Tax Internal Control Quality: The Role of Auditor-Provided Tax Services and Tax Department Integration

Transcription

1 Tax Internal Control Quality: The Role of Auditor-Provided Tax Services and Tax Department Integration Lisa De Simone University of Texas at Austin Matthew Ege* University of Texas at Austin and Bridget Stomberg University of Texas at Austin July 2012 Early Draft. Please do not cite without permission of the authors. *Corresponding author Abstract We propose that auditor-provided tax services (tax NAS) improve tax internal control quality by combining knowledge of business operations and tax reporting. Consistent with predictions, we find a positive and economically significant association between tax NAS and tax internal control quality from ; moving from the 25 th to the 75 th percentile of tax fees decreases the probability of a tax material weakness by 29 percent. This effect is strongest when companies pay less attention to tax such as in the years preceding the adoption of FIN 48 or when companies lack tax experts in top management or audit committee roles. We also provide evidence that tax NAS improve overall financial reporting quality through tax internal controls. Our research informs shareholders, boards, regulators, and auditors about the benefits of tax NAS and the mechanism through which tax NAS affect financial reporting quality. Acknowledgements: We thank Patrick Badolato for providing tax expertise data. We also appreciate helpful comments from Greg Capps, Dain Donelson, Tracie Majors, John McInnis, Lillian Mills, Jaime Schmidt and Laura Wang. All authors gratefully acknowledge support from the Accounting Doctoral Scholars Program as well as the Red McCombs School of Business.

2 I. Introduction Joint consideration of business operations, tax law and financial reporting for income taxes is crucial to maintaining effective internal controls over tax reporting. By providing tax services to their clients, audit firms combine their tax knowledge with knowledge of client business operations, allowing them to identify both audit and tax related risks. Therefore, we propose that auditor-provided tax non-audit services (tax NAS) improve tax internal control quality by facilitating the joint consideration of tax reporting and business operations within the client. Further, we predict the benefits of tax NAS on tax internal control quality are strongest in situations where the client is not already routinely paying attention to tax issues when making operating and financial reporting decisions. Companies with a high level of attention to tax at the director and officer level are more likely consider how operational changes affect tax risk and reporting. As such, we expect these companies to benefit less from tax NAS. 1 Demonstrating that tax NAS improve internal control quality has broad implications. First, Kinney et al. (2004) present evidence that tax NAS are associated with a reduced likelihood of financial restatements. We propose that improving tax internal control quality could be the mechanism through which tax NAS improve financial reporting quality. Gleason et al. (2011) document a link between tax internal control quality and tax financial reporting quality by providing evidence that tax material weaknesses facilitate earnings management through the tax expense. Although not specific to tax, Donelson et al. (2012) demonstrate that poor internal control quality is associated with an increased likelihood of future fraud revelation. Therefore, understanding the determinants of effective internal controls has implications for improving financial reporting quality. Additionally, prior to the passage of the Sarbanes-Oxley Act of Throughout this paper, we use client and company to refer to the entity being audited or receiving tax NAS. We use audit firm to refer to the financial statement audit or tax NAS provider. 1

3 (SOX) regulators and legislators extensively debated whether to allow NAS (Romano 2005; Caplan et al. 2009). Ultimately, regulators banned most NAS but specifically allowed tax NAS (U.S. Congress 2002). Identifying benefits of tax NAS with respect to internal control quality and financial reporting quality is of interest to shareholders, boards of directors, regulators, and auditors. Company personnel must be aware of how significant business transactions will affect tax reporting to ensure that tax-related internal controls are designed appropriately and operating effectively. If the income tax function is isolated from other departments, tax personnel may not be aware of issues that materially affect taxes. 2 Similarly, directors and officers may not be aware of the tax ramifications of business decisions. In these situations, having the financial audit firm provide tax services increases the likelihood that the company will jointly consider tax and business matters in the financial reporting process. This makes the company better able to identify potential tax risks, and design and implement appropriate tax internal controls. For example, the audit firm would become aware of a company s intention to expand operations into new countries during audit planning meetings. The audit firm can communicate potential tax implications of expansion to the company, such as the need to file tax returns in new jurisdictions, and prompt the company to ensure that adequate internal controls are in place to manage the new tax risks. 3 2 An informed tax function can itself be a control. For example, in a 2005 management report on internal controls Vetro, Inc. disclosed a material weakness of insufficient controls over the preparation of the income tax provision and stated that to improve internal controls over the calculation of the income tax provision they will include quarterly meetings between our tax and accounting team members to formalize communication between the two functions (Vetro 2006, 58). 3 Auditor independence rules prohibit the auditor from assuming the role of management or auditing their own work (PCAOB 2007). Although the tax team cannot design tax internal controls over financial reporting, the tax team can identify tax internal control risks and assess the design and operating effectiveness of tax internal controls over financial reporting throughout the year. The types of discussions and information exchange contemplated in this paper are within the scope of permissible auditor behavior and do not violate independence rules. 2

4 Tax NAS increase the likelihood that the company jointly considers tax and business matters by (1) fostering a strong relationship between the audit firm tax partner and company personnel, and (2) broadening the tax partner s knowledge of both client-specific tax and audit risks. First, relative to a tax partner on the audit team who does not provide any tax NAS, a tax partner who provides tax NAS develops a better relationship with the client through more frequent interactions and his or her role as client advocate on consulting projects. This relationship increases the likelihood that the tax partner shares information from discussions with company management or other audit team members with the client s tax personnel to prevent material weaknesses from arising. More frequent communications throughout the year also increase the likelihood that any potential material weaknesses are identified early in the reporting period and resolved by year-end. 4 Second, through involvement in both tax NAS and audit work, the audit firm s tax partner is more cognizant of both tax and audit risks than a tax provider who is not involved in the financial statement audit. The tax partner brings a deep knowledge of the client s tax issues to the audit as well as an understanding of potential audit risks when conducting tax NAS. Audit firm involvement during the year helps companies avoid surprises during the annual provision review by increasing the auditor s awareness of transaction details and audit risks early in the year (Larsen 2011). By being actively involved in the client s audit and tax functions, the tax partner is well-positioned to identify new tax issues that arise during the year and discuss them with the client. Therefore, we predict a positive association between tax NAS and tax internal control quality. 4 Auditor internal control opinions are as of a specific date, as opposed to covering an entire period. Accordingly, if a material weakness is discovered, but subsequently remediated and validated by the auditor prior to the as of date on the report, then the material weakness does not have to be disclosed. 3

5 We expect the benefits of tax NAS on tax internal control quality will be greatest when the company does not routinely pay attention to tax issues at the director and officer level when making business and financial reporting decisions. We test this prediction in two ways. The period immediately following the adoption of FIN 48 is our first proxy for attention to tax. From , the years preceding FIN 48, tax departments were charged primarily with monitoring tax compliance (Robinson et al. 2010). As a result tax risks were not well integrated into the overall risk assessment of the company. For example, a 2005 survey published by the Tax Executives Institute (TEI) reports that only 27 (43) percent of tax executives made presentations to the board (audit committee) in that year (Larsen 2011). This anecdotal and survey evidence supports our proposition that directors and officers, on average, did not pay much attention to taxes when making business and financial reporting decisions from Since that time, however, companies have increased their attention to income taxes, largely in response to SOX and the passage of FASB Interpretation No. 48 Accounting for Uncertainty in Income taxes An interpretation of FASB Statement No. 109 (FIN 48). We believe that both regulatory changes made tax issues more salient to directors and officers. 5 A 2011 TEI survey confirms this assertion, reporting that 81 percent of surveyed executives made presentations to the board or audit committee a threefold increase from 2005 (Larsen 2011). Additionally, over 58 percent of companies report that the implementation of FIN 48 was at least somewhat beneficial to improving the company s understanding of tax uncertainties (Deloitte 2007). Therefore, we predict that directors and officers are more cognizant of tax issues, on average, in recent years, and we use the period from 2007 through 2010 as our first proxy for attention to taxes in business and financial reporting decisions. 5 FIN 48 became effective for fiscal years beginning after December 15, The standard provides complex rules for measuring and recognizing tax benefits in financial statements and requires detailed annual disclosures of reserves related to uncertain tax positions. 4

6 Tax expertise of company managers and audit committee members is our second proxy for attention to tax. Top managers (i.e., CEOs and CFOs) are directly responsible for establishing and maintaining internal controls, and audit committees are charged with overseeing the accounting and financial reporting processes, which includes internal controls (U.S. Congress 2002). Top managers or audit committee members with tax expertise are more likely to have the requisite knowledge and experience to identify tax risks. Therefore, these individuals not only have the directive to maintain an effective internal control environment, but also have the taxrelated knowledge and expertise to evaluate internal controls over taxes and effectively communicate with tax department personnel about the income tax provision. Accordingly, we predict that companies with tax experts pay more attention to income taxes and benefit less from joint consideration of tax and business matters facilitated by tax NAS. Overall, results support our predictions that tax NAS improve internal control quality. Using pooled logistic regressions we find tax NAS are negatively associated with the disclosure of tax material weaknesses, our proxy for poor tax internal control quality, from 2004 to The marginal effect of moving from the 25 th to the 75 th percentile of the natural log of tax fees represents a 29 percent decrease in the likelihood of disclosing a tax material weakness, holding all other variables at their means. We also find that the benefits of tax NAS on tax internal control quality are weaker when the company is more likely to routinely pay attention to taxes when making business and financial reporting decisions. We find no association between tax NAS and tax material weakness from or for company-years where management or audit committee members have tax expertise. Together, these findings support our conjecture that tax NAS increase the likelihood that tax and business matters are jointly considered within the company, enabling companies to improve tax internal control quality. However, these 5

7 benefits are muted when the company routinely pays attention to tax at the director and officer level. Combined with evidence from prior literature that internal control quality affects financial reporting quality, our finding that tax NAS improve internal control quality motivates us to re-examine the negative relation between tax NAS and restatements documented in prior studies. We find the negative relation between tax NAS and all restatements becomes insignificant after controlling for tax material weaknesses. These results suggest that tax NAS improve financial reporting quality indirectly through tax internal control quality. In contrast, we find that the negative relation between tax NAS and tax-only restatements weakens but remains significant after controlling for tax material weaknesses. This suggests that tax NAS provide direct benefits to tax financial reporting quality incremental to those provided through tax internal control quality. This study makes several important contributions to the literature. First, although prior literature provides evidence that tax NAS improve financial reporting quality, there is limited research examining the link between tax NAS and internal control quality. Our findings that tax NAS are associated with enhanced tax internal control quality should be of interest to shareholders and audit committees assessing the costs and benefits of tax NAS, as well as to regulators. Further, our findings that tax NAS are less beneficial for companies with tax expertise in top management or on the audit committee can inform boards of directors and managers when establishing criteria for hiring top executives and appointing audit committee members. Second, this study extends the literature examining the determinants of financial reporting quality by highlighting a previously unexamined mechanism through which tax NAS improve financial reporting quality. Additionally, our finding that tax NAS benefit tax internal 6

8 control quality most in companies with little attention to tax extends the line of research examining the consequences of tax NAS. Despite the lack of evidence that tax NAS impair independence, auditor-provided tax fees declined sharply after the passage of SOX (Maydew and Shackelford 2005) and have not returned to the pre-sox level. 6 We help reconcile these seemingly conflicting facts by demonstrating that increased attention to tax at the director and officer level reduces the beneficial effects of tax NAS. Finally, our study uses new data that allow us to directly test how tax expertise affects the relation between tax NAS and tax internal control quality. The remainder of the paper proceeds as follows. Section II discusses related literature and develops hypotheses. Section III provides sample selection and research. Section IV provides results of tests, Section V describes additional supplemental analyses and results, and Section VI concludes. II. Related Literature and Hypothesis Development In July 2002, Congress passed SOX Section 201, which banned many auditor-provided NAS. However, tax NAS remain permissible as long as the audit committee pre-approves fees. Prior literature demonstrates that tax NAS are positively associated with financial reporting quality. Kinney et al. (2004) find a negative association between tax NAS and restatements, which supports the theory that tax NAS increase the information available to the auditor and lead to higher quality audits and improved financial reporting quality. The authors conclude that improved financial reporting quality outweighs any threat to independence created by the economic bond tax NAS generate. Paterson and Valencia (2011) extend Kinney et al. (2004) but 6 Maydew and Shackelford (2005) present evidence that auditor-provided tax fees as a percentage of audit fees decreased from 100 percent in 2001 to 25 percent in Based on our sample data, auditor-provided tax fees averaged $334,000 in 2004; this average declines to $278,000, or 12 percent of total audit fees, in

9 differentiate between recurring and nonrecurring tax NAS. Recurring tax NAS tend to encompass work that deepens the audit firm s knowledge of business operations. The authors find recurring tax NAS drive the negative relation between tax NAS and restatements. Gleason and Mills (2011) provide evidence that companies purchasing tax NAS more accurately estimate income tax expense, specifically when accruing reserves for potential future tax liabilities. Their results offer a direct link between tax NAS and tax financial reporting quality. Collectively, these studies suggest that tax NAS enhance financial reporting quality. However, to our knowledge, no published paper addresses the relation between tax NAS and tax internal control quality, which is a significant determinant of financial reporting quality. In unpublished work, Elder, Harris and Zhou (2008) examine the relation between tax NAS and reported material weaknesses in internal controls from The authors find that tax NAS are associated with fewer reported tax and non-tax material weaknesses, and attribute the result to a lack of auditor independence. This interpretation is somewhat unexpected given the limited evidence from prior studies that NAS are detrimental to auditor independence. For example, in addition to the literature cited above, Francis (2006) concludes there is no smoking gun evidence linking the provision of non-audit services with audit failures. We therefore propose and test an alternative explanation for the link between tax NAS and tax internal control quality. 7 We posit that tax NAS facilitate the joint consideration of tax and business operations creating a positive association between tax NAS and tax internal control quality. Because financial statement auditors are knowledgeable about the company s overall business organization and operations, they are well positioned to understand how general business risks 7 In an untabulated robustness test, we find no relation between tax NAS and the likelihood that a clean internal control opinion is subsequently restated. If tax NAS cause independence concerns, we would expect a positive association between tax NAS and clean internal control opinions that are subsequently restated, reflecting the auditors initial reluctance to issue a material weakness opinion due to economic incentives. 8

10 affect income tax reporting. In the course of providing tax compliance or consulting services, the audit firm can share this information with the tax department, whether through informal communications or directly through a specific tax engagement. For example, if a company intends to expand operations into new countries, the audit team which often includes the tax partner will be made aware of these plans during audit team meetings. The tax partner can communicate potential implications for tax reporting to the client s tax personnel, such as the need to file tax returns in new jurisdictions. This information could prompt tax personnel to ensure that adequate internal controls are in place to manage these risks. One such control might involve verifying with the legal department the nature and extent of each entity s activities to establish filing requirements. Additionally, increased interactions with the tax partner make it more likely that the client will become aware of potential control problems earlier in the year, which the client can remediate prior to year-end to avoid disclosure of a material weakness. These interactions are especially salient for the tax function, which according to Deloitte (2011), was not traditionally integrated with the rest of the organization or involved with the audit committee. We predict that tax NAS assist companies in identifying and mitigating potential internal control risks and therefore improve the client s internal controls over reporting for income taxes. We formally state this prediction in the alternative form below. H1: There is a negative relation between tax NAS and the probability of reporting a material weakness in internal controls over taxes. We expect the benefits of tax NAS will be greater for companies that are less likely to routinely pay attention to tax at the director and officer level. Companies that pay attention to tax are more likely to integrate tax considerations into changes in operations, general business risks, and strategic planning. A company that pays attention to tax acknowledges tax as a common 9

11 thread running through and impacting many business risks and incorporates tax risks into the overall risk framework of the company (Deloitte 2011). For these companies, the possibility for tax NAS to facilitate incremental consideration of tax matters is limited. Thus, we predict any positive association between tax NAS and tax internal control quality will be weakest for companies that are more likely paying attention to tax at the director and officer level. We formally state this prediction in the alternative form below. H2: The negative relation between tax NAS and the probability of an internal control weakness over income tax is weaker for companies more likely to pay attention to tax at the director and officer level. We test the effect of attention to tax on the benefits of tax NAS in two ways. First, we predict that the benefits of tax NAS on tax internal controls are more significant in the early years of SOX, when many companies assessed internal controls over taxes for the first time. We submit that companies, on average, pay more attention to tax subsequent to SOX. After the disclosure of a significant number of tax material weaknesses under SOX, boards of directors increased oversight of the tax function and companies committed resources to improve internal control quality (Larsen 2011). As a result, disclosed tax material weaknesses declined significantly, decreasing monotonically from 2005 through Additionally, we posit that the implementation of FIN 48 in 2007 fundamentally changed the way companies view tax risks and increased the visibility of income tax reporting. Prior to FIN 48, companies typically accrued contingent liabilities for uncertain tax positions, but little guidance existed governing the calculation of reserves and no separate disclosure was required. 8 FIN 48 provided new rules for the measurement and recognition of uncertain tax benefits and requires specific disclosures of the magnitude of tax uncertainties. This increased transparency of tax risk prompted 7 Because FAS 109 is silent on accounting for income tax contingencies, most companies followed the guidance in FAS 5 and recorded contingent liabilities for tax matters only when the likelihood of payment was probable. 10

12 management and boards of directors to increase their monitoring of income tax risk and reporting. Therefore, we expect companies pay more attention to tax, on average, in recent years and that there are fewer benefits of tax NAS in as compared to Second, we test our attention to tax hypothesis using information about the tax expertise of the CEO, CFO and audit committee members. Managers and audit committee members experienced in tax matters have more awareness of potential tax risks and can conduct more detailed analyses of tax amounts reported on the financial statements. For example, managers and audit committee members with tax expertise are likely able to ask probing questions about the measurement and recognition of tax contingencies, leading to more frequent communications between tax personnel and top management. In sum, we submit that management or audit committee tax expertise improves communications between the tax department and other company departments, and results in greater involvement from management and the board of directors in identifying and implementing controls over financial reporting for income tax. We therefore predict that having a tax expert in top management or an audit committee role weakens the positive relation between tax NAS and tax internal control quality. III. Sample and Research Design a. Tests of H1 To test H1, we begin with all SOX Section 302 and 404 internal control disclosures included in the Audit Analytics database from 2004 through Next, we require each observation to have data available from both CRSP and Compustat to calculate control variables 9 We begin our sample in 2004 because the Public Company Accounting Oversight Board did not issue guidance defining levels of internal control severity until it released Auditing Standard No. 2 on March 9, This standard included formal definitions of material weakness, significant deficiency, and control deficiency. As a result, the completeness of material weakness disclosures prior to 2004 is questionable. For example, the number of material weaknesses reported in 2004 significantly increased compared to 2002 and In our sample, the number of disclosed tax material weaknesses jumped from nine in 2003 to 184 in We find it unlikely that the material weaknesses reported in 2004 did not exist in earlier years. 11

13 shown in prior literature to affect the probability of a company (or its financial statement auditor) disclosing a material weakness. Finally, in the event of restated internal control reports, we retain only the most recently issued internal control report for each fiscal year. We do this because we are interested in whether or not a material weakness existed in a year, not in when the material weakness was disclosed. The full sample consists of 26,089 observations from 5,524 unique companies. Panel A of Table 1 provides descriptive statistics for the sample. All continuous variables are Winsorized at one and 99 percent. Nearly 72 percent of the sample purchases tax services from their auditor. The average (median) amount of tax fees paid to auditors is over $270,000 ($37,000), compared to average (median) audit fees of over $2.1 million ($851,000). [Insert Table 1 here.] H1 predicts a negative relation between tax NAS and the probability of a tax material weakness disclosure. To test H1, we estimate the following logistic regression from 2004 through Prob (ICW_TAX=1) = F(β 0 + β 1 TAXNAS + β 2 LNAUDIT + β 3 LNMKTCAP + β 4 AGGLOSS + β 5 SHUMWAY + β 6 LSEGCOUNT + β 7 FORTRANS + β 8 MERGER + β 9 EXTREMESG + β 10 RESTRUCTURE + β 11 BIG4 + β 12 AUDITOR_RESIGN + β 13 PCTFORSALES + β 14 TAXLOSS + Year FE) (1) We define all variables in detail in Appendix A, with references to Compustat variable names. ICW_TAX is a company-year indicator variable equal to one if a company discloses a material weakness in internal controls over income taxes in that year, and zero otherwise. We set ICW_TAX equal to one for a given company-year regardless of whether the material weakness was disclosed by the company in a SOX Section 302 or 404(a) disclosure, or by the financial statement auditor in a SOX Section 404(b) disclosure. We estimate equation (1) using two different measures of tax NAS: LNTAXNAS is the natural logarithm of tax fees reported in Audit Analytics in that year, and TAXNASIND is an 12

14 indicator variable equal to one if the company purchased tax NAS in that year, and zero otherwise. In equation (1), β 1 is the coefficient of interest. H1 predicts that β 1 is negative. Because audit fees increased dramatically after SOX and are positively associated with the probability of a material weakness disclosure (Hogan and Wilkins 2008), we avoid scaling tax fees by audit fees or total fees. 10 Instead we control for the natural logarithm of total audit fees LNAUDIT and expect the coefficient β 2 to be positive. We also include several control variables shown in Ashbaugh-Skaife et al. (2007) and Doyle et al. (2007b) to be significant determinants of material weaknesses. LNMKTCAP controls for size, as smaller companies exhibit an increased likelihood of material weaknesses. We therefore predict that β 3 will be negative. We include AGGLOSS and SHUMWAY as controls for financial distress. AGGLOSS is an indicator variable equal to one if earnings before extraordinary items in years t-1 and t sums to less than zero, and zero otherwise. SHUMWAY is the decile rank of the percentage probability of bankruptcy from the default hazard model prediction based on Shumway (2001). Higher scores indicate higher probability of bankruptcy. We expect companies in greater financial distress to be more likely to disclose material weaknesses and, therefore, β 4 and β 5 to be positive. Next we include variables to control for financial reporting complexity, all of which have been shown to increase the likelihood of material weaknesses. LSEGCOUNT captures the natural logarithm of the number of operating and geographic segments the company reports in the Segments database of Compustat in year t. FORTRANS is an indicator variable equal to one if the companies has non-zero foreign currency translation in year t, and zero otherwise. MERGER is an indicator variable equal to one if the company has non-zero acquisition expense in year t-1 10 Additionally, prior research shows a negative association between company size and material weakness disclosures (Ashbaugh-Sakife et al. 2007; Doyle et al. 2007b), so we do not consider measures of tax NAS that scale tax fees by total assets or market capitalization. 13

15 or year t-2, and zero otherwise. Consistent with prior literature, we expect β 6, β 7 and β 8 to be positive. EXTREMESG and RESTRUCTURE control for rapid growth, which increases the likelihood of reporting a material weakness. EXTREMESG is an indicator variable equal to one if the year-over-year, industry-adjusted sales growth is in the top quintile of the sample. RESTRUCTURE is the aggregate restructuring charges in years t-1 and t, scaled by the company s market capitalization in year t. We expect the coefficients on both of these variables to be positive. We also include control variables to capture auditor quality. BIG4 equals one if the financial statement auditor in year t is KPMG, Ernst and Young, Deloitte, PricewaterhouseCoopers or one of their predecessors, and zero otherwise. AUDITOR_RESIGN is an indicator variable equal to one if the company experienced an auditor resignation during the year as reported by Audit Analytics. We expect both of these variables to increase the likelihood of disclosing a material weakness, and expect β 11 and β 12 to be positive. Because we are examining tax material weaknesses specifically, we include two additional variables to control for income tax reporting complexity. PCTFORSALES is the percentage of total sales attributable for foreign segments as reported in the Compustat Segments database. As international operations expand, companies are more likely to face complex tax issues in multiple jurisdictions. Accurately calculating an income tax provision is more difficult as for multinational entities due to complex financial reporting issues such as permanently reinvested foreign earnings (APB 23 assertions), foreign tax credit calculations, foreign currency translation issues, valuation allowances in multiple jurisdictions and intraperiod tax allocations issues (e.g. related to other comprehensive income). Therefore, we expect β 13 to be positive. 14

16 Finally, TAXLOSS is a binary variable set equal to one if the company reports a tax net operating loss carryforward (NOL) in year t. 11 We make no prediction for the sign of β 14. On one hand, tax losses may reflect overall financial distress; companies with tax losses may not consider income tax reporting to be a priority and, therefore, be more likely to disclose material weaknesses. Additionally, companies must regularly evaluate NOLs to determine if they are realizable and establish a valuation allowance as contra asset to offset the benefit of any tax NOLs that are expected to expire unused. The rules governing accounting for valuation allowances are complex. Therefore, TAXLOSS could be positively associated with ICW_TAX. Conversely, companies with extensive tax losses may have simpler tax provisions, especially if the NOLs are completely offset with a valuation allowance, resulting in a zero tax provision. b. Tests of H2 i. By Period H2 predicts that attention to tax attenuates the negative relation between tax NAS and tax internal control weaknesses. Based on anecdotal evidence that SOX and FIN 48 prompted companies to better increase their focus on tax matters, our first proxy for attention to tax is the period from Our first test of H2 re-estimates equation (1) on the full sample separately from 2004 through 2006 and from 2007 through Panel B of Table 1 presents statistics for the sample period that pre-dates the adoption of FIN 48 (2004 through 2006), and Panel C presents statistics for the period subsequent to the effective date of FIN 48 (2007 through 2010). Consistent with a decline in tax NAS since the passage of SOX, mean and median tax fees are significantly lower in the later period although mean and median audit fees are higher. Sample companies have lower average market capitalization in the later period. The frequency of 11 We acknowledge the limitations of tax net operating loss data from Compustat. See, for example, and Kinney and Swanson (1993) and Guenther (2011). 15

17 aggregate losses is greater in the later period while the frequency of foreign transactions, mergers and restructuring are all significantly lower in the later period. Both trends likely reflect consequences of the economic downturn in 2008 and From 2007 through 2010 companies report a higher percentage of foreign sales and greater net operating losses. Sample companies are also less likely to use a Big 4 auditor in the later period. Table 1, Panel D presents the frequency of internal control weaknesses over tax by year. Disclosures of internal control weaknesses over income taxes peak in 2005 at 237 disclosures then decrease monotonically over the remaining sample period. In total, sample companies disclose 628 tax material weaknesses in the early period and 461 in the later period. As in our tests of H1, we use two measures of tax NAS: LNTAXNAS and TAXNASIND to test H2. Our coefficient of interest is β 1. We expect β 1 to be negative and significant in the early period ( ), but less significant in the late period ( ). Our predictions for control variables are consistent with our discussion of H1 above. ii. By Tax Expertise Our second proxy for attention to tax is top management or audit committee member tax expertise. We begin with the full sample detailed above and match company-year observations to Boardex, a professional business network database. Boardex provides biographical information and employment history for corporate directors and officers. 12 We utilize Boardex to search for employment titles that convey tax expertise. See Appendix B for a list of titles we consider representative of tax expertise. We set MGMT_EXPERT equal to one if a tax expert serves as the CEO or CFO during the year, and zero otherwise. Similarly, AC_EXPERT is an indicator variable set equal to one if a tax expert serves on the audit committee during the year, and zero otherwise. EXPERT equals one if either MGMT_EXPERT or AC_EXPERT equals one. 12 A t-test on average LMARKETCAP suggests that average company size does not significantly differ between our main sample and the Boardex sample (p-value =0.8870). 16

18 Table 2 provides descriptive statistics for the Boardex sample. All continuous variables are Winsorized at one and 99 percent. Panel A presents statistics for the entire sample, Panel B presents statistics for company-years that do not have a top manager or audit committee member with tax expertise (EXPERT=0), and Panel C presents statistics for company-years that have a top manager or audit committee member with tax expertise (EXPERT=1). Observations with tax expertise in top management or on the audit committee report significantly higher tax and audit fees. Expert company-years also have greater average market capitalization. Not surprisingly, these observations also reflect characteristics indicative of organizational and tax complexity such as more operating segments, and greater foreign transactions, foreign sales and mergers. [Insert Table 2 here.] Our second set of tests of H2 estimate equation (1) on the Boardex sample. We estimate the regression separately where EXPERT equals zero or one. To investigate whether management or audit committee expertise differentially affects the association between tax NAS and tax internal controls, we further split the sample by MGMT_EXPERT and AC_EXPERT, reestimating equation (1) separately where each indicator variable equals one. In these tests of H2, we restrict our tax NAS measure to LNTAXNAS. We do this because we do not have sufficient variation in the EXPERT subsamples to evaluate the effect of TAXNASIND on the probability of a tax material weakness disclosure. β 1 is the coefficient of interest. H2 predicts that β 1 will be negative and significant for company-years lacking tax expertise in a top management or audit committee position, but less significance for companyyears with tax experts in these roles. Our predictions for control variables are consistent with our discussion of H1 above. In alternative specifications, we also control for MGMT_EXPERT or AC_EXPERT, making no prediction on these control variables. 17

19 IV. Results a. Tests of H1 The first two columns of Table 3 provide results of testing H1 using two measures of auditor-provided tax services: LNTAXNAS and TAXNASIND. Results are consistent with our hypothesis that auditor-provided tax services decrease the probability of a tax internal control weakness, on average. Specifically, we find negative and statistically significant coefficients on LNTAXNAS and TAXNASIND over the full period, suggesting that the probability of disclosing an internal control weakness over income taxes is decreasing in purchases of non-audit tax services. Additionally, these results are economically significant. The coefficient of for LNTAXNAS in the first column corresponds to a marginal effect of approximately -1.3 percent. This marginal effect represent an approximate 29 percent decrease in the probability of disclosing a tax material weakness when moving from the 25 th to the 75 th percentile of LNTAXNAS, holding all other variables constant at their means. The economic significance of TAXNASIND is similar, representing a 25 percent decrease in the probability of disclosing a tax material weakness when moving from TAXNASIND equals zero to one. We interpret this evidence as supporting our theory that tax NAS improve tax internal control quality. [Insert Table 3 here.] The effects of control variables are mostly in line with predictions. We find positive and significant coefficients on LNAUDIT, AGGLOSS, FORTRANS, AUDITOR_RESIGN and PCTFORSALES across both measures of tax NAS. We also find a negative and significant coefficient on LMARKETCAP across both specifications. Contrary to predictions, we find negative and significant coefficients on SHUMWAY and BIG4. The coefficient on TAXLOSS, for 18

20 which we had no directional prediction, is significantly negative across both specifications. The remaining coefficients are not significantly different from zero. b. Tests of H2 i. By Period The remaining columns of Table 3 provide results for the first test of H2, which predicts that as companies increase their attention to tax following the passage of SOX and FIN 48, the benefits of tax NAS diminish. We find a negative and significant relation between both measures of tax NAS and tax material weaknesses in the early period, but an insignificant relation in the later period. We interpret these results as consistent with our theory that, in a period in which directors and officers paid little attention to taxes, the benefits of tax NAS on tax internal control quality were greater than in a period characterized by increased attention to tax. The coefficients on control variables are similar to those reported in testing H1 above. As in tests of H1, the coefficients in the early period are economically significant for both measures of tax NAS. The coefficient of for LNTAXNAS in the third column corresponds to a marginal effect of approximately -0.4 percent. This marginal effect represents an approximate 16 percent decrease in the probability of disclosing a tax material weakness when moving from the 25 th percentile to the 75 th percentile of LNTAXNAS, holding all other variables constant at their means. We estimate the marginal effect of TAXNASIND to be more than twice as large; moving from TAXNASIND equals zero to one is associated with an estimated 33 percent decrease in the probability of disclosing a tax material weakness. ii. By Tax Expertise Table 4 provides results for our second set of tests of H2. We estimate equation (1) separately by EXPERT, MGMT_EXPERT, and AC_EXPERT to investigate whether tax expertise 19

21 in a top management or audit committee position attenuates the negative relation between tax NAS and ICW_TAX. 13 Panel A presents results from bifurcating the sample by EXPERT. The first column of Panel A estimates equation (1) on the entire Boardex sample as a baseline; consistent with tests of H1 above, we find a negative and significant coefficient on LNTAXNAS when estimating equation (1) on this smaller sample. The second and third columns of Panel A present results by EXPERT. [Insert Table 4 here.] As predicted, we find a negative and significant relation between LNTAXNAS and ICW_TAX for company-years lacking tax expertise in a top management or audit committee role, but no statistically significant relation between LNTAXNAS and ICW_TAX for company-years having a tax expert in top management or on the audit committee. Additionally, the coefficient of for LNTAXNAS in column two (i.e., for company-years lacking tax expertise) corresponds to a marginal effect of approximately -1.5 percent. This marginal effect represents an approximate 32 percent decrease in the probability of disclosing a tax material weakness when moving from the 25 th percentile of LNTAXNAS, where the probability of disclosing a tax material weakness is , to the 75 th percentile of LNTAXNAS, where the probability of disclosing a tax material weakness is , assuming all other variables equal their means. The coefficients on control variables are similar to those reported in testing H1 above. We interpret these results as supporting our theory that attention to tax attenuates the benefits provided by tax NAS on tax internal control quality. 13 We choose to estimate Equation (1) on each subsample separately in lieu of reporting interactions because of the difficulty in interpreting interactions in logistic regressions. 20

22 Panels B and C of Table 4 report results of estimating equation (1) after bifurcating our sample by MGMT_EXPERT or AC_EXPERT, respectively. 14 We conduct these tests to shed light on whether the relation between tax NAS and internal control quality differs based on who in the company has tax expertise. The coefficient on LNTAXNAS is negative and significant for company-years lacking tax expertise across all specifications, supporting the conjecture that both top management and audit committee member tax expertise are important determinants of tax internal control quality. The latter columns of Panels B and C additionally control for AC_EXPERT or MGMT_EXPERT, respectively. The last column in Panel B indicates that company-years having a tax expert in a top management position receive no incremental tax internal control benefits from tax NAS, holding audit committee tax expertise constant. In contrast, the last column in Panel C suggests that company-years having a tax expert on the audit committee do receive incremental tax internal control benefits from purchasing tax NAS, holding top management expertise constant. Together, these results provide some evidence that management tax expertise may be more critical for mitigating tax internal control risks than audit committee expertise. We acknowledge that we cannot completely rule out the possibility that we find no significant relation in the later period as a result of lower power. However, given that there are still over 450 tax material weaknesses in the later period and that the magnitude of the coefficients are much less negative in the later period, we believe the results are supportive of our predictions that companies that are more likely to jointly consider business operations, tax law and tax financial reporting in business decisions benefit less from tax NAS. Additionally, our second set of tests of H2 related to tax expertise provides further support for our theory. 14 In the subsample for which MGMT_EXPERT=1, we drop the control variables BIG4 and AUDITOR_RESIGN due to lack of variation. 21

23 V. Additional Analyses on Tax NAS and Restatements Our first set of supplemental analyses re-examines the negative association between tax NAS and restatements documented in prior studies. Given our finding that tax NAS improve tax internal control quality and the evidence in Donelson et al. (2012) that internal control weaknesses are a significant determinant of restatements, we examine whether tax NAS are associated with restatements after controlling for tax internal control quality. We estimate the following logistic regression separately for all restatements and tax-only restatements. Prob (RESTATE=1) = F(β 0 + β 1 TAXNAS + [Β 2 ICW_TAX] + β 3 LNMKTCAP + β 4 AGGLOSS + β 5 EXTREMESG + β 6 MERGER + β 7 RESTRUCTURE + β 8 BIG4 + β 9 PCTFORSALES + β 10 TAXLOSS + β 11 BM + β 12 LEVERAGE + β 13 EXTERNALFINANCING + β 14 FREECASHFLOW + β 15 NICHANGE (2) We set RESTATE equal to one if the company misstated it financial statements in year t for any issue (ALL=1). We first estimate the effect of tax NAS on RESTATE without controlling for ICW_TAX. We then re-estimate equation (2) including ICW_TAX along with tax NAS to capture the incremental benefit tax NAS provide after controlling for the effect of internal control weaknesses. Following prior literature (e.g., Kinney et al. (2004); Seetharaman et al. 2011) we use a tax NAS ratio, TAXNASRATIO, defined as the ratio of tax fees to the sum of audit and audit-related fees in year t, as our measure of tax NAS. We include control variables shown to be significantly associated with the likelihood of a restatement. Several of the control variables overlap from equation (1); the predicted sign of these coefficients are identical in both equation (1) and equation (2) with the exception of BIG4. In equation (2), to the extent that BIG4 serves as proxy for audit quality, we expect the coefficient on BIG4 to be negative. We include five control variables from the restatement literature (e.g., Larcker et al. 2007, Carcello et al. 2011) that are unique to equation (2). BM captures the company s book-to-market ratio and controls for the possibility that rapidly growing 22

24 companies have higher incidences of restatements; thus, we expect β 11 to be negative. The next four variables capture pressure to manage earnings because of various financial constraints. LEVERAGE represents the debt-to-asset ratio. EXTERNALFINANCING is total debt and equity financing in year t, scaled by the market value equity at t-1. FREECASHFLOW is the difference between operating cash flows in year t and average capital expenditures in years t, t-1 and t-2. We expect the coefficients on these three variables to be positive. Finally, NICHANGE is the percent change in net income from year t to t-1 and is intended to measure deterioration of company profitability. We make no prediction for the sign of β 15. [Insert Table 5 here.] Table 5 presents results of this analysis. In the first two columns, we estimate equation (2) on the entire sample. Consistent with prior literature, we find a negative association between TAXNASRATIO and all restatements from 2004 through However, once we control for tax material weaknesses, the coefficient on tax NAS is no longer significantly different than zero. These results suggest that tax NAS improve financial reporting quality indirectly through its effect on tax internal control. In the last two columns we estimate equation (2) using tax-only misstatements (TAX=1). We use the taxonomy in Audit Analytics to determine if a restatement is predominantly attributed to accounting for income taxes. In the third column, we find that TAXNASRATIO is negatively related to tax-only restatements. The coefficient of is more negative and significant than the coefficient on TAXNASRATIO in the first column. This is not surprising; tax NAS should have a more direct and significant effect on accounting for income taxes than other accounts. In Column (d), we include ICW_TAX as a control variable and find that the coefficient on tax NAS is less negative and less significant than in Column (c). These 23