Journal of Accounting Research Vol 40 No September 2002 Printed in U.S.A Do Non-Audit Service Fees Impair Auditor Independence? Evidence from Going Concern Audit Opinions M A R K L D e F O N D ,∗ K R A G H U N A N D A N ,† A N D K R S U B R A M A N Y A M∗ Received July 2002; accepted 15 July 2002 ABSTRACT We find no significant association between non-audit service fees and impaired auditor independence, where auditor independence is surrogated by auditors’ propensity to issue going concern audit opinions We also find no association between going concern opinions and either total fees or audit fees In addition, our findings are robust to controlling for unexpected fees, to controlling for endogeneity among our variables, and to several alternative research design specifications Our results are consistent with market-based incentives, such as loss of reputation and litigation costs, dominating the expected benefits from compromising auditor independence Introduction Independent auditing is an essential feature of efficient capital markets and regulators have long been concerned with potential threats to auditor ∗ University of Southern California, Los Angeles; † Texas A&M International University The authors appreciate helpful suggestions from Ray Ball, Julia D’Souza, Ken Gaver, Scott Whisenant, Jerry Zimmerman, and participants at the University of Colorado-Boulder Winter Conference and the 2002 International Symposium on Auditing Research We also thank Liu Zheng for research assistance 1247 Copyright C , University of Chicago on behalf of the Institute of Professional Accounting, 2002 1248 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM independence.1 In the wake of the Enron bankruptcy, concerns about auditor independence have prompted Congress to enact legislation that bans most auditor-provided non-audit services.2 Regulators’ concerns about nonaudit services are based on the assumption that auditors are willing to sacrifice their independence in exchange for retaining clients that pay large non-audit fees A problem with this assumption, however, is that it ignores auditors’ expected costs of compromising their independence In particular, loss of reputation and litigation costs are likely to provide strong incentives for auditors to maintain their independence Therefore, the purpose of this article is to investigate the veracity of regulators’ concerns by empirically examining the association between non-audit (and audit) fees paid to incumbent auditors and auditor independence, where auditor independence is surrogated by the propensity of auditors to issue going concern audit opinions.3 Before the recently enacted legislation, the Securities and Exchange Commission (SEC) attempted to curtail auditors from providing non-audit services by requiring listed companies to publicly disclose non-audit (and audit) fees in proxy statements filed on or after February 5, 2001 (SEC [2000]) Although the SEC originally sought to ban non-audit services, it agreed to the compromise solution of public disclosure in the belief that such disclosures would attract the scrutiny of shareholders Several recent research papers use the newly available auditor fee data to empirically investigate whether non-audit services threaten auditor independence (e.g., Chung and Kallapur [2001], Francis and Ke [2002], Frankel, Johnson, and Nelson [2002], and Reynolds, Deis, and Francis [2002]) This line of research uses factors commonly associated with earnings management, such as discretionary accruals and managers’ propensity to meet earnings targets, as indicators of auditor independence The results from this research, however, yield mixed support for the contention that non-audit services impair auditor independence We attempt to provide additional evidence on this issue by investigating another indicator of auditor independence—the auditor’s willingness to issue a going concern audit opinion The audit report communicates the auditor’s findings to market participants and plays a crucial role in warning financial statement users of impending going concern problems Issuing a going concern opinion, however, means that the auditor must be able to objectively evaluate firm performance and withstand any client pressure to issue a clean opinion This For example, institutional incentives to maintain auditor independence have existed since the times of the English Merchant Guilds, nearly 800 years ago (Watts and Zimmerman [1983]) In April 2002 the U.S House of Representatives passed a bill that prohibits auditors from performing internal auditing and systems work, and in July 2002, the Senate approved a compromise version that that bans most non-audit services except tax work that is first approved by the audit committee The final legislation was signed into law by President Bush in July 2002 Henceforth we use “non-audit services” to mean all non-audit services provided by the incumbent auditor NON - AUDIT SERVICE FEES 1249 suggests that, ceteris paribus, the auditor’s propensity to issue a going concern opinion is positively correlated with the auditor’s level of independence Thus, if regulators’ concerns are justified, non-audit service fees will be inversely related to the probability of auditors’ issuing going concern audit reports We test this hypothesis We also recognize that the cost-benefit trade-off implied by regulators’ concerns about non-audit fees extends to total fees Implicitly, regulators are concerned that auditors find the benefits of retaining clients who purchase non-audit services exceed the costs of sacrificing auditor independence.4 Because the benefits of retaining these clients consist of the higher fees they generate, this implies that higher total fees, regardless of their origin, threaten auditor independence Therefore, if regulators’ concerns are justified, total fees will be inversely related to the probability of auditors’ issuing going concern audit reports We test this hypothesis, also We perform our analysis on 1,158 distressed firms with proxy statements that include audit fee disclosures, including 96 firms receiving first-time going concern audit reports As in prior research, we focus on distressed firms because the going concern problem is a more salient decision among this group We test the hypotheses by including measures of non-audit service fees and total fees in logistic regressions that explain the issuance of going concern opinions We investigate the first hypothesis by examining regressions that include two measures of non-audit fees: the log of non-audit service fees and the ratio of non-audit service fees to total fees (henceforth, the fee ratio) We also include fee ratio because the language in the SEC’s disclosure regulations suggests that the SEC is concerned with the proportion of non-audit fees to total client fees Observing a negative relation between our measures of non-audit service fees and going concern opinions would provide support for our first hypothesis Similarly, we test the second hypothesis by examining regressions that either include the log of total fees or that disaggregate total fees into its two components (the log of audit fees and the log of non-audit fees) We examine a disaggregated measure of total fees for two reasons First, the second hypothesis is predicated on the assumption that both audit and non-audit fees are likely to impair auditor independence, and this assumption can only be tested by looking separately at these components Second, a regression that examines the disaggregated components of total fees essentially tests both of our hypotheses simultaneously That is, we are able to see both the association with total fees and the association with non-audit service fees (after controlling for audit fees) in a single regression.5 To be plausible, these arguments must assume that the auditor receives economic rents from the non-audit services they provide; otherwise, the auditor will be indifferent between keeping or losing clients that pay higher fees Simunic [1984] suggests that such rents may come from “knowledge spillovers” that reduce auditors’ audit-related costs Another advantage of separately analyzing the components of total fees is that the signs of the associations with going concern opinions could be different across the two measures A regression that includes the disaggregated measures would thus be better specified 1250 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM Our results provide no support for either hypothesis That is, we find no evidence of a significant association between the auditor’s propensity to issue a going concern opinion and any of our fee measures This finding is robust to replacing all fee variables by their respective unexpected components, and after controlling for the simultaneity bias induced by endogeneity among non-audit fees, audit fees, and going concern opinions We contribute to the current debate on auditor independence by finding no evidence that non-audit service fees adversely affect the auditor’s opinion-formulation process These findings are consistent with marketbased institutional incentives, such as costly shareholder litigation and loss of reputation, dominating the expected benefits to auditors of compromising their independence Thus, we find no support for regulators’ concerns that non-audit services impair auditor independence We derive our conclusions from a research design that uses the auditor’s propensity to issue a going concern opinion to proxy for auditor independence In contrast to studies that use earnings management surrogates to proxy for auditor independence, we argue that our proxy is more direct and less noisy Specifically, although studies investigating earnings management assume that various earnings characteristics (e.g., discretionary accruals and propensity to meet earnings targets) are evidence of auditor independence, the auditor’s influence on client’s earnings characteristics is likely to be indirect, and there are empirical problems in measuring discretionary accruals (Guay, Kothari, and Watts [1996], Dechow, Sloan, and Sweeney [1995], Hribar and Collins [2002]).6 In contrast, the auditor clearly influences the type of audit opinion, and measuring the audit opinion is unambiguous We acknowledge, however, that a potential limitation of our research design is that our tests may lack the power to reject our null hypotheses For example, it is conceivable that auditors who perform non-audit services are more tolerant of earnings management, but draw the line at compromising the integrity of the audit opinion It is also possible that the costs and benefits to the auditor are stacked in favor of issuing the going concern opinion Thus, because we draw our conclusions based on the lack of finding a statistical association, we cannot rule out lack of power as an alternative explanation for our findings The next section discusses the motivation for our analysis, section describes our tests, section presents our results, and section summarizes our findings Non-Audit Services and Auditor Independence 2.1 INCENTIVES FOR AUDITOR INDEPENDENCE Auditor independence is often defined as the probability that the auditor will report a discovered breach in the financial reports (Watts and We note that this limitation applies to other studies that also link discretionary accruals to auditor behavior, such as Becker, DeFond, Jiambalvo, and Subramanyam [1998] NON - AUDIT SERVICE FEES 1251 Zimmerman [1983, 1986]).This suggests that auditor independence is synonymous with auditor objectivity and the ability to withstand client pressure to acquiesce to substandard reporting Jensen and Meckling [1976] conclude that managers have incentives to reduce agency costs by hiring independent auditors Supporting this conclusion, Watts and Zimmerman [1983] find evidence that 84% of New York Stock Exchange (NYSE) companies voluntarily engaged independent auditors in 1926, several years before the Securities Acts that mandated external auditing Thus, there is both theoretical and empirical evidence that managers find it in their best interests to engage independent auditors A large body of theoretical and empirical research also suggests that auditors have market-based institutional incentives to act independently For example, Benston [1975] conjectures that reputation concerns are likely to create incentives for independence, and Watts and Zimmerman [1983] document several historical examples of auditors taking costly actions to protect their reputation capital.7 More recently, reputation concerns are consistent with Arthur Andersen’s client losses in the months following the Enron collapse, as discussed in the following BusinessWeek quote (Weber, Little, Henry, and Lavelle [2001], p 32): The Enron meltdown gives present and prospective clients an excuse to flee They may want to avoid the heightened attention an Andersen audit might get in shareholder litigation or fear their financial reports could draw more scrutiny from regulators if they’re handled by Andersen The threat of class action lawsuits provides another incentive for auditor independence, particularly in U.S capital markets, where Big auditors incurred more than $1 billion in litigation-related costs in 1993 alone (Antle, Griffen, Teece, and Williamson [1997]).8 Palmrose [1988], drawing on theory from DeAngelo [1981], presents evidence consistent with Big auditors’ reducing litigation exposure by increasing their independence This evidence is also consistent with Francis, Maydew, and Sparks [1996] and Becker, DeFond, Jiambalvo, and Subramanyam [1998], who find that Big auditors appear to constrain managers’ ability to exercise accounting discretion Shu [2000] finds that auditors resign from clients (and hence forego fee revenues) in response to both increases in litigation risk and emerging mismatches with the clients In summary, a large body of research finds that outside auditors have market-based institutional incentives—particularly related to reputation and litigation costs—to remain independent of their publicly held clients 2.2 NON-AUDIT SERVICES AND AUDITOR INDEPENDENCE Although there are market-based incentives for auditors to remain independent, there are also forces that potentially threaten auditor Also, Antle [1984] indicates that reputation is a likely enforcement mechanism for auditor independence For expositional convenience, we use “Big 5” to refer to Big 5, Big 6, and Big auditors 1252 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM independence Specifically, regulators are concerned about two effects of non-audit services One is a fear that non-audit service fees make auditors financially dependent on their clients, and hence less willing to stand up to management pressure for fear of losing their business.9 The other is that the consulting nature of many non-audit services puts auditors in managerial roles, potentially threatening their objectivity about the transactions they audit These concerns are summarized in the following quote from the SEC regulations mandating fee disclosures (SEC [2000]): An auditor’s interest in establishing or preserving a non-audit services relationship raises two types of independence concerns First, the more the auditor has at stake in its dealings with the audit client, the greater the cost to the auditor should he or she displease the client, particularly when the non-audit services relationship has the potential to generate significant revenues on top of the audit relationship Second, certain types of non-audit services, when provided by the auditor, create inherent conflicts that are incompatible with objectivity.10 Regulators’ concerns that auditors become financially dependent on their clients are based on an intuitive cost-benefit trade-off Regulators fear that auditors will perceive that the benefits from retaining clients that pay large non-audit service fees outweigh the expected costs of sacrificing their independence As discussed earlier, the expected costs of sacrificed independence include the reputation loss and litigation costs associated with audit failures Although not explicitly stated, this argument is based on the assumption that non-audit service fees include economic rents Otherwise, auditors will be indifferent between keeping and losing their non-audit service clients Simunic [1984] argues that auditor-provided management advisory services can generate economic rents because of “knowledge spillovers.” Knowledge spillovers refer to information generated while performing management consulting services that can produce economic rents by reducing auditing costs Simunic [1984] investigates whether spillovers exist by examining audit fees Although he finds evidence that audit fees are higher in the presence of non-audit services—consistent with the existence of knowledge spillovers— a later study by Palmrose [1986] provides evidence that audit fees are higher even when clients engage consultants who are not incumbent auditors Even if the existence of spillovers is established, however, it is impossible to directly quantify the costs and benefits in the previously described trade-off However, Watts and Zimmerman [1983] report that requiring auditors to be financially independent of their clients is, in historical terms, relatively new As recently as 1844 the U.K Companies Act actually required auditors to be shareholders Thus, financial dependence, per se, does not necessarily threaten auditor independence 10 The timing of the fee disclosure regulations is partially a response to a recent increase in the amount of non-audit services provided by the Big Levitt [2000] asserts that consulting services of the Big now represent more than 50% of their revenues, up from just 12% in 1977 NON - AUDIT SERVICE FEES 1253 Therefore, it is ultimately an empirical question whether auditors compromise their independence to retain non-audit service clients One source of such empirical evidence is litigation against auditors Supporting the contention that auditor-supplied consulting services not result in substandard reporting, Palmrose [1999] finds no instances of lawsuits against auditors that allege non-audit services impair independence Similarly, of 610 cases of litigation against auditors examined in Antle, Griffen, Teece, and Williamson [1997], only 24 mention that the auditors also provide non-audit services, and only of those 24 allege that the non-audit services impaired auditor independence Thus, the evidence from class action lawsuits suggests that non-audit services are not an important source of litigation against auditors In addition to the evidence from class action lawsuits, several recent studies examine the association between non-audit service fees and evidence of earnings management (e.g., Chung and Kallapur [2001], Francis and Ke [2002], Frankel, Johnson, and Nelson [2002], and Reynolds, Deis, and Francis [2002]) Specifically, this line of research investigates whether companies that report higher levels of non-audit service fees are more likely to report larger discretionary accruals and meet analysts’ earnings forecasts The results from these investigations, however, are ambiguous For example, although Frankel, Johnson, and Nelson [2002] find a positive association between non-audit service fees and the magnitude of discretionary accruals, Chung and Kallapur [2001] not find this association, and Francis and Ke [2002] find that this relation is weakly significant, but only among non–Big auditors Similarly, although Frankel, Johnson, and Nelson find a positive association between non-audit service fees and managements’ propensity to meet analysts’ earnings forecasts, Reynolds, Deis, and Francis [2002] fail to find such a relation.11 Thus, the evidence on whether nonaudit services is associated with increased levels of earnings management is mixed Contrary to the concerns that fee dependency impairs auditor independence, Reynolds and Francis [2000] find evidence consistent with auditors increasing their independence in response to greater financial dependence Specifically, they find that relatively larger audit clients—those for which the auditor is expected to have the greatest financial dependence— tend to report significantly lower discretionary accruals when compared with smaller clients The authors conjecture that this is because the reputation and litigation damages associated with audit failure are likely to be greater for larger clients, providing incentives for auditors to be more conservative In addition, they also find no evidence that auditors are more lenient in issuing going concern reports to larger clients Thus, Reynolds 11 Using a sample of U.K companies, Gore, Pope, and Singh [2001] find a positive association between non-audit fees and earnings management for non-Big auditors’ clients but not for Big auditors’ clients 1254 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM and Francis find no evidence that financial dependency impairs auditor independence.12 In summary, theory and evidence suggests that although auditors have market-based incentives to remain independent, auditor independence may be threatened when auditors provide non-audit services to their clients The evidence on whether non-audit services actually impair independence, however, is inconclusive In the next section, we argue that examining the relation between non-audit service fees and the auditor’s propensity to issue a going concern opinion is likely to provide additional evidence on this issue 2.3 NON-AUDIT SERVICES AND GOING CONCERN AUDIT OPINIONS The auditor’s report plays a critical role in warning market participants of impending going concern problems Indeed, the term audit failure typically refers to cases in which auditors fail to issue going concern opinions to clients that subsequently file for bankruptcy (Blacconiere and DeFond [1997], Weil [2001]) Statement of Auditing Standard (SAS) No 59 (AICPA [1988]) requires auditors to issue going concern modified audit opinions when substantial doubt exists regarding the client’s ability to continue as a going concern for one year beyond the financial statement date.13 Because auditor independence is defined as the probability that the auditor will report a discovered breach in the financial reports (Watts and Zimmerman [1983]), auditors with impaired independence are less likely to issue going concern opinions when such opinions are warranted Thus, we test the following hypothesis (in the alternative form): H1: Ceteris paribus, non-audit service fees are inversely related to auditors’ propensity to issue going concern audit opinions Regulators are concerned that large non-audit service fees create incentives for auditors to reduce their independence However, auditors also receive audit fees, and DeAngelo [1981] argues that the bilateral monopoly created by nonzero auditor switching costs results in auditors’ receiving economic rents from providing audit services Thus, a logical extension of regulators’ concerns about high non-audit service fees is that high total fees potentially threaten auditor independence Therefore, we also test the following hypothesis (in the alternative form): H2: Ceteris paribus, total fees are inversely related to auditors’ propensity to issue going concern audit opinions 12 Further evidence that auditors tend to be conservative in response to client characteristics is provided by Francis and Krishnan [1999], who find that that auditors are more likely to issue going concern opinions for clients reporting larger total accruals This finding is consistent with auditor conservatism because there is more management judgment, and hence a greater chance of financial statement error, in the presence of larger accruals 13 SAS No 59 (AICPA [1988]) became effective for periods beginning after January 1, 1989, and provides specific guidance regarding the issuance of a going concern opinion NON - AUDIT SERVICE FEES 1255 Research Design 3.1 SAMPLE We obtain our sample by first identifying all available proxy statements filed with the SEC between February 5, 2001, and June 30, 2001, using the search phrase “audit fees” in the Edgar Online database To increase our sample of going concern observations, we extend this date to October 31, 2001, for firms listed in the Compact Disclosure-SEC database as receiving first-time going concern opinions.14 This process results in a sample of 4,105 firms with fee information, including 160 with going concern opinions for fiscal-year 2000 financial statements but not for prior-year financial statements.15 We then require the sample firms to have the necessary financial statement variables on the Compustat (industrial, full coverage, and research) databases, stock return variables on the Center for Research in Security Prices (CRSP) database, mergers and new issues variables on the SDC database, and institutional ownership variables on the Wharton WRDS 14F database After deleting financial institutions and all firms that change year-end, these procedures yield a preliminary sample of 2,428 firms, including 100 with first-time going concern opinions As in prior research, we limit our analysis to a sample of financially distressed firms in evaluating the auditor’s probability of issuing a first-time going concern opinion (Hopwood, McKeown, and Mutchler [1994], Mutchler, Hopwood, and McKeown [1997], Reynolds and Francis [2000]) This is because the going concern opinion decision is most salient among distressed firms As in Reynolds and Francis [2000], we define financially distressed firms as firms that report either negative earnings or operating cash flows during the current fiscal year After restricting the analysis to distressed firms (as defined earlier), we have a usable sample of 1,158 firms, including 96 with first-time going concern opinions 3.2 GOING CONCERN MODEL We test our hypotheses by estimating the coefficients in the following logistic regression that models the auditor’s probability of issuing a firsttime going concern opinion to a financially distressed client: OPINION = β0 + β1 (PROBANKZ) + β2 (log(ASSETS)) + β3 (log(AGE )) + β4 (BETA) + β5 (RETURN) + β6 (VOLATILITY ) + β7 (LEV ) + β8 (CLEV) + β9 (LLOSS) + β10 (INVESTMENTS ) + β11 (FUTURE FINANCE) + β12 (BIG ) + β13 (OP CASH FLOW ) + β14 (REPORT LAG ) 14 15 In general, distressed firms tend to file their proxies later in the year Some of our sample firms have fiscal year-ends occurring in early 2001 1256 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM + β15 (FEERATIO) + β16 (log(TOTAL FEE )) + β17 (log(AUDIT FEE)) + β18 (log(NON-AUDIT FEE )) + ε (1) where: OPINION = an indicator variable equal to for firms with going concern audit opinions, and otherwise PROBANKZ = probability of bankruptcy score (Zmijewski [1984]) log(ASSETS) = natural logarithm of total assets at the end of the year measured in millions of dollars log(AGE) = natural logarithm of the number of years since the company was listed on a stock exchange BETA = the firm’s beta estimated using a market model over the fiscal year RETURN = the firm’s stock return over the fiscal year VOLATILITY = the variance of the residual from the market model over the fiscal year LEV = total liabilities over total assets at the end of the fiscal year CLEV = change in LEV during the year LLOSS = an indicator variable equal to when the firm reports a bottom-line loss for the previous year, and otherwise INVESTMENTS = short- and long-term investment securities (including cash and cash equivalents) deflated by total assets at year-end FUTURE FINANCE = an indicator variable equal to when the firm issues equity or debt in the subsequent year (through October 31, 2001) BIG = an indicator variable equal to when the auditor is a member of the Big 5, and otherwise OP CASH FLOW = operating cash flows divided by total assets at fiscal year end REPORT LAG = number of days between fiscal year-end and earnings announcement date FEERATIO = the ratio of non-audit fees to total fees paid to the incumbent auditor log(TOTAL FEE) = the natural logarithm of the sum of audit and non-audit fees paid to the incumbent auditor log(AUDIT FEE) = the natural logarithm of the audit fee paid to the incumbent auditor log(NON-AUDIT FEE) = the natural logarithm of the sum total of all nonaudit fees paid to the incumbent auditor 1260 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM Row in table shows that 8% of our sample receives going concern opinions, which is comparable to Reynolds and Francis [2000], who find that 9% of their financially distressed sample receives going concern opinions Row indicates that the mean and median bankruptcy probability scores in our sample are 22 and 03, respectively Row indicates that mean and median firm sizes, measured in total assets, are $813 million and $110 million, indicating a skewed distribution and justifying our decision to log assets in our logit analysis Row indicates that our sample firms have been listed for to 38 years Row reports that the mean and median values for BETA are close to 1.0 Row 10 indicates that the mean and median stock returns during the prior year are –35% and –53%, respectively, indicating that our sample firms recently experienced significant loss of market value Row 11 reports that VOLATILITY is comparable to that reported in earlier research for distressed firms Rows 12 and 13 indicate that mean and median values of leverage are 0.48 and 0.44, respectively, and that the median change in leverage is relatively small It is not surprising that row 14 reports that prior-period losses are relatively frequent among our distressed sample Row 15 reports that mean and median values for INVESTMENTS are 31% and 22%, respectively, and row 16 indicates that an average of 8% of our sample firms obtain additional outside financing in the future Row 17 reports that 91% of our sample has a Big auditor Row 18 reports that mean and median operating cash flows divided by total assets (OP CASH FLOW ) are negative, which reflects our sample-selection criteria Finally, REPORT LAG averages 53.5 days, which is larger than average, consistent with the preponderance of losses in our sample (Chambers and Penman [1984]) 4.2 DESCRIPTIVE STATISTICS BY OPINION TYPE Table classifies the variables from Table by opinion type, along with p-values from t-tests and median tests of differences across the two groups Comparing the fee variables in the first four rows represents univariate tests of our hypotheses Row indicates that the mean and median values for FEERATIO are 38% and 40%, respectively, for the going concern sample, compared with 50% and 53% for the clean opinion sample, with the differences significant at p < 1% (two-tailed) Row shows that the mean and median values for TOTAL FEE are $540,000 and $234,000, respectively, for the going concern sample, and $954,000 and $393,000 for the clean opinion sample The difference between the means is significant at p < 10% (two-tailed), and the difference between the medians is significant at p < 1% (two-tailed) Row indicates that the mean and median values of NONAUDIT FEE are $274,000 and $83,000, respectively, for the going concern sample, and $646,000 and $198,000 for the clean opinion sample The difference between the means is significant at p < 5% (two-tailed), and the difference between the medians is significant at p < 1% (two-tailed) Row indicates that both the mean and median values of AUDIT FEE are not significantly different between the two opinion types Thus, univariate tests NON - AUDIT SERVICE FEES 1261 TABLE Comparison of Going Concern and Clean Opinion Samples for 1,158 Financially Distressed Firms (Including 96 Firms with Going Concern Opinions) with Available Auditor Fee Information for Fiscal Year 2000 a Mean Variables FEERATIO TOTAL FEE ($ thousands) NON-AUDIT FEE ($ thousands) AUDIT FEE ($ thousands) OPINION PROBANKZ ASSETS ($ millions) AGE (years listed) BETA 10 RETURN 11 VOLATILITY 12 LEV 13 CLEV 14 LLOSS 15 INVESTMENTS 16 FUTURE FINANCE 17 BIG 18 OP CASH FLOW 19 REPORT LAG Median GC Sample No GC Sample p-value GC Sample No GC Sample p-value 0.38 540 274 266 0.66 178 1.13 −0.68 0.01 0.75 0.20 0.82 0.20 0.06 0.88 −0.37 76 0.50 954 646 305 0.18 870 0.94 −0.32 0.01 0.46 −0.15 0.67 0.32 0.08 0.91 −0.12 51 00 06 04 44 N/A 00 00 47 03 00 00 00 00 00 00 52 24 00 00 0.40 234 83 138 0.84 43 0.99 −0.78 0.01 0.71 0.13 0.13 0.00 −0.23 89 0.53 393 198 164 0.02 118 0.97 −0.51 0.00 0.41 0.02 0.23 0.00 −0.05 46 00 00 00 12 N/A 00 00 31 09 00 00 00 00 00 00 52 23 00 00 a Financial distress is defined as a loss or negative operating cash flow in the current year All p-values are two-tailed See table for variable descriptions are consistent with regulators’ concerns that non-audit service fees impair auditor independence, and they provide some evidence that total fees also threaten independence A problem with drawing conclusions from univariate tests, however, is that they fail to control for the numerous contrary and mitigating factors associated with the auditor’s decision to issue a going concern opinion Thus, we rely on the multivariate tests to formally test our hypotheses It is not surprising that row in table indicates that the mean and median bankruptcy scores are significantly higher among the going concern sample at p < 1% (two-tailed) Row indicates that mean and median total assets are significantly smaller among the going concern sample at p < 1% (two-tailed), and row shows that there is no significant difference in age across the two opinion types Rows 9–11 indicate that mean and median values of BETA are higher among the going concern sample at p < 5% and p < 10% (two-tailed), respectively, and that mean and median values for RETURN are significantly lower and that mean and median values for VOLATILITY are significantly higher, among the going concern firms at p < 1% (two-tailed) This suggests that the going concern firms have lower stock returns and higher volatility than firms with clean opinions Consistent with expectations, rows 12 and 13 find that the going concern firms have higher leverage (LEV ) and a larger 1262 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM increase in leverage (CLEV ), both significant at p < 1% (two-tailed) for both the t-tests and the median tests Row 14 indicates that the mean relative frequency of prior-year losses is significantly higher among the going concern firms at p < 1% (two-tailed) Consistent with expectations, row 15 shows that INVESTMENTS is significantly lower at p < 1% (two-tailed) among the going concern firms, suggesting that firms with less liquidity are more likely to receive going concern opinions Rows 16 and 17 reveal that the likelihood of obtaining future financing (FUTURE FINANCE) and the relative frequency of Big auditors (BIG 5) is not different across the two groups As expected, rows 18 and 19 indicate that mean and median values of OP CASH FLOW are significantly lower and that REPORT LAG is significantly longer among the sample with going concern opinions at p < 1% (two-tailed) In summary, the descriptive statistics presented in tables and are consistent with the distressed nature of our total sample and with the going concern sample being even more distressed In addition, although univariate tests provide some support for our hypotheses, these results are unreliable because they not control for other factors affecting the auditor’s decision to issue a going concern opinion 4.3 MULTIVARIATE TEST RESULTS Table presents the results of estimating the logistic model in equation (1) using alternative fee measures to test our hypotheses Model presents a baseline case of our going concern model without including the fee variables, and models 2–5 sequentially introduce various combinations of our fee variables, with model simultaneously testing both of our hypotheses The Marginal Effect columns in table provide some evidence on the economic significance of each of the coefficients These statistics represent the change in probability of a going concern opinion in response to a onestandard-deviation change in each of the respective independent variables, evaluated at the base-rate probability of the going concern opinion (8%) The results indicate that model does a reasonably good job of explaining the going concern decision The pseudo R2 is 41%, and we find significance in the predicted direction at p < 10% (two-tailed) for the coefficients on PROBANKZ, RETURN, VOLATILITY, LLOSS, INVESTMENTS, BIG 5, and REPORT LAG Consistent with prior research, we not find significance in the predicted direction for the coefficients on BETA (Dopuch, Holthausen, and Leftwich [1987]) and CLEV (Reynolds and Francis [2000]) In addition, we not find significance for the coefficients on log(ASSETS), log(AGE), LEV, FUTURE FINANCE, or OP CASH FLOW The marginal effects suggest that PROBANKZ, RETURN, INVESTMENTS, and REPORT LAG are the most economically significant variables in the model Models 2–5 introduce the following combinations of our fee ratios to the model: FEERATIO alone, log(TOTAL FEE) alone, both FEE RATIO and log(TOTAL FEE), and both log(AUDIT FEE) and log(NON-AUDIT FEE), respectively The results in models 2–5 indicate that the estimated coefficients TABLE Going Concern Opinion Models, with Auditor Fee Variables Included as Independent Variables, for a Sample of 1,158 Financially Distressed Firms (Including 96 Firms with Going Concern Opinions) with Available Auditor Fee Information for Fiscal Year 2000 a Model Predicted Sign PseudoR2 % + − − + − + + + + − − + − + − − − − Model Marginal Effect Coefficient ( p-value) Marginal Effect −6.304 (.00) 1.773 (.00) −0.141 (.24) 0.075 (.64) 0.017 (.93) −1.209 (.00) 84.466 (.02) −0.194 (.71) 0.471 (.24) 0.567 (.09) −2.052 (.00) 0.222 (.68) 0.932 (.03) −0.710 (.18) 0.026 (.00) 4.6% −1.9% 0.6% 0.1% −5.8% 2.4% −0.5% 2.1% 2.0% −4.6% 0.5% 2.1% −1.4% 4.8% −6.212 (.00) 1.791 (.00) −0.113 (.37) 0.051 (.76) 0.020 (.92) −1.215 (.00) 82.947 (.02) −0.212 (.68) 0.460 (.26) 0.553 (.11) −2.033 (.00) 0.213 (.69) 0.962 (.03) −0.745 (.16) 0.025 (.00) −0.438 (.49) 4.6% −1.5% 0.4% 0.1% −5.7% 2.3% −0.5% 2.0% 1.9% −4.5% 0.4% 2.1% −1.5% 4.7% −0.7% Model Model Coefficient ( p-value) Marginal Effect −6.314 (.00) 1.772 (.00) −0.144 (.38) 0.075 (.64) 0.016 (.94) −1.208 (.00) 84.462 (.02) −0.194 (.71) 0.471 (.24) 0.567 (.09) −2.052 (.00) 0.223 (.67) 0.932 (.03) −0.708 (.19) 0.026 (.00) 4.6% −1.8% 0.6% 0.1% −5.7% 2.4% −0.5% 2.1% 2.0% −4.6% 0.5% 2.0% −1.4% 4.8% 0.0% −6.363 (.00) 1.782 (.00) −0.145 (.38) 0.052 (.75) 0.012 (.95) −1.204 (.00) 82.596 (.02) −0.221 (.67) 0.457 (.26) 0.551 (.11) −2.027 (.00) 0.219 (.68) 0.959 (.03) −0.728 (.17) 0.026 (.00) −0.513 (.45) 0.063 (.77) 4.6% −1.9% 0.4% 0.1% −5.7% 2.3% −0.6% 2.0% 1.9% −4.5% 0.5% 2.1% −1.5% 4.7% −0.9% 0.6% 41% 41% Coefficient ( p-value) Marginal Effect −6.701 (.00) 1.808 (.00) −0.135 (.41) 0.042 (.80) −0.007 (.97) −1.188 (.00) 82.770 (.02) −0.257 (.62) 0.453 (.26) 0.554 (.11) −1.988 (.00) 0.198 (.71) 0.981 (.02) −0.742 (.16) 0.026 (.00) 4.7% −1.8% 0.4% −0.1% −5.7% 2.3% −0.5% 2.0% 1.9% −4.5% 0.5% 2.1% −1.5% 4.7% 0.180 (.39) −0.099 (.28) 40% Marginal Effect 0.004 (.98) 41% Coefficient ( p-value) 1.2% −1.4% 40% NON - AUDIT SERVICE FEES Intercept PROBANKZ log(ASSETS) log(AGE) BETA RETURN VOLATILITY LEV CLEV LLOSS INVESTMENTS FUTURE FINANCE BIG OP CASH FLOW REPORT LAG FEERATIO log(TOTAL FEE) log(AUDIT FEE) log(NON-AUDIT FEE) Model Coefficient ( p-value) a 1263 Financial distress is defined as a loss or negative operating cash flow in the current year All p-values are two-tailed, and the marginal effect indicates the effect of a one-standarddeviation change in the respective variable on the probability of a going concern opinion See table for variable definitions 1264 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM on all of our fee variables are insignificant in every case The lack of significance on FEERATIO in models and and on log(NON-AUDIT FEE) in model indicates that we not find support for the first hypothesis The lack of significance on log(TOTAL FEE) in models and and on both log(AUDIT FEE) and log(NON-AUDIT FEE) in model indicates that we also not find support for the second hypothesis Because both hypotheses are tested in model 5, the lack of significance on the fee variables in that regression presents concise evidence that we are unable to support either hypothesis that fees impair auditor independence 4.4 CONTROLLING FOR EXPECTED FEES A potential limitation of the analysis in table is that auditor independence may be influenced by the amount of client fees relative to their expected amounts, rather than the nominal amounts we examine This notion is consistent with auditors’ being influenced by whether the client is a source of unusually high or low fees Therefore, we draw on prior research that models audit and non-audit fees to develop a model that extracts the unexpected portion of fees in our sample firms Specifically, we draw on Craswell, Francis, and Taylor [1995] and Whisenant, Sankaraguruswamy, and Raghunandan [2002] to identify variables explaining audit fees, and Firth [1997], Parkash and Venable [1993], Whisenant, Sankaraguruswamy, and Raghunandan [2002], and Frankel, Johnson, and Nelson [2002] to identify variables explaining non-audit fees and the ratio of non-audit fees to total audit fees Drawing on these sources, and considering additional variables we expect to be important fee determinants, we estimate the following models for audit and non-audit fees, respectively: log(AUDIT FEE) = β0 + β1 (log(ASSETS)) + β2 (BIG ) + β3 (ROA ) + β4 (RETURN) + β5 (VOLATILITY ) + β6 (LEV ) + β7 (INVREC) + β8 (INSTITUTIONAL ) + β9 (SPECIAL ITEMS) + β10 (BOOK TO MKT ) + β11 (SEGS) + β12 (FOROPS ) + β13 (EMPLAN ) + β14 (REPORT LAG) + β15 (INITIAL YEARS ) + ε where: ROA (2) = return on assets defined as operating income divided by total assets at fiscal year end INVREC = inventory plus accounts receivable divided by total assets at fiscal year-end INSTITUTIONAL = the percentage of institutional holdings at fiscal year end SPECIAL ITEMS = indicator variable equal to the absolute value of negative special items divided by total assets, and otherwise at fiscal year end NON - AUDIT SERVICE FEES BOOK TO MKT SEGS FOROPS EMPLAN INITIAL YEARS 1265 = the book-to-market ratio at the fiscal year end = the number of segments disclosed in the segment footnote = an indicator variable equal to if the company has foreign operations, and otherwise = an indicator variable equal to if the company has a pension or post-retirement plan, and otherwise = an indicator variable equal to if it is the initial two years of the audit engagement, and otherwise All other variables are as described in equation (1) log(NON-AUDIT FEE ) = β0 + β1 (log(ASSETS )) + β2 (BIG ) + β3 (ROA ) + β4 (RETURN) + β5 (LEV ) + β6 (INSTITUTIONAL ) + β7 (SPECIAL ITEMS ) + β8 (BOOK TO MKT ) + β9 (SEGS ) + β10 (FOROPS ) + β11 (EMPLAN ) + β12 (INITIAL YEARS ) + β13 (MERGER ) + β14 (FINANCE ) + β15 (SALES GROWTH ) + ε (3) where: log( NON-AUDIT FEE) = as defined in equation (1), and alternately substituted with FEERATIO MERGER = an indicator variable equal to if the client acquired a company during the fiscal year FINANCE = an indicator variable equal to if the firm issued equity or debt during the fiscal year SALES GROWTH = growth rate in sales over the prior year All other variables are as described in equations (1) or (2) Both estimated models also include 11 industry dummy variables (not reported in the model specification or in the tables) The model of audit fees in equation (2) is estimated with log(AUDIT FEE) as the dependent variable The model of non-audit fees in equation (3) is alternately estimated with log(NON-AUDIT FEE) and FEERATIO as the dependent variable Equation (3) is used to explain FEERATIO because FEERATIO is a scaled measure of non-audit fees The independent variables in equations (2) and (3) are combined in a single model and estimated with log(TOTAL FEE) as the dependent variable The coefficients based on these four models are shown in table The results indicate that each of the models has reasonable explanatory power, with adjusted R2 s ranging from 36% to 72%, and most of the coefficients are significant and in the expected direction at conventional levels 1266 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM TABLE Fee Expectation Models for a Sample of 1,158 Financially Distressed Firms (Including 96 Firms with Going Concern Opinions) with Available Auditor Fee Information for Fiscal Year 2000 a Predicted Sign log(ASSETS) BIG5 ROA RETURN VOLATILITY LEV INVREC INSTITUTIONAL SPECIAL ITEMS BOOK TO MKT SEGS FOROPS EMPLAN REPORT LAG INITIAL YEARS MERGER FINANCE SALES GROWTH Adjusted R2 % log (AUDIT FEE) log (NONAUDIT FEE) FEERATIO log (TOTAL FEE) + + ? − + + + + + − + + + + − + + + 0.424 (.00) 0.126 (.04) −0.274 (.00) −0.121 (.00) 7.945 (.14) 0.237 (.00) 0.584 (.00) 0.061 (.48) 0.290 (.09) −0.021 (.08) 0.034 (.00) 0.242 (.00) 0.112 (.02) 0.003 (.00) −0.105 (.03) 0.713 (.00) 0.515 (.00) −0.399 (.00) −0.247 (.00) 0.062 (.00) 0.064 (.00) −0.033 (.08) −0.019 (.08) −0.045 (.74) −0.057 (.00) 0.367 (.06) 0.669 (.09) −0.030 (.27) 0.034 (.03) 0.413 (.00) −0.024 (.83) 0.050 (.09) 0.041 (.49) −0.004 (.33) −0.000 (.86) 0.022 (.28) −0.023 (.18) −0.152 (.18) 0.073 (.41) 0.747 (.00) 0.015 (.36) −0.003 (.86) 0.003 (.82) 0.157 (.00) 0.007 (.00) 0.560 (.00) 0.213 (.00) −0.355 (.00) −0.160 (.00) 7.122 (.25) 0.129 (.06) 0.412 (.00) 0.115 (.24) 0.387 (.05) −0.029 (.04) 0.029 (.00) 0.297 (.00) 0.064 (.24) 0.002 (.04) −0.088 (.11) 0.037 (.40) 0.379 (.00) 0.005 (.53) 66% 56% 36% 72% a Financial distress is defined as a loss or negative operating cash flow in the current year All p-values are two-tailed Variables are defined as follows: ROA = return on assets defined as operating income divided by total assets at the fiscal year-end INVREC = inventory plus accounts receivable divided by total assets at fiscal year-end INSTITUTIONAL = the percentage of institutional holdings SPECIAL ITEMS = equal to the absolute value of negative special items divided by total assets, and otherwise BOOK TO MKT = the book-to-market ratio on the last day of the fiscal year SEGS = the number of segments disclosed in the segment footnote FOROPS = an indicator variable equal to if the company has foreign operations, and otherwise EMPLAN = an indicator variable equal to if the company has a pension or post-retirement plan, and otherwise INITIAL YEARS = an indicator variable equal to if it is the initial two years of the audit engagement, and otherwise MERGER = an indicator variable equal to if the client acquired a company during the fiscal year FINANCE = an indicator variable equal to if the firm issued equity or debt during the fiscal year SALES GROWTH = growth rate in sales over the prior year Other variables are defined as in table Consistent with the approach taken in Frankel, Johnson, and Nelson [2002] to estimate the unexpected FEERATIO, we use the error terms from the models in table to surrogate for the “unexpected” portion of each of our fee variables We then replace the fee variables used in table with the unexpected fee variables and repeat the analysis The resulting regressions are presented in table Model in table is a benchmark analysis identical to model in table 3, and models 2–5 introduce various combinations of our unexpected fee variables similar to table Consistent with the results in table 3, the results in table indicate that the estimated coefficients on all of our fee variables are insignificant at conventional levels in every case Thus, TABLE Going Concern Opinion Models, with Unexpected Fee Variables Included as Independent Variables, for a Sample of 1,158 Financially Distressed Firms (Including 96 Firms with Going Concern Opinions) with Available Auditor Fee Information for Fiscal Year 2000 a Model Predicted Sign PseudoR2 % + − − + − + + + + − − + − + − − − − Model Marginal Effect Coefficient ( p-value) Marginal Effect −6.304 (.00) 1.773 (.00) −0.141 (.24) 0.075 (.64) 0.017 (.93) −1.209 (.00) 84.466 (.02) −0.194 (.71) 0.471 (.24) 0.567 (.09) −2.052 (.00) 0.222 (.68) 0.932 (.03) −0.710 (.18) 0.026 (.00) 4.6% −1.9% 0.6% 0.1% −5.8% 2.4% −0.5% 2.1% 2.0% −4.6% 0.5% 2.1% −1.4% 4.8% −6.254 (.00) 1.771 (.00) −0.142 (.24) 0.059 (.72) 0.010 (.96) −1.204 (.00) 82.577 (.02) −0.173 (.74) 0.470 (.24) 0.554 (.11) −2.052 (.00) 0.213 (.69) 0.933 (.03) −0.728 (.17) 0.026 (.00) −0.485 (.46) 4.5% −1.8% 0.6% 0.0% −5.7% 2.3% −0.4% 2.1% 1.9% −4.6% 0.4% 2.0% −1.5% 4.7% −0.7% Model 41% Marginal Effect −6.314 (.00) 1.761 (.00) −0.140 (.24) 0.072 (.66) 0.019 (.92) −1.209 (.00) 84.106 (.02) −0.183 (.72) 0.469 (.24) 0.569 (.09) −2.049 (.00) 0.217 (.68) 0.928 (.03) −0.718 (.18) 0.026 (.00) 4.5% −1.8% 0.5% 0.1% −5.7% 2.4% −0.5% 2.1% 2.0% −4.6% 0.4% 2.0% −1.5% 4.8% −0.043 (.84) −0.2% Model 41% Coefficient ( p-value) Marginal Effect −6.249 (.00) 1.774 (.00) −0.142 (.24) 0.059 (.72) 0.009 (.96) −1.203 (.00) 82.637 (.02) −0.175 (.73) 0.470 (.24) 0.553 (.11) −2.052 (.00) 0.214 (.69) 0.934 (.03) −0.725 (.17) 0.025 (.00) −0.500 (.47) 0.014 (.95) 4.5% −1.8% 0.5% 0.1% −5.7% 2.3% −0.5% 2.1% 2.0% −4.6% 0.4% 2.0% −1.5% 4.7% −0.7% 0.0% 41% Coefficient ( p-value) Marginal Effect −6.277 (.00) 1.810 (.00) −0.133 (.27) 0.056 (.73) −0.013 (.95) −1.184 (.00) 84.257 (.02) −0.208 (.69) 0.471 (.24) 0.550 (.11) −2.044 (.00) 0.198 (.71) 0.946 (.03) −0.738 (.16) 0.026 (.00) 4.6% −1.7% 0.4% 0.0% −5.6% 2.4% −0.5% 2.1% 1.9% −4.6% 0.4% 2.1% −1.5% 4.8% 0.163 (.47) −0.108 (.25) 41% Coefficient ( p-value) 0.7% −1.0% 41% 1267 a Financial distress is defined as a loss or negative operating cash flow in the current year All p-values are two-tailed, and the marginal effect indicates the effect of a 1-standarddeviation change in the respective variable on the probability of a going concern opinion Variable definitions are as follows: UNEXPECTED FEERATIO = residual from applicable model in table UNEXPECTED log(TOTAL FEE) = residual from applicable model in table UNEXPECTED log(AUDIT FEE) = residual from applicable model in table UNEXPECTED log(NON-AUDIT FEE) = residual from applicable model in table Other variables are defined as in table NON - AUDIT SERVICE FEES Intercept PROBANKZ log(ASSETS) log(AGE) BETA RETURN VOLATILITY LEV CLEV LLOSS INVESTMENTS FUTURE FINANCE BIG OP CASH FLOW REPORT LAG Unexpected FEERATIO Unexpected log(TOTAL FEE) Unexpected log(AUDIT FEE) Unexpected log(NON-AUDIT FEE) Model Coefficient ( p-value) 1268 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM as in table 3, the results in table not support either of the hypotheses that fees impair auditor independence 4.5 CONTROLLING FOR ENDOGENEITY A potential limitation to our findings is that we expect the going concern opinion, audit fees, and non-audit fees to be endogenously determined That is, going concern opinions, audit fees, and non-audit fees are all likely to be associated with financial distress For example, in addition to a higher probability of receiving a going concern opinion, distressed firms are also more likely to command an audit risk premium and potentially involve more audit work, resulting in higher audit fees Similarly, we also expect non-audit fees to be associated with distress, although the sign of the relation is ambiguous For example, distressed firms may wish to decrease discretionary spending on consulting services to conserve cash or, alternatively, spend more on consulting services in an attempt to improve their financial condition Although our opinion and fee models include financial and market variables to control for financial distress, (e.g., PROBANKZ and RETURN ), we can never be sure that we have adequately controlled for financial distress In particular, because going concern opinions are likely to help predict distress beyond our control variables, endogeneity potentially confounds our results.19 The simultaneity bias induced by endogeneity may explain why we not observe a significant association between non-audit service fees and going concern opinions in tables and Specifically, if simultaneity bias induces a positive relation between non-audit fees and going concern opinions, it could swamp the hypothesized negative relation between non-audit fees and going concern opinions In addition, all of our fee variables (except for our measure of total fees) involve audit and non-audit fees together in some combination Whisenant, Sankaraguruswamy, and Raghunandan [2002] show that audit fees and nonaudit fees are endogenously determined Thus, it is difficult to interpret the results of our tests that use the FEERATIO or log(AUDIT FEE) and log(NONAUDIT FEE) together, in the presence of the endogeneity Because joint estimation of a simultaneous system of equations with one or more dichotomous dependent variables is problematic (Judge, Griffiths, Hill, Lutkepohl, and Lee [1985]), we implement a two-stage procedure 19 For example, Mutchler, Hopwood, and McKeown [1997] show that the going concern opinion has incremental explanatory power in explaining subsequent bankruptcy beyond various financial ratios We also test whether going concern opinions are informative in our sample by examining the incremental ability of going concern opinions to predict bankruptcy over the ensuing 12 months We estimate model in table with an indicator variable coded as if the firm filed for bankruptcy during the following year, and otherwise, as the dependent variable and include the going concern indicator variable as an additional explanatory variable The results (not presented) reveal that the going concern indicator variable is significant and positive at p < 1% (two-tailed) Thus, even in our distressed sample, the going concern opinion is incrementally informative in predicting bankruptcy beyond the variables in model in table NON - AUDIT SERVICE FEES 1269 recommended by Nelson and Olsen [1978].20 Accordingly, we implement a system of three structural models: one for each of our three endogenous variables—OPINION, log(AUDIT FEE), and log(NON-AUDIT FEE)—where each is a function of the other two and select exogenous variables We specify the three models as follows: (1) we model OPINION using all independent variables in equation (1) and include log(AUDIT FEE) and log(NON-AUDIT FEE) as endogenous independent variables, (2) we model log(AUDIT FEE) using all independent variables in equation (2) and include OPINION and log(NON-AUDIT FEE) as endogenous independent variables, and (3) we model log(NON-AUDIT FEE) using all independent variables in equation (3) and include OPINION and log(AUDIT FEE) as endogenous independent variables We separately model audit and non-audit service fees because this allows us to simultaneously test both of our hypotheses using the regression equation in model of table In the first stage we model each of our three endogenous variables as a function of all of the exogenous variables in the system (reduced-form models) We use a probit regression to estimate the going concern model and ordinary least squares regressions to estimate the two fee models We use probit instead of logistic regression to estimate the going concern model because the simultaneous system requires normally distributed residuals In the second stage we estimate the structural models after replacing each endogenous explanatory variable with the predicted value from the first stage, which we term GCHAT, AUDHAT, and NASHAT We then employ the “omitted variables” variant of the Hausman [1978] test to check for the presence of endogenity (Beatty, Chamberlain, and Magliolo [1995]) and find that the test is unable to reject the null of no endogeneity for the going concern model, but endogenity is detected in both fee models In particular, the two fee components appear to be endogenously determined Table reports the results of estimating the second-stage structural regressions in accordance with the previously described procedures The first regression in the table presents the results of estimating the going concern model, and the next two regressions present the results of estimating the two fee regressions The results in the going concern model report that the coefficients on AUDHAT and NASHAT are not significant at conventional levels In fact, the coefficient on NASHAT has a positive sign with a p-value of 0.20 Thus, even after controlling for potential simultaneity bias induced by endogeneity, we continue to find no support for either of the hypotheses that fees impair auditor independence.21 20 See D’Souza [1998] and Copley, Doucet, and Gaver [1994] for examples of its implementation in an accounting and auditing context 21 Amemiya [1978] shows that although the coefficients from the Nelson-Olsen secondstage equation are asymptotically unbiased, they are not the most efficient estimators He proposes a method for deriving more efficient estimators Although the Amemiya estimators are asymptotically equivalent to the Nelson-Olsen estimators, the covariance matrix of the Nelson-Olsen estimators is biased Amemiya proposes a correction to the covariance matrix 1270 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM TABLE Structural Models from Two-Stage Procedure After Controlling for Endogeneity, for a Sample of 1,158 Financially Distressed Firms (Including 96 Firms with Going Concern Opinions) with Available Auditor Fee Information for Fiscal Year 2000 a Dependent Variable = OPINION Intercept PROBANKZ log(ASSETS) log(AGE) BETA RETURN VOLATILITY LEV CLEV LLOSS INVESTMENTS FUTURE FINANCE BIG OP CASH FLOW REPORTLAG AUDHAT NASHAT −2.884 (.00) 0.806 (.00) −0.232 (.14) 0.112 (.29) −0.018 (.86) −0.604 (.00) 45.56 (.02) 0.005 (.99) 0.332 (.14) 0.314 (.09) −1.257 (.00) 0.134 (.61) 0.289 (.24) −0.373 (.21) 0.015 (.00) −0.295 (.46) 0.373 (.20) Model Chi-square = 215.86 p-value < 00 Dependent Variable = log(AUDIT FEE) Intercept log(ASSETS) BIG ROA RETURN VOLATILITY LEV INVREC INSTITUTIONAL SPECIAL ITEMS BOOK TO MKT SEGS FOROPS EMPLAN REPLAG INITIAL YEAR GCHAT NASHAT Model F = 128.43 p-value < 00 Adjusted R2 = 65 1.622 (.00) 0.414 (.00) 0.157 (.02) −0.384 (.00) −0.166 (.00) 13.222 (.02) 0.255 (.00) 0.740 (.00) 0.076 (.38) 0.313 (.08) −0.024 (.04) 0.037 (.00) 0.238 (.00) 0.146 (.00) 0.003 (.00) −0.078 (.11) 0.515 (.02) 0.006 (.90) Dependent Variable = log(NON-AUDIT FEE) Intercept log(ASSETS) BIG ROA RETURN LEV INSTITUTIONAL SPECIAL ITEMS BOOK TO MKT SEGS FOROPS EMPLAN INITIAL YEAR MERGER FINANCE SALES GROWTH GCHAT AUDHAT −1.217 (.20) 0.527 (.00) 0.561 (.00) −0.482 (.00) −0.237 (.00) −0.149 (.36) 0.422 (.03) 0.615 (.13) −0.033 (.20) 0.025 (.17) 0.296 (.06) −0.102 (.37) −0.088 (.45) 0.083 (.35) 0.738 (.00) 0.023 (.18) 0.763 (.07) 0.373 (.16) Model F = 83.43 p-value < 00 Adjusted R2 = 55 a Financial distress is defined as a loss or negative operating cash flow in the current year All p-values are two-tailed Variable definitions are as follows: AUDHAT = predicted value of audit fee NASHAT = predicted value of non-audit fee GCHAT = predicted value of going concern opinion Other variables are defined as in tables and 4.6 ROBUSTNESS CHECKS As in prior research we limit our sample to distressed firms (defined as having either negative cash flows or negative earnings) that receive first-time going concern opinions To test the robustness of our results to these sample restrictions, we replicate our analysis in tables 3, 5, and 6, using different samples based on the following selection criteria: (1) a distressed sample that includes multiple-year going concern opinions (with a dummy variable to control for firms with prior-year going concern opinions), (2) a sample of all available firms (i.e., not limited to distressed firms) that receive first-time going concern opinions, and (3) a sample of all available firms that includes multiple-year going concern opinions (with a dummy variable to control for firms with prior-year going concern opinions) The results from all of the that is unbiased We not implement this correction because the extent of this correction is expected to be small and is unlikely to affect our inferences NON - AUDIT SERVICE FEES 1271 regressions (not presented) report that all of the coefficients on all of the fee variables remain insignificant at conventional levels Thus, our findings are not sensitive to our sample-selection criteria In addition, we perform an analysis that considers alternative measures for our fee variables Specifically, as in Chung and Kallapur [2001], we compute measures equal to the client’s non-audit fees, audit fees, or total fees, divided by the total fees for that firm’s auditor across all clients These measures are designed to capture the relative importance of the client to the audit firm We then substitute these measures for the respective fee variables in the going concern models in table (except for FEERATIO) The results (not presented) are not sensitive to these alternative measures for our fee variables Finally, we perform an analysis that considers an alternative measure of auditor independence Specifically, we measure the ex post error in auditors’ going concern decisions by computing a variable that equals the going concern indicator variable used in equation (1), minus an indicator variable equal to if the client firm filed for bankruptcy during the 12 months following the respective year-end, and otherwise Hence, this new variable takes one of three values: –1, 0, or 1, where –1 indicates a Type II error (erroneous clean opinion), indicates a correct opinion, and indicates a Type I error (erroneous going concern opinion) We then use this new measure as the dependent variable in an ordered logistic regression with independent variables identical to those in the five models in tables and The results (not presented ) report that all of the coefficients on all of the fee variables are insignificant at conventional levels Thus, our findings are robust to this alternative measure of auditor independence Summary and Conclusion In the wake of the Enron bankruptcy and the subsequent indictment of its auditor, Arthur Andersen, regulators have taken actions to curtail auditorprovided non-audit services These actions are based on the premises that non-audit service fees impair auditor independence by making the auditor economically dependent on the client and that the consulting nature of non-audit services reduces the auditor’s objectivity In this study we attempt to empirically investigate the validity of the these concerns by gathering systematic evidence on the association between non-audit and audit fess paid to incumbent auditors and auditor independence measured as the propensity of auditors to issue going concern opinions Contrary to regulators’ concerns, we find no association between nonaudit service fees and the auditor’s propensity to issue a going concern opinion In addition, we find no relation between audit fees and the auditor’s propensity to issue a going concern opinion Our findings suggest 1272 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM that market-based institutional incentives, such as reputation loss and litigation costs, promote auditor independence and outweigh the economic dependency created by higher fees REFERENCES AMEMIYA, T “The Estimation of a Simultaneous Equation Generalized Probit Model.” Econometrica 46 (1978): 1193–1205 AMERICAN INSTITUTE OF CERTIfiED PUBLIC ACCOUNTANTS Statement on Auditing Standard No 59: The Auditor’s Consideration of an Entity’s Ability to Continue as a Going Concern New York: AICPA, 1988 ANTLE, R “Auditor Independence.” Journal of Accounting Research 22 (1984): 1–20 ANTLE, R.; P GRIFFEN; D TEECE; AND O WILLIAMSON “An Economic Analysis of Auditor Independence for a Multi-Client, Multi-Service Public Accounting Firm.” Report prepared for AICPA by the Law & Economics Consulting Group Inc (1997) AICPA, New York BEATTY, A.; S CHAMBERLAIN; AND J MAGLIOLO “Managing Financial Reports of Commercial Banks: The Influence of Taxes, Regulatory Capital and Earnings.” Journal of Accounting Research 33 (1995): 157–78 BECKER, C.; M DEFOND; J JIAMBALVO; AND K R SUBRAMANYAM “The Effect of Audit Quality on Earnings Management.” Contemporary Accounting Research 15 (1998): 1–24 BENEISH, D., AND E PRESS “Costs of Technical Default.” The Accounting Review 68 (1993): 233– 57 BENSTON, G “Accountant’s Integrity and Financial Reporting.” Financial Executive (August 1975): 10–14 BLACCONIERE, W., AND M DEFOND “An Investigation of Independent Audit Opinions and Subsequent Independent Auditor Litigation of Publicly Traded Failed Savings and Loans” Journal of Accounting and Public Policy 16 (1997): 415–54 CARCELLO, J V.; D HERMANSON; AND F HUSS “Temporal Changes in Bankruptcy-Related Reporting.” Auditing: A Journal of Practice and Theory 14 (1995): 133–43 CHAMBERS, A., AND S PENMAN “Timeliness of Reporting and the Stock Price Reaction to Earnings Announcements.” Journal of Accounting Research 22 (1984): 21–47 CHUNG, H., AND S KALLAPUR “Client Importance, Non-Audit Services, and Abnormal Accruals.” Working paper, Purdue University, 2001 COPLEY, P A.; M S DOUCET; AND K M GAVER “A Simultaneous Equations Analysis of Quality Control Review Outcomes and Engagement Fees for Audits of Recipients of Federal Financial Assistance.” The Accounting Review 69 (1994): 244–56 CRASWELL, A.; J FRANCIS; AND S TAYLOR “Auditor Brand Name Reputations and Industry Specializations.” Journal of Accounting and Economics 20 (1995): 297–312 DEANGELO, L “Auditor Size and Audit Quality.” Journal of Accounting and Economics (1981): 183–99 DECHOW, P.; R SLOAN; AND A SWEENEY “Detecting Earnings Management.” The Accounting Review 70 (1995): 193–225 DOPUCH, N.; R HOLTHAUSEN; AND R LEFTWICH “Predicting Audit Qualifications with Financial and Market Variables.” The Accounting Review 62 (1987): 431–54 D’SOUZA, J “Rate-Regulated Enterprises and Mandated Accounting Changes: The Case of Electric Utilities and Post-Retirement Benefits Other Than Pensions (SFAS No 106).” The Accounting Review 73 (1998): 387–410 FIRTH, M “The Provision of Non-Audit Services by Accounting Firms to Their Audit Clients.” Contemporary Accounting Research 14 (1997): 1–11 FRANCIS, J., AND B KE “Do Non-Audit Services Compromise Auditor Independence?” Working paper, University of Missouri, 2002 FRANCIS, J., AND J KRISHNAN “Accounting Accruals and Auditor Reporting Conservatism.” Contemporary Accounting Research 16 (1999): 135–65 NON - AUDIT SERVICE FEES 1273 FRANCIS, J.; E MAYDEW; AND C SPARKS “Earnings Management Opportunities, Auditor Quality, and External Monitoring.” Auditing: A Journal of Practice and Theory 18 (1999): 83–105 FRANKEL, R.; M JOHNSON; AND K NELSON “Auditor Independence and Earnings Quality.” Working paper, Massachusetts Institute of Technology, 2002 GORE, P.; P POPE; AND A SINGH “Non-Audit Services, Auditor Independence, and Earnings Management.” Working paper, Lancaster University, United Kingdom, 2001 GUAY, W.; S KOTHARI; AND R WATTS “A Market-Based Evaluation of Discretionary Accrual Models.” Journal of Accounting Research 37 Supplement (1996): 83–105 HAUSMAN, J “Specification Tests in Econometrics.” Econometrica 46 (1978): 1251–71 HOPWOOD, W.; J MCKEOWN; AND J MUTCHLER “Reexamination of Auditor Versus Model Accuracy Within the Context of the Going Concern Opinion Decision.” Contemporary Accounting Research 11 (1994): 295–310 HRIBAR, P., AND D W COLLINS “Errors in Estimating Accruals: Implications for Empirical Research.” Journal of Accounting Research 40 (2002): 105–34 JENSEN, M., AND W MECKLING “Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure.” Journal of Financial Economics (1976): 305–60 JUDGE, G G.; W E GRIFfiTHS; R C HILL; H LUTKEPHOL; AND T LEE The Theory and Practice of Econometrics, 2nd Edition New York: John Wiley and Sons, 1985 LEVITT, A “Renewing the Covenant with Investors.” Remarks delivered at the NYU Center for Law and Business, New York University, New York (2000) MUTCHLER, J.; W HOPWOOD; AND J MCKEOWN “The Influence of Contrary Information and Mitigating Factors in Audit Opinion Decisions on Bankrupt Companies.” Journal of Accounting Research 35 (1997): 295–310 NELSON, F., AND L OLSEN “Specification and Estimation of Simultaneous Equations Model with Limited Dependent Variables.” International Economic Review 19 (1978): 695– 710 OHLSON, J “Financial Ratios and the Prediction of Bankruptcy.” Journal of Accounting Research 18 (1980): 109–31 PALMROSE, Z “The Effect of Non-Audit Services on the Pricing of Audit Services: Further Evidence.” Journal of Accounting Research 24 (1986): 405–11 PALMROSE, Z “An Analysis of Auditor Litigation and Audit Service Quality,” The Accounting Review 63 (1988): 55–73 PALMROSE, Z “Empirical Research on Auditor Litigation: Considerations and Data.” Studies in Accounting Research #33 Sarasota, FL: American Accounting Association, 1999 PARKASH, M., AND C VENABLE “Auditee Incentives for Auditor Independence: The Case of Non-Audit Services.” The Accounting Review 68 (1993): 113–33 RAGHUNANDAN, K., AND D RAMA “Audit Opinions for Companies in Financial Distress: Before and After SAS No 59.” Auditing: A Journal of Practice and Theory 14 (1995): 50–63 REYNOLDS, K.; D DEIS; AND J FRANCIS “Professional Service Fees and Auditor Objectivity.” Working paper, Louisiana State University, 2002 REYNOLDS, K., AND J FRANCIS “Does Size Matter? The Influence of Large Clients on OfficeLevel Auditor Reporting Decisions.” Journal of Accounting and Economics 30 (2000): 375– 400 SECURITIES AND EXCHANGE COMMISSION “Final Rule: Revision of the Commission’s Auditor Independence Requirements.” Washington, DC: SEC, 2000 SHU, S “Auditor Resignations: Clientele Effects and Legal Liability.” Journal of Accounting and Economics 29 (2000): 173–205 SIMUNIC, D “Auditing, Consulting, and Auditor Independence.” Journal of Accounting Research 22 (1984): 679–702 WATTS, R., AND J ZIMMERMAN “Agency Problems, Auditing, and the Theory of the Firm: Some Evidence.” Journal of Law and Economics 26 (1983): 613–33 WATTS, R., AND J ZIMMERMAN Positive Accounting Theory Englewood Cliffs, NJ: Prentice Hall, 1986 WEBER, J.; D LITTLE; D HENRY; AND L LAVELLE “How Bad Will It Get?” Business Week (December 24, 2001): 30–32 1274 M L D e F ond, K RAGHUNANDAN, AND K R SUBRAMANYAM WEIL, J “‘Going Concerns’: Did Accountants Fail to Flag Problems at Dot-Com Casualties?” The Wall Street Journal (February 9, 2001): C1 WHISENANT, S.; S SANKARAGURUSWAMY; AND K RAGHUNANDAN “Evidence on the Joint Determination of Audit and Non-Audit Fees.” Working paper, University of Houston, 2002 ZMIJEWSKI, M “Methodological Issues Related to the Estimation of Financial Distress Prediction Models.” Journal of Accounting Research 22, Supplement (1984): 59–82 ... incumbent auditor TOTAL FEE = the sum of audit and non -audit fees paid to the incumbent auditor NON -AUDIT FEE = the sum total of all non -audit fees paid to the incumbent auditor AUDIT FEE = the audit. .. non -audit service fees are inversely related to auditors’ propensity to issue going concern audit opinions Regulators are concerned that large non -audit service fees create incentives for auditors... non -audit fees to total fees paid to the incumbent auditor log(TOTAL FEE) = the natural logarithm of the sum of audit and non -audit fees paid to the incumbent auditor log (AUDIT FEE) = the natural