Non-Audit Services and Earnings Management in the Pre-SOX and Post-SOX Eras

Transcription

1 Non-Audit Services and Earnings Management in the Pre-SOX and Post-SOX Eras Jayanthi Krishnan Fox School of Business and Management 13 th and Montgomery Streets, Speakman Hall, Temple University Philadelphia, PA Fax: Lixin (Nancy) Su School of Accounting and Finance The Hong Kong Polytechnic University Hung Hom, Kowloon, Hong Kong Yinqi Zhang Department of Accounting The Kogod School of Business American University Washington, DC, First Draft: September 2008 Current Draft: February 2009 We appreciate the comments of workshop participants at Shanghai University of Finance and Economics and Hong Kong Polytechnic University.

2 Non-Audit Services and Earnings Management in the Pre-SOX and Post-SOX Eras Abstract Extant research on the association between auditor-provided non-audit services (NAS) and earnings management assumes implicitly that all NAS can be equally harmful for auditor independence. However, the SEC prohibited specific kinds of NAS, arguing explicitly that these were potentially harmful to auditor independence. We argue that the decline in observed NAS from the pre-sox to post-sox eras resulting from the SEC s prohibition can be used to distinguish firms that had harmful NAS from those that did not. We examine earnings management (using discretionary accruals measures) during the pre-sox and post-sox eras for firms that underwent and did not undergo a decline in NAS following the new SEC rules prohibiting certain kinds of NAS. We find that firms that reduced (presumably harmful) NAS were associated with greater earnings management in the pre-sox era, but this was confined to downwards earnings management. Further, the downwards earnings management for the firms that reduced NAS was lower in the post-sox era than in the pre-sox era. These results obtain for firms that were more likely to have incentives to manipulate earnings downwards, firms with weaker corporate governance in the pre-sox than in the post-sox period, and clients of Big 5 auditors.

3 1. Introduction The impact of auditors joint provision of audit and nonaudit services (NAS) on auditor independence and the quality of their clients financial reporting has been in the spotlight in recent years. Motivated by SEC concerns about earnings management by companies (e.g., Levitt 2000), several recent studies have investigated whether clients of auditors that supply substantial non-audit services engaged in earnings management in the years prior to the Sarbanes-Oxley Act of 2002 (SOX hereafter), presumably because the auditors economic dependence on their clients arising from non-audit fees caused them to acquiesce to questionable accounting practices. The general consensus of this literature is that there is no association between accruals-based earnings management measures and nonaudit fees, except in some limited sub-samples. 1 The implicit assumption underlying this research is that all NAS can be harmful for auditor independence, because NAS fees, regardless of the nature of the service, creates economic dependence of the auditor on the client. Unlike this argument based on general economic dependence, the SEC articulates specific conditions under which NAS can harm auditor independence, that is, when the auditor functions in the role of management, audits its own work, or serves in an advocacy role for its client. This suggests that NAS has some bad components, which can affect auditor independence adversely. 2 Because the observed auditor fees comingle harmful and non-harmful elements of NAS, the lack of significant findings may indicate no association on average even if some NAS impairs independence. 3 1 See DeFond and Francis (2005) for a review of this literature, which we discuss in the next section. 2 However, even for the bad NAS, auditors concerns about litigation, and reputational loss could prevent them from compromising independence (DeFond et al. 2002). Presumably, the elements of NAS that are not bad would not affect auditor independence or audit quality. Further, if there are knowledge spillovers between audit and some nonaudit services, the latter may actually enhance the quality of audit and financial reporting (Simunic 1984; Lim and Tan 2008). 3 Recognizing the challenge of disentangling the effects of non-audit service provision and other mitigating factors (reputation concerns, litigation exposure, and knowledge spillover), Lim and Tan 1

4 We design our study using a harmful vs non-harmful NAS approach rather than the general economic dependence approach. Unfortunately the fee disclosures do not allow a direct separation of the two kinds of NAS, because other than information systems fees, fees relating to services which the SEC considered harmful were not identified in the disclosures. 4 Consequently we propose an indirect approach. Section 201 of the Sarbanes-Oxley Act of 2002 proscribed the provision of a number of nonaudit services by auditors, thus introducing an exogenous shock to the auditor-client relationship by forcibly eliminating one part of that relationship. 5 The forced reduction was in the kinds of NAS which were considered to be particularly harmful to auditor independence, for example, NAS relating to bookkeeping services. We argue that this reduction of NAS provides us with the opportunity to examine afresh the association between NAS and auditor independence. We use the change in the NAS fee between the pre-sox and post-sox periods as the proxy for harmful NAS provided in the pre-sox period, and examine its association with earnings management in both periods. If harmful NAS was in fact eliminated as a result of the policy change, the reduction would result in a reversal of any adverse earnings management effects it could have had in the pre-sox period. Other things equal, if in fact previous NAS had been a source of impairment of auditor independence, the elimination of harmful forms of NAS should (2008) take a different approach. They examine whether industry specialization moderates the association between NAS and several measures of auditor independence (in fact and in appearance). 4 The initial SEC fee disclosure rule (SEC 2000) required two NAS categories, information systems fees and other fees. Thus other than information system NAS, harmful NAS were included in the other fees categories. Therefore we are unable to separately identify most harmful NAS directly from the disclosures. The revised 2003 rules required three NAS categories, audit related fees, tax fees and other fees. 5 Specifically, SEC (2003), which implemented Section 201 of SOX, banned nine kinds of non-audit services. bookkeeping, financial information systems design and implementation, appraisal services and the like, actuarial services, internal audit outsourcing services, management functions or human resources, broker or dealer (including investment adviser), legal services and any other service that the Public Company Accounting Oversight Board determines, by regulation. One exception is tax-related non-audit services which can be provided by auditors as long as the provision is pre-approved by the client s audit committee. 2

5 lead to better earnings quality. 6 Thus while previous work used the levels of auditor non-audit fees in the pre-sox period as the proxy for the extent of impaired independence, we use the change in non-audit fees from the pre-sox to post-sox periods as the proxy because it reflects a forced reduction based specifically on principles of what constitutes harmful NAS. Following the above arguments, we explore whether (a) earnings management in the pre-sox period was greater for companies that subsequently had reductions in NAS fees and (b) whether companies with subsequent reductions in NAS fees had lower levels of earnings management in the post-sox period. We compare earnings management behavior by our sample firms for a pre-sox period, , and a post-sox period, We deliberately exclude the two years during which a number of rules, including the Sarbanes-Oxley Act, were introduced, and consequently was likely a period of transition. 7 Following prior work, we use discretionary accruals as our measure of earnings management. We find, consistent with Cohen et al (2008), that earnings management (measured by absolute discretionary accruals) were lower in the post-sox period than in the pre-sox period. More relevant to our investigation, we find that, during the pre-sox period, firms that subsequently had declines in NAS (which we argue was likely to have been harmful NAS), are associated with higher absolute discretionary accruals. Further, in the post-sox period these firms have lower levels of absolute discretionary accruals. This suggests, in contrast to the findings in the majority of previous studies, that harmful NAS could have affected auditor 6 Kinney et al. (2004) show a negative association between tax service fees and financial restatements suggesting, as DeFond and Francis (2005) argue, that particular nonaudit services such as taxation may actually reduce the incidence of earnings management and improve earnings quality. 7 The Sarbanes-Oxley Act was promulgated in During 2003, the SEC introduced a number of new proposed and final rules that affected manager and auditor reporting, including those to implement sections 302, 404, 406 and 407 of SOX, and new audit committee rules. 3

6 independence adversely in the pre-sox period. Next, we investigate this further by examining directional accrual measures. Interestingly, we find that the associations described above hold for negative discretionary accruals and not for positive discretionary accruals. Further, when we partition our sample into firms that had incentives for downwards earnings management and those that did not, we find the association between negative discretionary accruals and NAS holds for the former and not the latter. 8 Thus, if the harmful NAS caused impaired auditor independence, it occurred in the form of auditors allowing downward earnings management for firms with the incentive to do so, but not upward earnings management. Prior research has argued that litigation risk and fear of reputational loss can prevent auditors from compromising independence arising from NAS bonding (e.g. DeFond et al. 2002). Further, auditor litigation risk is greater in the presence of income-increasing earnings management (e.g. Lys and Watts 1994, Heninger 2001), and auditors work harder to detect income-increasing than income-decreasing misstatements (Barron et al. 2001). Therefore, our findings are consistent with the suggestion that auditors fear of litigation and/or reputational loss overshadows any NAS-driven incentives to compromise independence by allowing incomeincreasing earnings management. On the other hand, perhaps lower concerns about litigation and/or reputational loss associated with negative earnings management causes auditors to allow such management, thus giving in to NAS-driven incentives. Our study draws attention to a relatively little researched area, i.e., downwards earnings management. Prior research indicates downward earnings management is not infrequent. In a 8 In addition, the association between income-decreasing earnings management in the pre-sox period and the subsequent NAS reduction is more pronounced for firms audited by Big 5 auditors, and for firms which had a poorer quality of board governance in the pre-sox period than the post-sox period. 4

7 survey-based study, Nelson et al. (2003) find 31% of their sample adjustments were incomedecreasing in nature. Keune and Johnstone (2008) examine companies previously uncorrected accounting misstatements using disclosures mandated by Staff Accounting Bulletin No. 108 (SAB 108) and find that during the period , the mean misstatement was incomedecreasing. We also observe that, although prior research in the NAS area has not given this special attention, some studies (Ashbaugh et al. 2003; Reynolds et al. 2004) provide some evidence of an association between NAS and downwards earnings management. In general concerns about NAS focus on auditors role in allowing overstatements of client s financial performance. To the extent SOX helped reverse the tendency to allow downwards earnings management, our results indicate a positive effect of the new policy. The paper is organized as follows. We describe the background and our hypotheses in the next section. We then present our research design and models, followed by our empirical results. The final section contains our conclusions. 2. Background and Hypotheses The impact of auditors provision of NAS on their independence, both in fact and in appearance, has been the subject of considerable research. The research investigating independence in fact has focused overwhelmingly on earnings management. Because this is our focus as well, we discuss these other studies in some detail in the next subsection. A few studies examine outcomes other than earnings management. DeFond et al. (2002) study the association between the going concern opinion and nonaudit and total fees for the period prior to the promulgation of SOX, and find no association. Likewise, Kinney et al. (2003) investigate whether, prior to SOX, clients with greater NAS were more likely to have financial restatements 5

8 (indicating potentially lower audit quality and possible loss of auditor independence), and find no evidence of such an association. 9 Earnings Management and NAS A number of recent studies have examined whether auditors provision of NAS services could have adversely affected their independence, causing them to allow earnings management by their clients. These studies use different discretionary accruals measures, different proxies for auditor independence based on NAS fees, and often present results for partitions of their sample. Examining first the full sample results from these studies, we find that Frankel et al. (2002) report a positive association between absolute discretionary accruals and NAS fee ratio and total auditor fees. However, other studies using different measures of discretionary accruals (e.g., Ashbaugh et al. 2003, Larker and Richardson 2003), additional control variables (Reynolds et al. 2004) or different NAS-based measures of auditor independence (Chung and Kallapur 2003) find no association between absolute discretionary accruals and NAS-based independence measures. Some studies examine the results for subsets of their samples, based on the argument that contextual factors may cause the association to vary, even though on average there may be no association. In general, partitioning on factors such as firm size, corporate governance, client incentives and opportunities to manage earnings, indicates either that the absence of an association in the full sample also obtains in subsamples (e.g., Chung and Kallapur 2003) or that the association found in the full sample is eliminated in all subsamples (e.g., Reynolds et al 2004). 9 Unlike studies examining independence in fact, studies that examine the association between NAS and independence in appearance suggest somewhat more consistently, that NAS is negatively associated with perceived auditor independence. For example, Krishnan et al. (2005) and Francis and Ke (2006) show that client earnings response coefficients are negatively associated with NAS. 6

9 The combination of results from full sample analyses and sub-sample analyses discussed above suggests no support for an association between NAS and absolute discretionary accruals measures. We examine next the findings of studies that report results for positive and negative discretionary accruals separately. For positive discretionary accruals, Frankel et al. (2002) and Reynolds et al. (2004) report a positive association with NAS, Larcker and Richardson (2004) report a negative association with NAS, and Ashbaugh et al. (2003) find no association with NAS. However, there is less inconsistency in the results for negative discretionary accruals. Frankel et al. (2002), Ashbaugh et al. (2003), and Reynolds et al. (2004) find a negative association between negative discretionary accruals and at least some of their NAS measures, while Larcker and Richardson (2004) report a positive association. 10 Also, for a subset of their sample with weak corporate governance, Larcker and Richardson (2004) report a positive (negative) association between positive (negative) discretionary accruals and NAS. In sum, although prior work indicates that absolute discretionary accruals and incomeincreasing accruals are not associated with NAS-based auditor independence measures, we believe there is some indication that income-decreasing accruals were associated in the pre-sox period with NAS. An Alternative Approach: Not all NAS may be Harmful All of the studies discussed above focus on the period prior to the promulgation of the Sarbanes-Oxley Act of 2002, typically the years Further the tests are based on the assumption that it is the overall economic bonding arising from auditors provision of NAS, regardless of the kind of services being offered, that could impair independence. At that time, 10 Larcker and Richardson (2004) s regressions for positive and negative discretionary accruals differ from the other studies. For example, for their negative discretionary accruals regression, Larcker and Richardson (2004) estimate the regression for the full sample, but define negative discretionary accruals as equal to discretionary accruals when they are negative, and zero otherwise. The other studies cited above estimate their negative discretionary accruals regression for the subset of firms with negative discretionary accruals. 7

10 auditors offered a variety of non-audit services, including bookkeeping, information systems design, and tax advice. In providing its rationale for eliminating some kinds of NAS, the SEC argued, not that the overall economic bond would impair auditor independence, but that NAS with certain specific characteristics could impair auditor independence: The Commission's principles of independence with respect to services provided by auditors are largely predicated on three basic principles, violations of which would impair the auditor's independence: (1) an auditor cannot function in the role of management, (2) an auditor cannot audit his or her own work, and (3) an auditor cannot serve in an advocacy role for his or her client (SEC 2003). Following these principles, SEC (2003) prohibits the provision of bookkeeping services because they place the audit firm in the position of auditing its own work. Similarly information system services are prohibited as they put the auditors into a management role; other services that give the auditor management advisory roles are prohibited as well. Figure 1 shows the NAS fee composition in the pre-sox and post-sox periods. Eliminating the years as transitional years, we define the periods and as the pre-sox and post-sox periods respectively. In the pre-sox period, NAS fees are composed of Information systems (IS) fees and other fees. The latter include non-harmful services such as audit-related NAS and harmful services described above such as bookkeeping. Because the breakup of other fees is not available, we cannot determine the harmful and nonharmful portions of NAS fees as a whole. In the post-sox period, the alleged harmful portions of NAS fees were proscribed and the non-harmful portion was required to be reported in three categories: audit-related fees, tax service fees and a new other category. The elimination of bad NAS from the pre-sox to post-sox periods comes from IS fees and an unobserved portion of other fees. Thus we propose that, because the reduction was forced, the change in NAS fees from the pre- to post-sox periods is driven largely (although not completely) by the 8

11 elimination of the bad NAS, at least as defined by the SEC. Therefore we use the change in NAS fees as a proxy for the portion of NAS in the pre-sox period that could have impaired auditor independence. By contrast, previous studies use total pre-sox NAS fees, which comingles harmful and non-harmful NAS fees. 11 We expect that firms that had larger declines in NAS between and , were firms with more harmful NAS in the pre-sox period than firms that had no change in NAS or had increases in NAS. Then, if the forms of NAS identified by the SEC impaired auditor independence, firms with greater subsequent reduction in NAS from their pre-sox level would show greater earnings management in the pre-sox period. We test the following hypothesis: H1: Other things equal, earnings management in the pre-sox period is positively associated with the reduction in NAS fees from the pre- SOX to post-sox periods. Next we examine how the elimination of the alleged bad NAS could have affected earnings management in the post-sox period. Again, if the eliminated NAS services in fact represented those that can impair auditor independence, we would expect that firms whose NAS declined would show relatively lower earnings management in the post-sox period. We therefore test the following hypothesis: H2: The positive association between earnings management and the reduction in NAS fees is reduced in the post-sox period compared with the pre-sox period. 3. Research Design and Model Discretionary Accrual Measures Prior research documents that discretionary accrual estimates are correlated with firm performance (Dechow et al. 1995; Kasznik 1999; Kothari et al. 2005). Following prior work, we 11 However, the SEC regards tax services as unique and argues that auditors can provide tax services to their audit clients without impairing independence. Tax services are sometimes perceived as even providing benefits like knowledge spillover (e.g. the Panel on Audit Effectiveness 2000; Kinney et al. 2004). 9

12 employ two measures of performance-adjusted discretionary accruals (Ashbaugh et al. 2003). In our first measure, Portfolio Performance-adjusted Discretionary Accruals (PPDA), we control for firm performance through a portfolio technique. For our second measure, Performance Adjusted Discretionary Accruals (PADA), we adjust for firm performance by including a firm performance measure in the regression model used to estimate discretionary accruals. To construct PPDA, we first estimate the parameters of the following regression for each year and industry (using two-digit SIC codes), using all available firm-year observations on Compustat: TA it = 0 + β1 1/ ASSETSit 1) + β 2 ( ΔSales ΔAR) it + β 3 β ( PPE + ε (1) it it where SALES it is change in sales scaled by lagged total assets (ASSETS it-1 ), AR it is the change in accounts receivable scaled by lagged total assets, and PPE it is net property, plant and equipment scaled by lagged total assets. Following Kothari et al. (2005) we include a constant term in the estimation model. 12 A firms discretionary accrual (DA) is the estimated residual of equation (1). Next, we partition firms within each two-digit SIC code into deciles based on their prior year s ROA. PPDA is the difference between a sample firm s DA and the median DA for each ROA decile. 13. To construct our second measure, PADA, we follow Kothari et al. (2005) by including lagged ROA in the accrual regression to control for firm performance. We estimate the parameters of the following regression cross-sectionally for each year-industry combination 12 Kothari et al. (2005) provide three reasons for this. First, it provides an additional control for heteroskedasticity. Second, it mitigates problems arising from the omission of the size variable. Third, it makes the estimated discretionary accruals measures more symmetric increasing the power of tests. Following Kothari et al. (2005), we also exclude observations if the absolute value of total accruals scaled by total assets is greater than one, and winsorize the other variables in the accrual model at 1 and 99 percent. 13 The PPDA estimate of discretionary accruals controls for relative firm performance within industries. It does not impose a linear and constant relation between performance and accruals within a same industry. 10

13 (again based on two-digit SIC industry) using all firm-year observations available on the Compustat database: TA it = 0 + α1 1/ ASSETSit 1) + α 2 ( ΔSales ΔAR) it + α 3PPEit + α 4ROAit 1 α ( + ε (2) it ROA it-1 is the return on assets in the previous year. PADA is the estimated residual of equation (2). The PADA estimate of discretionary accruals controls for performance on a firm-specific basis and allows for the association between firm performance and accruals to differ across industries. Following most prior work in this area, we use the absolute magnitude of discretionary accruals to proxy for earnings management because we investigate the general quality of earnings without specific directional prediction. However, research on auditor litigation shows that lawsuits against auditors are related most often to overstatements of earnings or net assets (Kellogg 1984; St. Pierre and Anderson 1984; Lys and Watts 1994) or situations of significant income-increasing abnormal accruals (Heninger 2001). This suggests that auditors have asymmetric loss functions depending on the direction of earnings management, due to which their willingness to compromise their independence may differ for positive and negative earnings management efforts by management. Thus, the effects of harmful NAS on auditor independence may obtain only in situations where firms have incentives to manipulate earnings downward. To test this conjecture, we separately test income-increasing and income-decreasing discretionary accruals. Model We use the following OLS regression model to test the effect of nonaudit fee change and its interactive effect with SOX on earnings management. AB_DA = β 0 +β 1 NASDROP it +β 2 SOX it + β 3 NASDROP*SOX it + β m CONTROLS it + β n INDUSTRY it + ε it (3) 11

14 The dependent variable, AB_DA, is the absolute value of various discretionary accruals, measured as either the absolute value of PPDA (ABS_PPDA), the absolute value of PADA (ABS_PADA), the absolute value of positive PPDA (+PPDA), the absolute value of positive PADA (+PADA), the absolute value of negative PPDA (-PPDA), or the absolute value of negative PADA (-PADA). We use two alternative measures for NASDROP to capture the extent of decline in nonaudit fee from the pre- to post-sox period. NASDROP_DUM is a dummy variable coded 1 if nonaudit fee declines from the pre- to post-sox. NASDROP_PT is the percentile rank of the change in nonaudit fee (ranging from 0 to 1) from the pre- to post-sox (pre-sox nonaudit fee minus post-sox nonaudit fee) periods, scaled by pre-sox total assets. 14 The higher the percentile rank, the greater the drop of nonaudit fee from the pre-sox to post-sox period. SOX is an indicator variable that equals 1 for fiscal years 2004 and 2005, and 0 for fiscal years 2000 and SOX (2002) introduced a number of corporate governance changes in addition to changes regarding auditor services, which could have had a cautionary effect on manager reporting that also caused decline in earnings management. Thus it is important to control for its direct effect on earnings management. 15 We are interested in the coefficients β 1 and β 3 in equation (3). β 1 measures the association 14 We take the mean of nonaudit fees for years 2000 and 2001 as the pre-sox nonaudit fee; and the mean of nonaudit fees of years 2004 and 2005 as the post-sox nonaudit fee when data are available. We use fee for a single year if nonaudit fee for either 2000 or 2001 is missing for the pre-sox period or either 2004 or 2005 is missing for the post-sox period. We take the mean of total assets of year 2000 and 2001 as the pre-sox total assets and use total assets of a single year if data for either 2000 or 2001 is missing. As a robustness check, we also use alternative deflators to generate NASDROP_PT: (1) the mean of square root of total assets of year 2000 and 2001 (2) the mean of NAS fees of year 2000 and The results are qualitatively similar to those reported in the tables. Using change in NAS fees (undeflated) also yields similar results. 15 Variance inflation factors (VIFs) for NASDROP, SOX and NASDROP*SOX in equation (3) do not exceed 5.5, suggesting multicollinearity is not a serious concern. 12

15 between NASDROP and discretionary accruals (unsigned, positive or negative depending on the dependent variable) in the pre-sox period. If firms with greater decline in nonaudit fees were associated with greater earnings management in the pre-sox period, we would expect β 1 to be positive. The coefficient of the interaction NASDROP*SOX (β 3 ) captures the incremental effect of NASDROP on discretionary accruals in the post-sox period. If firms with greater declines in nonaudit fees were associated with greater improvement in earnings quality after SOX, β3 is expected to be negative. We include several control variables found in prior work to be associated with discretionary accruals (e.g. Ashbaugh et al. 2003; Hribar and Nichols 2007). BIG5 is an indicator variable that equals 1 when the client s auditor is a large accounting firm; TENURE measures the number of years the client is audited by the incumbent auditor, capped at nine. LOGTA is the logarithm transformation of total assets; LEV is the ratio of total debt to total assets; CFO is the cash flow from operations scaled by total assets; SGR is the sales growth measured as revenue at year t divided by revenue at year t-1; M/B is market-to-book ratio; LOSS is an indicator variable that equals 1 when the firm reports a net loss; LITIGATION is an indicator variable that equals 1 if the firm operates in a high-litigation industry (SIC codes of , , , , and ); MERGER is an indicator variable that equals 1 if the firm in engaged in a merger or acquisition in the fiscal year; FINANCING is an indicator variable that equals 1 if the firm obtains capital through either equity or debt market in the fiscal year. We control for the volatility of the operating environment of the firm using cash flow and sales volatility (SDCFO and SDREV). 16 In addition, we include fixed industry effects because although the level of discretionary accruals is estimated by industry, there may be significant 16 Hribar and Nichols (2007) find that absolute discretionary accruals are positively related to firms fundamental operating volatility, and argue that this could lead to bias in earnings management tests when the earnings management partitioning variable is correlated with operating volatility. 13

16 inter-industry variation in the absolute value. Sample Selection 4. Sample Selection We started with data on auditor fees from the Audit Analytics database for 20,388 firmyears for the fiscal years 2000, 2001, 2004 and We deleted 1,725 observations from the financial industry and 6,825 observations that lacked the necessary financial data on Compustat to calculate the variables in the model. To facilitate consistent comparisons between the pre-sox and post-sox periods we eliminated 233 observations for firms with data only in the pre-sox period or only in the post-sox period. This process yields a final sample of 3,328 unique firms and 11,605 observations. NAS Fee Distribution Table 1, Panel A presents descriptive statistics for NAS fees. About 69% of our sample firms show a decline in NAS fees from the pre-sox to post-sox period. The mean (median) decline in NAS fees for the firms that had a decline was $1,293 ($184) thousand. By contrast the mean (median) increase in NAS for firms that saw an increase was significantly lower, $223 ($67) thousand. In Table 1 Panel B, we examine whether our sample firms would be ranked differently based on their pre-sox fees and the change in their fees from the pre-sox period to the post- SOX period. If firms reduced NAS in general (regardless of the kind of service), we would expect a strong positive correlation between the level of pre-sox fees and the extent of change in NAS fees. If on the other hand, the reduction in NAS was predominantly due to elimination 17 Audit Analytics provides two databases on auditor fees, one containing original numbers and the second containing restated numbers. Because SEC (2003) changed the rules concerning the definition and categorization of auditor fees, we used the restated database. 14

17 of the kinds of NAS banned by the SEC, we would expect a weaker correlation because firms would have different amounts of the banned services. Thus we present correlations between our two measures of decline in NAS, the dummy variable, NASDROP_DUM, and the percentile ranking of the change in NAS fees from pre-sox to post-sox period, scaled by assets (NASDROP_PT) and percentile ranking of the pre-sox fees, and the percentile ranking of pre- SOX fees scaled by assets. We find that the correlation between the percentile ranking of pre- SOX NAS fees and NASDROP_PT (NASDROP_DUM) is positive and significant (p<0.001) but of relatively small magnitude, 0.34 (0.37). The percentile ranking of pre-sox NAS fees scaled by assets has a higher correlation with NASDROP_PT (0.54) but this correlation does not suggest a close association between pre-sox NAS fees and the change in fees. Likewise the correlation between the percentile ranking of pre-sox NAS fees scaled by assets and NASDROP_DUM is As we discussed before, the fee disclosures in the pre-sox period consisted of two categories, IS fees and other fees. IS services were one of those subsequently proscribed by the SEC; the bulk of the other types of banned services were included in the other category. We also checked for correlations between pre-sox other fees and our two test variables. The four correlations reported in the bottom two rows of Table 1 Panel B range between 0.21 and 0.5, suggesting that larger other NAS fees in the pre-sox period was associated with larger declines in NAS from the pre-sox to post-sox periods (as reflected in the positive correlations). However,the magnitude of the correlations indicate firms would not be similarly ranked on the basis of their pre-sox NAS fees and the change in NAS fees. Descriptive Statistics 15

18 Table 2 reports the descriptive statistics for the full sample. 18 The first two rows report the absolute value of the two discretionary accruals measures (ABS_PPDA and ABS_PADA). Despite the different procedures used to control for firm performance, the two measures have similar means, medians and standard deviations. The mean for SOX is 0.54, indicating that we have slightly more firm-year observations in the post-sox period. 19 In our sample, about 83% of firm-years are audited by Big 5 auditors. The mean and median value for auditor tenure is 5.6 and 6 years respectively. Total asset is about $287 (e ) million on average. Table 3 reports the Pearson correlation among the variables used in the study. The two discretionary accruals measures have a correlation of Both absolute discretionary accruals have positive and significant correlations with NASDROP_PT, the percentile rank of NASROP, but have no significant correlation with NASDROP_DUM. SOX is negatively associated with the absolute value of discretionary accruals, suggesting there was a decline of earnings management in the post-sox period, without considering other controls. Correlations between the control variables and the two absolute discretionary accruals are generally consistent with prior literature. For example, Big 5 auditor and longer auditor tenure are associated with lower earnings management; higher leverage, higher growth, loss firms and more financing are associated with greater earnings management. Absolute Discretionary Accruals In Table 4, we present the results of estimating equation (3) with the unsigned value of 18 We winsorize all continuous variables at the one and ninety-nine percent levels, except for ABS_PPDA, ABS_PADA, SDREV, and SDCFO, which are winsorized at the 99 percent only. 19 We also construct a subsample of firms that have observations available for each of the pre-sox and post-sox years (a balanced sample), and find similar results (see additional tests). 20 Ashbaugh et al. (2003) also generated performance-matched discretionary accruals by portfolio matching or by controlling prior year s ROA in the estimation of expected accruals. However, they examine discretionary current accruals, while we examine discretionary total accruals. Nevertheless, they too report high correlations between the discretionary accruals measures generated under the two procedures- Pearson correlation of 0.91 and Spearman correlation of

19 discretionary accruals as the dependent variable. Panel A reports results for our full sample, and Panel B reports results for firms for which NAS declined from the pre-sox period to the post- SOX period. In Panel A, columns (1) and (4), we present base-line results of regressing ABS_PPDA and ABS_PADA on SOX and the control variables only. We find consistent with Cohen et al. (2008), that SOX is negatively associated with absolute discretionary accruals, suggesting a general decline in earnings management in the post-sox period. In Column (2) and (5), we present the models including NASDROP_DUM, and its interaction with SOX. We find a positive relation between both absolute discretionary accruals measures (ABS_PADA and ABS_PPDA) and NASDROP_DUM, suggesting earnings management is higher in pre-sox period for firms that have subsequent declines in nonaudit fee compared with firms that have subsequent nonaudit fee increases or no change. Further, we find a negative and significant sign on NASDROP_DUM*SOX for both ABS_PPDA and ABS_PADA, suggesting that compared with firms that have zero or positive subsequent change in nonaudit fee, firms that have subsequent decline in nonaudit fee have incrementally lower earnings management in the post-sox period. When we replace NASDROP_DUM with NASDROP_PT (Column (3) and (6)), we find similar results. NASDROP_PT has a significant and positive sign, confirming that earnings management in the pre-sox period is greater for larger subsequent declines in nonaudit fee. Further, we find a negative and significant sign on NASDROP_PT*SOX. Once again, this suggests that firms that have larger subsequent nonaudit fee declines have greater declines in earnings management compared with other firms in the post-sox period. The results for the control variables are in general consistent with prior work, and with those reported in the univariate correlations, except that M/B is no longer significant. 17

20 In Table 4, Panel B, we present a stronger test of our hypotheses by estimating the models for firms with decline in NAS (i.e., NASDROP_DUM=1) from the pre-sox to post-sox periods. For brevity we report only the coefficients for the test variables. The results are similar to those in Panel A. The interaction NASDROP_PT*SOX has a significant negative coefficient in both columns 1 and 2. Income-increasing and Income-decreasing Discretionary Accruals Table 5 reports the regression results estimated conditional on whether discretionary accruals are income-increasing or income-decreasing. Columns (1) through (4) report results estimated using observations with PPDA or PADA greater than zero. We find no association between either of our measures of NASDROP (NASDROP_DUM and NASDROP_PT) with +PADA or +PPDA. Further the interactions of these variables with SOX are also not significant. Overall, positive discretionary accruals in the two periods have no association with declines in NAS. The results for negative discretionary accruals, using observations with PPDA or PADA less than or equal to zero are different (columns (5)-(8)). For ease of interpretation and to facilitate comparison with the positive discretionary accruals results, we use the absolute value of the negative discretionary accruals as the dependent variable. We find a positive relation between NASDROP_DUM and both -PPDA and -PADA. (columns (5) and (7)). Furthermore, the interaction NASDROP_DUM*SOX is negatively associated with both negative PADA and negative PPDA. The results for NASDROP_PT and NASDROP_PT*SOX in columns (6) and (8) are similar. Thus we find that, in the pre-sox period, firms with greater subsequent declines in NAS fees had greater downward earnings management. In the post-sox period, firms with greater declines in NAS fees had larger declines in downward earnings management. 18

21 As in Table 4, Panel B, we also estimated the models in Table 5 for the subset of firms with decline in NAS from the pre-sox to post-sox periods. The results were similar to those reported in Table The different results for positive and negative discretionary accruals suggest that the results in Table 4 for all discretionary accruals are primarily driven by observations with negative discretionary accruals. The finding that NAS was associated with negative earnings management and not positive earnings management in the pre-sox period is consistent with auditors facing an asymmetric loss function, arising from litigation exposure and concerns about reputational loss, making them more likely to acquiesce to negative earnings management than to positive earnings management by managers. Further, we find that with the decrease of the nonaudit services, the greater income-decreasing earnings management by high NAS firms was eliminated in the post-sox period. 22 Alternative Explanations 5. Alternative Explanations and Additional Analyses As discussed above, our full sample results are primarily driven by firm-years with negative discretionary accruals. Firms with more questionable NAS (as measured by its subsequent decline) were associated with greater downward earnings management than other firms. Although this is consistent with impaired auditor independence, we examine alternative 21 We also ran our models for a balanced sample with firms that have data for all four years, to alleviate the concern that the results may be influenced by differences in sample composition across the years. For the absolute discretionary accruals model (N=7,112), NASDROP_DUM and NASDROP_PT and their interactions with SOX are significant in the expected directions in all cases, except that the interaction between NASDROP_DUM and SOX becomes insignificant (p=0.140) when ABS_PADA is used as dependent variable. For the negative discretionary accrual sample, our results with respect to our NASDROP variables are similar for all specifications as reported in Table As a sensitivity test, we control for auditor change to ensure that the effect of decline in nonaudit fee on discretionary accruals is not driven primarily by firms that had auditor changes during the sample period. We include AUDITORCH, coded 1 for all observations of firms with auditor change during the period , and 0 otherwise, and the interaction of AUDITORCH and SOX. AUDITORCH captures all former Arthur Anderson clients as well as firms that voluntarily switched auditors in the sample period. Neither AUDITORCH nor its interaction with SOX is significant. Results for our test variables remain unchanged. 19

22 explanations and contextual factors that could affect the interpretability of this result. Following previous work we further examine the results in Table 5 by partitioning the sample in three ways: (1) firms with and without the incentive to engage in downward manipulation of earnings (2) firms with lower, and higher or equivalent quality of corporate governance in the pre-sox compared with the post-sox period; and (3) clients of Big 5 and non-big 5 auditors. The rationale and results for each partitioning are presented below. Our first partition is motivated by the possibility that the association between negative discretionary accruals and NASDROP is consistent, as Ashbaugh et al. (2003) point out, with the conservative application of generally accepted accounting principles. To examine this possibility, we partition our negative discretionary accruals sample into two subgroups, based on their incentives to engage in downward earnings management incentive. If accounting conservatism is the explanation for our results, then we would expect to see a similar association between the decline in NAS and income-decreasing accruals for both firms with and without downward earnings management incentives. However, if income-decreasing earning management is the explanation for our results, we would expect a more pronounced effect of our NASDROP variables on negative discretionary accruals for firms with downward earnings management incentives than for those without such incentives. Riedl (2004) suggests that managers have incentives to take a big bath when pre-managed earnings are unexpectedly low or smooth reported earnings when pre-managed earnings are unexpectedly high. Following Riedl (2004), we classify firms as having downward manipulation incentives if the change in cash flow from operations from years t-1 to t, scaled by total assets at the end of t-1, is below the median of nonzero negative values or above the median of nonzero positive values Unlike Riedl (2004), which employs pre-write-off earnings as a proxy for unmanaged earnings, we use cash flow from operations because our study does not examine specific techniques (e.g. write-offs) to 20

23 Table 6 Panel A reports the results for our negative discretionary accruals sample for firms with and without the incentive to manage earnings downward. For brevity, we only report the results for our test variables in all Panels in Table 6. We find that for the firms with downward manipulation incentives (columns (1) through (4)), both NASDROP_DUM and NASDROP_PT are positively related with both discretionary accrual measures. The interactions NASDROP_DUM*SOX and NASDROP_PT*SOX are both negatively related with both discretionary accrual measures. By contrast, for the firms without downward earnings manipulation incentives (columns (5) through (8)), the significance of NASDROP_DUM and NASDROP_PT is sensitive to the measure of discretionary accruals and NASDROP_DUM*SOX are not significant when either measure of discretionary accruals is used. These findings suggest that auditors allowed more downward earnings management for clients that purchased more prohibited NAS when clients had more incentives to manipulate earnings downward. 24 Larcker and Richardson (2004) find that the negative relation between level of fees and signed discretionary accruals are driven by firms with weak corporate governance. Consequently, we examine whether our results for negative discretionary accruals hold only for firms with weak corporate governance. The Sarbanes-Oxley Act introduced a number of changes aimed at improving corporate governance. For example, the audit committees of public firms must now be wholly independent, and have at least one member designated as a financial expert. manage earnings. 24 We also conducted an alternative test for conservatism. Following Ball and Shivakumar (2005), we interacted our test variables (SOX, NASDROP and SOX*NASDROP) with CFO (cash flow from operations), DCFO (coded 1 if CFO<0, and 0 otherwise), and CFO*DCFO. The coefficient on CFO*DCFO captures conditional conservatism, reflecting quicker recognition of losses than gains. Consequently the interaction of NASDROP with CFO*DCFO would capture any differential conservatism for companies that underwent a decline in SOX. None of the interactions with our test variables were significant, suggesting conditional conservatism does not confound our results. 21

24 Because our sample comprises both pre- and post-sox observations, we partition our sample on the change in the quality of corporate governance over this period. Following DeFond et al. (2005) and Baber and Liang (2008) we measure corporate governance quality by a summary measure that combines five governance characteristics into a single index (B-index). 25 Firms are classified as having improved corporate governance if their B-index is higher in the post-sox period than in the pre-sox period. Table 6 Panel B reports the results for our negative discretionary accruals sample, partitioned on firms with and without improvement in B-index. We find that the results for the sample with improvement in B-index are similar to those in the full sample in Table 4: both NASDROP_DUM and NASDROP_PT are positive and significant and the coefficients for NASDROP_DUM*SOX and NASDROP_PT*SOX are negative and significant for both discretionary accrual measures (except for NASDROP_PT*SOX in the PPDA sample). However, for firms with no improvement in B-index, none of the coefficients for these variables is significant in the expected direction. These findings suggest, consistent with Larcker and Richardson (2004), that harmful NAS may have impaired auditor independence only in situations of poor corporate governance. Last, Kim et al. (2003) show that Big 5 auditors are less effective in deterring opportunistic downward earnings management than non-big 5 auditors, because concerns about losses from litigation are less likely with income-decreasing accruals management. We partition our negative discretionary accruals sample on auditor type. The results are reported in Table 6 Panel C. We find the associations between the NASDROP variables (and their interactions with SOX) and 25 The B-index is the sum of the following 5 dummy variables: (1) 1, if the board size is less than median of the distribution for all sample firms, zero otherwise (2) 1, if the proportion of independent directors is greater than the sample median, zero otherwise, (3) 1, if the proportion of the number of audit committee members to the total number of directors on board is greater than the sample median, zero otherwise; (4) 1, if all audit committee members are independent directors, zero otherwise; (5) 1 if the CEO is not the board Chair, zero otherwise. 22