defond et al - 2002 - do non–audit service fees impair auditor independence - evidence from going concern audit opinions

Therefore, the purpose ofthis article is to investigate the veracity of regulators’ concerns by empir-ically examining the association between non-audit and audit fees paidau-to incumben

Printed in U.S.A.

Do Non-Audit Service Fees Impair Auditor Independence? Evidence from Going Concern Audit

im-to controlling for endogeneity among our variables, and im-to several alternativeresearch design specifications Our results are consistent with market-basedincentives, such as loss of reputation and litigation costs, dominating the ex-pected benefits from compromising auditor independence

1 Introduction

Independent auditing is an essential feature of efficient capital marketsand 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



independence.1In the wake of the Enron bankruptcy, concerns about ditor independence have prompted Congress to enact legislation that bansmost auditor-provided non-audit services.2Regulators’ concerns about non-audit services are based on the assumption that auditors are willing to sac-rifice their independence in exchange for retaining clients that pay largenon-audit fees A problem with this assumption, however, is that it ignoresauditors’ expected costs of compromising their independence In particular,loss of reputation and litigation costs are likely to provide strong incentivesfor auditors to maintain their independence Therefore, the purpose ofthis article is to investigate the veracity of regulators’ concerns by empir-ically examining the association between non-audit (and audit) fees paid

au-to incumbent audiau-tors and audiau-tor independence, where audiau-tor dence is surrogated by the propensity of auditors to issue going concernaudit opinions.3

indepen-Before the recently enacted legislation, the Securities and Exchange mission (SEC) attempted to curtail auditors from providing non-audit ser-vices by requiring listed companies to publicly disclose non-audit (and audit)fees in proxy statements filed on or after February 5, 2001 (SEC [2000]) Al-though the SEC originally sought to ban non-audit services, it agreed to thecompromise solution of public disclosure in the belief that such disclosureswould attract the scrutiny of shareholders Several recent research papersuse the newly available auditor fee data to empirically investigate whethernon-audit services threaten auditor independence (e.g., Chung and Kalla-pur [2001], Francis and Ke [2002], Frankel, Johnson, and Nelson [2002],and Reynolds, Deis, and Francis [2002]) This line of research uses factorscommonly associated with earnings management, such as discretionary ac-cruals and managers’ propensity to meet earnings targets, as indicators ofauditor independence The results from this research, however, yield mixedsupport for the contention that non-audit services impair auditor indepen-dence We attempt to provide additional evidence on this issue by investigat-ing another indicator of auditor independence—the auditor’s willingness

Com-to issue a going concern audit opinion

The audit report communicates the auditor’s findings to market ticipants and plays a crucial role in warning financial statement users ofimpending going concern problems Issuing a going concern opinion, how-ever, means that the auditor must be able to objectively evaluate firm per-formance and withstand any client pressure to issue a clean opinion This

par-1 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]).

2 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 com- promise 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.

3 Henceforth we use “non-audit services” to mean all non-audit services provided by the

incumbent auditor.

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 in-versely related to the probability of auditors’ issuing going concern auditreports 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 fied, total fees will be inversely related to the probability of auditors’ issuinggoing concern audit reports We test this hypothesis, also

justi-We perform our analysis on 1,158 distressed firms with proxy statementsthat include audit fee disclosures, including 96 firms receiving first-timegoing concern audit reports As in prior research, we focus on distressedfirms because the going concern problem is a more salient decision amongthis group We test the hypotheses by including measures of non-audit ser-vice fees and total fees in logistic regressions that explain the issuance ofgoing concern opinions We investigate the first hypothesis by examiningregressions that include two measures of non-audit fees: the log of non-auditservice 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 betweenour measures of non-audit service fees and going concern opinions wouldprovide support for our first hypothesis

Similarly, we test the second hypothesis by examining regressions that ther include the log of total fees or that disaggregate total fees into its twocomponents (the log of audit fees and the log of non-audit fees) We exam-ine a disaggregated measure of total fees for two reasons First, the second

ei-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 regressionthat examines the disaggregated components of total fees essentially testsboth of our hypotheses simultaneously That is, we are able to see both theassociation with total fees and the association with non-audit service fees(after controlling for audit fees) in a single regression.5

4 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.

5 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.

Our results provide no support for either hypothesis That is, we find noevidence of a significant association between the auditor’s propensity to is-sue a going concern opinion and any of our fee measures This finding isrobust to replacing all fee variables by their respective unexpected compo-nents, and after controlling for the simultaneity bias induced by endogeneityamong non-audit fees, audit fees, and going concern opinions.

We contribute to the current debate on auditor independence by ing no evidence that non-audit service fees adversely affect the auditor’sopinion-formulation process These findings are consistent with market-based institutional incentives, such as costly shareholder litigation and loss

find-of reputation, dominating the expected benefits to auditors find-of ing their independence Thus, we find no support for regulators’ concernsthat non-audit services impair auditor independence

compromis-We derive our conclusions from a research design that uses the auditor’spropensity to issue a going concern opinion to proxy for auditor indepen-dence In contrast to studies that use earnings management surrogates toproxy for auditor independence, we argue that our proxy is more direct andless noisy Specifically, although studies investigating earnings managementassume that various earnings characteristics (e.g., discretionary accruals andpropensity to meet earnings targets) are evidence of auditor independence,the auditor’s influence on client’s earnings characteristics is likely to be indi-rect, and there are empirical problems in measuring discretionary accruals(Guay, Kothari, and Watts [1996], Dechow, Sloan, and Sweeney [1995],Hribar and Collins [2002]).6In contrast, the auditor clearly influences thetype of audit opinion, and measuring the audit opinion is unambiguous

We acknowledge, however, that a potential limitation of our research sign is that our tests may lack the power to reject our null hypotheses Forexample, it is conceivable that auditors who perform non-audit services aremore tolerant of earnings management, but draw the line at compromis-ing the integrity of the audit opinion It is also possible that the costs andbenefits to the auditor are stacked in favor of issuing the going concernopinion Thus, because we draw our conclusions based on the lack of find-ing a statistical association, we cannot rule out lack of power as an alternativeexplanation for our findings

The next section discusses the motivation for our analysis, section 3 scribes our tests, section 4 presents our results, and section 5 summarizesour findings

de-2 Non-Audit Services and Auditor Independence

2.1 INCENTIVES FOR AUDITOR INDEPENDENCE

Auditor independence is often defined as the probability that the ditor will report a discovered breach in the financial reports (Watts and

au-6 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].

Zimmerman [1983, 1986]).This suggests that auditor independence is onymous with auditor objectivity and the ability to withstand client pres-sure to acquiesce to substandard reporting Jensen and Meckling [1976]conclude that managers have incentives to reduce agency costs by hiringindependent auditors Supporting this conclusion, Watts and Zimmerman[1983] find evidence that 84% of New York Stock Exchange (NYSE) compa-

syn-nies voluntarily engaged independent auditors in 1926, several years before

the Securities Acts that mandated external auditing Thus, there is both oretical and empirical evidence that managers find it in their best interests

the-to engage independent audithe-tors

A large body of theoretical and empirical research also suggests that ditors have market-based institutional incentives to act independently Forexample, Benston [1975] conjectures that reputation concerns are likely

au-to create incentives for independence, and Watts and Zimmerman [1983]document several historical examples of auditors taking costly actions toprotect their reputation capital.7 More recently, reputation concerns areconsistent 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 mightget in shareholder litigation or fear their financial reports could draw morescrutiny from regulators if they’re handled by Andersen

The threat of class action lawsuits provides another incentive for auditorindependence, particularly in U.S capital markets, where Big 5 auditorsincurred more than $1 billion in litigation-related costs in 1993 alone (An-tle, Griffen, Teece, and Williamson [1997]).8Palmrose [1988], drawing ontheory from DeAngelo [1981], presents evidence consistent with Big 5 au-ditors’ reducing litigation exposure by increasing their independence Thisevidence is also consistent with Francis, Maydew, and Sparks [1996] andBecker, DeFond, Jiambalvo, and Subramanyam [1998], who find that Big 5auditors appear to constrain managers’ ability to exercise accounting discre-tion Shu [2000] finds that auditors resign from clients (and hence foregofee revenues) in response to both increases in litigation risk and emergingmismatches with the clients In summary, a large body of research finds thatoutside auditors have market-based institutional incentives—particularly re-lated to reputation and litigation costs—to remain independent of theirpublicly held clients

2.2 NON-AUDIT SERVICES AND AUDITOR INDEPENDENCE

Although there are market-based incentives for auditors to remainindependent, there are also forces that potentially threaten auditor

7 Also, Antle [1984] indicates that reputation is a likely enforcement mechanism for auditor independence.

8 For expositional convenience, we use “Big 5” to refer to Big 5, Big 6, and Big 8 auditors.

independence Specifically, regulators are concerned about two effects ofnon-audit services One is a fear that non-audit service fees make auditorsfinancially dependent on their clients, and hence less willing to stand up tomanagement pressure for fear of losing their business.9The other is that theconsulting nature of many non-audit services puts auditors in managerialroles, potentially threatening their objectivity about the transactions theyaudit These concerns are summarized in the following quote from the SECregulations mandating fee disclosures (SEC [2000]):

An auditor’s interest in establishing or preserving a non-audit services ship raises two types of independence concerns First, the more the auditorhas at stake in its dealings with the audit client, the greater the cost to theauditor should he or she displease the client, particularly when the non-auditservices relationship has the potential to generate significant revenues on top

relation-of the audit relationship Second, certain types relation-of non-audit services, whenprovided by the auditor, create inherent conflicts that are incompatible withobjectivity.10

Regulators’ concerns that auditors become financially dependent on theirclients are based on an intuitive cost-benefit trade-off Regulators fear thatauditors will perceive that the benefits from retaining clients that pay largenon-audit service fees outweigh the expected costs of sacrificing their in-dependence As discussed earlier, the expected costs of sacrificed indepen-dence include the reputation loss and litigation costs associated with auditfailures Although not explicitly stated, this argument is based on the as-sumption that non-audit service fees include economic rents Otherwise,auditors will be indifferent between keeping and losing their non-audit ser-vice clients Simunic [1984] argues that auditor-provided management advi-sory services can generate economic rents because of “knowledge spillovers.”Knowledge spillovers refer to information generated while performing man-agement consulting services that can produce economic rents by reducingauditing costs

Simunic [1984] investigates whether spillovers exist by examining auditfees Although he finds evidence that audit fees are higher in the presence ofnon-audit services—consistent with the existence of knowledge spillovers—

a later study by Palmrose [1986] provides evidence that audit fees are highereven when clients engage consultants who are not incumbent auditors Even

if the existence of spillovers is established, however, it is impossible to rectly quantify the costs and benefits in the previously described trade-off

di-9 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 5 Levitt [2000] asserts that consulting services of the Big 5 now represent more than 50% of their revenues, up from just 12% in 1977.

Therefore, it is ultimately an empirical question whether auditors mise their independence to retain non-audit service clients.

compro-One source of such empirical evidence is litigation against auditors porting the contention that auditor-supplied consulting services do not re-sult in substandard reporting, Palmrose [1999] finds no instances of lawsuitsagainst auditors that allege non-audit services impair independence Simi-larly, of 610 cases of litigation against auditors examined in Antle, Griffen,Teece, and Williamson [1997], only 24 mention that the auditors also pro-vide non-audit services, and only 3 of those 24 allege that the non-auditservices impaired auditor independence Thus, the evidence from class ac-tion lawsuits suggests that non-audit services are not an important source oflitigation against auditors

Sup-In addition to the evidence from class action lawsuits, several recent ies examine the association between non-audit service fees and evidence

stud-of earnings management (e.g., Chung and Kallapur [2001], Francis and

Ke [2002], Frankel, Johnson, and Nelson [2002], and Reynolds, Deis, andFrancis [2002]) Specifically, this line of research investigates whether com-panies that report higher levels of non-audit service fees are more likely toreport 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 associationbetween non-audit service fees and the magnitude of discretionary accru-als, Chung and Kallapur [2001] do not find this association, and Francisand Ke [2002] find that this relation is weakly significant, but only amongnon–Big 5 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 non-audit services is associated with increased levels of earnings management ismixed

Contrary to the concerns that fee dependency impairs auditor dence, Reynolds and Francis [2000] find evidence consistent with audi-

intors increasing their independence in response to greater financial

depen-dence Specifically, they find that relatively larger audit clients—those forwhich 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 utation and litigation damages associated with audit failure are likely to begreater for larger clients, providing incentives for auditors to be more con-servative In addition, they also find no evidence that auditors are morelenient in issuing going concern reports to larger clients Thus, Reynolds

rep-11 Using a sample of U.K companies, Gore, Pope, and Singh [2001] find a positive tion between non-audit fees and earnings management for non-Big 5 auditors’ clients but not for Big 5 auditors’ clients.

associa-and Francis find no evidence that financial dependency impairs auditorindependence.12

In summary, theory and evidence suggests that although auditors havemarket-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 therelation 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 ically refers to cases in which auditors fail to issue going concern opinions

typ-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 opinionswhen substantial doubt exists regarding the client’s ability to continue as agoing concern for one year beyond the financial statement date.13Becauseauditor independence is defined as the probability that the auditor will re-port a discovered breach in the financial reports (Watts and Zimmerman[1983]), auditors with impaired independence are less likely to issue go-ing concern opinions when such opinions are warranted Thus, we test thefollowing 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

incen-tives for auditors to reduce their independence However, auditors also

re-ceive audit fees, and DeAngelo [1981] argues that the bilateral monopoly

created by nonzero auditor switching costs results in auditors’ receivingeconomic 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 lowing hypothesis (in the alternative form):

fol-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.

13SAS 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.

3 Research Design

3.1 SAMPLE

We obtain our sample by first identifying all available proxy statementsfiled 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 ing first-time going concern opinions.14 This process results in a sample of4,105 firms with fee information, including 160 with going concern opin-ions for fiscal-year 2000 financial statements but not for prior-year financialstatements.15We then require the sample firms to have the necessary finan-cial statement variables on the Compustat (industrial, full coverage, andresearch) databases, stock return variables on the Center for Research inSecurity Prices (CRSP) database, mergers and new issues variables on theSDC database, and institutional ownership variables on the Wharton WRDS14F database After deleting financial institutions and all firms that changeyear-end, these procedures yield a preliminary sample of 2,428 firms, in-cluding 100 with first-time going concern opinions

receiv-As in prior research, we limit our analysis to a sample of financially tressed firms in evaluating the auditor’s probability of issuing a first-time go-ing concern opinion (Hopwood, McKeown, and Mutchler [1994], Mutchler,Hopwood, and McKeown [1997], Reynolds and Francis [2000]) This is be-cause the going concern opinion decision is most salient among distressedfirms As in Reynolds and Francis [2000], we define financially distressedfirms as firms that report either negative earnings or operating cash flowsduring the current fiscal year After restricting the analysis to distressed firms(as defined earlier), we have a usable sample of 1,158 firms, including 96with first-time going concern opinions

dis-3.2 GOING CONCERN MODEL

We test our hypotheses by estimating the coefficients in the followinglogistic regression that models the auditor’s probability of issuing a first-time 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 5 )

+ β13(OP CASH FLOW ) + β14(REPORT LAG )

14 In general, distressed firms tend to file their proxies later in the year.

15 Some of our sample firms have fiscal year-ends occurring in early 2001.

+ β15(FEERATIO ) + β16(log(TOTAL FEE ))

+ β17(log(AUDIT FEE )) + β18(log(NON-AUDIT FEE )) + ε

(1)where:

OPINION = an indicator variable equal to 1 for firms with

going concern audit opinions, and 0 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 1 when the firm

reports a bottom-line loss for the previous year,and 0 otherwise

INVESTMENTS = short- and long-term investment securities

(in-cluding cash and cash equivalents) deflated bytotal assets at year-end

FUTURE FINANCE = an indicator variable equal to 1 when the firm

issues equity or debt in the subsequent year(through October 31, 2001)

BIG 5 = an indicator variable equal to 1 when the

audi-tor is a member of the Big 5, and 0 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

non-audit fees paid to the incumbent non-auditor

The last four variables in the model—FEERATIO, log(TOTAL FEE), log(AUDIT FEE) and log(NON-AUDIT FEE)—are added to the model in

various combinations to test the two hypotheses If non-audit service fees

impair auditor independence, we expect a negative coefficient on TIO and log(NON-AUDIT FEE) If total fees impair auditor independence,

FEERA-we expect a negative coefficient on log(TOTAL FEE) and on its separate components, log(AUDIT FEE) and log(NON-AUDIT FEE).

The choice of independent variables in the going concern model is

mo-tivated by the “contrary” and “mitigating” factors identified in SAS No.

59 (AICPA [1988]) and includes many variables used in prior research

(e.g., Dopuch, Holthausen, and Leftwich [1987], Mutchler, Hopwood, and

McKeown [1997], Reynolds and Francis [2000]) SAS No 59 defines

“con-trary” factors as those suggesting a going concern opinion is appropriate and

“mitigating” factors as those mitigating the circumstances that suggest a ing concern opinion Financial distress is an important contrary factor men-

go-tioned in SAS No 59, and we capture financial distress using several financial statement and market variables PROBANKZ is the probability of bankruptcy

score from Zmijewski [1984], with higher values indicating a higher

proba-bility of bankruptcy The log(AGE) variable is the log of the number of years

the company has been publicly traded and is included because younger firmsare more prone to failure (Dopuch, Holthausen, and Leftwich [1987]) Wealso include the following three market-based measures following Dopuch,

Holthausen, and Leftwich, [1987]: BETA, which is the systematic risk of the firm’s stock return; RETURN, which is the stock return over the fiscal year; and VOLATILITY, which is the return volatility of the company’s stock We predict that RETURN is negatively associated with OPINION, and we predict that BETA and VOLATILITY are positively associated with OPINION Other contrary factors in our model include LEV and CLEV because

Mutchler, Hopwood, and McKeown [1997] find that debt covenant tions are positively associated with the probability of issuing a going concern

viola-opinion We include LEV to capture proximity to covenant violation because

firms close to violation are likely to have high leverage (Beneish and Press

[1993]), and we include CLEV because increases in leverage are likely to

move firms closer to covenant violation (Reynolds and Francis [2000]) As

in Reynolds and Francis [2000], we also include LLOSS (a dummy

indicat-ing a loss in the prior year) because firms with multiple-year losses are morelikely to fail.16We include OP CASH FLOW (operating cash flows divided by

total assets) because poor operating cash flows are often associated with theprobability of bankruptcy, and the Zmijewski [1984] bankruptcy score doesnot include a cash-flow measure.17 In addition, we include REPORT LAG

16 We do not include a dummy for current-year loss because our sample-selection criterion

is based on the firm incurring a loss (or negative operating cash flow) during the current year.

17 An alternative measure of the probability of bankruptcy that considers cash flows is that

in Ohlson [1980] However, the Ohlson measure yields an extremely narrow distribution of high bankruptcy probabilities in our sample of distressed firms Therefore, we use the score

(number of days between the fiscal year-end and the earnings ment date) because prior research finds that going concern opinions areassociated with longer reporting delays (Raghunandan and Rama [1995],Carcello, Hermanson, and Huss [1995]).

announce-In addition, we include several factors that are likely to mitigate theprobability of receiving a going concern opinion The log of total assets

(log(ASSETS)) is included because large firms have more negotiating power

in the event of financial difficulties and hence are more likely to avoid

bankruptcy (Reynolds and Francis [2000]) INVESTMENTS is the sum of

the firm’s cash and investment securities (long and short term), scaled bytotal assets, and is a liquidity measure that captures the ability to quicklyraise cash It is included because firms with large cash and investment secu-rities have more resources to stave off bankruptcy in the event of financialdifficulties.18FUTURE FINANCE captures firms in our sample that issue new

debt or equity securities (public or private) through October 31, 2001 (thelast day of available information for our sample firms at the time of our anal-ysis) This variable is included because Mutchler, Hopwood, and McKeown[1997] find that new financing (and refinancing) is a mitigating factor that

reduces the probability of bankruptcy We also include BIG 5, an indicator

variable equal to 1 if the auditor is a member of the Big 5, and 0 otherwise.This variable is included because prior research argues that Big 5 auditorsare more likely to issue going concern audit opinions (Mutchler, Hopwood,and McKeown [1997])

4 Results

4.1 DESCRIPTIVE STATISTICS FOR FULL SAMPLE

Table 1 presents descriptive statistics on the full sample for the variablesused in our going concern model in equation (1) We winsorize all continu-ous variables at the 99th percentile of their absolute values prior to limitingthe sample to distressed firms The first four rows present the measures ofaudit and non-audit fees used in our analysis Row 1 indicates that the mean

and median values of FEERATIO are 49% and 51%, respectively These

ra-tios are comparable to the mean and median fee rara-tios reported in Frankel,Johnson, and Nelson [2002] Row 2 reports that the mean and median

values of TOTAL FEE are $920,000 and $380,000, respectively Our sample

from Zmijewski [1984], which yields a wider distribution of scores and thus appears better at distinguishing the relative degree of distress among our sample firms.

18 Because the 2001 Compustat database is still unavailable, we do not have financial ment data to measure the magnitude of subsequent asset sales, a mitigating factor suggested in Mutchler, Hopwood, and McKeown [1997] and used in Reynolds and Francis [2000] However,

state-we believe that INVESTMENTS is a variable that captures the underlying economics of the “sales

of fixed assets.” That is, our INVESTMENTS variable captures the ability of the firm to quickly raise cash in the event of financial difficulties In addition, an advantage of INVESTMENTS over assets sales is that INVESTMENTS is an ex ante measure and thus likely to be more relevant

to the auditor’s opinion-formulation decision.

T A B L E 1

Descriptive Statistics for 1,158 Financially Distressed Firms (Including 96 Firms with Going Concern

Opinions) with Available Auditor Fee Information for Fiscal Year 2000a

Full Sample (n= 1,158)

FEERATIO= the ratio of non-audit fees to total fees paid to the 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 fee paid to the incumbent auditor

OPINION= an indicator variable equal to 1 for firms with going concern audit opinions, and 0 otherwise

PROBANKZ= probability of bankruptcy score (Zmijewski [1984])

ASSETS= total assets at the end of the year measured in millions of dollars

AGE= number of years since the company was listed in 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 1 when the firm reports a bottom-line loss for the previous year, and 0 otherwise

INVESTMENTS= short- and long-term investment securities (including cash and cash equivalents) deflated by total assets at the end of the fiscal year

FUTURE FINANCE= an indicator variable equal to 1 when the firm issues equity or debt in the subsequent year (as of October 31, 2001)

BIG 5= an indicator variable equal to 1 when the auditor is a member of the Big 5, and 0 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.

firms report smaller total fees compared with those reported in Frankel,Johnson, and Nelson, consistent with the distressed nature of our sample

firms Rows 3 and 4 report mean and median values of NON-AUDIT FEE

of $616,000 and $183,000, and mean and median values of AUDIT FEE of

$302,000 and $160,000, respectively As with total fees, the means and dians of the fee variables in our sample tend to be lower than those found

me-in Frankel, Johnson, and Nelson

Row 5 in table 1 shows that 8% of our sample receives going concern ions, which is comparable to Reynolds and Francis [2000], who find that 9%

opin-of their financially distressed sample receives going concern opinions Row

6 indicates that the mean and median bankruptcy probability scores in oursample are 22 and 03, respectively Row 7 indicates that mean and medianfirm sizes, measured in total assets, are $813 million and $110 million, in-dicating a skewed distribution and justifying our decision to log assets inour logit analysis Row 8 indicates that our sample firms have been listed

for 1 to 38 years Row 9 reports that the mean and median values for BETA

are close to 1.0 Row 10 indicates that the mean and median stock returnsduring the prior year are –35% and –53%, respectively, indicating that oursample 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 0.48and 0.44, respectively, and that the median change in leverage is relativelysmall It is not surprising that row 14 reports that prior-period losses arerelatively 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 additionaloutside financing in the future Row 17 reports that 91% of our sample has

a Big 5 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 oursample (Chambers and Penman [1984])

4.2 DESCRIPTIVE STATISTICS BY OPINION TYPE

Table 2 classifies the variables from Table 1 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 1 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

dif-ferences significant at p < 1% (two-tailed) Row 2 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

opin-ion 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 3 indicates that the mean and median values of NON- AUDIT 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

dif-ference between the means is significant at p < 5% (two-tailed), and the difference between the medians is significant at p < 1% (two-tailed) Row 4 indicates that both the mean and median values of AUDIT FEE are not sig-

nificantly different between the two opinion types Thus, univariate tests