-
Managerial Auditing JournalBig 4 fee premium and audit quality:
latest evidence from UK listed companiesDomenico Campa
Article information:To cite this document:Domenico Campa,
(2013),"Big 4 fee premium and audit quality: latest evidence from
UK listedcompanies", Managerial Auditing Journal, Vol. 28 Iss 8 pp.
680 - 707Permanent link to this
document:http://dx.doi.org/10.1108/MAJ-11-2012-0784
Downloaded on: 08 October 2014, At: 08:20 (PT)References: this
document contains references to 67 other documents.To copy this
document: [email protected] fulltext of this
document has been downloaded 1227 times since 2013*
Users who downloaded this article also downloaded:Rani Hoitash,
Ariel Markelevich, Charles A. Barragato, (2007),"Auditor fees and
audit quality", ManagerialAuditing Journal, Vol. 22 Iss 8 pp.
761-786Ahsan Habib, Haiyan Jiang, Donghua Zhou, (2014),"Audit
quality and market pricing of earningsand earnings components in
China", Asian Review of Accounting, Vol. 22 Iss 1 pp. 20-34
http://dx.doi.org/10.1108/ARA-05-2013-0034Robert Houmes, Maggie
Foley, Richard J. Cebula, (2013),"Audit quality and overvalued
equity", AccountingResearch Journal, Vol. 26 Iss 1 pp. 56-74
Access to this document was granted through an Emerald
subscription provided by 434496 []
For AuthorsIf you would like to write for this, or any other
Emerald publication, then please use our Emerald forAuthors service
information about how to choose which publication to write for and
submission guidelinesare available for all. Please visit
www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.comEmerald is a global
publisher linking research and practice to the benefit of society.
The companymanages a portfolio of more than 290 journals and over
2,350 books and book series volumes, as well asproviding an
extensive range of online products and additional customer
resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The
organization is a partner of the Committeeon Publication Ethics
(COPE) and also works with Portico and the LOCKSS initiative for
digital archivepreservation.
*Related content and download information correct at time of
download.
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
Big 4 fee premium and auditquality: latest evidence from UK
listed companiesDomenico Campa
School of Business, Trinity College Dublin, Dublin, Ireland
Abstract
Purpose Using the most recent observations (2005-2011) from a
sample of UK listed companies,This paper aims to investigate
whether Big 4 audit firms exhibit a fee premium and, if this is
thecase, whether the premium is related to the delivery of a better
audit service.
Design/methodology/approach Univariate tests, multivariate
regressions and twomethodologies that control for self-selection
bias are used to answer the proposed researchquestions. Data are
collected from DataStream.
Findings Findings provide consistent evidence about the
existence of an audit fee premiumcharged by Big 4 firms while they
do not highlight any significant relationship between audit
qualityand type of auditor with respect to the audit quality
proxies investigated.
Research limitations/implications Evidence from this paper might
signal the need forlegislative intervention to improve the
competitiveness of the audit market on the basis that
itsconcentrated structure is leading to excessive fees for Big 4
clients. Findings might also enhance Big4 client bargaining power.
However, as the paper analyses only one country, generalizability
of theresults might be a limitation.
Originality/value This study joins two streams of the extant
literature that investigate theexistence of a Big 4 audit fee
premium and different levels of audit quality among Big 4 and
non-Big4 clients. Evidence supports the concerns raised by the UK
House of Lords in 2010 that theconcentrated structure of the audit
market could be the driver of excessive fees for Big 4 clients as
itdoes not find differences in audit quality between Big 4 and
non-Big 4 clients.
Keywords Audit fee premium, Big 4, Audit quality, Discretionary
accruals, Accounting conservatism,Value relevance, Auditing,
Auditors
Paper type Research paper
1. IntroductionSince DeAngelos (1981) research, academics have
been interested in investigatingwhether the audit service provided
by different audit firms could be, in some way,differentiated.
Studies have suggested that larger audit firms provide higher
qualityauditing because they have more expertise than smaller
competitors as they usuallydeal with larger clients from different
industries which enhances the level of auditorsskills (OKeefe and
Westort, 1992). Bigger audit firms also have more incentives
toprovide better auditing as they have a strong brand reputation to
maintain given theirlarge base of clients (Dopuch and Simunic,
1980; Francis and Wilson, 1988).
The current issue and full text archive of this journal is
available at
www.emeraldinsight.com/0268-6902.htm
The author would like to thank Professor Pearse Colbert and Dr
Gerard McHugh for their helpand advices, the editors and the
editorial assistant of the journal for their support and,
inparticular, two anonymous reviewers for their invaluable help,
comments and suggestions thatsignificantly improved this paper. All
remaining errors are the authors own.
Managerial Auditing JournalVol. 28 No. 8, 2013pp. 680-707q
Emerald Group Publishing Limited0268-6902DOI
10.1108/MAJ-11-2012-0784
MAJ28,8
680
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
Within this group of bigger audit firms, the Big N have been
usually consideredseparately by the extant literature and compared
with all the other companies thatprovide the same service. This is
due to the fact that the group of the Big N has adominant position
in the audit market, especially among listed companies.
Despiteexceptions, the majority of studies indicate that the Big N
provide better auditing thanother audit firms as their clients
exhibit, for example, less earnings management thannon-Big N
clients (Becker et al., 1998). This better service comes at a cost
to the clients;indeed, it has been shown that the Big N exhibit a
fee premium in comparison withtheir smaller competitors
(Gonthier-Besacier and Schatt, 2007).
A recent study (Lawrence et al., 2011) has made a significant
contribution to thistopic. It suggests that differences in audit
quality, measured by proxies for clientsearnings quality, between
Big and non-Big audit firms might not depend on the type ofauditor
but could be the reflection of their respective clients
characteristics. Indeed,Big N audit firms serve bigger companies
that are under closer media scrutiny; havemore resources to invest
in implementing better accounting systems and strongerinternal
controls; could appoint highly qualified directors in their audit
committees;and can hire highly skilled professionals in their
internal audit departments. Thus, theimprovement in the quality of
their earnings might be the consequence of these factorsrather than
the conduct of the audit by a Big N firm.
Furthermore, replying to a call for evidence from the Economic
Affairs Committeeof the UK House of Lords in 2010, some of the
interviewees expressed their belief in theexistence of a Big 4
audit premium that related to the oligopolistic structure of the
auditmarket rather than to the delivery of a superior service. The
report issued as result of thiscall for evidence indicates that
some institutional investors believe the (audit) market(in the UK)
is not competitive and that most witnesses believe that the
dominance ofthe Big 4 limits competition and choice in the audit
market (House of Lords, 2010b).
In addition, the possible lack of a relationship between fee
premium and better auditquality might also be due to the fact that,
after well-known accounting scandals(e.g. Worldcom, Enron,
Parmalat), regulatory reforms have led to an improvement in
themonitoring of auditor performance in the conduct of audits. The
Sarbanes-Oxley act,national regulations and local corporate
governance codes set more rigid requirements foraudit firms such as
greater emphasis on independence, firm and partner rotation.
Theseinstruments, together with the implementation of updated
international standards onauditing, might have made the quality of
the audit service more homogeneous across auditfirms.
Using a sample of companies listed on the UK stock market and
the most recentdata more specifically, observations from annual
reports prepared under IFRS(2005-2011) the aim of this paper is to
provide evidence on whether the group of theBig N audit firms (Big
4, in this case) exhibits a fee premium and, if this is the
case,whether the premium is related to the delivery of a better
audit service. Threedeterminants of clients earnings quality are
used to measure audit quality: earningsmanagement, accounting
conservatism and the value relevance of earnings.
After using two methodologies that control for self-selection
bias the Heckman(1976) procedure and propensity-score matching
models and controlling for severalfactors that affect the level of
the audit fees, findings indicate that the Big 4 do charge afee
premium to their clients. This premium is not followed by an
improvement in the
Big 4 feepremium and
audit quality
681
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
level of service provided as multivariate analyses highlights
that audit quality is notbetter among Big 4 clients with respect to
the three proxies investigated.
Findings from this study have several implications. They support
the concerns of theUK House of Lords that the oligopolistic
structure of the audit market results inexcessive fees charged by
the audit companies that dominate it. Furthermore, theyindicate
that these higher fees are not associated with any improvement in
the quality ofthe audit service delivered. This might signal the
need for legislative intervention aimedat enhancing the
competitiveness of the audit market. Results might also increase
Big 4client bargaining power at the moment of contracting audit
fees as firms would not bewilling to pay a premium to the Big 4,
especially if they were aware that it is not followedby an
enhancement of the quality of the audit. This would favour smaller
audit firmswhich do not charge any premium and provide the same
level of audit quality.
The rest of the paper is organized as follows. Section 2 frames
the study into theextant literature and presents the research
questions. Section 3 describes themethodology used to gather
evidence in order to answer the research questions andthe methods
employed to control for self-selection bias. It also details the
sampleselection procedure. Section 4 discusses the empirical
results and, finally, Section 5concludes this research by
highlighting its main implications and limitations.
2. Background and research questionsA consistent body of
literature in the past investigated whether all audit firms deliver
thesame level of audit service. DeAngelo (1981) was one of the
first to provide evidenceagainst the assertion made by regulators
and small audit firms that audit firm size doesnot affect audit
quality. Indeed, she suggests that bigger audit firms have more
tolose by failing to report a discovered breach in clients records.
Big audit firms candeliver better service as their employees engage
in greater degree of specialization, auditteams backgrounds are
more extensive and they offer a higher level of
continuingprofessional education (OKeefe and Westort, 1992). They
also need to maintain theirestablished brand reputation (Francis
and Wilson, 1988) and are able to put morepressure on management,
given their larger client portfolios (Lys and Watts, 1994).
Several studies have examined the link between bigger audit
firms (usually identifiedby the Big N) and audit quality, using
measures of clients earnings quality as proxies foraudit quality.
Despite some exceptions ( Jeong and Rho, 2004), most previous
studies finda direct association between earnings quality and the
conduct of audit by a Big N firm.There is evidence that non-Big 6
clients exhibit discretionary accruals that increaseincome
relatively more than Big 6 clients do (Becker et al., 1998). In the
same way, theconduct of audit by a Big 5 audit firm is associated
with lower discretionary accrualsamong companies going for an
initial public offering (Chen et al., 2005), as well ascompanies
making seasoned equity offerings (Zhou and Elder, 2008). Francis et
al. (1999)provide evidence that the presence of the Big 6 mitigates
aggressive reporting as theirclients exhibit higher total accruals
but lower discretionary accruals. Krishnan (2003)suggests that
there is a greater association between discretionary accruals and
futureearnings for Big 6 clients in comparison to those audited by
non-Big 6 audit firms. Thehigher audit quality provided by the Big
N comes at a price. Indeed, evidence indicatesthat these companies
charge an audit fee premium, as found in many countries such asthe
UK (Ireland and Lennox, 2002), Australia (Craswell et al., 1995),
Hong Kong(DeFond et al., 2000), France (Gonthier-Besacier and
Schatt, 2007).
MAJ28,8
682
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
Recently, there has been a school of thought that linked the
perception of a feepremium applied by the Big 4, in the UK, to the
concentration of the audit marketrather than to the provision of a
better audit service, as highlighted by some of therespondents to a
call for evidence from the House of Lords in 2010. More
specifically,on 27 July 2010, the Economic Affairs Committee of the
UK House of Lords (2010a)issued a call for evidence which states
that:
[. . .] audit is dominated globally by the Big Four. The narrow
field of choice raises concernsabout competition and the quality of
audited accounts, and about possible conflict of interestbetween
audit and consulting arms.
The attached questionnaire included the following question
(number 2): Does a lack ofcompetition mean clients are charged
excessive fees? (House of Lords, 2010a). Theevidence from the
answers to this question is not straightforward. While some of
therespondents state that there is no evidence of higher audit fees
charged by the Big 4(House of Lords, 2010c, pp. 6, 34, 56, 62, 65),
others believe, instead, that they do chargea premium (House of
Lords, 2010c, pp. 2, 28). However, many of the respondents
alsoindicate that there is not enough evidence to permit a precise
answer to this question,so a study that investigates this further
would appear relevant.
In addition, recent studies have shown that the link between
audit quality and thepresence of a Big N audit firm is not as
straightforward as it might seem, and that thisrelationship might
depend on other factors such as the presence of auditors
withindustry-related expertise (Francis, 2004) or on client
characteristics (Lawrence et al.,2011). As a matter of fact, the
latter study, which uses US companies and three proxiesfor audit
quality (discretionary accruals, the ex ante cost-of-equity capital
and analystforecast accuracy), provides evidence that there is no
difference between the presenceof Big 4 and non-Big 4 audit firms
with respect to the three proxies investigated.Lawrence et al.
indicate that the differences found between Big 4 and non-Big 4
clientslargely reflect client characteristics and, more
specifically, size (more differentiated andbigger among Big 4 audit
firms).
The findings of US studies cannot be extended to the UK. Indeed,
there are severaldifferences in terms of auditing and financial
reporting regulation between these twoinstitutional settings. For
example, Li et al. (2009), referring to Huijgen and
Lubberink(2005), state that the Security Exchange Commission (SEC)
in the US engages inproactive reviewing of registrants accounts to
penalise companies for earningsmanipulation. This approach is
believed to be more influential than that of theFinancial Reporting
Review Panel (FRRP) in the UK, which operates in a responsivemode,
doing little or no monitoring of its own and only responding to
complaints madeto its. Li et al. (2009) recognise also that there
is a different level of litigation forauditors in the USA and the
UK. They indicate that, because class actions arepermitted in the
USA, there is a huge pool of plaintiffs for US auditors. In
addition,Hughes and Snyder (1995) state that, as each party is
liable for their own legal fees inthe USA, little disincentive
exists for instigating legal action against an audit firm.
Bycontrast, in the UK the losing party bears all the fees for
taking a matter to court andthis can act as a deterrent (Coffee,
1999). Finally, in accordance with Armour et al.(2002) and Li et
al. (2009) point out that in the UK litigation risk and frequency
oflawsuits against auditors are much lower than the USA especially
because of firmdebt structure. They state that UK firms tend to
obtain more private loans from banks
Big 4 feepremium and
audit quality
683
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
and have more concentrated debt structures than is typical in
the US which has a widepublic debt market.
All these differences affect the perceived audit risk in the two
institutional settingsdiscussed which might influence both the
level of the audit fees charged to clients andaudit strategies,
which in turn has an impact on audit quality.
Bearing all these in mind, the main research questions of this
study can be stated asfollows:
RQ1. Do Big 4 audit firms charge an audit fee premium to UK
listed companies incomparison with their smaller competitors?
RQ2. Is the presence (absence) of an audit fee premium related
to the delivery of higher(comparable) audit quality?
3. Methodology3.1 Investigation of the audit fee premiumThe
existence of a fee premium among Big 4 clients is investigated
using aregression model which relates the amount of audit fees
charged by audit firms to theirclients to several control
variables, as explained by the following equation (1):
LNFEEit a b1BIG4it b2MBV it b3LEV it b4LOSSit b5SIZEit
b6INVRECit b7ROAit b8QUICKRATIOit 1it 1
where:
LNFEEit natural logarithm of audit fees.BIG4it 1 for Big 4
clients and 0 otherwise.MBVit market to book ratio.LEVit leverage
measured as total liabilities over total equity.LOSSit 1 when firm
i reports a net loss in year t, and 0 otherwise.SIZEit natural
logarithm of total assets.INVRECit sum of inventories and
receivables divided by total assets of
firm i in year t.
ROAit return on asset calculated as operating profit over total
assets.QUICKRATIOit quick ratio of company i in year t.
The variable BIG4 is included to test the existence of an audit
fee premium amongthe group of the Big 4.The existence of a premium
would be evidenced by a positiveand significant coefficient b1.
Other variables are included to control for additional factors
that may affect theamount of the audit fees. As other studies (Choi
and Wong, 2007; Choi et al., 2008, 2010)suggest that SIZE, MBV and
INVREC are included to control, respectively, for clientsize,
growth and complexity. SIZE affects the number of hours needed to
complete theaudit, so a positive coefficient related to this
variable is expected. Higher fees are
MAJ28,8
684
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
expected to be sought from firms with higher levels of growth
(Reynolds et al., 2004) soa positive b2 is predicted. A positive
relation is also expected between fees and thecomplexity of clients
business; for this reason a positive and significant b6 should
beobserved. Following other research (Simunic, 1980; Seetharaman et
al., 2002; Antle et al.,2006), the model also includes the
variables LEV, LOSS and QUICKRATIO to measureclient-related
litigation risks to be tolerated by auditors and ROA to control
foroperating performance. Simunic and Stein (1996) suggest that
auditors charge higherfees to risky clients so a positive
(negative) coefficient associated with the variablesLEV, LOSS and
QUICKRATIO (ROA) is expected.
3.2 Measures of audit qualityThree determinants of clients
earnings quality are used as proxies for audit quality:
(1) earnings management;
(2) accounting conservatism; and
(3) value relevance of earnings.
3.2.1 Earnings management. As far as earnings management is
concerned, theanalysis focuses on the most widely used measure of
this dimension: discretionaryaccruals.
Discretionary accruals are based on abnormal working capital
accruals (AWCA)following DeFond and Park (2001) methodology. Kim et
al. (2003), have observed thatalthough previous research has widely
used discretionary accruals under Jones (1991)model or its other
variants, these models have been criticized as the
parameterestimates are biased and measurement errors associated
therewith could potentiallyinduce erroneous conclusions about the
existence of earnings management (Bernardand Skinner, 1996; Guay et
al., 1996; Healy, 1996). The methodology used by DeFondand Park
(2001) is independent from potential measurement errors associated
withJones (1991) model parameters.
DeFond and Park (2001) estimate abnormal working capital
accruals (AWCA) asthe difference between the current years realized
working capital accruals and theexpected level of working capital
accruals, using the following formula:
AWCAit WCit 2 WCit21=Sit21*Sitwhere:
WCit operating working capital. It is calculated as current
assets(DataStream/Worldscope code: WC02201) after subtracting cash
andcash equivalent (WC02005), less current liabilities (WC03101)
net of thecurrent portion of long term debt (WC18232).
Sit net sales or revenue (WC01001).AWCA captures the deviation
of the current years working capital accruals from thenormal level
of working capital accruals required to support current sales
activitiesand is interpreted as an outcome of opportunistic
earnings management (Kim et al.,2003). DeFond and Park also find
their measure to be a more powerful test incomparison to a test
that uses total accruals. The focus on working capital accruals
is
Big 4 feepremium and
audit quality
685
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
further supported by other research that suggests that
management has the mostdiscretion over such accruals (Becker et
al., 1998; Ashbaugh et al., 2003).
AWCAs calculated above are divided by total assets (WC02999) to
adjust for firmsize. The absolute value of AWCA is used as the
focus of the research on earningsmanagement per se rather than
income-increasing or income-decreasing decisions. Toremove
potential bias due to the presence of outliers, abnormal accruals
are winsorisedusing a top 98 per cent winsorisation: all the data
above the 98th percentile are set tothe 98th percentile. No actions
are taken at the bottom percentiles because abnormalaccruals are
considered in their absolute values; consequently, figures very
close tozero cannot be seen as outliers.
AWCAs are finally included in the following regression model (2)
to control forother factors that might affect earnings
management:
AWCAit a b1BIG4it b2CFOit b3LEV it b4SIZEit b5ROAit b6EISSUEit
b7DISSUEit b8MBV it b9GROWTHit 1it 2
where:
AWCAit winsorised absolute value of the abnormal working capital
accruals.BIG4it 1 for Big 4 clients and 0 otherwise.CFOit cash flow
from operations divided by total assets.LEVit leverage measured as
total liabilities over total equity.SIZEit natural logarithm of
total assets.ROAit return on asset calculated as operating profit
over total assets.EISSUEit annual change in shareholder
equity.DISSUEit annual change in liabilities.MBVit market to book
ratio.GROWTHit annual change in net sales.
The sign and magnitude of the coefficient b1 associated with
BIG4 providesinformation about the relation between earnings
management and auditor type. Inparticular, a positive (negative)
coefficient means that Big 4 audit firms are associatedwith more
(less) earnings management and provides evidence of worse (better)
auditquality, as it would indicate, ceteris paribus, higher (lower)
discretionary accrualsamong Big 4 clients.
A set of control variables is included in the regression to
control for other firm-levelfactors that can influence earnings
management. Earlier studies have found thatfinancial leverage (LEV)
is positively related to earnings management (Dechow et al.,1995).
Cash flow from operations (CFO) and return on assets (ROA) are
included in themodel to control for extreme performance, which may
affect the level of earningsmanagement (Kothari et al., 2005).
Growth of the firm can affect the extent of earningsmanagement
(Carey and Simnett, 2006); for this reason the variables GROWTH
andMBV are introduced. Shan et al. (2011) show that failure to
control for changes infinancing can result in significant earnings
management measurement errors and
MAJ28,8
686
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
erroneous inferences. DISSUE and EISSUE are included in the
model to control fordebt and equity issuance, respectively. SIZE,
measured as the natural logarithm oftotal assets, is included as
earnings management is negatively related to firmdimension, which
also captures political costs (Bartov et al., 2000; Park and Shin,
2004).
3.2.2 Accounting conservatism. Beekes et al. (2004) describe
accountingconservatism as the asymmetric timeliness in earnings
between bad and goodnews. They also indicate that:
[. . .] prior research suggests that managers have significant
incentives to disclose bad newson a timely basis, for example to
protect themselves against potential legal liability (Skinner,1994;
Trueman, 1997). Hence, there is an expectation that bad news will
be incorporatedquickly into earnings.
Using a sample of UK companies and analysing the effect of
outside directors onaccounting conservatism, Beekes et al. (2004)
provide evidence that firms that areactively monitored are expected
be more conservative and report bad news on an evenmore timely
basis. On the other hand, they also expect that management might
haveincentives to accelerate the recognition of good news for
opportunistic motivations, suchas an increase in their compensation
(Healy, 1985). Effective monitoring should alsoreduce this tendency
(Beekes et al., 2004). As Big 4 audit firms could be perceived
asstronger monitors, similar expectations to those provided by
Beekes et al. (2004) can bepredicted.
Following their methodology, an extension of Basu (1997) model
is used toinvestigate accounting conservatism. It is represented by
the following equation (3):
Def Eit a b1RETit b2NEGit b3RET*NEGit b4BIG4it b5RET*BIG4it
b6NEG*BIG4it b7RET*NEG*BIG4it 1it 3
where:
Def (E)it earnings per share scaled by prior year-end
price.RETit 12-month raw returns beginning eight months before the
fiscal year-end
and ending four months after the year-end.
NEGit 1 if returns are negative and 0 otherwise.BIG4it 1 for Big
4 clients and 0 otherwise.
In accordance with Beekes et al. (2004), under conservative
accounting earnings willhave a higher sensitivity to bad news as
compared with good news. However, asexplained above the extent of
accounting conservatism in accounting earnings mightdiffer
depending on the strength of monitors. If it is believed that the
Big 4 better monitorclients activities, the latter are expected to
incorporate bad news into earnings on a moretimely basis. In model
(3), b7 captures the sensitivity to bad news (exclusive of the
effectof good news) for Big 4 clients. For this reason, if they are
more conservative thannon-Big 4 clients, the coefficient b7 is
expected to be positive and significant.
Regarding good news, Beekes et al. (2004) indicate that there is
a tendency formanagers to emphasise their availability for their
own bonus and promotionalprospects and that firms under stronger
monitoring activity might adopt aconservative approach to recording
good news in earnings because of the greater
Big 4 feepremium and
audit quality
687
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
constraint placed on managers opportunism. On the hypothesis
that a Big 4 firmexercises stronger monitoring activity, a more
conservative approach to therecognition of good news in earnings is
expected among Big 4 clients. In model (3),the timeliness of good
news is captured by the coefficient b1 in relation to non-Big
4clients, while the interaction term b5 captures the marginal
timeliness effect for Big 4clients. If the former are aggressive
reporters, the coefficient b1 will be significantlylarger than (b1
b5), therefore b5 is expected to be negative and significant.
3.2.3 Value relevance of earnings. The third measure of audit
quality takes intoaccount the value relevance of net incomes using
the methodology developed byOhlson (1995), which models the value
of a firm as a function of its book value andearnings:
Pit a b1BV it b2Eit b3BIG4it b4BV*BIG4it b5E*BIG4it 1it
4where:
Pit price of a share for a firm six months after the fiscal
year-end.BVit firm book value per share.Eit firm earnings per
share.BIG4it 1 for Big 4 clients and 0 otherwise.
This model highlights how the market perceives the information
content of netincomes disclosed by Big 4 clients. If earnings
included in their financial statements aremore informative than
those reported by non-Big 4 clients, the coefficient b5 should
bepositive and significant.
Although the variable of interest is the net income of the
company, the model alsoincludes the book value per share as its
omission could cause model misspecificationproblems (Collins et
al., 1999).
3.3 Control for self-selection biasIn all models reported above,
Big 4 and non-Big 4 clients are distinguished by adummy variable,
BIG4. The way this variable is included in the models assumes
that:
[. . .] auditors are randomly allocated to client firms, which
rationalizes the inclusion of BIG4as an exogenous variable in the
regression. However, it is widely accepted that clientsself-select
their auditor and, from an econometric point of view,
self-selection might introducebias when using standard OLS (Chaney
et al., 2004).
As in many other studies that investigate audit fees and audit
quality (Chaney et al.,2004), I control for self-selection bias. I
use two methodologies that address this issue:the two-stage model
in accordance with the procedure developed by Heckman (1979)and the
propensity-score matching models introduced by Rosenbaum and
Rubin(1983).
3.3.1 Heckman (1979) procedure. The first control for
self-selection bias is based onthe two-stage model following
Heckmans (1979) procedure. I will also deal with themain
limitations of this methodology, as highlighted in detail by Lennox
et al. (2012),later in Section 3.3.1.1.
In the first stage, the following logistic regression equation
(5) is used to estimatethe probability of selecting a Big 4 audit
firm. In accordance with Chaney et al. (2004),
MAJ28,8
688
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
the independent variables are proxies for client size, growth,
risk and performance andare the same as those used in model
(1):
BIG4it a b1MBV it b2LEV it b3LOSSit b4SIZEit b5INVRECit b6ROAit
b7QUICKRATIOit 1it
5
Heckmans (1979) methodology requires that the inverse Mills
ratio (INVMILLS) fromequation (5) be included in all regression
models because it controls for potentialself-selection bias in the
second stage. Accordingly, this ratio will be calculated using
Stataand added to all models presented thus far. The significance
of the coefficient on INVMILLSindicates the importance of
controlling for self-selection bias (Chaney et al., 2004).
3.3.1.1 Dealing with the limitations of the two stage Heckman
(1979) procedure.A detailed analysis of the two stage Heckman
(1979) procedure is presented byLennox et al. (2012). In their
paper, the authors report the main weaknesses of thismethodology
and, reviewing 75 articles that employ it, they investigate whether
andhow these studies dealt with the limitations highlighted.
Among others, Lennox et al. (2012) identify two main problems.
First of all, theyexplain that the sign and the significance of the
coefficients of the models in the secondstage might be affected by
multicollinearity caused by the introduction of the inverseMills
ratio as an additional independent variable. They also point out
that thisproblem has been often underestimated by the extant
literature, as 72 out of the 75studies reviewed do not include any
multicollinearity diagnosis and discussion.Second, they indicate
that, in the absence of exclusion restrictions, the results
dependentirely on the inverse Mills ratio nonlinearity.
Accordingly, in their Section VIsuggestions for better
implementation of selection models, Lennox et al. (2012)highlight
that one of the best papers that uses selection models is that
prepared byFeng et al. (2009), which emphasises the importance of
having at least one exclusionrestriction, reports diagnostic tests
for multicollinearity and investigates whether thefindings are
sensitive to alternative model specifications.
Taking all the above into account, I aim to follow these best
practices and I report anddiscuss a diagnostic test for
multicollinearity by presenting the variance inflation factor(VIF)
coefficients for all regression models. Furthermore, I run several
sensitivity checks toinvestigate, in particular, whether results
are sensitive to any specific exclusion restrictions.
3.3.2 Propensity-score matching models. The second control for
self-selection bias isbased on propensity-score matching models
developed by Rosenbaum and Rubin(1983). They highlight that:
[. . .] in non-randomized experiments, direct comparisons
between treatment and controlgroups may be misleading because the
units exposed to one treatment generally differsystematically from
the units exposed to the other treatment. Balancing scores, such
aspropensity-scores, can be used to group treated and control units
so that directcomparisons are more meaningful (Rosenbaum and Rubin,
1983).
Recently, propensity-score matching models have been employed in
studies focused onauditing and, more precisely, on the effect of
the presence of Big 4 and non-Big 4 auditfirms on particular
variables of interest (Boone et al., 2010; Lawrence et al.,
2011).Despite their limitations (Lawrence et al., 2011, p. 262),
Lawrence et al. (2011) state thatmatching models are appropriate in
such studies as they:
Big 4 feepremium and
audit quality
689
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
[. . .] generate samples in which the clients of Big 4 and
non-Big 4 auditors are similar,providing a natural framework to
parse out the effects of auditor and client characteristics onthe
variables of interest, such as audit quality proxies.
Lennox et al. (2012) indicate that these models also have the
advantage that they do notneed the inverse Mills ratio variable so
that the researcher is not required to find avalid set of control
variables as it is needed in the first stage regression of
Heckman(1979) procedure.
When employing this methodology, I use the logistic model (5) to
estimate theprobability of selecting a Big 4 audit firm, as this is
the most prevalent approach toestimating propensity-scores
(Lawrence et al., 2011). I then match withoutreplacement a Big 4
client with the non-Big 4 client that has the closest
predictedvalue from model (5) following the nearest neighbor
matching procedure. I finallyestimate models (1)-(4) based on this
propensity-score matched sample.
All models included in the paper are estimated using OLS when
the dependentvariable is continuous and a logit regression model
when the dependent variable isdichotomous. p-values are calculated
using t-statistics from robust standard errorsclustered by firm to
correct for serial correlation. Models also include year and
industrydummy variables. The latter are based on the Industry
Classification Benchmark (ICB),an industry classification taxonomy
developed by Dow Jones and FTSE.
3.4 Sample selectionThe most recent data on companies listed on
the UK stock market (FTSE All-Share) areused for the analyses. The
time period covers 2005-2011, choosing as a starting pointthe first
year that International Accounting Standards (IFRS) were made
mandatoryfor all companies listed on any European market. Firms
operating in the banking andfinancial services industries have been
excluded. All data used in the analyses has beengathered from
DataStream.
After imposing all the necessary requirements to obtain the
annual fees andcalculate earnings management regression variables,
I obtained 5,663 firm-yearobservations, in which 3,403 are Big 4
clients and 2,260 are non-Big 4 clients. Afteremploying the
propensity-score matching model, I obtained a
propensity-scorematched sample of 2,260 Big 4 client and 2,260
non-Big 4 client observations, for a totalof 4,520 firm-year
observations. Once I imposed all the necessary requirements for
thecalculation of the variables included in the accounting
conservatism and valuerelevance models I obtained 3,941 firm-year
observations, in which 2,760 are Big 4 and1,181 are non-Big 4
clients, because some of the required data was not available.
Afteremploying the propensity-score matching model I obtained a
propensity-score matchedsample of 1,181 Big 4 client and 1,181
non-Big 4 client observations, for a total of 2,362firm-year
observations.
4. Results and discussion4.1 Descriptive statisticsTable I
presents the descriptive statistics for audit fees (LNFEE) as well
as for theearnings management measure (AWCA) and for all the other
variables used in theregression models. They are presented in
aggregate (Panel A) and separated betweenBig 4 and non-Big 4
clients (Panel B). In Panel B, tests of differences in means,
mediansand standard deviations between the two groups of companies
have been also
MAJ28,8
690
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
n Mean Median SD n Mean Median SD
Panel A: aggregate descriptive statisticsLNFEE 5,663 5.335 5.056
1.707AWCA 5,663 0.107 0.041 0.256Def(E) 3,941 0.104 0.078 0.156CFO
5,663 20.008 0.064 1.024LEV 5,663 1.394 0.884 3.369LOSS 5,663 0.331
0.000 0.471SIZE 5,663 11.258 11.067 2.472INVREC 5,663 0.281 0.247
0.203ROA 5,663 20.022 0.063 0.642QUICKRATIO 5,663 1.857 0.990
3.835EISSUE 5,663 0.236 0.003 1.121DISSUE 5,663 0.247 0.052
0.808MBV 5,663 2.836 1.662 5.412GROWTH 5,663 0.318 0.089 1.165RET
3,941 0.112 0.037 0.654NEG 3,941 0.464 0.000 0.499BV 3,941 1.542
0.851 2.192E 3,941 23.830 12.560 34.328Panel B: Big 4 vs non-Big 4
clients
Big 4 clients Non-Big 4 clients
LNFEE 3,403 6.155 * * * 6.014 * * * 1.615 * * * 2,260 4.100
4.060 0.911AWCA 3,403 0.072 * * * 0.033 * * * 0.175 * * * 2,260
0.159 0.061 0.336Def(E) 2,760 0.096 * * * 0.076 * * * 0.133 * * *
1,181 0.124 0.084 0.198CFO 3,403 0.062 * * * 0.080 * * * 0.209 * *
* 2,260 20.114 0.018 1.595LEV 3,403 1.638 * * * 1.098 * * * 3.510 *
* * 2,260 1.026 0.601 3.110LOSS 3,403 0.231 * * * 0.000 * * * 0.421
* * * 2,260 0.481 0.000 0.500SIZE 3,403 12.485 * * * 12.332 * * *
2.183 * * * 2,260 9.410 9.453 1.565INVREC 3,403 0.275 * * * 0.244 *
0.195 * * * 2,260 0.290 0.252 0.214ROA 3,403 0.055 * * * 0.076 * *
* 0.219 * * * 2,260 20.138 0.019 0.968QUICKRATIO 3,403 1.661 * * *
0.930 * * * 3.861 2,260 2.150 1.115 3.776EISSUE 3,403 0.199 * * *
0.002 * * * 1.046 * * * 2,260 0.293 0.005 1.222DISSUE 3,403 0.221 *
* * 0.051 0.716 * * * 2,260 0.288 0.056 0.928MBV 3,403 3.940 1.940
* * * 62.18 2,260 4.970 1.340 63.24GROWTH 3,403 0.261 * * * 0.082 *
1.008 * * * 2,260 0.404 0.106 1.363RET 2,760 0.108 0.059 * * *
0.516 * * * 1,181 0.122 20.011 0.897NEG 2,760 0.446 * * * 0.000 * *
* 0.497 * * * 1,181 0.504 1.000 0.500BV 2,760 1.848 * * * 1.074 * *
* 2.394 * * * 1,181 0.825 0.432 1.382E 2,760 29.738 * * * 17.210 *
* * 38.170 * * * 1,181 10.023 4.840 15.994
Notes: Significant at: *10, * *5 and * * *1 per cent levels
(two-tailed) of differences between means, mediansand standard
deviations between Big 4 and non-Big 4 clients
Variables definition:LNFEE is the natural logarithm of audit
fees; AWCA is the winsorised absolute value of the abnormal
workingcapital accruals; Def(E) is the earnings per share scaled by
prior year-end price; CFO is the cash flow fromoperations divided
by total assets; LEV is leverage measured as total liabilities over
total equity; LOSS is 1 if thefirm reports a net loss and 0
otherwise; SIZE is the natural logarithm of total assets; INVREC is
the sum ofinventories and receivables divided by total assets; ROA
is the return on asset calculated as operating profit overtotal
assets; QUICKRATIO is quick ratio of the company; EISSUE is the
annual change in shareholder equity;DISSUE is the annual change in
liabilities; MBV is the market to book ratio; GROWTH is the annual
change in netsales; RET is 12-month raw returns beginning eight
months before the fiscal year-end and ending four monthsafter the
year-end; NEG is 1 if returns are negative and 0 otherwise; P is
the price of a share for a firm six monthsafter the fiscal
year-end; BV is the firms book value per share; E is the firms
earnings per share
Table I.Descriptive analysis
Big 4 feepremium and
audit quality
691
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
carried out. Differences in means have been calculated using a
t-test, differences inmedians have been calculated using the
Kruskal-Wallis test and differences in thestandard deviations
follow Levenes test.
With respect to the descriptive statistics, it is worth noting
that the average LNFEEis 5.335 which corresponds to an average
audit fee of 207.4 thousand pounds.Abnormal working capital
accruals represent around 10 per cent of firms total assets.Panel B
of Table I indicates that the Big 4 charge higher audit fees and
that their clientsexhibit lower discretionary accruals. However, in
accordance with Lawrence et al.(2011), it also highlights that Big
4 clients are significantly different from those auditedby non-Big
4 firms. Big 4 clients are bigger as suggested by the variable
SIZE, moreprofitable (higher ROA and lower frequency of LOSS),
exhibit higher cash flow fromoperations and leverage and have lower
levels of growth.
As all the features described above might potentially affect the
level of fees andabnormal accruals, more reliable conclusions can
only be made after a multivariateanalysis.
Table II reports a Pearson correlation matrix.The results mainly
support the descriptive statistics. The level of fees (abnormal
accruals) is positively (negatively) related with the variable
BIG4. However, aspreviously explained, these variables have
significant correlations with other factorsthat will be used as
control variables in the regression models; for this reason, only
amultivariate analysis can provide statistically reliable evidence
about the researchquestions. Indeed, the level of audit fees is
also positively related with the cash flow fromoperations,
leverage, company size and profitability, while it is negatively
related to thelevel of current assets (receivables and
inventories), quick ratio and the issuance of newfinancing, both in
terms of equity and external debt. The level of abnormal
accrualsseems higher among firms that report a loss, have higher
levels of current assets and areinvolved in either equity or debt
issuance. On the other hand, a negative association isfound with
company size, leverage and cash flow, the presence of a Big 4 firm
and thereturn on assets.
No comments can be made at this stage in relation to the other
dimensions of auditquality investigated accounting conservatism and
value relevance of earnings which will be analysed in detail when
the multivariate analyses are presented.
4.2 Multivariate analyses4.2.1 Probability of selecting a Big 4
audit firm. Table III reports the estimation of model(5). The
reason this model is presented is that, first of all, it provides
the basis for thecomputation of the variable INVMILLS and for the
implementation of thepropensity-score matching model. Furthermore,
Lennox et al. (2012) indicate thatone of the problems with the use
of Heckman (1979) procedure from previous studies isthe lack of
information about the first stage of this methodology. In the
followingTable III, the paper also deals with this observation.
The model is highly statistically significant ( p-value 0.000)
and exhibits an R 2 of39.7 per cent. As analysis of the
determinants of auditor choice is not the aim of thispaper, the
discussion of the results is limited and only points out that the
findings aremainly in line with the extant literature where they
indicate that the choice of a Big 4 firmis associated with client
size, growth and complexity (Simunic and Stein, 1987; Hay andDavis,
2004).
MAJ28,8
692
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
Notes:
Coe
ffici
ent
isst
atis
tica
lly
sig
nifi
can
tat
:* 1
0,*
* 5,
and
**
* 1p
erce
nt
lev
els
orb
ette
r,tw
o-ta
iled
Vari
abl
esdefi
nit
ion:
LN
FE
Eis
the
nat
ura
llog
arit
hm
ofau
dit
fees
;AW
CA
isth
ew
inso
rise
dab
solu
tev
alu
eof
the
abn
orm
alw
ork
ing
cap
ital
accr
ual
s;D
ef(E
)is
the
earn
ing
sp
ersh
are
scal
edb
yp
rior
yea
r-en
dp
rice
;BIG
4is
1fo
rB
ig4
clie
nts
and
0ot
her
wis
e;C
FO
isth
eca
shfl
owfr
omop
erat
ion
sd
ivid
edb
yto
tala
sset
s;L
EV
isle
ver
age
mea
sure
das
tota
llia
bil
itie
sov
erto
tale
qu
ity
;LO
SS
is1
ifth
efi
rmre
por
tsa
net
loss
and
0ot
her
wis
e;S
IZE
isth
en
atu
rall
ogar
ith
mof
tota
lass
ets;
INV
RE
Cis
the
sum
ofin
ven
tori
esan
dre
ceiv
able
sd
ivid
edb
yto
tala
sset
s;R
OA
isth
ere
turn
onas
set
calc
ula
ted
asop
erat
ing
pro
fit
over
tota
lass
ets;
QU
ICK
RA
TIO
isq
uic
kra
tio
ofth
eco
mp
any
;EIS
SU
Eis
the
ann
ual
chan
ge
insh
areh
old
ereq
uit
y;D
ISS
UE
isth
ean
nu
alch
ang
ein
liab
ilit
ies;
MB
Vis
the
mar
ket
tob
ook
rati
o;G
RO
WT
His
the
ann
ual
chan
ge
inn
etsa
les;
RE
Tis
12-m
onth
raw
retu
rns
beg
inn
ing
eig
htm
onth
sb
efor
eth
efi
scal
yea
r-en
dan
den
din
gfo
ur
mon
ths
afte
rth
ey
ear-
end
;NE
Gis
1if
retu
rns
are
neg
ativ
ean
d0
oth
erw
ise;
Pis
the
pri
ceof
ash
are
for
afi
rmsi
xm
onth
saf
ter
the
fisc
aly
ear-
end
;BV
isth
efi
rms
boo
kv
alu
ep
ersh
are;
Eis
the
firm
sea
rnin
gs
per
shar
e
Table II.Correlation matrix
LN
FE
EA
WC
AD
ef(E
)P
BIG
4C
FO
LE
VLO
SS
SIZ
EIN
VR
EC
RO
A
QU
ICK
RA
TIO
EIS
SU
ED
ISS
UE
MB
VG
RO
WT
HR
ET
NE
GB
VE
LN
FE
EA
WC
A-0.190
***
Def
(E)
-0.108
***
0.13
0***
P0.01
5-0.006
-0.006
BIG
40.59
0***
-0.166
***
-0.083
***
0.04
7***
CF
O0.10
6***
-0.234
***
-0.107
***
-0.005
0.08
4***
LE
V0.14
9***
-0.068
***
-0.008
-0.001
0.08
9***
0.03
3**
LOS
S-0.316
***
0.20
5***
0.08
2***
-0.014
-0.261
***
-0.149
***
-0.054
***
SIZ
E0.89
2***
-0.286
***
-0.128
***
0.01
80.60
9***
0.17
3***
0.14
2***
-0.398
***
INV
RE
C-0.040
***
0.03
7***
0.02
20.00
2-0.038
***
0.00
10.05
0***
-0.144
***
-0.113
***
RO
A
0.18
0***
-0.331
***
-0.110
***
0.00
10.14
7***
0.83
4***
0.04
7***
-0.278
***
0.27
9***
0.03
0**
QU
ICK
RA
TIO
-0.183
***
-0.018
0.04
7***
-0.010
-0.062
***
-0.018
-0.099
***
0.16
8***
-0.158
***
-0.177
***
-0.044
***
EIS
SU
E-0.025
*0.05
3***
-0.003
0.06
4***
-0.041
***
-0.018
-0.003
0.10
1***
-0.066
***
-0.066
***
-0.020
0.07
2***
DIS
SU
E-0.059
***
0.07
1***
-0.058
***
0.05
6***
-0.043
***
0.00
5-0.011
0.05
6***
-0.037
***
-0.099
***
-0.013
0.00
50.10
1***
MB
V0.38
7***
0.04
2***
-0.028
*0.00
10.00
8-0.034
***
0.22
2***
0.01
7-0.062
***
-0.007
-0.004
-0.004
-0.001
-0.000
GR
OW
TH
-0.095
***
0.41
7***
-0.012
0.03
5**
-0.060
***
0.00
2-0.040
***
0.11
7***
-0.088
***
-0.116
***
-0.008
0.07
8***
0.11
0***
0.33
0***
0.00
2R
ET
-0.043
***
0.02
10.13
1***
0.02
0-0.010
0.04
0**
0.00
2-0.060
***
-0.048
***
0.04
9***
-0.019
0.01
00.04
3***
0.04
8***
0.00
90.04
8***
NE
G-0.050
***
0.05
5***
-0.051
***
-0.017
-0.053
***
-0.095
***
-0.016
0.10
9***
-0.060
***
-0.011
-0.054
***
0.00
90.02
40.02
8*-0.018
0.03
2**
-0.592
***
BV
0.38
7***
-0.096
***
-0.072
***
0.27
3***
0.21
4***
-0.007
-0.013
-0.049
***
0.46
1***
-0.065
***
-0.003
0.01
8-0.006
0.00
7-0.018
-0.005
-0.044
***
-0.011
E0.34
6***
-0.045
***
0.10
2***
0.28
6***
0.26
3****
0.06
9***
0.09
5***
-0.075
***
0.39
7***
-0.008
0.10
6***
-0.022
-0.024
-0.015
0.00
2-0.010
-0.052
***
-0.008
0.57
9***
Big 4 feepremium and
audit quality
693
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
4.2.2 Big 4 audit firms and audit fee premium. Table IV presents
the results of model(1) which investigates the existence of an
audit fee premium among Big 4 clients. Thetable includes the
estimation of the model using standard OLS before any control
forself-selection bias (column A) and after a control for
self-selection bias using Heckman(1979) procedure (column B) and
the propensity-score matched sample (column C). Thisstructure will
also be used consistently for the presentation of the other
regressionanalyses.
Model (1) is statistically significant ( p-value 0.000 in all
cases) with an R 2 thatgoes from 70.2 (column C) to 82.9 per cent
(column B).
Column (A) shows a coefficient b1, associated with the variable
BIG4, which ispositive and significant at the 1 per cent level (b
0.216; p-value 0.000). This resultis also supported when a control
for self-selection bias is introduced using bothHeckman (1979)
procedure (b 0.131; p-value 0.000) and the propensity-score
Dependent variable BIG4
INTERCEPT 26.409 * * *
(224.50)MBV 0.001 * * *
(3.58)LEV 20.002
(20.73)LOSS 0.009
(0.18)SIZE 0.599 * * *
(38.26)INVREC 0.070
(0.58)ROA 20.153 * * *
(24.29)QUICKRATIO 0.020 * * *
(3.66)Observations 5,663R 2 0.397F-stat. 3020.52 * * *
year dummies YesIndustries dummies Yes
Notes: Coefficient is statistically significant at: *10, * *5
and * * *1 per cent levels or better; t-statistics(in parentheses
below the coefficients) are calculated using standard errors
clustered by firm; forclarity, the year-specific and
industry-specific intercepts are omittedRegression models:
BIG4it a b1MBV it b2LEV it b3LOSSit b4SIZEit b5INVRECit b6ROAit
b7QUICKRATIOit 1it
Variables definition:BIG4 is 1 for Big 4 clients and 0
otherwise; MBV is the market to book ratio; LEV is leverage
measured astotal liabilities over total equity; LOSS is 1 if the
firm reports a net loss and 0 otherwise; SIZE is thenatural
logarithm of total assets; INVREC is the sum of inventories and
receivables divided by totalassets; ROA is the return on asset
calculated as operating profit over total assets; QUICKRATIO is
quickratio of the company
Table III.Probability of selecting aBig 4 audit firm
MAJ28,8
694
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
Standard OLS
Control for SSB:Heckman (1979)
procedure
Control for SSB:propensity-scorematched sample
(A) (B) (C)Dependent variable LNFEE LNFEE LNFEE
INTERCEPT 22.699 * * * 25.457 * * * 21.928 * * *
(212.77) (214.32) (28.85)BIG4 0.216 * * * 0.131 * * * 0.300 * *
*
(4.65) (7.00) (6.41)MBV 0.000 0.001 * * 0.000
(1.27) (2.28) (0.92)LEV 20.000 20.001 * * * 20.000
(20.84) (24.57) (20.38)LOSS 0.216 * * * 0.177 * * * 0.193 * *
*
(6.44) (5.70) (6.43)SIZE 0.633 * * * 0.800 * * * 0.548 * * *
(47.06) (32.28) (35.71)INVREC 0.517 * * * 0.511 * * * 0.522 * *
*
(3.67) (3.77) (4.71)ROA 20.188 * * * 20.194 * * * 20.144 * *
*
(25.26) (29.27) (25.24)QUICKRATIO 20.015 * * * 20.007 20.013
(23.28) (21.50) (21.32)INVMILLS 0.701 * * *
(0.000)Average VIF of the model 1.90 3.79 4.37VIF BIG4 1.66 1.73
1.47VIF INVMILLS 7.25Observations 5,663 5,663 4,520R 2 0.818 0.829
0.702F-stat. 239.24 * * * 275.58 * * * 133.69 * * *
Year dummies Yes Yes YesIndustries dummies Yes Yes Yes
Notes: Coefficient is statistically significant at: *10, * *5
and * * *1 per cent levels or better; t-statistics (inparentheses
below the coefficients) are calculated using standard errors
clustered by firm; for clarity, theyear-specific and
industry-specific intercepts are omittedRegression models:
Col: A and C : LNFEEit a b1BIG4it b2MBVit b3LEV it b4LOSSit
b5SIZEit b6INVRECit b7ROAit b8QUICKRATIOit 1it
Col: B : LNFEEit a b1BIG4it b2MBVit b3LEV it b4LOSSit b5SIZEit
b6INVRECit b7ROAit b8QUICKRATIOit b9INVMILLSit 1it
Variables definition:LNFEE is the natural logarithm of audit
fees; BIG4 is 1 for Big 4 clients and 0 otherwise; MBV is the
marketto book ratio; LEV is leverage measured as total liabilities
over total equity; LOSS is 1 if the firm reports a netloss and 0
otherwise; SIZE is the natural logarithm of total assets; INVREC is
the sum of inventories andreceivables divided by total assets; ROA
is the return on asset calculated as operating profit over total
assets;QUICKRATIO is quick ratio of the company; INVMILLS is the
inverse Mills ratio from equation (5)
Table IV.The investigation of the
audit fee premium
Big 4 feepremium and
audit quality
695
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
matched sample (b 0.300; p-value 0.000). This is evidence that
the level of auditfees, after controlling for other factors that
affect it, is higher among Big 4 clients.
The other control variables are mainly in line with the extant
literature. Simunic andStein (1996) suggest that auditors charge
higher fees to risky clients and a positive(negative) coefficient
is found here associated with the variable LOSS (ROA) both withand
without a control for self-selection bias. In accordance with
Simunic (1980) andChoi et al. (2008), a positive coefficient,
irrespective of the methodology used, isobserved between the level
of fees and the size of the client, as this is a proxy for
thenumber of hours needed to carry out the audit, and between fees
and the complexity ofthe client, measured here by the variable
INVREC. Higher fees are expected to besought from firms with higher
levels of growth (Reynolds et al., 2004) and, accordingly,a
significant coefficient is found between LNFEE and MBV when a
control forself-selection bias using Heckman (1979) procedure is
employed. Contrary to theexpectations that fees are higher with
respect to the level of debt of the company, anegative coefficient
is observed between fees and firm leverage under Heckman
(1979)procedure, while the same coefficient is not significant in
the absence of a control forself-selection bias and when the
propensity-score matched sample is employed. Finally,a
non-significant relation is found between fees and the variable
QUICKRATIO whena control for self-selection bias is introduced.
The estimation of model (1) presented in column (B) is
potentially exposed to boththe main limitations of Heckman (1979)
procedure presented in Section 3.3.1.1, namelymulticollinearity and
exclusion restriction.
In relation to the multicollinearity issue, the relevant VIF are
disclosed in Table IV.The average VIF coefficient of model (1)
reported in column (B) is 3.79 while thoseassociated to the main
variables of interest, BIG4 and INVMILLS, measure 1.73 and7.25,
respectively. On the basis of these coefficients, it can be stated
thatmulticollinearity does not represent a concern. Indeed, they
are all below the criticalvalue of 10. Values above 10 would
indicate that multicollinearity significantly affectsthe stability
of the parameter estimates (Dielman, 1991). It is also worth
pointing outthat the relevant VIF coefficients are lower than the
critical value of 10 in columns (A)and (C) of Table IV as well,
which provides assurance that multicollinearity does notaffect the
reliability of the analysis of the existence of a fee premium among
Big 4clients that has been presented thus far.
Exclusion restriction might be another potential problem when a
control forself-selection bias on the basis of Heckman (1979)
methodology is used. Indeed, theindependent variables of model (1)
are the same as those included in the first stageregression of
Heckman (1979) procedure. Feng et al. (2009) stress the importance
ofhaving at least one exclusion restriction in the second stage to
assess the robustness ofthe results. To investigate the robustness
of the coefficient b1 reported in column (B) ofTable IV, I
excluded, in turn, each of the independent variables included in
model (1).The results (untabled) indicate that none of those
variables has a significant effect onthe coefficient associated
with BIG4 or on its VIF. Indeed, the former always remainspositive
and significant at the 1 per cent level ( p-value 0.000 in all
cases); the latterfluctuates between 1.28 and 1.73.
To summarise, evidence reported thus far consistently indicates
that, in so far asUK listed company audits are concerned, the Big 4
audit firms charge higher fees totheir clients than their smaller
competitors.
MAJ28,8
696
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
4.2.3 Big 4 audit firms and audit quality. Evidence in Table IV
indicates theexistence of a Big 4 audit premium. The rest of the
analyses investigate whether thisadditional charge is accompanied
by an improvement in audit quality.
Table V reports the evidence from the first audit quality proxy:
earningsmanagement, and, more specifically discretionary
accruals.
The model is statistically robust ( p-value 0.000 both with and
without a controlfor self-selection bias) and the R 2 goes from
21.7(column A) to 28.9 per cent (column C).
The sign and significance of the coefficient associated with the
variable BIG4 is notconsistent across the three columns. More
precisely, column (A), which includes theestimation of model (2)
without controlling for self-selection bias, exhibits
anon-significant b1 (b 0.004; p-value 0.689). It suggests that
earnings managementthrough the use of discretionary accruals is not
less pervasive among Big 4 clients thannon-Big 4 clients.
Column (B) of Table V reports the estimation of model (2) after
controlling forself-selection bias using Heckman (1979) procedure.
The coefficient b1 becomes,surprisingly, positive and significant
at the 5 per cent level (b 0.028;p-value 0.020). This would
indicate that the use of abnormal accruals is morepervasive among
Big 4 clients. The conduct of an audit by a Big 4 firm would
beassociated with higher earnings management and, consequently,
lower audit quality.However, it is worth pointing out that this
result is not corroborated when thepropensity-score matched sample
is used (Table V, column C). In the latter case, thecoefficient,
consistent with column (A), is not significant (b 0.017; p-value
0.128).It indicates that the evidence from Heckman (1979)
methodology does not hold in asample where client characteristics
are balanced across the two auditor groupings.This supports the
conclusion that Big 4 clients do not exhibit lower levels
ofdiscretionary accruals than non-Big 4 clients and, accordingly,
audit quality is nothigher among the former group of companies.
When a control for self-selection bias is introduced, results
from Table V alsoindicate that discretionary accruals significantly
depend on the size of the company(b 20.017; p-value 0.006, when
Heckman (1979) procedure is employed;b 20.036; p-value 0.000, when
the propensity-score matched sample is used),as smaller firms are
less monitored and exhibit more earnings management(Bartov et al.,
2000; Park and Shin, 2004). The use of abnormal accruals is also
morepervasive among less profitable companies and higher growth
firms (Carey andSimnett, 2006), as indicated by the negative and
significant coefficient between AWCAand ROA (b 20.128; p-value
0.000, when Heckman (1979) procedure isintroduced; b 20.127;
p-value 0.000, when the propensity-score matchedsample is used),
and by the positive and significant coefficient between AWCA
andGROWTH (b 0.001; p-value 0.023, under Heckman (1979) procedure;
b 0.004;p-value 0.000, when the propensity-score matched sample is
employed).
The evidence provided here is not affected by multicollinearity
as the relevant VIFcoefficients reported in Table V are all below
the critical value of 10. The exclusionrestriction problem, which
might arise when Heckman (1979) procedure is employed, isnot a
concern in the analysis of audit quality as the independent
variables of models(2)-(4) are not exactly the same as those used
in the first stage regression of theselection model.
Big 4 feepremium and
audit quality
697
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
Standard OLS
Control for SSB:Heckman (1979)
procedure
Control for SSB:propensity-scorematched sample
(A) (B) (C)Dependent variable AWCA AWCA AWCA
INTERCEPT 0.420 * * * 20.212 * 0.534 * * *
(13.73) (21.92) (5.62)BIG4 0.004 0.028 * * 0.017
(0.40) (2.32) (1.52)CFO 0.023 * 0.017 0.022
(1.70) (1.34) (1.59)LEV 0.000 * * * 0.000 0.000 * * *
(3.12) (1.22) (4.84)SIZE 20.023 * * * 20.017 * * * 20.036 * *
*
(25.92) (22.77) (25.06)ROA 20.137 * * * 20.128 * * * 20.127 * *
*
(23.62) (23.94) (23.48)EISSUE 0.000 0.000 20.000
(0.78) (0.70) (20.74)DISSUE 20.000 * * * 20.000 20.000
(22.80) (20.22) (20.30)MBV 0.000 0.000 * * * 0.000
(1.47) (3.36) (1.11)GROWTH 0.001 * * 0.001 * * 0.004 * * *
(2.31) (2.33) (5.69)INVMILLS 0.166 * * *
(4.56)Average VIF of the model 3.45 3.45 4.28VIF BIG4 1.66 1.73
1.47VIF INVMILLS 7.07Observations 5,663 5,663 4,520R 2 0.217 0.244
0.289F-stat. 11.30 * * * 11.64 * * * 9.68 * * *
Year dummies Yes Yes YesIndustry dummies Yes Yes Yes
Notes: Coefficient is statistically significant at: *10, * *5
and * * *1 per cent levels or better; t-statistics (inparentheses
below the coefficients) are calculated using standard errors
clustered by firm; for clarity, the year-specific and
industry-specific intercepts are omittedRegression models:
Col A and C : AWCAit a b1BIG4it b2CFOit b3LEVit b4SIZEit b5ROAit
b6EISSUEit b7DISSUEit b8MBVit b9GROWTHit 1it
Col B : AWCAit a b1BIG4it b2CFOit b3LEV it b4SIZEit b5ROAit
b6EISSUEit b7DISSUEit b8MBV it b9GROWTHit b10INVMILLSit 1it
Variables definition:AWCA is the winsorised absolute value of
the abnormal working capital accruals; BIG4 is 1 for Big 4 clients
and0 otherwise; CFO is the cash flow from operations divided by
total assets; LEV is leverage measured as totalliabilities over
total equity; SIZE is the natural logarithm of total assets; ROA is
the return on asset calculatedas operating profit over total
assets; EISSUE is the annual change in shareholder equity; DISSUE
is the annualchange in liabilities; MBV is the market to book
ratio; GROWTH is the annual change in net sales; INVMILLS isthe
inverse Mills ratio from equation (5)
Table V.Big 4 audit firms andaudit quality:
earningsmanagement
MAJ28,8
698
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
Overall, the evidence in Table V indicates that the conduct of
audit by a Big 4 auditfirm is not associated with better audit
quality, as measured by less earningsmanagement. Thus, the
significant relation between discretionary accruals and type
ofauditor, highlighted in the descriptive analysis and correlation
table, is notcorroborated here. The results support those of
Lawrence et al. (2011): as earningsmanagement is affected by
several firm-level factors, and given that the characteristicsof
Big 4 and non-Big 4 clients are significantly different,
differences in audit qualityproxies reflect these different
features. Indeed, the factors that, on the basis of Table V,are
strongly associated with lower abnormal accruals bigger size,
higherprofitability, lower levels of growth are exactly some of the
main characteristics ofBig 4 clients.
Findings from the accounting conservatism dimension are reported
in Table VI.The model is always statistically significant ( p-value
0.000) and the R 2
fluctuates from 4.3 (column C) to 6.9 per cent (column B).Column
(A) presents the estimation of model (3) without any control for
self-selection
bias. The coefficient b1 is positive and significant at the 1
per cent level (b 0.032;p-value 0.002), indicating that, on
average, companies report good news on a timelybasis. The
insignificant b3 (b 20.053; p-value 0.177) suggests that firms in
thesample do not have a higher propensity to report bad news as
compared with good news.The interaction terms RET*BIG4 and
RET*NEG*BIG4 are both not significant. Theformer (b 0.010; p-value
0.535) indicates that there is no incremental improvementin the
speed of incorporating good news into earnings among Big 4 clients,
while thelatter (b 0.017; p-value 0.674) suggests that the conduct
of an audit by a Big 4 firmdoes not influence the speed of
recognition of bad news either.
All evidence provided thus far holds also when either Heckman
(1979) methodologyor the propensity-score matched sample is
employed. Indeed, the inferences from thecoefficients presented
above remains exactly the same as that reported in the
previousparagraph if columns (B) and (C) of Table VI are taken into
account.
The results reported in Table VI are not affected by
multicollinearity problems sincethe relevant VIF coefficients are,
in all cases, lower than the critical value of 10.
Finally, Table VII reports the findings related to the last
measure of audit quality:the value relevance of earnings.
The statistical significance of the model is never in question (
p-value 0.000) andits R 2 goes from 50.9 (column C) to 67.5 per
cent (column A).
As expected, the coefficients b2 and b3 are, in all columns,
positive and significantas market prices are directly sensitive to
the earnings and book value of companies.
For the purposes of this paper, the focus in this table is on
the coefficient b5, whichsignals whether the market interprets
earnings disclosed by Big 4 clients to be moreinformative than
those presented by other firms.
In column (A), which reports the estimation of model (4) without
any control forself-selection bias,b5 is not significant (b 0.289;
p-value 0.726). It provides evidencethat the market does not
perceive earnings reported by Big 4 clients to be moreinformative
than those presented by non-Big 4 clients. However, this
coefficient is notstatistically robust not only because a control
for self-selection bias has not been carriedout, but also because
its associated VIF (471.26) highlights serious
multicollinearityissues.
Big 4 feepremium and
audit quality
699
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
StandardOLS
Control for SSB: Heckman (1979)procedure
Control for SSB:propensity-score matched
sample(A) (B) (C)
Dependent variable Def(E) Def(E) Def(E)
INTERCEPT 0.086 * * * 0.050 * * * 0.063 * * *
(6.46) (3.39) (5.21)RET 0.032 * * * 0.028 * * * 0.031 * * *
(3.06) (2.62) (2.90)NEG 20.007 20.011 20.007
(20.45) (20.72) (20.47)RET*NEG 20.053 20.046 20.049
(21.35) (21.19) (21.22)BIG4 20.035 * * * 20.006 20.015
(23.23) (20.47) (21.18)RET*BIG4 0.010 0.010 20.007
(0.62) (0.57) ( 2 0.37)NEG*BIG4 0.023 0.010 0.022
(1.25) (0.57) (0.88)RET*NEG*BIG4 0.017 0.021 0.070
(0.42) (0.51) (1.41)INVMILLS 0.041 * * *
(3.25)Average VIF of themodel
4.80 4.70 5.71
VIF RET*BIG4 3.51 3.51 2.91VIF RET*NEG* BIG4 7.45 7.46 5.62VIF
INVMILLS 1.76Observations 3,941 3,941 2,362R 2 0.057 0.069
0.043F-stat. 10.16 * * * 9.93 * * * 6.81 * * *
Year dummies Yes Yes YesIndustries dummies Yes Yes Yes
Notes: Coefficient is statistically significant at: *10, * *5
and * * *1 per cent levels or better; t-statistics (inparentheses
below the coefficients) are calculated using standard errors
clustered by firm; for clarity, theyear-specific and
industry-specific intercepts are omittedRegression models:
Col A and C : Def Eit a b1RETit b2NEGit b3RET*NEGit b4BIG4it
b5RET*BIG4it b6NEG*BIG4it b7RET*NEG*BIG4it 1it
Col B : Def Eit a b1RETit b2NEGit b3RET*NEGit b4BIG4it
b5RET*BIG4it b6NEG*BIG4it b7RET*NEG*BIG4it b8INVMILLSit 1it
Variables definition:Def(E) is the earnings per share scaled by
prior year-end price; RET is 12-month raw returns beginning
eightmonths before the fiscal year-end and ending four months after
the year-end; NEG is 1 if returns are negativeand 0 otherwise; BIG4
is 1 for Big 4 clients and 0 otherwise; INVMILLS is the inverse
Mills ratio fromequation (5)
Table VI.Big 4 audit firms andaudit quality:
accountingconservatism
MAJ28,8
700
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
Column (B) of Table VII includes the estimation of the same
model (4) with a control forself-selection bias based on Heckman
(1979) methodology. Under this scenario, b5becomes positive and
significant (b 6.068; p-value 0.000). This would seem tocontradict
the previous finding and would suggest that earnings disclosed by
Big 4clients are perceived to be more informative by the market. On
the other hand,additional evidence in the same column questions the
statistical robustness of thiscoefficient. First of all, its
disproportionate VIF (417.35) highlights
significantmulticollinearity problems. Furthermore, the coefficient
associated to INVMILLS is not
Standard OLS
Control for SSB:Heckman (1979)
procedure
Control for SSB:propensity-scorematched sample
(A) (B) (C)Dependent variable P P P
INTERCEPT 811.51 * * * 51.617 281.787 * * *
(2.58) (0.29) (2.60)BV 0.903 * * 108.826 * * 56.889 * * *
(2.01) (2.22) (3.44)E 2.382 * * * 2.043 * * * 2.255 * * *
(11.97) (7.57) (19.89)BIG4 66.824 * * 230.056 23.557
(2.19) (20.16) (1.18)BV*BIG4 93.812 * * * 174.303 28.951
(5.26) (1.25) (1.04)E*BIG4 0.289 6.068 * * * 1.757
(0.35) (13.53) (1.08)INVMILLS 393.182
(1.15)Average VIF of the model 48.72 45.85 5.35VIF E*BIG4 417.26
417.35 1.95VIF INVMILLS 1.88Observations 3,941 3,941 2,362R 2 0.675
0.605 0.509F-stat. 72.65 * * * 123.32 * * * 50.91 * * *
Year dummies Yes Yes YesIndustries dummies Yes Yes Yes
Notes: Coefficient is statistically significant at: *10, * *5
and * * *1 per cent levels or better; t-statistics(in parentheses
below the coefficients) are calculated using standard errors
clustered by firm; forclarity, the year-specific and
industry-specific intercepts are omittedRegression models:
Col A and C : Pit a b1BVit b2Eit b3BIG4it b4BV*BIG4it b5E*BIG4it
1it
Col B : Pit a b1BV it b2Eit b3BIG4it b4BV*BIG4it b 5E*BIG4it
b6INVMILLSit 1it
Variables definition:P is the price of a share for a firm six
months after the fiscal year-end; BV is the firms book value
pershare; E is the firms earnings per share; BIG4 is 1 for Big 4
clients and 0 otherwise; INVMILLS is theinverse Mills ratio from
equation (5)
Table VII.Big 4 audit firms and
audit quality: valuerelevance of earnings
Big 4 feepremium and
audit quality
701
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
significant (b 393.182; p-value 0.250). This is evidence that a
control forself-selection bias, using Heckman (1979) procedure, is
not effective in this particularcase and that OLS that ignore
self-selection (i.e. OLS that do not include the variableINVMILLS)
do not pose a problem with the self-selection bias (Krishnan et
al., 2008). Itleads back to the inference made on the basis of
column (A) of Table VII which still hasthe multicollinearity
problem.
All these issues are eventually overcome in column (C) of Table
VII, which reportsthe estimation of model (4) using the
propensity-score matched sample. Under thisscenario, b5 is not
significant (b 1.757; p-value 0.282), supporting the view thatthe
market does not perceive that earnings disclosed by Big 4 clients
are moreinformative than those reported by non-Big 4 clients. In
addition, the coefficient is nowstatistically robust since it is
not affected by multicollinearity problems as itsassociated VIF
(1.95) is lower than the critical value of 10.
Finally, in relation to the value relevance of the book value,
while its informationcontent seems higher among Big 4 clients
without a control for self-selection bias(b 93.812; p-value 0.000),
its VIF coefficient (19.70, untabled) does not make itsuitable for
a reliable inference. Once the propensity-score matched sample is
used, thesame coefficient becomes non-significant (b 28.951;
p-value 0.299) andstatistically robust (VIF 1.71, untabled).
To summarize, multivariate analyses support the concerns raised
by the UK Houseof Lords, as they indicate that the concentrated
structure of the audit market in the UKresults in higher fees for
Big 4 clients. In addition, the analyses consistently suggestthat
this is happening without any improvement in the level of audit
quality providedby the Big 4, at least with respect to the three
proxies investigated.
4.3 Robustness testsTo increase the reliability of the findings
presented above, some robustness tests(untabled) have been carried
out. The relation between type of auditors and auditquality has
been repeated using other dependent variables and/or models with
andwithout a control for self-selection bias. More precisely, the
absolute values ofincome-increasing and income-decreasing abnormal
accruals are used separately inmodel (2). Results are consistent
with those reported in the main analyses: when I usethe
propensity-score matched sample, the relation between the level of
abnormalaccruals and the conduct of audits by Big 4 firms is not
significant in any circumstance(b 0.015; p-value 0.355, when
income-increasing abnormal accruals are used;b 0.019; p-value
0.138, when income-decreasing abnormal accruals are used). Thesame
coefficients are also not significant without a control for
self-selection bias(b 0.005; p-value 0.588 and b 0.000; p-value
0.983, respectively). Nomulticollinearity problems are
observed.
The analysis of audit quality using the propensity-score
matching model has beenrepeated by estimating propensity-scores
using an extended version of equation (5)where, in accordance with
Lawrence et al. (2011), I included additional independentvariables;
more precisely, those variables used in the respective audit
qualityregressions. Evidence is unchanged.
Finally, another model for the estimation of accounting
conservatism, namely theaccruals-cash flow model developed by Ball
and Shivakumar (2005), is also used.Evidence consistently shows no
differences between Big 4 and non-Big 4 clients as the
MAJ28,8
702
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
relevant coefficient from that model is not significant either
with a control forself-selection bias (b 20.275; p-value 0.912,
when Heckman (1979) procedure isemployed; b 0.283; p-value 0.914
when the propensity-score matched sample isused) or without (b
0.241; p-value 0.924). No multicollinearity issues are observedin
this case as well.
5. ConclusionsInspired by the concerns raised by the UK House of
Lords that the oligopolisticstructure of the UK audit market might
lead to excessive fees for Big 4 clients, theaim of this study is
the empirical investigation of the existence of a Big 4 audit
feepremium and, in that case, whether it is related to the delivery
of a better quality audit.
Using the most recent data (2005-2011) from non-financial
companies listed in theUK and controlling for self-selection bias,
results reveal that Big 4 audit firms docharge their clients an
audit fee premium. This premium, however, is not related tothe
delivery of a superior audit service as measured using three
proxies for clientsearnings quality discretionary accruals,
accounting conservatism and valuerelevance of earnings. Indeed,
multivariate regressions indicate that Big 4 audit firms,ceteris
paribus, do not reduce the use of abnormal accruals and do not
improve thespeed of recognition of both good and bad news in
earnings. Furthermore, the financialmarket does not perceive
earnings disclosed by Big 4 clients to be more informativethan
those reported by non-Big 4 clients.
The overall evidence of the paper supports the concerns
expressed by the EconomicAffairs Committee of the UK House of Lords
and by some respondents to their call forevidence that the
concentrated structure of the audit market results in
excessiveaudit fees, at least for Big 4 clients, to such an extent
that a legislative interventioncould be even required (House of
Lords, 2010c). This intervention might also be moreurgent as the
findings of this research reveal that these higher audit fees are
notaccompanied by an enhancement of the quality of the service
provided.
This evidence might also be of interest to Big 4 clients in the
event of a renegotiationof the audit fees as it may improve their
bargaining power. Big 4 clients would not beready to pay a premium
especially if they were aware that it does not result in
anyincremental benefit in terms of audit quality. They could be
then more willing to hire anon-Big 4 firm, paying lower fees
without suffering a decrease in the level of the auditservice
received. This situation would consequently decrease the
concentration of theaudit market, which is a big concern for many
countries, such as the UK, and wouldimprove the overall level of
auditing and the price/quality relationship.
This research is not free from limitations. Although the
measures employed areamong those commonly used in the literature
focused on audit quality, at the sametime, they are not direct
measures of this phenomenon as they do not take directly
intoaccount factors such as audit hours, audit strategy
deficiencies, audit team skills andindependence, etc. For these
reasons, the findings reported are not conclusive butrather
suggestive in that they carry out comparisons in audit quality
among differenttypes of auditors. Second, while the paper explores
audit quality using several and themost appropriate dimensions of
earnings quality, many other proxies have beendeveloped in the
academic literature to analyse the same topic that, if employed,
mightchange the evidence here reported. Finally, findings are based
on only one country, theUK. This may affect the generalizability of
the reported results.
Big 4 feepremium and
audit quality
703
Dow
nloa
ded
by U
nive
rsiti
Tek
nolo
gi M
ARA
At 0
8:20
08
Oct
ober
201
4 (P
T)
-
Future research could extend this analysis to other countries to
verify whether Big 4audit firms charge higher fees without
providing higher levels of audit quality indifferent institutional
settings.
References
Antle, R., Gordon, E.A., Narayanamoorthy, G. and Zhou, L.
(2006), The joint determination ofaudit fees, non-audit fees and
abnormal accruals, Review of Quantitative Finance andAccounting,
Vol. 27 No. 3, pp. 235-266.
Armour, J., Cheffins, B.R. and Skeel, D.A. (2002), Corporate
ownership structure and theevolution of bankruptcy law: lessons
from the United Kingdom, Vanderbilt Law Review,Vol. 55 No. 6, pp.
1699-1785.
Ashbaugh, H., LaFond, R. and Mayhew, B. (2003), Do non-audit
services compromise auditorindependence? Further evidence, The
Accounting Review, Vol. 78 No. 3, pp. 611-639.
Ball, R. and Shivakumar, L. (2005), Earnings quality in UK
private firms: comparative lossrecognition timeliness, Journal of
Accounting and Economics, Vol. 39 No. 1, pp. 83-128.
Bartov, E., Gul, F. and Tsui, J. (2000), Discretionary accruals
model and audit qualification,Journal of Accounting and Economics,
Vol. 30 No. 3, pp. 421-452.
Basu, S. (1997), The conservatism principle and the asymmetric
timeliness of earnings, Journalof Accounting and Economics, Vol. 24
No. 1, pp. 3-37.
Becker, C.L., DeFond, M.L., Jiambalvo, J. and Subramanyam, K.R.
(1998), The effect of auditquality on earnings management,
Contemporary Accounting Research, Vol. 15 No. 1,pp. 1-24.
Beekes, W., Pope, P. and Young, S. (2004), The link between
earnings timeliness, earningsconservatism and board composition:
evidence from the UK, Corporate Governance:An International Review,
Vol. 12 No. 1, pp. 47-59.
Bernard, V. and Skinner, D.J. (1996), What motivates managers
choice of discretionaryaccruals?, Journal of Accounting and
Economics, Vol. 22 Nos 1-3, pp. 313-325.
Boone, J.P., Khurana, I.K. and Raman, K.K. (2010), Do the Big 4
and the second-tier firms provideaudits of similar quality?,
Journal of Accounting & Public Policy, Vol. 29 No. 4, pp.
330-352.
Carey, P. and Simnett, R. (2006), Audit partner tenure and audit
quality, The AccountingReview, Vol. 81 No. 3, pp. 653-676.
Chaney, P.K., Jeter, D.C. and Shivakumar, L. (2004),
Self-selection of auditors and au