You are on page 1of 23

J. Account.

Public Policy 29 (2010) 330352

Contents lists available at ScienceDirect

J. Account. Public Policy


journal homepage: www.elsevier.com/locate/jaccpubpol

Do the Big 4 and the Second-tier rms provide audits


of similar quality?
Jeff P. Boone a, Inder K. Khurana b, K. K. Raman c,*
a
Department of Accounting, College of Business Administration, University of Texas at San Antonio,
P.O. Box 305219 Denton, TX 76203 940/565-3089, United States
b
School of Accountancy, University of Missouri-Columbia, Columbia, MO 65211, United States
c
Department of Accounting, College of Business Administration, University of North Texas, P.O. Box 305219,
Denton, TX 76203, United States

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, we examine audit quality for Big 4 and Second-tier
auditors during 20032006. We utilize the auditors propensity
to issue a going concern audit report for distressed clients as a
measure of audit quality. In addition, since the purpose of an audit
is to improve nancial reporting quality, we utilize abnormal
accruals as an observable proxy for audit quality. Further, we uti-
lize the client- and year-specic ex ante equity risk premium as
a proxy for audit quality as perceived by investors. We control for
auditor self-selection bias using the matched-pairs sample
approach discussed by Francis and Lennox (2008). We nd weak
evidence that the Big 4 have a higher propensity to issue going con-
cern audit opinions for distressed companies. However, the level of
performance-adjusted abnormal accruals for Big 4 and Second-tier
audit rm clients appears to be similar. With respect to investor
perceptions, we nd the client-specic ex ante equity risk pre-
mium to be lower for Big 4 clients than for Second-tier audit rm
clients. Overall, our ndings suggest little difference in actual audit
quality but a more pronounced difference in perceived audit qual-
ity. Collectively, the evidence we provide informs the current dis-
course on audit quality, auditor choice, and the viability of
Second-tier auditors as an alternative to the Big 4.
2010 Elsevier Inc. All rights reserved.

* Corresponding author. Tel.: +1 940 565 3089.


E-mail addresses: Jeff.Boone@utsa.edu (J.P. Boone), khurana@missouri.edu (I.K. Khurana), Raman@unt.edu (K. K. Raman).

0278-4254/$ - see front matter 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.jaccpubpol.2010.06.007
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 331

1. Introduction

In recent years, a Second-tier of large international audit rms appear to have emerged as an alter-
native to the Big 4 (Byrnes, 2005; GAO, 2006; Jopson, 2006).1 The emergence of these Second-tier rms
has been facilitated by the increase in workload (i.e., the increase in the number of hours required for a
typical audit) and capacity constraints brought on by the enactment of the 2002 Sarbanes Oxley Act
(GAO, 2006). Moreover, regulators (such as the PCAOB) and interest groups (such as the American
Assembly and the US Chamber of Commerce) appear to be promoting these Second-tier rms as a viable
alternative to the Big 4 based on the underlying belief that audit quality for Big 4 and Second-tier rms is
similar.2 Still, whether or not the Second-tier rms provide audit quality similar to that of the Big 4 re-
mains an interesting and open empirical question.
In this paper, we investigate audit quality differentials between Big 4 and Second-tier auditors dur-
ing 20032006.3 Specically, we examine whether the auditors propensity to issue a going concern
audit opinion for distressed companies is similar for Big 4 and Second-tier auditors. Further, since the
purpose of an external audit is to improve nancial reporting quality, we examine whether abnormal
accruals (as an observable proxy for accounting information quality) are similar for Big 4 and Second-tier
audit rm clients.4 In addition, from an investor perspective, we examine whether the perceived account-
ing information quality of Big 4 and Second-tier rm audit clients is similar as reected in the client- and
year-specic ex ante equity risk premium. Collectively, our research attempts to address the actual as
well as perceived audit quality of Big 4 and Second-tier auditors.
Prior theoretical and empirical research in auditing suggests that the two primary drivers of audit
quality are litigation costs and reputation loss. Specically, Palmrose (1988) and Simunic and Stein
(1987) suggest that given their deep pockets and heavy spending on building their brand names,
the large audit rms have an incentive to lower litigation risk and protect their reputational capital
by providing more credible nancial reports. Their ndings are consistent with the notion (articulated
by DeAngelo 1981) that audit rm size is an important determinant of audit quality. Specically, the
argument is that large size (1) enables auditors to spend heavily on training and audit technology
(thereby contributing to their competence), and (2) makes them less dependent on an individual cli-
ent, and thus better able to resist client pressure to issue a clean (unqualied) audit opinion and/or
to restrain earnings management.
Further, DeAngelo (1981) argues that the value of an audit rm consists of the present value of
future quasi rents, i.e., the excess of audit fees over marginal costs. Consequently, the auditor with
a greater number of clients has more to lose in terms of aggregate quasi rents by failing to report a

1
We ranked all US audit rms in 2006 by market share, i.e., the totals of their audit clients assets or revenues. Either approach
to measuring market share provided similar rankings. The four largest rms (the Big 4) are PwC, Deloitte & Touche, Ernst & Young,
and KPMG. We classify the fth and sixth largest audit rms (Grant Thornton and BDO Seidman) as Second-tier auditors. Relative
to Grant Thornton and BDO Seidman, the seventh and eighth largest audit rms (Crowe Chizek, and McGladrey Pullen) are much
smaller, i.e., during 20032006, the number of public audit clients for Crowe Chizek, and McGladrey Pullen was less than a 10th of
the number of such clients for Grant Thornton and BDO Seidman. Further, as discussed below, Carson (2009) identies Grant
Thornton and BDO Seidman (along with the Big 4) as comprising the six global audit rm networks. Hence, for the purpose of this
study, we identify only Grant Thornton and BDO Seidman as Second-tier auditors.
2
Thus, Kayla Gillan, a member of the PCAOB has indicated that audit committees should consider the so-called second-tier of
audit rms. I dislike using that term because it implies that the rms are secondary in quality which I strongly believe is false . . .
(Grant Thornton, 2007). Similarly, the PCAOB Chairman Mark Olson has noted the growth in size and skill taking place in (audit)
rms following the Big 4, and Johnson (2007) suggests that the PCAOBs inspection of the larger non-Big 4 rms found fewer
deciencies than those reported at the Big 4. Separately, the American Assembly (2005) indicates that mid-tier rms could
satisfactorily serve a large number of those companies that seem to be principally served by the Big 4 (p. 11). The US Chamber of
Commerce (2006) suggests that all parties should actively encourage public companies to consider high quality rms outside the
Big 4 . . . (p. 18).
3
Essentially, we examine audit quality differentials since the demise of Arthur Andersen in 2002. We focus on the 20032006
time period to avoid the potentially confounding effects of various events (such as the stock market collapse, the various
accounting scandals including Enron and WorldCom, and the criminal conviction of Arthur Andersen) during 2002.
4
In other words, we recognize that accrual choices are jointly affected by client and auditor preferences, and that it is difcult to
distinguish audit quality from nancial reporting quality. Consistent with prior research (e.g., Francis et al., 1999; Khurana and
Raman, 2004), since higher audit quality implies more credible nancial reporting, we view accounting information quality as a
consequence of audit quality.
332 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

discovered breach, i.e., subsequent detection of the failure to report might entail loss of reputation and
clients and/or lower future audit fees. Her argument is that this collateral aspect of the auditorclient
relationship raises the penalty for compromising independence in proportion to the size of the audi-
tor. Consequently, audit quality is higher for large auditors. Put differently, litigation costs and repu-
tation loss (and the resulting loss of aggregate quasi rents) provide large auditors the necessary
market-based incentive to maintain their independence and provide audit quality consistent with
their brand name reputation.
It is worth noting that the theory discussed above was developed in the 1980s and draws on the
(then) distinction between the Big 8 and other smaller audit rms.5 By contrast, the Second-tier audi-
tors today represent the fth and sixth largest audit rms in the US and worldwide, with correspondingly
developed signicant national and international brand name reputation capital. Also, as discussed by
Carson (2009, p. 363), the two Second-tier rms along with the Big 4 comprise the extant six global
audit rm networks dened as the founding members of the international Forum of Firms which oper-
ates as the Transnational Auditors Committee of IFAC.6 Moreover, the Second-tier rms today are likely
of the size contemplated by Palmrose (1988) and Simunic and Stein (1987) for the Big 8 in the 1980s.
Potentially, the Second-tier rms today have an incentive level (based on litigation costs and reputation
loss) similar to the Big 4 to maintain auditor independence and audit quality.7 Further, Nelson (2006)
indicates that reforms mandated by the 2002 Sarbanes Oxley Act (SOX) have increased both client
and auditor incentives for accurate nancial reporting. In turn, stronger client and auditor incentives
for accurate reporting may be expected to contribute to more uniform audit quality generally.8 For all
these reasons, Big 4 and Second-tier rm audit quality today could be similar.
However, there is also an important reason why Big 4 and Second-tier audit quality may not be
similar. Specically, differences in audit quality between Big 4 and Second-tier auditors may exist
to the extent that these auditors vary in their independence, i.e., their ability and willingness to with-
stand client pressure for a clean (unqualied) audit opinion or to allow aggressive reporting. As noted
by the GAO (2008) and the US Department of the Treasury (2008), the largest Second-tier rm is less
than one-third the size of the smallest Big 4 rm as measured by aggregate revenues and/or the num-
ber of partners, professional staff, and public company clients. By virtue of their much smaller size, the
Second-tier rms are likely to have considerably less invested capital (partner equity) and smaller
aggregate quasi rents than the Big 4 rms. Hence, to the extent that the Big 4 by virtue of their con-
siderably larger size continue to have deeper pockets (and are subject to greater scrutiny for those
reasons) and considerably larger aggregate quasi rents than the Second-tier rms, the Big 4 could very
well have a greater market-based economic incentive to pushback harder against client pressures to
allow aggressive and potentially opportunistic reporting. Consequently, differential audit quality

5
The Big 8 subsequently merged (or demised, in the case of Arthur Andersen) to form the Big 6, the Big 5, and the extant Big 4.
6
Carson (2009) also notes that these six rms issued a joint statement on recommendations for changes in auditing and
nancial reporting standards (DiPiazza et al., 2006).
7
We thank a reviewer for the line of reasoning discussed here.
8
Specically, with respect to client incentives for more accurate reporting, SOXs Section 302 requires CEOs/CFOs to certify that
internal controls are effective and that the nancial statements present fairly the rms nancial position and results. Further,
CEOs/CFOs now face a variety of severe nancial, criminal, and civil penalties in the event that their certications are later proved
incorrect. Thus, managers are now responsible for fair presentation (which goes beyond mere technical GAAP compliance) and can
be visibly penalized for violations. Nelson (2006) suggests that these new requirements and penalties for CEOs/CFOs make it less
likely that the client will misstate the nancial statements, or pressure the auditor to allow aggressive earnings management.
With respect to auditor incentives, SOXs Section 104 created the PCAOB and an ongoing program of quality inspections of audit
rms. The PCAOB has broad investigative and disciplinary powers, and is required to conduct inspections of audit rms. Closer
monitoring (by the PCAOB) is likely to induce auditors to increase audit effort, tighten materiality, and make more conservative
judgments and decisions in their audit engagements. Further, SOX Sections 104 and 404 require auditors to do more in terms of
testing the control system and attesting to the accuracy of CEO/CFO certications of the effectiveness of internal controls over
nancial reporting. Nelson (2006) suggests that these and other SOX-mandated reforms (such as independent audit committees
and audit partner rotation) and the continuing threat of PCAOB annual inspections and sanctions, make it more likely that the
auditor will detect nancial statement misstatements and resist client pressure to allow opportunistic reporting. Thus, the
cumulative effect of the reforms in terms of (1) less aggressive reporting by the client, (2) less client pressure to waive corrections
to the misstatements, (3) increased auditor effort to identify misstatements, and (4) greater auditor resistance to client pressure to
allow aggressive reporting may be expected to contribute to more uniform audit quality across auditors.
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 333

between Big 4 and Second-tier auditors could exist. Hence, whether Big 4 and Second-tier audit qual-
ity is similar remains an interesting and ultimately an empirical question.
In our study, we recognize that not all Big 4 clients are auditable by Second-tier auditors.9 Hence, in
examining our research questions, we focus on Big 4 clients for whom the Second-tier rms are poten-
tially viable as auditors. Specically, we exclude Big 4 clients whose revenues exceed the revenues of the
largest Second-tier client in each of the four years.10 Also, as discussed below, in our analyses we control
for auditor self-selection by using the matched-pairs sample approach discussed by Francis and Lennox
(2008).
Our results broadly suggest that nancial reporting quality is similar for Big 4 and Second-tier audit
rm clients. Specically, we nd weak evidence that Big 4 auditors have a higher propensity to issue a
going concern audit opinion for distressed clients relative to Second-tier auditors. Also, we nd the
level of performance-adjusted abnormal accruals for Big 4 audit clients to be no different from those
of Second-tier audit rm clients.
With respect to investor perceptions, we nd the client-specic ex ante equity risk premium to be
lower for Big 4 clients than for Second-tier audit rm clients. This nding is consistent with the notion
that investors perceive Big 4 audit quality (and by implication the accounting information quality of
Big 4 clients) to be higher. An alternative explanation is that investment bankers and institutional
investors prefer Big 4 auditors (for their own clients and investees) primarily for insurance reasons,
i.e., possible access to the Big 4s greater nancial resources in the event of an audit failure (GAO,
2008, p. 44).11 Put differently, underwriters and investors prefer Big 4 auditors primarily for their dee-
per pockets and resulting greater ability to share more of the settlement costs and shareholder losses in
the event the deal falters or the client fails. In turn, this preference appears to be translated into a lower
cost of equity capital for Big 4 clients.
Overall, the evidence we document in this study suggests little difference in actual audit quality but
a more pronounced difference in perceived audit quality. This suggests that the choice between Big 4
and Second-tier auditors is not primarily a quality-based choice but, rather, is primarily driven by cost
savings in the form of lower ex ante equity risk premiums due to hiring a Big 4 auditor. Thus, clients of
Second-tier auditors potentially trade off higher ex ante equity risk premiums for possibly lower audit
fees. From the perspective of public policy, the ability to choose between Big 4 and Second-tier audit
rms is benecial because it allows rms to enhance economic efciency (by matching marginal costs
and benets) while still obtaining largely similar audit quality. Collectively, our ndings are important
and timely, and inform the discourse on audit quality and auditor choice (i.e., the viability of Second-
tier auditors as an alternative to the Big 4) both of which appear to be of signicant public policy con-
cern at this time for regulators as well as users of nancial statements.
The rest of the paper proceeds as follows: Section 2 describes and develops our hypotheses. Sec-
tion 3 discusses our methodology and sample. The empirical ndings are reported in Section 4, and
Section 5 concludes the paper.

2. Hypotheses

2.1. Going concern audit opinion

Consistent with prior research (e.g., Carey and Simnett, 2006; Reynolds and Francis, 2001), we uti-
lize the auditors propensity to issue a going concern audit report for distressed clients as a measure of

9
As noted previously, the largest Second-tier rm is about a third the size of the smallest Big 4 rm. The GAO (2008) indicates
that the very large Big 4 clients have global businesses and complex operations that the Second-tier rms may not have the
capacity to audit due to their lack of geographical reach and the necessary depth in staff resources and technical expertise.
Specically, it is not uncommon for an audit of a very large client to require hundreds of staff, and it may be difcult (given limited
staff resources) for a Second-tier rm to commit hundreds of employees to a single client.
10
Our results are not sensitive to this imposed rule. Our ndings are similar if the rule is based on assets (rather than revenues)
or if we simply exclude the top 10% of Big 4 audit clients.
11
As noted previously, despite their emergence in recent years as a possible alternative to the Big 4, the Second-tier auditors
remain considerably smaller in size than the Big 4 and thus are likely to have to have less nancial resources (partner equity)
invested in the rm.
334 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

audit quality. Since the issuance of SAS No. 59, the auditor is required to evaluate the clients ability to
continue as a going concern for one year beyond the date of the nancial statements. Indeed, as noted
by DeFond et al. (2002), the auditors failure to issue a going concern opinion to a client who subse-
quently les for bankruptcy is typically referred to as a case of audit failure.
Understandably, the issuance of a going concern audit report is viewed unfavorably by clients. Put
differently, clients view a going concern audit opinion as an adverse development, i.e., as a potential
self-fullling prophesy that is likely to cause bankruptcy because of the reports possible adverse ef-
fects on investors, creditors, and suppliers. Consequently, it is important for the auditor to objectively
evaluate a clients nancial circumstances and to withstand client pressure to issue a clean (unqual-
ied) opinion.
Prior research (e.g., DeFond et al., 2002) also suggests that the decision to issue a going concern
opinion is more salient for nancially distressed companies. Consequently, prior research (e.g., Carey
and Simnett, 2006; Reynolds and Francis, 2001) has focused on nancially distressed clients, and sug-
gests a correlation between the issuance of a going concern report for distressed clients and auditor
independence. Stated differently, after controlling for other factors that are likely to inuence the pro-
pensity to issue a going concern audit report for a client, the higher the auditors probability of issuing
a going concern report, the higher the auditors independence and (by implication) audit quality.
Consistent with the notion (discussed previously) that audit quality for Big 4 and Second-tier audi-
tors could be similar, our rst hypothesis (stated in the null form) is as follows:

H1. Ceteris paribus, the propensity of Big 4 auditors to issue a going concern audit report for
distressed clients is not different from that of Second-tier auditors.

2.2. Accruals-based earnings management

Prior research (e.g., Becker et al., 1998; Francis et al., 1999) has sought to link audit quality to earn-
ings management, and suggests consistent with the theory discussed previously (DeAngelo, 1981;
Palmrose, 1988; Simunic and Stein, 1987) that large audit rms are more effective at constraining
the clients ability to manipulate earnings. These studies indicate that the net result of a large rm
audit is lower abnormal accruals, i.e., reported earnings that are more reective of the clients under-
lying economics. In essence, prior research suggests that large auditors provide audits of higher qual-
ity relative to other auditors.
Put differently, prior research views the auditors role as one of limiting the potential for client-dri-
ven opportunistic management of accrual-based earnings. Given managers incentives to manipulate
reported earnings for private gain (such as executive compensation contracts) or to meet or beat earn-
ings benchmarks (Bartov et al., 2002; Graham et al., 2005), investors may be expected to be uncertain
as to whether or not the reported earnings are the outcome of a reasonable and unbiased application
of GAAP. In this milieu, prior research suggests that large auditors are more effective in restraining
managerial opportunism in terms of accruals-based earnings management, and thus are able lend
greater credibility to reported earnings. Consistent with this view, Becker et al. (1998) report lower
abnormal accruals for Big 6 clients than for non-Big 6 audit clients. Further, Francis et al. (1999) par-
tition their sample into three auditor groups (Big 6, mid-tier rms, and other audit rms) during
19881994, and nd signicant differences in abnormal accruals across all three auditor groups.
Once again, consistent with the notion (discussed previously) that audit quality for Big 4 and Sec-
ond-tier auditors could be similar, our second hypothesis (stated in the null form) is as follows:

H2. Ceteris paribus, the level of accruals-based earnings management by Big 4 audit clients is not
different from that of Second-tier audit rm clients.

2.3. Investor-perspective analysis

As noted by Dopuch and Simunic (1982), the higher the perceived quality of the audit, the more
credible the clients reported earnings as judged by users (p. 407). In turn, more credible earnings
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 335

may be expected to be associated with higher information (accruals) quality, i.e., increased likelihood
that the reported earnings map well with the clients underlying economic performance.
In recent research, Bhattacharya et al. (2003, pp. 643644) argue that although investors in an
efcient market can rationally anticipate earnings management, they cannot see through it, i.e., they
cannot undo the effects of earnings manipulations to arrive at the underlying (but unobservable)
economic earnings number. In other words, information asymmetry persists. Further, the information
asymmetry created by earnings management is not completely resolved through other communica-
tion mechanisms such as disclosures. For these reasons, earnings manipulations can be expected to
lower information quality for investors.
Separately, traditional asset-pricing theory takes the position that information quality is diversi-
able and should not affect expected returns. More recently, Lambert et al. (2007) suggest that although
accounting information quality may not be an independent risk factor, it can affect the cost of equity
capital because higher information quality reduces a rms beta by dampening the conditional covari-
ance of the rms cash ows with the market. They indicate that this effect on the assessed covariance
of the rms cash ows is not diversiable and is thus priced.
Consistent with the theory of Lambert et al. (2007), empirical work presents evidence that informa-
tion quality matters for expected returns. Both Francis et al. (2005) and Core et al. (2008) nd accrual
quality to be priced in the ex ante cost of equity capital. Other research (e.g., Botosan, 1997; Botosan
and Plumlee, 2002; Bhattacharya et al., 2003) also shows that rms with lower information quality
exhibit a higher cost of equity nancing.
Stein (1989) argues that investors rationally expect earnings management, but cannot completely
undo its effects, and therefore price the expected earnings manipulation. Given this pricing behavior,
i.e., the fact that rational investors anticipate and price earnings manipulation, it is optimal for man-
agers to manage earnings. Put differently, managers may be expected to continue to manage earnings
even when investors cannot be systematically fooled (Stein 1989, p. 668).12 Consistent with the notion
of rational investors, Bhattacharya et al. (2003), Core et al. (2008), Francis et al. (2005) show that lower
earnings quality is priced by investors in a higher cost of equity capital for the rm.
To the extent that investors are cognizant of the linkage between audit quality and earnings man-
agement and perceive Big 4 and Second-tier rm audit quality to be similar, Big 4 and Second-tier audit
rm clients may be expected to be associated with similar client-specic ex ante equity risk premi-
ums. By contrast, to the extent that investors expect differentials in earnings management across
Big 4 and Second-tier audit rm clients, and to the extent that they cannot undo (or diversify away)
the adverse effects of increased earnings management for individual clients, they may be expected to
respond with price protection by demanding differentials in the expected returns going forward, e.g., a
higher ex ante equity risk premium for accepting lower information quality, and vice versa.13
Once again, consistent with the notion (discussed previously) that audit quality for Big 4 and Sec-
ond-tier auditors could be similar, our third (and nal) hypothesis (stated in the null form) is as
follows:

H3. Ceteris paribus, the ex ante equity risk premium for Big 4 clients is not different from that of
Second-tier audit rm clients.

3. Research design and data

3.1. Going concern opinion model

To test Hypothesis 1, we estimate the following going concern opinion probit model:

12
Similarly, Graham et al. (2005) suggest that investors are rational and assume (and price the notion) that companies manage
earnings. Consequently, they suggest that it is rational for managers to manage earnings to meet or beat earnings benchmarks
particularly to avoid the severe negative stock price reaction typically associated with an earnings disappointment.
13
Note that price protection benets potential shareholders but adversely affects current shareholders. Other things being equal,
an increase in the ex ante equity risk premium (i.e., higher expected returns going forward) implies a lower share price and a loss of
wealth for current shareholders.
336 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

Table 1
Denition of variables.

Variable Denition
Panel A. Going concern analysis
Dependent variable
GC =1 if the auditor issued a going concern opinion in the current year; 0 otherwise
Test variable
BIG4 =1 if auditor is a Big 4 auditor; 0 if auditor is a Second-tier auditor
Control variables
AUDTEN Auditor tenure, i.e., the number of years for which the current audit rm has audited the client
SIZE Natural log of total assets (in millions of $) at the balance sheet date
LEV The ratio of debt to total assets at the balance sheet date
DLEV Change in LEV (i.e., LEV in year t less LEV in year t  1)
AGE Age of the rm based on the listing date reported in CRSP
DISTRESS Financial distress measure based on Zmijewski (1984)
RET Market-adjusted stock return for the 12 month period ending on the balance sheet date
PRIORLOSS =1 if net income in the prior year was <0; 0 otherwise
INVEST Asset-deated value of investments, calculated as total current assets less trade receivables and total
inventory, divided by total assets
CFFO Cash ow from operations deated by total assets
FEE The ratio of non-audit to audit fees
LAGGC =1 if the auditor issued a going concern opinion in the prior year; 0 otherwise
Panel B. Earnings management analysis
Dependent variable
ABACC Accruals-based earnings management metric, based on signed asset-deated performance-adjusted
abnormal accruals estimated using the cross-sectional modied Jones (1991) model. The higher the
metric, the higher the reported earnings
Test variable
BIG4 =1 if auditor is a Big 4 auditor; 0 if auditor is a Second-tier auditor
Control variables
AUDTEN Auditor tenure, i.e., the number of years for which the current audit rm has audited the client
SIZE Natural log of total assets (in millions of $) at the balance sheet date.
BM Book-to-market ratio at the balance sheet date
DISTRESS Financial distress measure based on Zmijewski (1984)
CFFO Cash ow from operations deated by total assets
GROWTH Change in sales from prior year to current year deated by prior year sales
FINANCE =1 if number of shares outstanding increased by at least 10% or long-term debt increased by at least 20%
during the year; 0 otherwise
LEV The ratio of debt to total assets at the balance sheet date
LITIGATION =1 if SIC is 28332836, 35703577, 73707374, 36003674, 52005961, 87318734; 0 otherwise
IMPLICIT Factor score based on factor analysis of measures related to production, R&D spending, and labor
intensity, consistent with Matsumoto (2002)
Panel C. Ex ante equity risk premium (ravg) analysis
Dependent variable
ravg Average of four client-specic ex ante equity risk premium metrics (rgls, rct, rojn, and rmpeg) measured four
months after scal year-end. The equity risk premium is the excess of the ex ante cost of equity capital
over the yield on the 10-year US Treasury bond. The four ex ante cost of equity capital estimates are based
on Gebhardt et al. (2001), Claus and Thomas (2001), Gode and Mohanram (2003), and Easton (2004),
respectively. The higher the metric, the higher the client-specic equity risk premium
Test variable
BIG4 =1 if auditor is a Big 4 auditor; 0 if auditor is a Second-tier auditor
Control
variables
BM Book-to-market ratio at the balance sheet date
LEV The ratio of debt to total assets at the balance sheet date
SIZE Natural log of total assets (in millions of $) at the balance sheet date
BETA Systematic risk (stock beta, obtained from rm-specic regressions of weekly stock returns on the equal-
weighted market index across the 52 week period ended on the balance sheet date
SDROA Standard deviation of asset-deated earnings before extraordinary items and discontinued operations,
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 337

Table 1 (continued)

Variable Denition
calculated over the current and prior 4 years
LTG Mean analysts forecast of long-term earnings growth (in percent) as reported by IBES
REC_RET Recent one-year stock return calculated over the 12 month period preceding the measurement of the
dependent variable ravg

GC a b1 BIG4 b2 AUDTEN b3 SIZE b4 LEV b5 DLEV b6 AGE b7 DISTRESS b8 RET


X
b9 PRIORLOSS b10 INVEST b11 CFFO b12 FEE b13 LAGGC bxx YRxx e
xx

Consistent with prior research (e.g., Carey and Simnett, 2006; DeFond et al., 2002; Reynolds and
Francis, 2001), we model the auditors probability of issuing a going concern opinion to a nancially
distressed client. The dependent and independent variables are dened in Table 1 panel A, and dis-
cussed below. Specically, the dependent variable GC is a dummy variable equal to 1 for clients with
a going concern audit opinion, and 0 otherwise.
In the probit regression, the test variable BIG4 is a dummy variable equal to 1 if the auditor is a Big 4
rm, and 0 if the auditor is a Second-tier auditor. As hypothesized previously, Big 4 and Second-tier audi-
tors could be similar in their ability to withstand client pressure to issue a clean audit opinion (when in
fact a going concern qualication is appropriate). Hence, we do not predict the sign for variable BIG4. Sep-
arately, the variables AUDTEN through LAGGC represent control variables in the model. Specically,
AUDTEN is the auditors tenure in years. To the extent that tenure impairs auditor objectivity, variable
AUDTEN is expected to have a negative sign. Variable SIZE is the size of the client (as measured by the
log of total assets). The expected sign for this variable is negative, since larger clients are less likely to
end up in bankruptcy given their greater negotiating power. Variables LEV and DLEV capture the clients
leverage (as measured by the debt to total assets ratio) and change in leverage. The greater the leverage
(and increase in leverage), the higher the probability of bankruptcy. Hence, the predicted signs for LEV
and DLEV are positive.
Variable AGE measures the age of the client company, and the expected sign for this variable is neg-
ative since younger companies are more likely to encounter bankruptcy. Variable DISTRESS measures the
clients nancial distress based on Zmijewski (1984), such that higher values of the variable represent
higher levels of nancial distress for the company. The predicted sign for DISTRESS is positive, since
the greater the nancial distress, the higher the probability of failure. Variable RET is a market-based
measure of risk and client performance. The predicted sign is negative, since the lower the clients mar-
ket-adjusted stock return, the greater the risk of failure and the higher the probability that the auditor
will issue a going concern opinion. Variable PRIORLOSS is a dummy variable that captures whether
the client reported a loss in the prior year. Since clients with continued losses are more likely to fail,
the predicted sign for this variable is positive. Separately, the greater the clients ability to quickly raise
cash and/or operating cash ows, the lower the risk of failure. Hence, the expected sign for variables IN-
VEST and CFFO is negative. Variable FEE captures the ratio of non-audit fees paid to the incumbent audi-
tor to audit fees. To the extent that non-audit fees impair auditor objectivity, the expected sign for FEE is
negative. Finally, variable LAGGC is a dummy variable equal to 1 if the auditor issued a going concern
opinion in the prior year. Consistent with Reynolds and Francis (2001), the predicted sign for LAGGC
in the regression is positive.14

3.2. Accruals-based earnings management model

To test Hypothesis 2, we estimate the following accruals-based earnings management model:


ABACC a b1 BIG4 b2 AUDTEN b3 SIZE b4 BM b5 DISTRESS b6 CFFO b7 GROWTH
b8 FINANCE b9 LEV b10 LITIGATION b11 IMPLICIT b12 SDSALES b13 SDROA
b14 SDCFFO e

14
As discussed later in the paper, we separately also examine rst-time going-concern audit opinions.
338 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

The dependent and independent variables in the earnings management model are dened in Ta-
ble 1 panel B. The dependent variable ABACC represents the signed asset-deated performance-ad-
justed abnormal accruals estimated using the cross-sectional modied Jones (1991) model.15 The
higher the metric, the higher the reported earnings.
To estimate normal accruals (consistent with DeFond and Jiambalvo (1994) and Subramanyam
(1996)), we utilize the modied Jones (1991) model below in the cross-section by industry.

TAit 1 DSALESit  DARit PPEit


a b1 b2 ei;t
Assetsi;t1 Assetsi;t1 Assetsi;t1 Assetsi;t1
In this model, TA is total accruals calculated as income from continuing operations less operating
cash ows from continuing operations, DSALES is change in sales revenue, DAR is the change in ac-
counts receivables, PPE is gross property and equipment, and the subscripts i and t denote rm and
year, respectively. The abnormal accruals represent the difference between total accruals and the esti-
mated (tted) normal accruals. As suggested by Kothari et al. (2005) and consistent with Francis et al.
(2005), we array rms in each industry (based on 2-digit SIC codes) into deciles based on the prior year
return on assets (ROA), and obtain the performance-adjusted abnormal accruals by subtracting from
each rms abnormal accrual the median abnormal accrual from the corresponding ROA-industry dec-
ile to which the rm belongs.
In the regression, the test variable BIG4 is equal to 1 if the auditor is a Big 4 rm, and 0 if the auditor
is a Second-tier rm. As hypothesized previously, Big 4 and Second-tier auditors could be equally
effective in restraining managerial opportunism in terms of accruals-based earnings management
by the client. Hence, we do not predict the sign for variable BIG4. Also in the regression, the variables
AUDTEN through IMPLICIT represent control variables. Prior research suggests tenure may affect the
auditors objectivity and knowledge of the client favorably in the early years, but also adversely in la-
ter years (Johnson et al., 2002; Myers et al., 2003; POB, 2002). Hence, we control for auditor tenure
(AUDTEN) in the regressions although we do not predict the sign for this variable.
The other control variables included in the model are also based on prior research. Lang and Lund-
holm (1993) suggest that larger clients have an incentive to manage earnings less (report more accu-
rately) in an attempt to avoid litigation. We utilize variable SIZE as a proxy for the size of the audit
client, and the predicted sign for this variable is negative. Consistent with prior research (e.g., Ashb-
augh et al., 2003), the book-to-market ratio (variable BM) represents the inverse of the rms growth
opportunities, and the predicted sign in negative.
In the regressions, variable DISTRESS represents the rms nancial condition based on Zmijewski
(1984). Prior research (e.g., DeFond and Jiambalvo, 1994; Reynolds and Francis, 2000; Sweeney, 1994)
suggests that nancial distress is positively associated with abnormal accruals. However, DeAngelo
et al. (1994) nd that more troubled rms tend to manage earnings downwards to facilitate debt rene-
gotiation. Hence, in the regressions, we do not predict the sign for variable DISTRESS. Based on prior
research (Ashbaugh et al., 2003; Chung and Kallapur, 2003), the audit clients cash ow from opera-
tions is expected to be negatively related to abnormal accruals. In the regressions, variable CFFO rep-
resents the cash ow from operations scaled by total assets, and the predicted sign for this variable is
negative.
Menon and Williams (2004) suggest that the clients growth is positively associated with abnormal
accruals. Also, Barth et al. (1999) and Dechow and Skinner (2000) suggest that growth rms have a
strong incentive to manage earnings in an attempt to meet or beat earnings benchmarks and thus sus-
tain the stock price. Hence, the predicted sign for GROWTH in the regressions is positive. Also, prior
research (Ashbaugh et al., 2003; Chung and Kallapur, 2003; Rangan, 1998) suggests that signicant

15
We thank the reviewers for suggesting that we examine signed accruals. Also, Hribar and Nichols (2007) suggest that
researchers examine signed (rather than absolute) abnormal accruals to mitigate the threat of bias in favor of rejecting the null
hypothesis of no earnings management. Separately, Kothari et al. (2005) indicate that accruals are affected by the clients
performance as measured by the return on assets (ROA). Hence, we examine performance-adjusted accruals as discussed below.
As discussed later in the paper, we separately also examine positive (i.e., income increasing) performance-adjusted abnormal
accruals, i.e., observations with ABACC > 0, since the prior literature suggests that managing earnings upward is more of a concern
for auditors, i.e., that the auditors legal exposure primarily emanates from rms overstating income (e.g., Heninger, 2001).
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 339

changes in company nancing are positively related to the clients incentive to manage earnings.
Hence, in the regressions, the predicted sign for the variable FINANCE is positive.
Separately, DeFond and Jiambalvo (1994) suggest that more leveraged rms are more likely to use
income-increasing accruals to avoid violating debt covenants. Hence, in the regressions, the predicted
sign of variable LEV is positive. Also, rms in certain industries (such as biotechnology or electronics)
have a higher incidence of shareholder litigation (Francis et al., 1994). We utilize an indicator variable
LITIGATION to represent rms in high litigation risk industries, and consistent with Heninger (2001)
the predicted sign for variable LITIGATION is positive. Finally, Bowen et al. (1995) suggest that rms
have implicit claims with their stakeholders including customers, suppliers, and employees. Following
Matsumoto (2002), we create variable IMPLICIT by factor analyzing three measures related to produc-
tion, R&D spending, and labor intensity.

3.3. Investor perspective (ex ante equity risk premium) analysis model

Finally, to test Hypothesis 3, our investor-perspective analysis is based on the following model:

rav g a b1 BIG4 b2 BM b3 LEV b4 SIZE b5 BETA b6 SDROA b7 LTG b8 REC RET


X
dxx INDUSTRYxx e
xx

In this model, the dependent variable ravg is the client-specic ex ante equity risk premium, dened
as the average of four client-specic ex ante equity risk premium metrics (rgls, rct, rojn, and rmpeg) mea-
sured four months after scal year-end. The equity risk premium is the excess of the ex ante cost of
equity capital over the yield on the 10-year US Treasury bond. Variable ravg is essentially an ex ante
metric, i.e., a measure of expected rather than realized returns, and thus not directly observable.16
The higher the metric, the higher the client-specic equity risk premium.
As noted by Dhaliwal et al. (2006), there are four alternative approaches discussed in the prior
literature for estimating the ex ante cost of equity capital based on Gebhardt et al. (2001), Claus
and Thomas (2001), Gode and Mohanram (2003), and Easton (2004). We refer to these estimates as
rgls, rct, rojn, and rmpeg, respectively. All four approaches utilize models based on forecasted earnings,
dividends, and book values to estimate the return required by shareholders as implied by the observed
share price. The details of each of the four models (and their underlying denitions and assumptions)
is discussed in the appendix in Dhaliwal et al. (2006), and we refer the reader to that summary for
convenience.
Given the lack of consensus as to which approach (model) is superior in estimating the cost of equi-
ty, we follow Dhaliwal et al. (2006) and Hail and Leuz (2006) in using the average (ravg) of the four
estimates in our empirical tests to mitigate the effects of particular assumptions that underlie each
model on our results. Consistent with prior research (Gebhardt et al., 2001; Dhaliwal et al., 2006),
we focus on the equity risk premium (i.e., the excess of the ex ante cost of equity capital over the yield
on the 10-year US Treasury bond) as our research metric.17
As before, the test variable BIG4 is equal to 1 if the auditor is a Big 4 rm, and 0 if the auditor is a
Second-tier rm. As hypothesized previously, Big 4 and Second-tier auditors could be perceived by
investors as similarly effective in constraining managerial opportunism in terms of accruals-based
earnings management (and thus equally effective in lending greater credibility to reported earnings
for their clients). Hence, we do not predict the sign for variable BIG4.
In the regression, we control for factors that may be expected to affect the client-specic ex ante
equity risk premium. These control variables (BM through REC_RET) are based on prior research
(e.g., Khurana and Raman 2004, 2006) and are dened in Table 1 panel C. Consistent with Fama
and French (1997), we control for the book-to-market ratio (variable BM) as an equity risk factor.
Variable BM is predicted to have a positive sign in the regressions. Also, a higher degree of nancial

16
Consistent with the prior literature (Basu, 2004; Botosan and Plumlee, 2005; Francis et al., 2004), Dhaliwal et al. (2006, p. 699)
notes that at a theoretical level, ex ante estimation is more appropriate for estimating the return demanded by stockholders than
methods that rely on ex post realizations.
17
Consistent with Gebhardt et al. (2001, p. 143), the fourth month is chosen to facilitate alignment between IBES and Compustat.
340 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

leverage (variable LEV) is expected to raise the perceived risk and increase the rms equity risk pre-
mium (Gebhardt et al., 2001). Hence, variable LEV is also expected to have a positive sign in the
regressions.
Other things being equal, the larger the client, the greater the information available about the cli-
ent, the more liquid the stock, and the lower the perceived risk (Brennan and Subrahmanyam, 1996;
Fama and French, 1997; Gebhardt et al., 2001). Hence, we control for the size of the client. Variable
SIZE is predicted to have a negative sign in the regressions. Also, in the context of the capital asset pric-
ing model, systematic risk (as measured by the stock beta) is predicted to be positively correlated with
the equity risk premium. Consequently, variable BETA is expected to have a positive sign.
Separately, Gebhardt et al. (2001) suggest that the volatility in reported earnings may be perceived
as a source of risk in valuing stocks. Hence, variable SDROA (the standard deviation of asset-deated
earnings) is predicted to have a positive sign in the regressions. Also, as suggested by Beaver et al.
(1970) and La Porta (1996), there may be a positive association between growth and equity risk since
growth opportunities are more uncertain than normal earnings. Hence, variable LTG (based on the
mean analysts forecast of long-term earnings growth as reported by IBES) is expected to have a po-
sitive sign. Further, to control for sluggishness in analysts earnings forecasts, we include the recent
one-year stock return (variable REC_RET) calculated over the 12 months preceding the measurement
of the dependent variable ravg. Consistent with Guay et al. (2005), the predicted sign for variable RE-
C_RET is negative.18 Finally, our regressions include dummy variables to control for the clients industry
membership (variable INDUSTRYxx, based on 2-digit SIC classications).

3.4. Control for self-selection bias and econometric estimations

In a recent paper, Francis and Lennox (2008) examine the pitfalls associated with using the two-
step procedure developed by Heckman (1979) to control for auditor self-selection bias. They docu-
ment that the Heckman (1979) two-step procedure often results in (1) severe collinearity problems,
and (2) inferences that are extremely fragile even when collinearity is not a problem. By contrast, they
show that OLS results are much more robust. Hence, Francis and Lennox (2008) recommend that
researchers report OLS results. To control for selection bias, they (p. 27) recommend that researchers
also report OLS results based on a matched-pairs sample technique (developed by Rosenbaum and Ru-
bin 1983) using matched propensity scores.
To implement the matched-pairs sample technique, we follow exactly the approach used by Fran-
cis and Lennox (2008, p. 28) by rst estimating the auditor choice model for each sample year.19 Based
on the model, for each client we predict the propensity to choose a Big 4 auditor and sort the sample by
the predicted propensity (probability) score of selecting a Big 4 rm. For each Big 4 client, we identify the
two client companies that have the closest predicted probabilities (as per the sorted scores) and adopt
the following matching rule: (a) if only one of the two potential matches is a Second-tier auditor client,
we choose that client as the match, (b) if both potential matches are Second-tier auditor clients, we
choose the client whose predicted probability is nearest, and (c) if both potential matches are Big 4 cli-
ents, we determine that there is no suitable match. This rule ensures that we obtain extremely close
matches, implying that the scored distributions are virtually identical for Big 4 and Second-tier clients.
We follow this approach for testing each of our three hypotheses. However, as noted by Francis and Len-
nox (2008), this rule is also quite costly in the sense that many Big 4 clients lack such close matches. As
an example, for the going concern audit opinion analysis (discussed below), the full sample of 2219

18
If recent stock returns have been high (low), and if analysts forecasts of future earnings are too low (high) due to sluggish
updates of the information that has been recently impounded in stock price, the imputed discount rate will be articially low
(high) in order to maintain the pricing equation (Guay et al. 2005, p. 18). Hence, in the regressions, the predicted sign for recent
stock returns (REC_RET) is negative. Note also that variables BM and SIZE in the model (discussed previously) may be expected to
control for any systematic optimism in analysts earnings forecasts (Guay et al. 2005, p. 25).
19
Our probit auditor choice model (based on Francis et al. (1999)) is as follows: BIG4 a b1 TA b2 ln SALES b3 LEV b4 E=P
b5 ISSUE b6 LOSS e, where BIG4 = 1 if the client uses a Big 4 auditor; 0 if auditor is a Second-tier auditor; TA is the absolute value
total accruals, scaled by sales; SALES is sales revenues in millions of dollars; LEV is the ratio of debt to total assets; E/P is the inverse
of the price-earnings ratio; ISSUE = 1 if change in equity > 10%, 0 otherwise; and, LOSS = 1 if current net income is <0 and the
absolute change in income >10%, 0 otherwise.
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 341

Table 2
Sample selection and composition.

Going concern Earnings Investor perspective (ex ante


opinion analysis management equity risk premium) analysis
analysis
Panel A. Sample selection
Client-year observations during 20032006 12,726 12,726 12,726
(excluding utilities and nancial services
rms) with complete data on control
variables
Exclude observations not in IBES or CRSP 6858a
Exclude clients not using a Big 4 or Second-tier 2539 2539 213
auditor (Grant Thornton or BDO Seidman)
Exclude clients not nancially distressed 7902b
Exclude Big 4 clients with revenues > revenues 0.66c 877c 643c
of the largest Second-tier client in that year
Full sample 2219 9310 5012

Year Total
2003 2004 2005 2006
Panel B. Sample composition
Going concern analysis
Big 4 clients 653 457 415 334 1859
Second-tier clients 88 79 109 84 360
Full sample 741 536 524 418 2219
Earnings management analysis
Big 4 clients 2376 2171 1921 1606 8074
Second-tier clients 281 307 346 302 1236
Full sample 2657 2478 2267 1908 9310
Investor-perspective analysis
Big 4 clients 1201 1230 1166 1086 4683
Second-tier clients 40 66 100 123 329
Full sample 1241 1296 1266 1209 5012
a
Client-years for which analysts earnings forecasts (required for estimating ex ante cost of equity capital) not available on
IBES; also, consistent with Easton (2004), we exclude client-years in which the restriction EPS2 > EPS1 > 0 is not met, where
EPS2 and EPS1 are mean analysts earnings forecasts for year t + 2 and t + 1.
b
Consistent with Reynolds and Francis (2001) and Carey and Simnett (2006), nancially distressed clients are those with
negative earnings and/or negative cash ows from operations. Thus, non-distressed clients are those with both positive
earnings and positive cash ows from operations.
c
As discussed in the paper, all tabulated analyses are based on a sample which excludes Big 4 audit clients whose revenues
exceed the revenues of the largest Second-tier audit client in that year. For 2003 through 2006, the largest Second-tier client
revenues were $5.27 billion, $5.95 billion, $6.17 billion, and $7.86 billion, respectively. Thus, the Big 4 clients in our sample are
potentially auditable by Second-tier auditors.

observations drops to a matched-pairs sample of 656. As suggested by Francis and Lennox (2008), in our
study we report OLS results for the full (unmatched) sample as well as OLS results for the matched-pairs
sample.
Separately, to estimate the various models discussed previously, we utilize a pooled time-series,
cross-sectional approach. Statistical inferences for the pooled regressions are based on robust t-sta-
tistics that are adjusted for residual correlation arising from pooling cross-sectional observations
across time, i.e., the t-statistics are based on White (1980) heteroskedasticity-adjusted robust variance
estimates that are adjusted for within-cluster correlation where the rm and scal year comprise the
cluster (Petersen, 2009).

3.5. Data and sample

Table 2 panel A summarizes the sample selection process. The sample is formed from the merged
Compustat annual industrial les, including the primary, secondary, tertiary and full coverage
342 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

Table 3
Descriptive statistics (going concern opinion analysis).

Full sample Differences in means


(Big 4 vs. Second-tier)
Big 4 (n = 1859) Second-tier (n = 360)
Mean Median Std. dev. Mean Median Std. dev.
GC 0.0580 0.0000 0.2338 0.0889 0.0000 0.2850 0.0309**
AUDTEN 8.5644 7.0000 7.3808 3.0556 2.0000 3.8281 5.5089***
SIZE 5.1290 5.0823 1.5571 3.9621 3.8788 1.1873 1.1669***
LEV 0.2116 0.1514 0.2328 0.1877 0.1161 0.2130 0.0239*
DLEV 0.0183 0.0000 0.2119 0.0063 0.0000 0.2046 0.0246**
AGE 13.1396 9.0000 11.6131 14.7083 11.0000 10.5567 1.5687**
DISTRESS 0.1164 0.0029 0.2557 0.1007 0.0031 0.2428 0.0157
RET 0.1235 0.1423 0.9461 0.0368 0.2233 0.8035 0.1602***
PRIORLOSS 0.5290 1.0000 0.4993 0.5778 1.0000 0.4946 0.0488*
INVEST 0.3144 0.2554 0.2379 0.2672 0.0231 0.2208 0.0471***
CFFO 0.0727 0.0184 0.2090 0.0851 0.0231 0.2089 0.0123
FEE 0.3623 0.2269 0.4207 0.2881 0.1578 0.3668 0.0743***
LAGGC 0.0371 0.0000 0.1890 0.0722 0.0000 0.2592 0.0352***

Matched-pairs sample
Big 4 (n = 328) Second-tier (n = 328)
GC 0.0793 0.0000 0.2706 0.0884 0.0000 0.2843 0.0091
AUDTEN 8.5183 7.0000 6.7810 3.0640 2.0000 3.8455 5.4543***
SIZE 4.3425 4.2894 1.2918 3.9834 3.8788 1.2024 0.3591***
LEV 0.1650 0.0626 0.2263 0.1896 0.1219 0.2110 0.0246
DLEV 0.0100 0.0000 0.1966 0.0106 0.0000 0.2092 0.0206
AGE 11.7531 8.0000 10.1922 14.7866 12.0000 10.5638 3.0335***
DISTRESS 0.1275 0.0018 0.2816 0.1035 0.0034 0.2458 0.0240
RET 0.1006 0.1584 0.9481 0.0223 0.2222 0.8333 0.1229*
PRIORLOSS 0.6159 1.0000 0.4871 0.5793 1.0000 0.4944 0.0366
INVEST 0.3924 0.3295 0.2598 0.2633 0.1909 0.2183 0.1290***
CFFO 0.1331 0.0632 0.2356 0.0852 0.0231 0.2084 0.0478***
FEE 0.2989 0.1663 0.3710 0.2948 0.1578 0.3785 0.0041
LAGGC 0.0488 0.0000 0.2157 0.0762 0.0000 0.2658 0.0274

Notes: GC = 1 if the auditor issued a going concern opinion in the current year; = 0 otherwise. Please see Table 1 (panel A) for
other variable denitions.
Consistent with Francis and Lennox (2008), the matched-pairs sample is formed by matching each Big 4 audit client with a
Second-tier audit client whose probability of selecting a Big 4 auditor is similar.
*
Denotes signicant difference in the means in a two-tailed test at the 0.10 levels.
**
Denotes signicant difference in the means in a two-tailed test at the 0.05 levels.
***
Denotes signicant difference in the means in a two-tailed test at the 0.01 levels.

research les. Excluded from our sample are utility and nancial services clients, and industries
(2-digit SIC code) with fewer than 10 client-year observations available to estimate the industry-spe-
cic modied Jones (1991) model.20 After excluding observations with missing data on control vari-
ables, we are left with a sample of 12,726 client-years. For the investor-perspective ex ante equity
risk premium (ravg) analysis, we exclude observations (client-years) not found in IBES and CRSP. We then
exclude clients not using a Big 4 or Second-tier auditor. For the going concern opinion analysis, we also
exclude clients that are not nancially distressed.
As discussed previously, we seek to include in our analysis only those Big 4 clients for whom the
Second-tier rms are potentially viable as auditors.21 During 2003 through 2006, the largest Second-
tier client had revenues of $5.27 billion, $5.95 billion, $6.17 billion, and $7.86 billion, respectively.

20
Consistent with prior research (e.g., Fields et al., 2004), utilities and nancial institutions are excluded because of their unique
regulatory and operating characteristics.
21
In other words, as discussed previously (Footnote 9), the Second-tier rms may not have the capacity to audit the very large Big
4 clients.
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 343

Table 4
Probit regression results (going concern opinion analysis) GC a b1 BIG4 b2 AUDTEN b3 SIZE b4 LEV b5 DLEV
P
b6 AGE b7 DISTRESS b8 RET b9 PRIORLOSS b10 INVEST b11 CFFO b12 FEE b13 LAGGC xx bxx YRxx e.

Expected sign Full sample Matched-pairs sample


Coef. z-Stat p-Value VIF Coef. z-Stat p-Value VIF
Panel A. All observations
Intercept ? 0.8699 3.33 0.001 0.3224 0.57 0.566
BIG4 0.1954 1.68 0.092 1.26 0.5822 2.25 0.025 1.49
AUDTEN  0.0071 0.74 0.231 1.39 0.0176 0.74 0.230 1.46
SIZE  0.1716 3.69 0.000 1.65 0.5034 4.09 0.000 1.39
LEV + 0.5834 1.78 0.038 2.80 1.2258 1.86 0.031 2.57
DLEV + 0.0254 0.09 0.464 1.38 0.5262 1.02 0.846 1.38
AGE  0.0026 0.43 0.335 1.35 0.0008 0.06 0.477 1.26
DISTRESS + 0.5689 2.23 0.013 2.67 0.9995 2.11 0.018 2.63
RET  0.4780 4.92 0.000 1.16 0.9475 3.56 0.000 1.14
PRIORLOSS + 0.0369 0.30 0.384 1.17 0.0108 0.04 0.483 1.19
INVEST  1.2867 4.34 0.000 1.42 1.1162 2.14 0.016 1.47
CFFO  0.7841 2.74 0.003 1.88 0.5178 0.97 0.166 1.86
FEE  0.1599 0.97 0.167 1.16 0.1143 0.38 0.354 1.12
LAGGC + 1.9971 12.21 0.000 1.07 1.8849 6.45 0.000 1.10
Nobs 2219 656
Pseudo R2 40.51% 51.56%

Panel B. First-time going concern opinions only


Intercept ? 1.0666 3.77 0.000 0.00 0.3073 0.57 0.569 0.00
BIG4 ? 0.1833 1.05 0.294 1.25 0.5314 1.93 0.053 1.48
AUDTEN  0.0070 0.68 0.248 1.39 0.0186 0.73 0.230 1.46
SIZE  0.1731 3.42 0.000 1.62 0.4658 3.67 0.000 1.35
LEV + 0.6782 1.86 0.032 2.85 1.0733 1.45 0.074 2.78
DLEV + 0.0373 0.10 0.547 1.37 0.1975 0.33 0.632 1.37
AGE  0.0019 0.28 0.387 1.35 0.0013 0.10 0.463 1.26
DISTRESS + 0.5753 1.98 0.024 2.70 0.9462 1.66 0.049 2.87
RET  0.5900 4.10 0.001 1.16 0.8919 2.78 0.003 1.14
PRIORLOSS + 0.0786 0.58 0.279 1.15 0.0654 0.24 0.400 1.18
INVEST  1.3710 4.14 0.001 1.42 0.8759 1.58 0.057 1.48
CFFO  0.8838 2.70 0.004 1.87 0.3696 0.62 0.270 1.92
FEE  0.2246 1.25 0.106 1.16 0.1303 0.40 0.345 1.13
Nobs 2124 615
Pseudo R2 24.30% 23.43%

Notes: The dependent variable GC = 1 if the auditor issued a going concern opinion in the current year; = 0 otherwise. Please see
Table 1 (panel A) for other variable denitions. The probit model is estimated by maximum likelihood from cross-sectional
observations for the period 20032006. Reported signicance levels are based on two-tailed tests (where the sign is not
predicted) and one-tail otherwise. VIF denotes the variance ination factor. Coefcient estimates for yearly dummies are
omitted for brevity.
Results reported in Panel A are based on all observations, while results in Panel B are based on rst-time going concern opinions
only (i.e., excludes recurring going concern opinions).
Consistent with Francis and Lennox (2008), the matched-pairs sample is formed by matching each Big 4 audit client with a
Second-tier audit client whose probability of selecting a Big 4 auditor is similar.

Thus, we exclude Big 4 clients with revenues exceeding these threshold amounts in each of those
years. Our full (unmatched) sample consists of 2219, 9310, and 5012 observations for the going con-
cern opinion, earnings management, and investor-perspective (ex ante equity risk premium) analyses,
respectively. As discussed previously, to control for auditor self-selection bias, we follow the approach
used by Francis and Lennox (2008, p. 28) which reduces our matched-pairs sample to 656, 2154, and
618 observations for the going concern opinion, earnings management, and investor-perspective anal-
yses, respectively. Finally, to reduce the inuence of outliers, all variables are truncated at the 1st and
99th percentiles.
Table 2 panel B provides information on sample composition by year and auditor group for all three
of our analyses. Consistent with the notion that the Big 4 dominate the audit market (GAO, 2003,
2008), a large majority of our observations have a Big 4 rm for auditor.
344 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

4. Empirical ndings

4.1. Going concern audit opinion analysis

Table 3 provides descriptive statistics for the dependent and explanatory variables in the going
concern opinion regressions for the full and matched-pair samples. Variable GC, the dependent var-
iable in this analysis, is an indicator variable equal to 1 if the auditor issued a going concern audit
opinion in the current year, and 0 otherwise. For variable GC, the mean is smaller for Big 4 clients
than for Second-tier clients in both the full and matched-pair samples, indicating (on an univariate
basis) that Big 4 rms have a lower propensity to issue a going concern opinion (although the dif-
ference in means is not signicant in the matched-pairs sample).22 Similarly, for a number of the
control variables AUDTEN through LAGGC in the model, the univariate means are different for Big
4 and Second-tier clients, although the difference in means is signicant for fewer of these control
variables in the matched-pairs sample than in the full sample. In our multivariate analysis discussed
below, we control for these differences in client characteristics in examining the difference (if any) in
the propensity of Big 4 and Second-tier auditors to issue a going concern opinion for distressed
clients.
For brevity, we do not show correlation matrices. However, the highest pairwise correlation be-
tween the test variable BIG4 and any of the control variables (in either the full or matched-pair sam-
ples) was only 0.28 (with variable SIZE). As discussed below, we diagnose multicollinearity in the
regressions using variance ination factors (VIFs).
Table 4 panel A reports the regression results used to test Hypothesis 1. In both regressions in
Table 4, the VIFs are very low (all below 3) which suggests that collinearity is not likely to be a
problem in interpreting the results. The control variables AUDTEN through LAGGC, where signi-
cant, have the predicted signs and are consistent with prior research (e.g., Carey and Simnett,
2006; Reynolds and Francis, 2001). The test variable BIG4 is statistically signicant at the 0.10
and 0.025 levels, with a positive sign in the regressions for the full and matched-pair samples.
Since the matched-pair sample results (unlike the full sample results) control for auditor self-selec-
tion bias, we place more reliance on the matched-pair ndings. The statistically signicant results
are also economically meaningful, i.e., for audit clients, the use of a Big 4 auditor translates into a
nearly 3-fold increase in the likelihood of receiving a going concern qualication in the matched-
pair analysis.23
Separately, as part of sensitivity analysis, we also examine rst-time going concern audit
opinions.24 Table 4 panel B reports the regression results used to test Hypothesis 1 based on rst-
time going concern modications only. Once again, the control variables AUDTEN through FEE, where
signicant, have the predicted signs. The test variable BIG4 is not signicant in the regression for the
full sample but is signicant at the 0.06 level with a positive sign in the regression for the matched-
pair sample. Once again, since the matched-pair sample results (unlike the full sample results) con-
trol for auditor self-selection bias, we place more reliance on the matched-pair ndings. The statis-
tically signicant results are also economically meaningful, i.e., for audit clients, the use of a Big 4

22
In the full sample, the means for variable GC indicate that 5.8 (8.89) percent of Big 4 (Second-tier) distressed clients received a
going concern opinion. The difference in means is statistically signicant at the 5% level, indicating that on an univariate basis Big 4
auditors have a lower propensity to issue a going concern opinion. By contrast, in the matched-pairs sample, the means for variable
GC indicate that 7.93 (8.84) percent of Big 4 (Second-tier) distressed clients received a going concern opinion. Here, the difference
is not statistically signicant. Although we discuss the univariate ndings for completeness, note that univariate comparisons are
potentially misleading without controlling for other client characteristics that are likely to inuence the auditors decision to issue
a going concern opinion. Our multivariate analysis (which controls for differences in other client characteristics) is discussed
below.
23
In the matched-pair analysis, the probability of receiving a going concern opinion for a Big 4 audit is 0.043 whereas the
probability for a Second-tier auditor is only 0.0108. The 0.043 equals Uy ^, where U is the standard normal cumulative
probability distribution function and y^ is the tted value of the probit model determined using all estimated coefcients and the
mean values of the control variables. The 0.0108 equals Uy ^, where y
^ is the tted value of the probit model using all estimated
coefcients (except the BIG4 coefcient) and the mean values of the control variables. The increase in the probability is 0.0322,
which represents a nearly 3-fold increase in the probability of receiving a going concern opinion.
24
We thank a reviewer for suggesting the additional analysis discussed here.
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 345

auditor translates into a nearly 3-fold increase in the likelihood of receiving a going concern quali-
cation in the matched-pair analysis.25
Collectively, the ndings from panels A and B indicate that after controlling for the various client
factors that are likely to inuence the auditors propensity to issue a going concern report, Big 4 audi-
tors have a higher probability than Second-tier auditors for issuing a going concern opinion for dis-
tressed clients. Thus, the results based on the matched-pair sample do not support our rst null
hypothesis (H1), and indicate that Big 4 auditors are better able to objectively evaluate a clients nan-
cial circumstances and withstand client pressure to issue a clean (unqualied) opinion, i.e., Big 4 audi-
tors demonstrate greater auditor independence and (by implication) higher audit quality than Second-
tier auditors.

4.2. Earnings management analysis

Table 5 provides descriptive statistics for the dependent and explanatory variables in the earnings
management regressions for the full and matched-pair samples. Variable ABACC, the dependent
variable in this analysis, is the signed value of the asset-deated performance-adjusted abnormal
accruals. Other things being equal, the higher the metric, the higher the reported earnings (and earn-
ings management). For variable ABACC, the mean is smaller (more negative) for Big 4 clients than for
Second-tier clients in both the full and matched-pair samples, indicating (on an univariate basis) that
Big 4 clients report lower abnormal accruals than Second-tier audit rm clients. For a number of the
control variables AUDTEN through IMPLICIT also, the univariate means are different for Big 4 and Sec-
ond-tier clients, although (once again) the difference in means is signicant for fewer of these control
variables in the matched-pairs sample than in the full sample. In our multivariate analyses discussed
below, we control for these differences in client characteristics in examining the difference (if any) in
abnormal accruals for Big 4 and Second-tier audit clients.
Once again, for brevity, we do not show correlation matrices. However, the highest pairwise corre-
lation between the test variable BIG4 and any of the control variables (in either the full or matched-
pair samples) was only 0.35 (with variable SIZE). As discussed below, we diagnose multicollinearity in
the regressions using variance ination factors (VIFs).
Table 6 panel A reports the regression model results used to test Hypothesis 2. Once again, in
both the regressions in Table 6 panel A, the VIFs are low which suggests that collinearity is not
likely to be a problem in interpreting the results. In particular, the VIFs for the test variable
BIG4 are very low (below 2). The control variables AUDTEN through IMPLICIT, where signicant,
have the predicted signs. The ndings for the control variables are consistent with prior research
(e.g., Ashbaugh et al., 2003; Wareld et al., 1995) which basically suggests that the relation be-
tween earnings management metrics and the control variables are not always clear cut with coef-
cients differing in signs across studies.
In the regressions for both the full and matched-pair samples, the test variable BIG4 is not signif-
icant.26 These results indicate that after controlling for various client characteristics that are likely to
inuence abnormal accruals, Big 4 audit clients have signed abnormal accruals that are no different from
those of Second-tier audit clients. Thus, the ndings suggest that Big 4 auditors and Second-tier auditors

25
In this analysis, the probability of receiving a going concern opinion for a Big 4 audit is 0.034 whereas the probability for a
Second-tier auditor is only 0.009. The increase in the probability is 0.024, which represents a nearly 3-fold increase in the
probability of receiving a going concern opinion. The probability values were derived by evaluating the standard normal
cumulative probability distribution model at the tted value of the probit model as described in fn. 23 above.
26
We evaluate the non-rejection of the null hypothesis using the approach described in Greenwald (1975). This approach entails
specifying a null hypothesis range based on the treatment effect magnitude that would be considered non-trivial, then calculating
the area under the posterior distribution that lies within the null range. This area represents the posterior probability of accepting
the null range (i.e., the probably that the null is true). In our case, we specify the null range based on a .015 to 0 range, where the
minimum value of .015 represents the treatment effect documented in prior research (i.e., Becker et al., 1998) and the maximum
value of 0 was chosen under the assumption that any adverse Big 4 treatment effect (i.e., greater earnings management among Big
4 as compared to Second-tier) would be non-trivial. For the matched-pair sample regression in Table 6 Panel A the area under the
posterior distribution that lies within the null range was 0.34 (i.e., a 0.34 probability in support of the null). Based on this analysis,
we conclude that our non-signicant results potentially are due to sampling error or inadequate power.
346 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

Table 5
Descriptive statistics (earnings management analysis).

Full sample Differences in means


(Big 4 vs. Second-tier)
Big 4 (n = 8074) Second-tier (n = 1236)
Mean Median Std. dev. Mean Median Std. dev.
ABACC 0.0064 0.0019 0.0894 0.0015 0.0029 0.1152 0.0049***
AUDTEN 9.6354 7.0000 8.5701 3.0542 2.0000 3.6829 6.5812***
SIZE 5.7819 5.8851 1.6400 4.0206 4.0608 1.5111 1.7613***
BM 0.4035 0.4220 1.0069 0.2832 0.4063 1.4847 0.1203***
DISTRESS 0.0543 0.0002 0.1839 0.1114 0.0004 0.2731 0.0572***
CFFO 0.0639 0.0841 0.1929 0.0032 0.0503 0.2750 0.0607***
GROWTH 0.1593 0.1019 0.3801 0.1467 0.0726 0.4673 0.0126
FINANCE 0.2901 0.0000 0.4538 0.3511 0.0000 0.4775 0.0611***
LEV 0.1876 0.1329 0.2379 0.2354 0.1102 0.4005 0.0478***
LITIGATION 0.4061 0.0000 0.4911 0.4749 0.0000 0.4996 0.0688***
IMPLICIT 0.0016 0.0144 0.9203 0.0506 0.0585 0.8964 0.0521*

Matched-pairs sample
Big 4 (n = 1077) Second-tier (n = 1077)
ABACC 0.0013 0.0002 0.1071 0.0005 0.0042 0.1131 0.0018**
AUDTEN 7.9359 6.0000 7.0558 3.0409 2.0000 3.7355 4.8951***
SIZE 4.5118 4.4880 1.5315 4.1403 4.1561 1.5019 0.3715***
BM 0.3288 0.3909 1.1367 0.3471 0.4177 1.3073 0.0183
DISTRESS 0.1074 0.0003 0.2685 0.0973 0.0004 0.2545 0.0102
CFFO 0.0290 0.0441 0.3120 0.0089 0.0528 0.2712 0.0379***
GROWTH 0.1770 0.0973 0.5138 0.1463 0.0779 0.4379 0.0307
FINANCE 0.3649 0.0000 0.4816 0.3426 0.0000 0.4748 0.0223
LEV 0.1865 0.0558 0.3414 0.2222 0.1106 0.3665 0.0357**
LITIGATION 0.4921 0.0000 0.5002 0.4782 0.0000 0.4998 0.0139
IMPLICIT 0.2640 0.3360 0.9916 0.0292 0.0108 0.8922 0.2348***

Notes: ABACC is the accruals-based earnings management metric based on signed asset-deated performance-adjusted
abnormal accruals. Please see Table 1 (panel B) for other variable denitions.
Consistent with Francis and Lennox (2008), the matched-pairs sample is formed by matching each Big 4 audit client with a
Second-tier audit client whose probability of selecting a Big 4 auditor is similar.
*
Denotes signicant difference in the means in a two-tailed test at the 0.10 levels.
**
Denotes signicant difference in the means in a two-tailed test at the 0.05 levels.
***
Denotes signicant difference in the means in a two-tailed test at the 0.01 levels.

are similar at restraining the clients ability to manipulate reported earnings, i.e., in restraining manage-
rial opportunism in terms of accruals-based earnings management.
For completeness, Table 6 panel B presents a separate analysis of income-increasing abnormal
accruals, i.e., observations with ABACC > 0 only. Managing earnings upward to meet or beat bench-
marks may be more common, and (as noted previously) may be of more concern to auditors. Note that
in Table 6 panel B, the model is estimated using a truncated regression approach since only observa-
tions with a positive value of the dependent variable ABACC are included (Greene, 1997).
In panel B also, the VIFs are low, and the control variables AUDTEN through IMPLICIT, where sig-
nicant, have the predicted signs. In the regressions for both the full and matched-pair samples, the
test variable BIG4 is not signicant.27 Once again, these results indicate that after controlling for various
client characteristics that are likely to inuence abnormal accruals, Big 4 audit clients have income-
increasing abnormal accruals that are no different from those of Second-tier rm audit clients. Thus,
the ndings suggest that Big 4 auditors and Second-tier auditors are similar at restraining the clients

27
For the matched-pair sample regression in Table 6 Panel B, the area under the posterior distribution that lies within the null
range was 0.30 (i.e., a 0.30 probability in support of the null). Based on this analysis, we conclude that our non-signicant results
potentially are due to sampling error or inadequate power.
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 347

Table 6
Regression Results (earnings management analysis) ABACC a b1 BIG4 b2 AUDTEN b3 SIZE b4 BM b5 DISTRESS b6 CFFO
b7 GROWTH b8 FINANCE b9 LEV b10 LITIGATION b11 IMPLICIT e.

Expected sign Full sample Matched-pairs sample


Coef. Robust t-stat p-Value VIF Coef. Robust t-stat p-Value VIF
Panel A. All observations
Intercept ? 0.0349 8.16 0.000 0.0507 5.86 0.000
BIG4 ? 0.0027 0.80 0.426 1.19 0.0027 0.38 0.701 1.25
AUDTEN ? 0.0003 2.72 0.007 1.15 0.0001 0.13 0.894 1.24
SIZE  0.0058 8.47 0.000 1.54 0.0089 5.39 0.000 1.31
BM  0.0030 2.22 0.013 1.20 0.0045 1.53 0.125 1.35
DISTRESS ? 0.1987 13.60 0.000 2.17 0.1947 6.99 0.000 2.37
CFFO  0.1534 9.30 0.000 1.57 0.1145 5.15 0.000 1.85
GROWTH + 0.0015 0.19 0.423 1.08 0.0009 0.07 0.945 1.09
FINANCE + 0.0031 1.68 0.953 1.70 0.0087 1.70 0.089 1.14
LEV + 0.0432 4.40 0.000 1.93 0.0409 1.81 0.071 1.97
LITIGATION + 0.0017 0.54 0.295 1.11 0.0051 0.99 0.322 1.07
IMPLICIT ? 0.0158 10.55 0.000 1.24 0.0099 4.25 0.000 1.26
Nobs 9310 2154
Adj. R2 14.59% 11.81%

Panel B. Observations with ABACC > 0 only


Intercept ? 0.1233 10.53 0.001 0.00 0.1466 8.14 0.001 0.00
BIG4 ? 0.0004 0.04 0.967 1.21 0.0040 0.35 0.728 1.29
AUDTEN ? 0.0005 1.09 0.276 1.15 0.0025 2.41 0.016 1.25
SIZE  0.0440 13.99 0.001 1.51 0.0446 8.80 0.001 1.23
BM  0.0110 3.59 0.000 1.20 0.0077 1.66 0.049 1.48
DISTRESS ? 0.1072 4.57 0.001 2.36 0.0432 1.40 0.162 2.78
CFFO  0.1657 10.79 0.001 1.58 0.1160 5.48 0.001 1.90
GROWTH + 0.0361 5.99 0.001 1.05 0.0330 3.80 0.000 1.05
FINANCE + 0.0549 7.43 0.001 1.11 0.0451 3.83 0.000 1.18
LEV + 0.0034 0.26 0.399 2.13 0.0013 0.08 0.469 2.34
LITIGATION + 0.0205 3.09 0.001 1.10 0.0142 1.35 0.089 1.07
IMPLICIT ? 0.0142 3.64 0.000 1.33 0.0078 1.26 0.206 1.36
Nobs 4547 1108
Adj. R2 24.82% 27.03%

Notes: The dependent variable ABACC is the client- and year-specic signed performance-adjusted abnormal accruals. Please
see Table 1 (panel B) for variable denitions. The model is estimated from cross-sectional observations for the period 2003
2006. Statistical inferences are based on robust t-statistics that are adjusted for residual correlation arising from pooling
cross-sectional observations across time, i.e., the t-statistics are based on White (1980) heteroskedasticity-adjusted robust
variance estimates that are adjusted for client and time clustering (Petersen, 2009). In Panel B, the model is estimated as a
truncated regression by maximum likelihood with statistical inferences based on the resulting covariance matrix (Greene, 1997,
pp. 954958). Reported signicance levels are based on two-tailed tests (where the sign is not predicted) and one-tail other-
wise. VIF denotes the variance ination factor.
Consistent with Francis and Lennox (2008), the matched-pairs sample is formed by matching each Big 4 audit client with a
Second-tier audit client whose probability of selecting a Big 4 auditor is similar.

ability to increase reported earnings, i.e., in restraining managerial opportunism in terms of income-
increasing earnings management. It is worth noting that recent research by Cohen et al. (2008) docu-
ments a decline in accruals-based earnings management and a shift towards real earnings management
in the post-SOX era which covers the time period of our study (20032006). Potentially, this shift may
explain our nding of no signicant difference in accruals-based earnings management between Big 4
and Second-tier auditors during 20032006.

4.3. Investor perspective (ex ante equity risk premium) analysis

Table 7 provides descriptive statistics for the dependent and explanatory variables in the ex ante
equity risk premium regressions for the full and matched-pair samples. Variable ravg, the dependent
variable in this analysis, is the client-specic ex ante equity risk premium, i.e., the excess of the ex ante
348 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

Table 7
Descriptive statistics (investor-perspective ex ante equity risk premium analysis).

Full sample Differences in means (Big 4 vs. Second-tier)


Big 4 (n = 4683) Second-tier (n = 329)
Mean Median Std. dev. Mean Median Std. dev.
ravg 0.0567 0.0503 0.0292 0.0647 0.0564 0.0348 0.0080***
BM 0.4348 0.3883 0.2588 0.4430 0.3924 0.2563 0.0082
LEV 0.1632 0.1330 0.1662 0.1305 0.0571 0.1600 0.0327***
SIZE 6.4232 6.4589 1.2561 5.4743 5.2346 1.1708 0.9489***
BETA 1.3433 1.2554 0.7374 1.2752 1.2378 0.8321 0.0681
SDROA 0.1120 0.0452 0.3518 0.1320 0.0659 0.3049 0.0200
LTG 0.2044 0.1638 0.1674 0.2457 0.1900 0.2242 0.0413***
REC RET 0.3404 0.1860 0.6513 0.3280 0.1860 0.6187 0.0124

Matched-pairs sample
Big 4 (n = 309) Second-tier (n = 309)
ravg 0.0562 0.0511 0.0306 0.0639 0.0563 0.0344 0.0077***
BM 0.4128 0.3780 0.2242 0.4462 0.3934 0.2601 0.0335*
LEV 0.1185 0.0459 0.1526 0.1330 0.0618 0.1628 0.0145
SIZE 5.7025 5.7087 1.2196 5.5245 5.2734 1.1779 0.1781*
BETA 1.2830 1.1768 0.7381 1.2904 1.2521 0.8203 0.0074
SDROA 0.1201 0.0621 0.1869 0.1173 0.0653 0.1828 0.0029
LTG 0.2405 0.1833 0.1996 0.2385 0.1850 0.2174 0.0020
REC_RET 0.3084 0.1475 0.6016 0.3406 0.2033 0.6297 0.0322

Notes: ravg is the client- and year-specic ex ante equity risk premium. Please see Table 1 panel C for other variable denitions.
Consistent with Francis and Lennox (2008), the matched-pairs sample is formed by matching each Big 4 audit client with a
Second-tier audit client whose probability of selecting a Big 4 auditor is similar.
*
Denotes signicant difference in the means in a two-tailed test at the 0.10 levels.
***
Denotes signicant difference in the means in a two-tailed test at the 0.01 levels.

cost of equity capital over the yield on the 10-year US Treasury bond. The descriptive statistics for var-
iable ravg (the average of the four ex ante equity risk premium metrics) reported in Table 7 are com-
parable to those reported by Dhaliwal et al. (2006).28
Other things being equal, the higher the ravg metric, the lower the clients accounting information
quality as perceived by investors. For variable ravg, the mean is smaller for Big 4 clients than for Sec-
ond-tier clients in both the full and matched-pair samples, indicating (on an univariate basis) that
investors perceive Big 4 clients as reporting higher quality accounting information. For a number of
the control variables BM through REC_RET also, the univariate means are different for Big 4 and Sec-
ond-tier clients, although (once again) the difference in means is signicant for fewer of these control
variables in the matched-pairs sample than in the full sample. In our multivariate analyses (discussed
below), we control for these differences in client characteristics in examining the difference (if any) in
the ravg metric for Big 4 and Second-tier audit clients.
Once again, for brevity, we do not show correlation matrices. However, the highest pairwise corre-
lation between the test variable BIG4 and any of the control variables in the ex ante equity risk pre-
mium analysis (for either the full or matched-pair samples) was only 0.18 (with variable SIZE). As
discussed below, we diagnose multicollinearity in the regressions using variance ination factors
(VIFs).
Table 8 reports the regression results used to test Hypothesis 3. Once again, in both the regressions
in the Table, the VIFs are very low (below 2) which suggests that collinearity is not likely to be a prob-
lem in interpreting the results. The control variables BM through REC_RET, where signicant, have the

28
For brevity, we do not report the mean, median, and standard deviation for the four client-specic ex ante equity risk premium
metrics (rgls, rct, rojn, and rmpeg). These descriptive statistics were comparable to those reported by Dhaliwal et al. (2006). The pair-
wise correlations among the ex ante equity risk premium metrics (rgls, rct, rojn, rmpeg, and ravg) were similar to those reported by
Dhaliwal et al. (2006).
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 349

Table 8
Regression results (investor-perspective ex ante equity risk premium analysis) r av g a b1 BIG4 b2 BM b3 LEV b4 SIZE
P
b5 BETA b6 SDROA b7 LTG b8 REC RET xx dxx INDUSTRYxx e.

Expected sign Full sample Matched-pairs sample


Coef. Robust t-stat p-Value VIF Coef. Robust t-stat p-Value VIF
Constant 0.0461 16.12 0.000 0.0101 8.22 0.000
BIG4 ? 0.0033 4.09 0.000 1.06 0.0060 6.07 0.000 1.20
BM + 0.0259 21.41 0.000 1.13 0.0219 6.15 0.000 1.25
LEV + 0.0385 11.16 0.000 1.29 0.0282 7.25 0.000 1.40
SIZE  0.0035 6.56 0.000 1.41 0.0035 4.45 0.000 1.72
BETA + 0.0019 4.24 0.000 1.17 0.0017 3.79 0.000 1.22
SDROA + 0.0018 3.35 0.001 1.11 0.0009 0.30 0.764 1.28
LTG + 0.1031 32.37 0.000 1.15 0.1075 72.00 0.000 1.24
REC_RET  0.0067 2.89 0.004 1.10 0.0086 3.12 0.002 1.17
Nobs 5012 618
Adj. R2 53.37% 61.49%

Notes: The dependent variable ravg is the client- and year-specic ex ante equity risk premium. Please see Table 1 panel D for
other variable denitions. The model is estimated from cross-sectional observations pooled across the period 20032006.
Statistical inferences are based on robust t-statistics that are adjusted for residual correlation arising from pooling cross-
sectional observations across time, i.e., the t-statistics are based on White (1980) heteroskedasticity-adjusted robust variance
estimates that are adjusted for client and time clustering (Petersen, 2009). Reported signicance levels are based on two-tailed
tests (where the sign is not predicted) and one-tail otherwise. Coefcient estimates for INDUSTRYxx are omitted for brevity. VIF
denotes the variance ination factor.
Consistent with Francis and Lennox (2008), the matched-pairs sample is formed by matching each Big 4 audit client with a
Second-tier audit client whose probability of selecting a Big 4 auditor is similar.

predicted signs and are consistent with prior research (e.g., Khurana and Raman, 2004, 2006). Note
also that the model includes industry dummies, although the industry-specic intercepts are not re-
ported in the Table for brevity. In the regressions, the test variable BIG4 is signicant with a negative
sign for both the full and matched-pairs samples.29 These ndings appear to be economically signi-
cant, i.e., the slope coefcients for variable BIG4 imply a reduction in the ex ante equity risk premium
of between 33 and 60 basis points for Big 4 (relative to Second-tier) audit clients. Overall, the results indi-
cate that after controlling for various client characteristics that are likely to inuence the client-specic
ex ante equity risk premium, Big 4 audit clients have a lower equity risk premium than Second-tier audit
clients. Thus, the ndings suggest that Big 4 auditors are perceived by investors as providing more cred-
ible accounting information for their clients than Second-tier auditors. Alternatively, as discussed previ-
ously, the lower equity risk premium for Big 4 clients may reect the deeper pockets, i.e., the greater
ability of the Big 4 auditor to share more of stockholder losses in the event of client failure.

5. Concluding remarks

In recent years, regulators and public interest groups have encouraged public companies to utilize
Second-tier auditors as an alternative to the Big 4 as a way of increasing the number of suppliers of
audit services for large clients. Still, whether or not Second-tier rm audit quality is similar to that
of the Big 4 remains an empirical question.
Our study examines audit quality for Big 4 and Second-tier auditors during 20032006. We utilize
the auditors propensity to issue a going concern audit report for distressed clients as a measure of
auditor independence and, by implication, audit quality. In addition, since the purpose of an audit
is to improve the credibility of nancial reporting, we utilize measures of nancial reporting quality
as observable proxies for audit quality. Thus, we utilize the clients signed performance-adjusted
abnormal accruals as an observable proxy for auditor quality. In addition, we utilize a market-based

29
We also analyzed the results using annual regressions, and the results hold based on Fama-Macbeth t-statistics. In particular,
the test variable BIG4 was signicant with a negative sign at the 0.01 level in the annual regressions.
350 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

representation of accounting information quality the client-specic ex ante equity risk premium as
an observable proxy for audit quality as perceived by investors. We control for auditor self-selection
bias using the matched-pair sample approach discussed by Francis and Lennox (2008). Still, we recog-
nize that our various regression models may not fully control for differences in client characteristics.
Hence, one cannot fully discount the possibility that the differences identied are due to differences in
client characteristics rather than auditor characteristics. We acknowledge this limitation of our paper.
Overall, our results suggest that nancial reporting quality is broadly similar for Big 4 and Second-
tier audit rm clients. We nd some evidence that relative to Second-tier auditors, Big 4 auditors have
a higher propensity to issue a going concern opinion for distressed clients. However, we nd that
signed (as well income-increasing) performance-adjusted abnormal accruals are similar for Big 4
and Second-tier audit rm clients. Consistent with prior research, this nding suggests that Big 4
and Second-tier auditors are equally effective in restraining aggressive and potentially opportunistic
reporting, i.e., similar in terms of mitigating accruals-based earnings management by their audit
clients.30
Finally, from an investor perspective, our analysis indicates that the ex ante equity risk premium is
lower for Big 4 clients than for Second-tier audit clients. These ndings imply that investors perceive
accounting information quality to be higher for Big 4 clients than for Second-tier audit rm clients.
Alternatively, the lower equity risk premium for Big 4 clients may be related to insurance reasons,
i.e., the Big 4 auditors deeper pockets (greater nancial resources) and ability to share more of
the settlement costs and shareholder losses in the event of an audit failure.31 Collectively, the evidence
we provide (albeit mixed) makes a contribution to the literature on audit quality and auditor choice, and
informs the current discourse on the viability of Second-tier auditors as an alternative to the Big 4.

Acknowldgements

Authors names are listed alphabetically. We gratefully acknowledge the helpful comments and
suggestions of two reviewers, Jere Francis, and participants at the 2008 AAA Annual Meeting, the
2008 EAA annual congress and workshops at the University of Missouri-Columbia, the University of
North Texas, the University of Texas at San Antonio, Hong Kong Polytechnic University, and Chul-
alongkorn University.

References

American Assembly, 2005. The Future of the Accounting Profession: Auditor Concentration, Columbia University, New York, NY.
Ashbaugh, H., LaFond, R., Mayhew, B., 2003. Do nonaudit services compromise auditor independence? Further evidence. The
Accounting Review 78 (3), 611639.
Barth, M., Elliott, J., Finn, M., 1999. Market rewards associated with increasing earnings patterns. Journal of Accounting Research
37, 387413.
Bartov, E., Givoly, D., Hayn, C., 2002. The rewards to meeting or beating earnings expectations. Journal of Accounting and
Economics 33, 173204.
Basu, S., 2004. What do we learn from two new accounting-based stock market anomalies? Journal of Accounting and
Economics 38, 333348.
Beaver, W., Kettler, P., Scholes, M., 1970. The association between market determined and accounting determined risk
measures. The Accounting Review (October) 654, 681.
Becker, C., Defond, M., Jiambalvo, J., Subramanyam, K., 1998. The effects of audit quality on earnings management.
Contemporary Accounting Research 15 (1), 124.
Bhattacharya, U., Daouk, H., Welker, M., 2003. The world price of earnings opacity. The Accounting Review 78 (3), 641678.
Botosan, C., 1997. Disclosure level and the cost of equity capital. The Accounting Review 72, 323349.
Botosan, C., Plumlee, M., 2005. Assessing alternative proxies for the expected risk premium. The Accounting Review 80 (1), 21
54.

30
Cohen et al. (2008) document that the passage of SOX in 2002 (which was intended to restore the integrity of nancial
statements and was followed by highly publicized enforcement action by the Department of Justice) resulted in a signicant
decline in accruals-based earnings management. The absence of a signicant difference in accruals-based earnings management
between Big 4 and Second-tier auditors during the period of our study (20032006) may be related to the fact that clients shifted
from using accruals-based to real earnings management after SOX (p. 785).
31
As noted previously, although the Second-tier rms have emerged as large international rms with correspondingly developed
brand name reputation capital, they are still less than a third the size of the smallest Big 4 rm. Consequently, their nancial
resources (such as invested partner equity) are likely considerably smaller than those of the Big 4 rms.
J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352 351

Botosan, C., Plumlee, M., 2002. A re-examination of disclosure level and the expected cost of equity capital. Journal of
Accounting Research 40, 2140.
Bowen, R., DuCharme, L., Shores, D., 1995. Stakeholders implicit claims and accounting method choice. Journal of Accounting
and Economics 20 (December), 255295.
Brennan, M., Subrahmanyam, A., 1996. Market microstructure and asset pricing: on the compensation for illiquidity in stock
returns. Journal of Financial Economics 41 (3), 441464.
Byrnes, N., 2005. The little guys doing large audits. Business Week (August 22), 39.
Carey, P., Simnett, R., 2006. Audit partner tenure and audit quality. The Accounting Review 81 (3), 653676.
Carson, E., 2009. Industry specialization by global audit rm networks. The Accounting Review 84 (2), 355382.
Chung, H., Kallapur, S., 2003. Client importance, nonaudit services, and abnormal accruals. The Accounting Review 78 (4), 931
956.
Claus, J., Thomas, J., 2001. Equity premia as low as three percent? Empirical evidence from analysts earnings forecasts for
domestic and international stock markets. Journal of Finance (October), 16291665.
Cohen, D., Dey, A., Lys, T., 2008. Real and accrual-based earnings management in the pre- and post-Sarbanes Oxley periods. The
Accounting Review 83, 757787.
Core, J., Guay, W., Verdi, R., 2008. Is accruals quality a priced risk factor? Journal of Accounting and Economics 46 (September),
222.
DeAngelo, L., 1981. Auditor size and audit quality. Journal of Accounting and Economics (December), 183199.
DeAngelo, H., DeAngelo, L., Skinner, D., 1994. Accounting choice in troubled companies. Journal of Accounting and Economics
18, 113143.
Dechow, P., Skinner, D., 2000. Earnings management: reconciling the views of accounting academics, practitioners, and
regulators. Accounting Horizons 14 (2), 235250.
DeFond, M., Jiambalvo, J., 1994. Debt covenant effects and the manipulation of accruals. Journal of Accounting and Economics
17, 145176.
DeFond, M., Raghunandan, K., Subramanyam, K., 2002. Do non-audit service fees impair auditor independence? Evidence from
going concern audit opinions. Journal of Accounting Research 40 (4), 12471274.
Dhaliwal, D., Heitzman, S., Li, O., 2006. Taxes, leverage, and the cost of equity capital. Journal of Accounting Research 44 (4),
691723.
DiPiazza, S., McDonnell, D., Parrett, W., Rake, M., Samyn, F., Turley, J., 2006. Global Capital Markets and the Global Economy: A
Vision from the CEOs of the International Audit Networks. <http://globalpublicpolicysymposium.com/CEO_Vision.pdf>.
Dopuch, N., Simunic, D., 1982. Competition in auditing research: an assessment. In: Fourth Symposium on Auditing Research,
University of Illinois, pp. 403450.
Easton, P., 2004. PE ratios, PEG ratios, and estimating the implied expected rate of return on equity capital. The Accounting
Review 79, 7395.
Fama, E., French, K., 1997. Industry costs of equity. Journal of Financial Economics (43), 153193.
Fields, L., Fraser, D., Wilkins, M., 2004. An investigation of the pricing of audit services for nancial institutions. Journal of
Accounting and Public Policy 23, 5377.
Francis, J., Lennox, C., 2008. Selection Models in Accounting Research, SSRN.com.
Francis, J., Philbrick, D., Schipper, K., 1994. Shareholder litigation and corporate disclosures. Journal of Accounting Research 32
(Autumn), 137164.
Francis, J., La Fond, R., Olsson, P., Schipper, K., 2004. Cost of capital and earnings attributes. The Accounting Review 79 (4), 967
1010.
Francis, J., La Fond, R., Olsson, P., Schipper, K., 2005. The market pricing of accruals quality. Journal of Accounting and Economics
39, 295327.
Francis, J., Maydew, E., Sparks, H., 1999. The role of Big 6 auditors in the credible reporting of accruals. Auditing, A Journal of
Practice and Theory 18 (2), 1734.
Gebhardt, W., Lee, C., Swaminathan, B., 2001. Toward an implied cost of capital. Journal of Accounting Research 39 (1), 135176.
Gode, D., Mohanram, P., 2003. Inferring the cost of capital using the OhlsonJuettner model. Review of Accounting Studies 8,
399433.
Government Accountability Ofce (GAO), 2008. Audits of Public Companies: Continued Concentration in Audit Market for Large
Public Companies Does Not Call for Immediate Action, GAO, Washington, DC.
Government Accountability Ofce (GAO), 2006. Sarbanes Oxley Act: Consideration of Key Principles Needed in Addressing
Implementation for Smaller Public Companies, GAO, Washington, DC.
Government Accountability Ofce (GAO), 2003. Public Accounting Firms: Mandated Study on Consolidation and Competition,
Washington, DC.
Graham, J., Harvey, C., Rajagopal, S., 2005. The economic implications of corporate nancial reporting. Journal of Accounting and
Economics 40, 373.
Grant Thornton, 2007. After Decades of Consolidation, the Onset of Sarbanes-Oxley and the Fall of Andersen, the Audit
Environment is Changing. <http://www.GrantThornton.com>.
Greene, W., 1997. Econometric Analysis, third ed. Upper Saddle River, NJ, Prentice Hall.
Greenwald, A., 1975. Consequences of prejudice against the null hypothesis. Psychological Bulletin 82 (1), 120.
Guay, W., Kothari, S., Shu, S., 2005. Properties of Implied Cost of Capital using Analysts Forecasts, SSRN.com.
Hail, L., Leuz, C., 2006. International differences in cost of equity: do legal institutions and securities regulation matter? Journal
of Accounting Research 44, 485531.
Heckman, J., 1979. The sample selection bias as a specication error. Econometrica 47 (1), 153162.
Heninger, W., 2001. The association between auditor litigation and abnormal accruals. The Accounting Review 76 (1), 111126.
Hribar, P., Nichols, D., 2007. The use of unsigned earnings quality measures in tests of earnings management. Journal of
Accounting Research 20, 10171053.
Johnson, S., Ovtober 1, 2007. PCAOB gives Crowe Chizek Higher Marks, CFO.Com.
352 J.P. Boone et al. / J. Account. Public Policy 29 (2010) 330352

Johnson, V., Khurana, I., Reynolds, J., 2002. Audit-rm tenure and the quality of nancial reports. Contemporary Accounting
Research 19 (4), 637660.
Jones, J., 1991. Earnings management during import relief investigations. Journal of Accounting Research 29 (2), 193228.
Jopson, B., 2006. Audit Firms Fight Back against Big 4, Financial Times, March 27, p. 20.
Khurana, I., Raman, K., 2006. Do investors care about the auditors economic dependence on the client? Contemporary
Accounting Research 23 (4), 9771016.
Khurana, I., Raman, K., 2004. Litigation risk and the nancial reporting credibility of Big 4 versus non-Big 4 audits: evidence
from Anglo-American countries. The Accounting Review 79 (2), 473495.
Kothari, S., Leone, A., Wasley, C., 2005. Performance matched discretionary accrual measures. Journal of Accounting and
Economics 39 (1), 163197.
La Porta, R., 1996. Expectations and the cross-section of stock returns. Journal of Finance 51, 17151742.
Lambert, R., Leuz, C., Verrecchia, R., 2007. Accounting information, disclosure, and the cost of capital. Journal of Accounting
Research 31 (May), 385420.
Lang, M., Lundholm, R., 1993. Cross-sectional determinants of analyst ratings of corporate disclosures. Journal of Accounting
Research 31 (Autumn), 246271.
Matsumoto, D., 2002. Managements incentives to avoid negative earnings surprises. The Accounting Review 77 (3), 483514.
Menon, K., Williams, D., 2004. Former audit partners and abnormal accruals. The Accounting Review 79 (4), 10951118.
Myers, J., Myers, L., Omer, T., 2003. Exploring the term of auditorclient relationship and the quality of earnings: a case for
mandatory auditor rotation? The Accounting Review 78 (3), 779799.
Nelson, M., 2006. Ameliorating conicts of interest in auditing: effects of recent reforms on auditors and their clients. Academy
of Management Review 31 (1), 3042.
Palmrose, Z., 1988. An analysis of auditor litigation and audit service quality. The Accounting Review 63, 5573.
Petersen, M., 2009. Estimating standard errors in nance panel data sets: comparing approaches. Review of Financial Studies 22
(1), 435480.
Public Oversight Board (POB), 2002. Final Annual Report 2001, Stamford, CT.
Rangan, S., 1998. Earnings management and the performance of seasoned equity offerings. Journal of Financial Economics 50,
101122.
Reynolds, J., Francis, J., 2001. Does size matter? The inuence of large clients on ofce-level auditor reporting decisions. Journal
of Accounting and Economics (30), 375400.
Rosenbaum, P., Rubin, D., 1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70
(1), 4155.
Simunic, D., Stein, M., 1987. Product Differentiation in Auditing: Auditor Choice in the Market for Unseasoned New Issues. The
Canadian Certied General Accountants Research Foundation, Vancouver, Canada.
Stein, J., 1989. Efcient capital markets, inefcient rms: a model of myopic corporate behavior. The Quarterly Journal of
Economics (November), 655669.
Subramanyam, K., 1996. The pricing of discretionary accruals. Journal of Accounting and Economics 22, 249281.
Sweeney, A., 1994. Debt-covenant violations and managers accounting responses. Journal of Accounting and Economics 17 (3),
281308.
US Chamber of Commerce, 2006. Auditing: A Profession at Risk. US Chamber of Commerce, Washington, DC.
US Department of the Treasury, 2008. Final Report of the Advisory Committee on the Auditing Profession. Department of the
Treasury, Washington, DC.
Wareld, T.D., Wild, J.J., Wild, K.L., 1995. Managerial ownership, accounting choices, and informativeness of earnings. Journal of
Accounting and Economics 20 (1), 6191.
White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity.
Econometrica 48, 817838.
Zmijewski, M., 1984. Methodological issues related to the estimation of nancial distress. Journal of Accounting Research 22
(Suppl.), 5982.