regressions, and at least one tax avoidance variable are available. N stands for number of observations. Appendix A provides detailed information on all the variables.

Variable

N

Mean

Std. Dev

P25

Median

P75

Panel A. Measures of Tax Avoidance SHELTERit

6,839

0.2677

0.4428

0.0000

0.0000

1.0000

DTAXit

6,393

0.0080

0.1163

-0.0326

0.0000

0.0418

DD_BTit

4,191

-0.0001

0.0536

-0.0183

0.0034

0.0248

CETRit

9,147

0.2533

0.1918

0.1185

0.2400

0.3382

Panel B. CSR Measures and Control Variables NEG_CSRit

11,006

1.9219

1.8530

1.0000

2.0000

2.0000

HIGH_NEG_CSRit

11,006

0.1350

0.3418

0.0000

0.0000

0.0000

POS_CSRit

11,006

1.3543

2.1276

0.0000

1.0000

2.0000

ABS_DAit

11,006

0.1451

0.2217

0.0309

0.0708

0.1740

IOit

11,006

0.7465

0.2129

0.6291

0.7908

0.9143

CASHit

11,006

0.2218

0.2882

0.0413

0.1264

0.3066

ROAit

11,006

0.0496

0.1595

0.0200

0.0607

0.1069

LEVit

11,006

0.2005

0.2523

0.0015

0.1473

0.2973

NOLit

11,006

0.4839

0.4998

0.0000

0.0000

1.0000

∆NOLit

11,006

0.0221

0.3401

0.0000

0.0000

0.0013

FIit

11,006

0.0203

0.0508

0.0000

0.0000

0.0302

PPEit

11,006

0.2826

0.2825

0.0929

0.1933

0.3758

INTANGit

11,006

0.2327

0.2825

0.0333

0.1536

0.3456

EQINCit

11,006

0.0008

0.0039

0.0000

0.0000

0.0000

R&Dit

11,006

0.0443

0.0912

0.0000

0.0049

0.0549

EMPit

11,006

1.7595

1.2520

0.7419

1.5333

2.5257

∆SALEit

11,006

0.1368

0.3217

0.0038

0.0917

0.2031

SIZEit-1

11,006

7.0046

1.5107

5.9619

6.8297

7.9063

MBit-1

11,006

3.3020

5.7165

1.5947

2.4093

3.8543

42

TABLE 2 Irresponsible CSR Activities by Year and by Industry Panel A presents the mean values of NEG_CSRit and HIGH_NEG_CSRit by years and the frequency distribution of sample firms by years and across different levels of NEG_CSR. Panel B presents the number of observations and the mean values of CSR measures by selected two-digit SIC industries with more than 100 firm-year observations during our sample period from 2003 to 2009. The sample consists of 11,006 firm-year observations for which data for CSR variables, control variables of our baseline regressions, and at least one tax avoidance variable are available. N stands for number of observations.

Panel A. Mean Values and Frequency Distribution Year 2003 2004 2005 2006 2007 2008 2009

N 1,438 1,508 1,581 1,581 1,596 1,660 1,642

NEG_CSRit Mean 1.2364 1.7115 1.8254 2.0936 2.1924 2.1759 2.1236

0 582 330 274 180 183 204 241

Level of NEG_CSRit 1 2 3 425 235 87 464 392 168 536 424 162 490 480 207 483 453 213 508 463 205 485 456 190

≥4 109 154 185 224 264 280 270

HIGH_NEG_CSRit Mean 0.0758 0.1021 0.1170 0.1416 0.1654 0.1687 0.1644

Panel B. Mean Values of Irresponsible CSR Activities by Selected 2-digit SIC Industries 2-digit SIC 13 20 23 26 27 28 30 33 34 35 36 37 38 39 42 48 50 51 53 55 56 58 59 73 79 80 87

Industry

Oil & Gas Food, Beverage Apparel & Other Textile Products Paper & Allied Products Printing & Publishing Chemicals & Allied Products Rubber Primary Metal Industries Fabricated Metal Products Industrial Machinery & Computer Equipment Electronic & Other Electric Equipment Transportation Equipment Instruments & Related Products Miscellaneous Manufacturing Trucking and Warehousing Communication Wholesale: Durable Goods Wholesale: Non-durable Goods General Merchandise Store Auto Dealers, Gas Stations Apparel & Accessory Stores Eating & Drinking Miscellaneous Retail Business Services Amusement & Recreation Services Health Services Engineering & Management Services Other

43

N 467 322 142 160 190 1,067 119 178 185 785 1,023 287 849 103 118 205 289 143 106 108 237 197 272 1,397 127 265 262 1,403

NEG_CSRit 2.0942 2.6211 2.2253 2.5187 1.5000 2.3861 1.7142 2.4775 2.2378 1.9108 1.6080 2.7073 1.389 1.6601 1.5677 2.0195 1.2387 2.0909 4.0943 2.5555 1.6329 2.3147 1.6323 1.3908 1.8346 1.6943 1.3740 2.1799

HIGH_NEG_CSRit 0.1306 0.2484 0.1690 0.2625 0.1158 0.2202 0.1261 0.2528 0.1946 0.1287 0.0694 0.2230 0.0459 0.1262 0.0847 0.1463 0.0242 0.2028 0.4717 0.2685 0.0717 0.1827 0.0809 0.0451 0.0866 0.1132 0.0649 0.2046

TABLE 3 CSR and Aggressive Tax Avoidance: Evidence from Firm-Year Level Regressions Our sample consists of 11,006 firm-year observations covering the period 2003-2009 for which data for CSR variables, control variables of our regressions, and at least one tax avoidance variable are available. The number of observations used in each regression varies due to additional data requirements for estimating the sheltering probability, book-tax differences and the tax rates. The regression models are: AGGRESSIVEit = β0 + β1 NEG_CSRit/HIGH_ NEG_CSRit + β2 POS_CSRit + β3 ABS_DAit + β4 IOit + β5 CASHit + β6 ROAit + β7 LEVit + β8 NOLit + β9 ∆NOLit+ β10 FIit + β11 PPEit + β12 INTANGit + β13 EQINCit + β14 R&Dit + β15 EMPit+ β16 ∆SALEit + β17 SIZEit-1 + β18MBit-1 + β19 Lag(Dependent Variable) + Year Dummies + Industry Dummies + εit; where AGGRESSIVEit are the several measures of aggressive tax avoidance and NEG_CSRit/HIGH_ NEG_CSRit are the two measures of irresponsible CSR activities. Appendix A provides detailed information on all variables. Year and industry dummies are included in each specification. Panel A presents the baseline OLS regression results. Panel B presents the results from the Heckman two-stage procedure. In Panel B, INVERSE_MILLS_RATIOit is the inverse Mills ratio from a first-stage probit regression, where the dependent variable is HIGH_NEG_CSRit and the independent variables are political preference, firm visibility, industry-adjusted advertisement expenditure, KZ score (financial constraint measure), institutional ownership (IOit), ROAit, LEVit, SIZEit−1, MBit−1, stock return volatility, firm age, year and industry dummies. The t-statistics and z-statistics, reported in parentheses, are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

Panel A. Baseline Regression Results Dependent Variable SHELTERit Logit Model 1

Model 2

Irresponsible CSR activities NEG_CSRit 0.1522***

Control variables POS_CSRit ABS_DAit IOit CASHit

Model 3

0.4018** (2.48) 0.0185 (0.53) 0.3298 (1.54) 0.3688 (1.36) 0.1474 (0.60)

DD_BTit OLS

Model 4

0.0041*** (4.28)

(4.09)

HIGH_NEG_CSRit

DTAXit OLS

0.0156 (0.45) 0.3480 (1.63) 0.3880 (1.43) 0.1617 (0.66)

Model 5

Model 6

0.0022*** (5.35) 0.0143*** (3.37)

-0.0008 (-1.09) 0.0058 (0.65) -0.0086 (-1.06) 0.0083 (0.80)

-0.0007 (-0.87) 0.0062 (0.70) -0.0081 (-0.99) 0.0087 (0.83)

44

CETRit OLS Model 7 -0.0061*** (-4.83)

0.0081*** (3.92) 0.0000 (0.07) -0.0120*** (-3.32) 0.0064 (1.56) -0.0266*** (-4.50)

Model 8

0.0002 (0.50) -0.0119*** (-3.32) 0.0064 (1.55) -0.0264*** (-4.42)

-0.0304*** (-5.28) -0.0002 (-0.24) 0.0132 (1.33) -0.0488*** (-4.65) -0.0185 (-1.47)

-0.0004 (-0.36) 0.0129 (1.31) -0.0492*** (-4.69) -0.0160 (-1.27)

ROAit LEVit NOLit ∆NOLit FIit PPEit INTANGit EQINCit R&Dit EMPit ∆SALEit SIZEit-1 MBit-1 Lag(Dependent Variable) Intercept

(Pseudo) R-Square Number of Observations

8.2913*** (15.04) -1.3249*** (-4.81) 1.0314*** (7.74) 5.6978*** (15.50) 7.2540*** (6.72) 0.4761 (1.38) 1.2774*** (5.98) 12.8920 (1.04) 3.0773*** (3.98) 0.5764*** (7.26) 0.2001 (1.25) 0.9544*** (14.08) -0.0304*** (-3.33) 2.4838*** (22.82) -11.9217*** (-17.23)

8.2148*** (15.00) -1.3867*** (-5.01) 1.0572*** (7.95) 5.6400*** (15.47) 7.1209*** (6.62) 0.5236 (1.52) 1.3033*** (6.09) 12.8859 (1.04) 3.0419*** (3.98) 0.6129*** (7.79) 0.1920 (1.20) 0.9679*** (14.30) -0.0311*** (-3.42) 2.4879*** (22.88) -11.7663*** (-17.16)

0.1730*** (5.74) 0.0303* (1.88) 0.0156*** (4.44) 0.0534** (2.54) 0.1363** (2.46) -0.0179 (-1.29) -0.0041 (-0.36) -0.9120** (-1.96) 0.1459*** (3.30) 0.0027 (1.05) 0.0199*** (3.04) -0.0081*** (-4.14) -0.0001 (-0.30) -0.0947*** (-4.69) 0.0185 (1.26)

0.1724*** (5.69) 0.0300* (1.85) 0.0163*** (4.66) 0.0533** (2.53) 0.1359** (2.45) -0.0183 (-1.32) -0.0044 (-0.39) -0.8391** (-1.82) 0.1461*** (3.30) 0.0035 (1.39) 0.0197*** (3.01) -0.0076*** (-3.92) -0.0001 (-0.40) -0.0937*** (-4.63) 0.0222 (1.53)

0.4261*** (22.70) 0.0084* (1.80) 0.0035** (2.43) -0.0261 (-1.42) -0.3080*** (-13.93) 0.0132** (2.20) -0.0001 (-0.02) -1.4068*** (-4.80) -0.0345* (-1.65) 0.0033** (2.27) -0.0035 (-0.76) -0.0055*** (-5.30) -0.0003* (-1.91) 0.1723*** (8.06) 0.0093 (1.14)

0.4245*** (22.49) 0.0083* (1.75) 0.0036** (2.44) -0.0263 (-1.45) -0.3084*** (-13.86) 0.0130*** (2.14) -0.0001 (-0.04) -0.9621*** (-4.38) -0.0332 (-1.57) 0.0037** (2.48) -0.0034 (-0.74) -0.0053*** (-5.07) -0.0003* (-1.89) 0.1720*** (8.03) 0.0131* (1.70)

-0.1152*** (-3.25) -0.0255** (-2.43) -0.0170*** (-4.14) 0.0509** (2.22) -0.1307** (-2.36) -0.0222 (-1.59) 0.0055 (0.57) -0.6262 (-1.23) -0.1515*** (-3.01) -0.0023 (-0.70) -0.0628*** (-5.17) -0.0011 (-0.42) 0.0004 (1.16) 0.3319*** (20.08) 0.2339*** (8.63)

-0.1166*** (-3.28) -0.0263** (-2.50) -0.0173*** (-4.24) 0.0491** (2.15) -0.1300** (-2.34) -0.0221 (-1.57) 0.0041 (0.43) -0.4994 (-0.98) -0.1441*** (-2.86) 0.0009 (0.26) -0.0639*** (-5.24) 0.0008 (0.28) 0.0004 (1.08) 0.3303*** (19.93) 0.2227*** (8.25)

0.6340 6,839

0.6323 6,839

0.1210 6,393

0.1197 6,393

0.4689 4,191

0.4670 4,191

0.1983 9,147

0.1983 9,147

Panel B: Results from the Heckman Two-Stage Procedure Dependent Variable

SHELTERit

45

DTAXit

DD_BTit

CETRit

HIGH_NEG_CSRit INVERSE_MILLS_RATIOit

(Pseudo) R-Square Number of Observations

Logit

OLS

OLS

OLS

0.3425** (2.03) -0.5413** (-2.23)

0.0121*** (2.82) -0.0031 (-0.38)

0.0086*** (4.09) -0.0033 (-0.97)

-0.0323***

0.6372 6,447

0.1214 6,050

0.4639 4,014

0.2025 8,677

46

(-5.52) 0.0225*** (2.61)

TABLE 4 CSR and Uncertain Tax Positions: Evidence from FIN 48 This table presents the cross-sectional regression results of the following regression models: LN(UTB_BEGIN2007), PCT_SETTLE or DECREASE = β0 + β1 HIGH_NEG_CSRpre + β2 POS_CSR + β3 ABS_DA + β4 IO + β5 CASH + β6 ROA + β7 LEV + β8 NOL + β9 ∆NOL + β10 FI + β11 PPE + β12 INTANG + β13 EQINC + β14 R&D + β15 EMP + β16 ∆SALE + β17SIZE + β18MB + β19 UTB_BEGIN2007 + Industry Dummies + ε; where LN(UTB_BEGIN2007) is the natural logarithm of the beginning UTB balance at the onset of FIN 48 in year 2007. PCT_SETTLE is the sum of all settlements with tax authorities over the period 2007-2009 divided by beginning UTB balance for year 2007. DECREASE is a dummy variable that equals one if ending UTB balance in 2009 minus beginning UTB balance in 2007 is less than zero and equals zero otherwise. UTB_BEGIN2007 is the beginning UTB balance at the onset of FIN 48 in year 2007. HIGH_NEG_CSRpre is a dummy variable that equals one if the firm has four or more irresponsible CSR activities in year 2006. All other independent variables are as defined in Appendix A and they are calculated using data in year 2006. Industry dummies are included in each specification. Panel A presents the OLS regression results. Panel B presents the results from the Heckman two-stage procedure. In Panel B, the INVERSE_MILLS_RATIOpre is the inverse Mills ratio from a first-stage probit regression, where the dependent variable is HIGH_NEG_CSRpre and the independent variables are political preference, firm visibility, industry-adjusted advertisement expenditure, KZ score (financial constraint measure), institutional ownership (IO), ROA, LEV, SIZE, MB, stock return volatility, firm age, year and industry dummies. The t-statistics/z-statistics based on heteroskedasticity-robust standard errors are reported in parenthesis. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

Panel A. FIN 48 Results Dependent Variable

CSR measures HIGH_NEG_CSRpre Control Variables POS_CSRit ABS_DA IO CASH ROA LEV NOL ∆NOL FI PPE

LN(UTB_BEGIN2007) OLS

PCT_SETTLE Tobit

DECREASE Logit

0.3945*** (3.26)

0.1807*** (3.24)

0.6174*** (2.72)

0.0081 (0.40) -0.5559** (-2.38) 0.5750*** (3.20) -0.1464 (-1.23) -0.45253909 (-1.42) 0.1633 (1.10) 0.0242 (0.34) -0.2330 (-0.83) 0.7509 (1.06) -0.2494

-0.0067 (-0.69) -0.2284* (-1.88) 0.0222 (0.25) -0.1702** (-2.21) 0.2331 (1.31) 0.1141 (1.55) -0.0677** (-1.95) -0.2285 (-1.23) 0.0470 (0.13) 0.0986

0.0002 (0.87) -1.2086** (-2.20) 0.0456 (0.13) -0.4375 (-1.40) 1.5779** (2.19) 0.0456 (0.16) -0.1153 (-0.83) 0.2824 (0.43) -0.4980 (-0.36) 0.6633**

47

(1.14) 0.0879 (1.31) -3.5891 (-1.30) 0.3323 (1.16) 0.0891*** (3.21) -0.2289*** (-3.13) 0.0439* (1.93) -0.0004 (-0.37) -0.0002***

(2.00) -0.0393 (-0.14) 11.1050 (0.97) -1.2979 (-1.00) 0.1573 (1.43) -1.3283*** (-3.92) -0.1727* (-1.91) -0.0032 (-0.70) 0.0002

-3.5067*** (-9.51)

(-2.99) -0.2115 (-1.16)

(0.87) 0.6453 (0.98)

0.5935 1,137

0.147 1,121

0.0922 1,121

PCT_SETTLE Tobit

DECREASE Logit

INVERSE_MILLS_RATIOpre

(2.95) -0.1139 (-0.90)

0.1924*** (3.42) 0.1052* (1.69)

0.5937** (2.56) 0.0022 (0.10)

(Pseudo) R-Square Number of Observations

0.5865 1,062

0.1404 1,047

0.0943 1,047

INTANG EQINC R&D EMP ∆SALE SIZE MB

(-1.58) -0.2743** (-2.12) 0.8880 (0.16) -0.0537 (-0.12) 0.3780*** (6.75) -0.0692 (-1.49) 0.6424*** (13.83) -0.0008 (-0.34)

UTB_BEGIN2007

Intercept

(Pseudo) R-Square Number of Observations

Panel B. Results from the Heckman Two-Stage Procedure Dependent Variable Ln(UTB_BEGIN2007) OLS HIGH_NEG_CSRpre 0.3663***

48

TABLE 5 Incorporating Moderate Level of Irresponsible CSR Activities This table reports regression results from two samples. Panel A presents results that correspond to those reported in Panel A of Table 3 using 11,006 firm-year observations covering the period 2003-2009. Panel B uses firm-level data around the implementation of FIN 48 and presents results that correspond to those reported in Table 4, Panel A. The regression models are: AGGRESSIVE = β0 + β1 HIGH_ NEG_CSR + β2 SOME_NEG_CSR + β3 POS_CSR + β4 ABS_DA + β5 IO + β6 CASH + β7 ROA + β8 LEV + β9 NOL + β10 ∆NOL + β11 FI + β12 PPE+ β13 INTANG + β14 EQINC + β15 R&D + β16 EMP+ β17 ∆SALE + β18 SIZE + β19 MB + β20 Lag(Dependent Variable) + Year Dummies + Industry Dummies + ε; where all variables, except SOME_NEG_CSR, are as defined in Appendix A. SOME_NEG_CSR is an indicator variable that equals 1 if NEG_CSR = 2 or 3. In the Panel A, suscript it stands for firm i in year t; in Panel B, suscript pre indicates that the variable is pre-determined (as of 2006) in relation to the tax avoidance measures (as of 2007 or as of 2007-2009). Regressions in Panel B do not include year dummies. The t-statistics and z-statistics are reported in parenthesis. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

Panel A.Results from the Baseline Analysis Dependent Variable SHELTERit Logit

HIGH_NEG_CSRit SOME_NEG_CSRit

(Pseudo) R-Square Number of Observations

DTAXit OLS

DD_BTit OLS

CETRit OLS

Model 1

Model 2

Model 3

Model 4

0.5914*** (3.40) 0.3242*** (2.91)

0.0216*** (4.36) 0.0113*** (3.37)

0.0109*** (4.73) 0.0041*** (2.64)

-0.0345***

0.6334 6,839

0.1215 6,393

0.4681 4,191

0.1985 9,147

Panel B. Results from the FIN 48 Anlaysis Dependent Variable Ln(UTB_BEGIN2007) OLS

HIGH_NEG_CSRpre SOME_NEG_CSRpre

(Pseudo) R-Square Number of Observations

(-5.39) -0.0063 (-1.48)

PCT_SETTLE Tobit

DECREASE Logit

Model 1

Model 2

Model 3

0.5639*** (4.30) 0.2567*** (3.31)

0.2526*** (4.11) 0.1046*** (2.74)

0.8122*** (3.25) 0.2819* (1.86)

0.5972 1,137

0.1489 1,121

0.0945 1,121

49

TABLE 6 Additional Analyses of the Baseline Regression Results In this table, we report the results of several additional analyses based on the baseline regressions. Panel A presents the baseline regression results based on the propensity score matched samples. Propensity scores are calculated using a logit model where the dependent variable is HIGH_NEG_CSRit and the independent variables are political preference, firm visibility, industry-adjusted advertisement expenditure, KZ score (financial constraint measure), institutional ownership (IOit), ROAit, LEVit, SIZEit−1, MBit−1, stock return volatility, firm age, year and industry dummies. Panel B presents the results of the firm-level regressions. Specifically, we first calculate firm-level empirical measures for all the variables in the baseline regression model of equation (1) by using the average of the variables over the seven-year sampling period. For dummy variables, such as SHELTERit, HIGH_NEG_CSRit and NOLit, we construct their firm-level dummy variables, which equal one if these dummy variables equal one in at least half of the years during 2003-2009 and they equal zero otherwise. Then we use these firm-level variables to run the firm-level regressions. Panel C presents the results of Fama-MacBeth regressions. The t-statistics and z-statistics are reported in parenthesis. ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

DTAXit OLS

DD_BTit OLS

CETRit OLS

0.0143*** (2.71) 0.2627 1,180

0.0059** (2.22) 0.5795 962

-0.0279***

0.0145* (1.70) 0.0617 1,792

0.0077** (2.15) 0.5257 1,178

-0.0275** (-2.19) 0.1074 2,249

0.0150** (3.31) 0.2632 6,393

0.0069** (3.47) 0.4657 4,191

-0.0277***

SHELTERit Logit Panel A. Propensity Score Matching Results Dependent Variable

HIGH_NEG_CSRit (Pseudo) R-Square Number of Observations

0.1542** (2.40) 0.6549 1,258

(-3.78) 0.1910 1,894

Panel B. Firm-Level Regression Results HIGH_NEG_CSRi (Pseudo) R-Square Number of Observations

0.6171** (2.50) 0.5400 1,916

Panel C. Fama-MacBeth Regression Results HIGH_NEG_CSRit Average (Pseudo) R-Square Number of Observations

0.5382*** (4.53) 0.6840 6,839

50

(-4.69) 0.2407 9,147