SSRN-Id299843Opinion Shopping, Audit Firm Dismissals, And Audit Committees

7/28/2019 SSRN-Id299843Opinion Shopping, Audit Firm Dismissals, And Audit Committees

1/48

Opinion Shopping, Audit Firm Dismissals, and Audit Committees*

Clive S. Lennox

Hong Kong University of Science and Technology

February 2002

Abstract

This paper tests whether SEC registrants engage in opinion shopping and examines the

role of audit committees when auditors are dismissed (1996-98). There are three main

findings. First, companies strategically dismiss incumbent auditors if they are more

likely to issue unfavorable audit opinions compared to newly appointed auditors. Using

the methodology of Lennox (2000), it is estimated that opinion shopping motivates 17%

of auditor dismissals. Opinion shopping dismissals are also found to occur significantly

later in the fiscal year compared to other types of dismissals. Second, I use the level and

change in meeting activity during the auditor dismissal year to investigate the extent to

which committees participate in auditor dismissal decisions. It is estimated that 15% of

audit committees do not participate in auditor dismissal decisions but opinion shopping

is not significantly associated with participation. Third, Accounting Series Release No.

247 requires that companies disclose whether audit committees approve audit firm

changes. After controlling for non-participation, I find audit committees are more likely

to disapprove of auditor dismissals that are motivated by opinion shopping. However,

independent audit committee members are more likely to leave committees that

disapprove of opinion shopping.

* I owe thanks to Eric Falkenstein at Moodys for providing data on corporate defaults. I am grateful for comments from MikeAdams, Gary Biddle, Kevin Chen, Elisabeth Dedman, Jim Frederickson, Chul Park, TJ Wong and seminar participants at BristolUniversity, City University Business School (London), Edinburgh University, Lancaster University, Hong Kong University ofScience and Technology, Southampton University and Tuck Business School. The research is carried out under the auspices of the

Institute of Chartered Accountants in Scotland and is funded by The Scottish Accountancy Research Trust and Hong Kong Universityof Science and Technology. I would also like to thank Angel Chu, Licardo Chan and Dammi Sze for their research assistance. Anyremaining errors are my own. Tel: +852-23587571. E-mail: [email protected]


2/48

2

1. Introduction

Audit opinions attest to whether financial statements are presented fairly and they can alert

investors to significant problems affecting the interpretation of accounting information.

Unfavorable audit opinions are associated with falling share prices (Fleak and Wilson, 1994),

perhaps because they signal impending bankruptcy (Chen and Church, 1996) and make it harder

for companies to raise capital (Firth, 1980). Unfavorable opinions might also cause falls in

executive remuneration or motivate management changes. These factors mean managers may try

to suppress the release of negative news by avoiding unfavorable audit opinions (e.g., Securities

and Exchange Commission (SEC), 1988). Unfavorable opinions might be avoided through the

strategic use of auditor dismissals this practice is known as opinion shopping. This paper tests

whether companies engage in opinion shopping and investigates the role of audit committees

when auditors are dismissed.

The extant literature draws conflicting conclusions about whether companies engage in

opinion shopping and no previous study examines the mitigating influence of audit committees.

Some studies claim opinion shopping is futile since incoming auditors first opinions are not

generally more favorable than outgoing auditors final opinions (e.g. Krishnan, 1994; Krishnan

and Stephens, 1995).1 Lennox (2000) argues a comparison of opinions issued by outgoing and

incoming auditors is a flawed test for opinion shopping. Using an alternative methodology, he

predicts the probabilities that companies receive unfavorable opinions if audit firm changes are

different to those actually observed. His results for the UK show that opinion shopping is a

highly significant predictor of audit firm changes. One contribution of this paper is to apply the

methodology of Lennox (2000) to SEC registrants in order to determine whether extant

conclusions differ because of institutional differences between the UK and US or because of

differences in empirical methodologies.2

1 DeFond and Subramanyam (1998) examine the association between earnings management and audit firm changes. They finddiscretionary accruals are income decreasing during the last year with the outgoing auditor and insignificant during the first year with

the incoming auditor. Their results are consistent with managers dismissing incumbent auditors in response to auditor conservatism.2Opinion shopping might be more prevalent in the UK where auditor change disclosure requirements are less strict compared to theUS. SEC registrants are required to disclose audit firm changes within five business days of an incumbent auditors termination and


3/48

3

There would be no scope for opinion shopping if a company would receive the same

opinion from a newly appointed auditor as from a retained incumbent auditor (Magee and Tseng,

1990). Therefore, a necessary condition for opinion shopping is that new and retained auditors

report differently. It is found that a new auditor is more likely than a retained incumbent to issue

an opinion that differs from that of the previous year. A sufficient condition for opinion shopping

is that a company dismisses (retains) its incumbent auditor if a new auditor is less (more) likely

to issue an unfavorable opinion. The evidence shows companies use dismissal decisions to avoid

unfavorable opinions and so it is concluded that companies successfully engage in opinion

shopping. The lack of improvement in audit opinions after companies change auditors does not

imply that opinion shopping is futile. Rather, audit opinions do not generally improve because

opinion shopping does not motivate most audit firm changes. Opinion shopping is a highly

significant predictor of auditor dismissals, but is estimated to motivate only 17% of dismissals. I

also find opinion shopping dismissals occur significantly later in the fiscal year compared to

other types of dismissals. This is consistent with companies dismissing when they believe

incumbent auditors are likely to give unfavourable opinions and when information asymmetries

are high for incoming auditors.

A second contribution of this study is to investigate the role of audit committees when

companies dismiss their auditors. As a sub-committee of the board of directors, the audit

committee is considered an important corporate governance mechanism for monitoring senior

management and ensuring accountability to stakeholders (Cadbury Committee, 1992). The audit

committees duties include reviewing the financial statements, hiring and firing external audit

firms, and acting as an intermediary between senior management, internal and external auditors.

The audit committee is part of a regulatory framework that emphasises disclosure and

transparency in financial reporting (Blue Ribbon Committee, 1999). Since opinion shopping

makes financial reporting less transparent it is expected the audit committee should deter opinion

shopping. Barrons National Business and Financial Weekly (1992) argues,

they should disclose auditor-client disagreements and modified opinions in the two previous years. In the UK, auditor changes and


4/48

4

Managers want their audit reports to be positive (unqualified) . . . Excluding

management from auditor hiring-paying-firing could deter opinion shopping.

This might possibly be accomplished through an effective audit committee. 3

An effective audit committee should participate in the companys decision to retain or dismiss

the incumbent audit firm and it should disapprove of dismissals that are motivated by opinion

shopping. This paper provides evidence regarding audit committee participation in, and approval

of, auditor dismissal decisions and examines the influence of opinion shopping on participation

and approval.4

Some studies posit that audit committees are associated with higher quality financial

reporting. DeFond and Jiambalvo (1991) find earnings overstatements are less prevalent in

companies that have audit committees. Similarly Dechow et al., (1996) demonstrate that

companies with audit committees are less likely to commit fraud compared to a control sample

matched by industry and size. For a sample of distressed companies, Carcello and Neal (2000)

find auditors issue going concern opinions more often when audit committee members are

independent of senior management. Chtourou et al., (2001) find less income-increasing earnings

management when audit committees meet more than twice a year and are composed only of

independent directors. They also report less income-decreasing earnings management when the

audit committee has at least one member with financial expertise.

Other studies provide evidence that audit committee effectiveness is rather limited. Klein

(2000) finds a negative relation between audit committee independence and earnings

management but no significant relation between earnings management and 100% independence.

Beasley (1996) finds no significant association between financial statement fraud and audit

committee existence. Menon and Williams (1994) show that voluntarily formed audit committees

disagreements do not have to be disclosed to the London Stock Exchange.3Barry Vinocur (16 March, 1992), SEC Earthquake: Commission Delists Realty Pension Advisers, Barrons National Business

and Financial Weekly (Boston).4There is no extant evidence on the role of audit committees in audit firm dismissals and little evidence regarding the hiring of newaudit firms. Eichenseher and Shields (1985) find a positive association between audit committee existence and the hiring of largeaudit firms, but Cottell and Rankin (1988) find no significant relation. Abbott and Parker (2000) find companies with both active andindependent audit committees are more likely to hire industry-specialist auditors, but the effects of activity and independence are


5/48

5

hold meetings infrequently and they conclude committees are often formed for appearance rather

than because companies rely on them. Similar sentiments have been expressed by the Public

Oversight Board (1993) who state,

in too many instances, audit committee members do not perform their duties

adequately.

Consistent with Menon and Williams (1994), I find many audit committees are inactive,

but there is a significant increase in meeting activity during auditor dismissal years. Audit

committee participation in auditor dismissal decisions is estimated using both the level of, and

increase in, meeting activity during the dismissal year. The increase in meeting activity captures

the possibility that a committee has one or more extra meetings in order to discuss the dismissal.

The level of meeting activity captures the possibility that an active audit committee discusses the

dismissal as part of its regular meeting schedule. It is estimated that only 85% of audit

committees participate in auditor dismissal decisions but no significant association is found

between participation and opinion shopping.

Accounting Series Release No. 247 requires that companies disclose in 8-K filings

whether audit committees approve audit firm changes. An audit committee approves a change if

it participates and does not disapprove, whereas a change is not approved either because the

committee does not participate or because it disapproves. The participation and disapproval

decisions are modelled both jointly and separately using a simultaneous equation system that

takes into account the partial observability of participation and disapproval. After controlling for

non-participation, it is found that audit committees are more likely to disapprove of auditor

dismissals that are motivated by opinion shopping. This is consistent with the view that audit

committees help maintain the integrity of the financial reporting process. However, audit

committee disapproval of opinion shopping is associated with a significantly higher departure

rate for independent audit committee members. It appears that either audit committee members

resign because they do not wish to be associated with opinion shopping dismissals, or managers

insignificant when tested separately. Beasley and Petroni (2001) find independent audit committees are associated with the hiring of


6/48

6

change audit committee members as well as auditors in their attempts to avoid unfavorable

opinions. It is concluded that audit committee effectiveness is diminished by a lack of

participation in auditor dismissal decisions and by a high rate of member departure when

committees disapprove of opinion shopping.

The rest of this paper is structured as follows. Section 2 first explains how the

methodology identifies whether an auditor dismissal is motivated by opinion shopping. It then

describes the data sources and provides descriptive statistics for auditor dismissals, audit

opinions and the control variables. Finally, it is tested whether companies engage in opinion

shopping. Section 3 uses the results of Section 2 to test the effects of opinion shopping on audit

committee participation in, and approval of, auditor dismissal decisions. It then estimates the

effect of audit committee disapproval of opinion shopping on the departure of audit committee

members. Section 4 summarises the papers main results and includes suggestions for future

research on opinion shopping.

2. Identifying opinion shopping companies

This section is structured as follows. Sub-section 2.1 explains the methodology employed to test

for opinion shopping. Sub-section 2.2 explains the data sources and provides descriptive

statistics. Sub-section 2.3 tests whether significant reporting differences exist between retained

and newly appointed audit firms, and sub-section 2.4 tests whether companies exploit reporting

differences in order to avoid unfavorable audit opinions.

2.1 Methodology for testing opinion shopping

Whether there is scope for opinion shopping depends on whether significant reporting

differences exist between retained and newly appointed auditors. Reporting differences make

opinion shopping possible (but not inevitable) since a company can condition its dismissal

industry specialist auditors but not with brand name auditors.


7/48

7

decision on whether a new auditor or the incumbent would be more likely to give a favorable

opinion.

For the lack of a better terminology, an unfavorable audit opinion is usually described as

modified and a favorable opinion is termed unmodified (e.g., Frost, 1994; Carcello and

Palmrose, 1994). An audit opinion dummy )( itM equals one if company i receives a modified

report in period t or zero if it receives an unmodified report. The final audit opinion of an

outgoing auditor is given at t-1 )( 1itM . An auditor dismissal dummy )( itD equals one if

company i appoints a new auditor or zero if it retains its incumbent. The opinion that company i

receives in period t )( itM depends on the auditor dismissal decision )( itD , the opinion received

prior to the dismissal decision )1( itM and a vector of other explanatory variables )( itX . The

conditional probability that company i receives a modified report is ),1,|1Pr( itXitMitDitM =

and, for notational convenience, this probability is written as )1Pr( =Dit

M where the D

superscript denotes the dismissal decision.

Reporting differences exist between retained and new auditors if )10Pr( =it

M

)11

Pr( =itM . Company i receives a modified report with probability )10

Pr( =itM if it retains its

incumbent auditor and with probability )11Pr( =it

M if it hires a new auditor. Company i uses the

auditor dismissal decision to avoid a modified opinion if it dismisses its auditor when )10Pr( =it

M

> )11Pr( =it

M and if it retains its auditor when )10Pr( =it

M )11Pr( =it

M . Estimation takes place

in two stages. First, the modified opinion probabilities are predicted by estimating a probit model

of audit reporting. Next, the difference between opinion probabilities ( )10Pr( =itM - )11Pr( =itM ) is

included in a probit model of auditor dismissal in order to test whether companies engage in

opinion shopping.

The audit reporting model is:


8/48

8

itvitXitDitMitDitDitXitMitM ++++++= 51432110 (1)

The 1 coefficient on prior opinions )1( itM is expected to be positive because extant research

finds strong persistence in audit reporting (e.g., Monroe and Teh, 1993; Krishnan et al., 1996;

Lennox, 2000). As discussed later, the itX vector includes control variables that previous studies

show are associated with modified audit opinions (e.g., profitability, liquidity, leverage, default,

company size and growth). The auditor dismissal dummy ( itD ) and interaction terms

( 1 itMitD and itXitD ) capture reporting asymmetries between retained and new auditors.5

The 3 coefficient measures the sensitivity of audit opinions to auditor dismissal when prior

opinions are unmodified. The 4 coefficient measures the incremental sensitivity of audit

opinions to dismissal when prior opinions are modified. The total sensitivity of audit opinions to

dismissal is 43 + when prior opinions are modified. Interpretation of the 5 coefficient is

similar to 4 . Reporting differences exist between new and retained auditors if the coefficients

3 , 4 , or 5 are statistically significant.

The difference between modified opinion probabilities ( )10Pr( =it

M - )11Pr( =it

M ) is

predicted from eq. (1) and included in an auditor dismissal model as follows:

ituitZitM

itM

itD ++==+= 2))1

1Pr(-)10Pr((10 (2)

The opinion shopping variable ( )10Pr( =it

M - )11Pr( =it

M ) captures the effect of reporting

differences on auditor dismissal decisions. The 1 coefficient is positive if companies

strategically dismiss auditors in order to avoid modified opinions. The itZ vector includes

control variables that previous studies show are associated with audit firm changes (these are

discussed later).

2.2. Data sources and descriptive statistics


9/48

9

2.2.1 Audit firm dismissals

COMPUSTAT provides audit firm names for clients of the big five and 18 medium-sized audit

firms, but it does not identify smaller firms (they are grouped into an other category) and so it

does not provide information on changes between small auditors.

6

AUDITOR-TRAK is a more

comprehensive database on audit firm changes as its source is companies 8-K filings. A few

discrepancies are found between AUDITOR-TRAK and COMPUSTAT regarding the timing of

auditor changes and these are resolved by checking auditor change dates and audit firm names in

8-K and 10-K filings. AUDITOR-TRAK includes audit firm mergers as changes but these are

classified in the paper as non-dismissals. The largest audit firm merger during the sample period

(1996-98) involves Price Waterhouse and Coopers and Lybrand.7

Because of data collection costs, 8-K filings are obtained only where financial statement

data and audit opinions are available from COMPUSTAT and 10-K filings. An 8-K filing is

supposed to disclose whether an auditor change is a dismissal or resignation. Audit firm

resignations are dropped from the sample because opinion shopping concerns auditor dismissal

decisions.8 Audit firm changes are also dropped when: (a) an 8-K filing does not disclose

whether the change is a dismissal or resignation, (b) the change is mutually agreed by the auditor

and client, or (c) the client and auditor disagree about whether the change is a dismissal or a

resignation. This leaves a sample of 828 auditor dismissals and 18,445 auditor retentions (N =

19,273).

2.2.2 Audit opinions

COMPUSTAT codes audit opinions into four types: (1) unqualified with no explanatory

language, (2) unqualified with explanatory language, (3) qualified, and (4) opinion

5The inclusion of interaction terms to capture reporting asymmetries is similar to the modelling approach of Basu (1997) who testsfor asymmetry in the timeliness of accounting income.6 The big five audit firms are Arthur Andersen, Ernst and Young, PricewaterhouseCoopers (Coopers and Lybrand or PriceWaterhouse before 1st July 1998), Deloitte and Touche, and KPMG. The 18 other identified audit firms are: Altschuler, Melvoin andGlasser; BDO Seidman; Baird, Kurtz and Dobson; Cherry, Bekaert and Holland; Clarkson Gordon; Clifton Gunderson; CroweChizek; Grant Thornton; J H Cohn; Kenneth Leventhal; Laventhol & Horwath; McGladery & Pullen; Moore Stephens; Moss Adams;Pannell Kerr Forster; Plante & Moran; Richard A. Eisner; Spicer & Oppenheim.7I identify other audit firm mergers by reading 8-K filings to see why the change occurs.


10/48

10

disclaimer. COMPUSTAT does not provide further information on explanatory language,

qualifications or disclaimers, so opinion types (2)-(4) are collected by hand from 10-K filings.9

Table 1 details the different types of audit opinions. Panel A shows the most frequent

opinion is unqualified with no explanatory language (87.40%). The unqualified with explanatory

language opinions are shown in Panels B-E, the qualified opinions are in Panel F and the opinion

disclaimers are in panel G. Unqualified opinions with explanatory language are grouped into

Panels B-E according to the type of language used. Panel B contains opinions that are harmless

whereas Panels C and D contain unfavorable opinions. Panel E contains opinions that are not

easily classified as harmless or unfavorable.

[INSERT TABLE 1 HERE]

Panel B shows 5.75% of opinions are unqualified with harmless explanatory language.

Explanatory language is assumed to be harmless if it contains one or more of the following

statements: (a) the financial statements comply with SEC regulations, (b) the opinion is shared

with another audit firm, (c) there is a change in accounting principles, or (d) certain events affect

the comparability of current and prior year financial statements. Auditors refer to changes in

accounting principles when there is a material effect on the financial statements. Most changes in

accounting principles occur because of new accounting standards rather than for voluntary

reasons, so (c) is assumed to be harmless.10 Financial statement comparability is usually

mentioned when there is a change in accounting principles or when the company engages in

merger or acquisition activity, so (d) is also viewed as harmless. An exception is when the

opinion refers to the correction of accounting errors, in which case it is shown in Panel E rather

than Panel B.

8A limitation of the Lennox (2000) study is its failure to distinguish between auditor dismissals and resignations.

9In general, 10-K filings are not collected for unqualified opinions with no explanatory language. An exception is that all opinionsbefore and after dismissals and all changes in opinions are checked against 10-K filings.10

Voluntary changes in accounting principles might be viewed as harmless or unfavorable (e.g., opportunistic earningsmanagement), so it is unclear how these opinions should be coded. In unreported results, the papers conclusions are found to berobust when these opinions are dropped from the sample.


11/48

11

Panel C contains unqualified opinions where the explanatory language refers to going

concern difficulties (6.10% of opinions).11 A going concern opinion is viewed as unfavorable

because it is issued if the auditor concludes there is substantial doubt about the companys ability

to continue trading.

12

Audit opinions sometimes contain multiple bad news disclosures (e.g.,

going concern plus litigation uncertainty) and these are shown separately in Panels C-G. Panel D

contains unqualified opinions where the explanatory language refers to a fundamental

uncertainty other than going concern. The most common fundamental uncertainty is litigation

against the company. Fundamental uncertainties are viewed as unfavorable because they indicate

high risk and usually involve potential future cash outflows.

Panel E contains unqualified opinions where the explanatory language does not refer to

going concern or fundamental uncertainty but where the language might be viewed as

unfavorable. Some opinions disclose financial distress (e.g., sale of a significant part of the

companys operations, debt refinancing) but presumably the auditor believes the distress is not so

severe that it warrants a going concern opinion. Some opinions indicate existing problems (e.g.,

lack of compliance with SEC filing requirements, criminal investigation) while others hint at

potential problems (e.g., related party transactions).

Panel F contains qualified (except for) opinions which are issued in one of two

circumstances: (a) there is a reporting disagreement if an auditor believes that part of the

financial statements is not fairly presented, or (b) there is a limitation on audit scope if an auditor

is unable to collect sufficient evidence on part of the financial statements. An auditor can issue

an except for opinion only if he/she concludes the overall financial statements are fair.13 Except

for opinions are rarely issued as shown by the 0.06% frequency. Panel G contains opinion

disclaimers (0.07%), which are issued when the scope limitation is so severe that the auditor is

unable to form an opinion about whether the financial statements are fair.

11There are 171 reports where auditors disclose the company is in bankruptcy (these nearly always contain statements about going

concern uncertainty). Bankruptcy is a public signal of poor performance irrespective of the audit opinion, so bankrupt observationsare dropped from the sample.12

Some dismissal companies mistakenly refer to going concern opinions as audit qualifications in their 8-K filings, which isconsistent with going concern opinions being viewed as negative.


12/48

12

It is necessary to map the different types of opinions into a quantitative variable. Since

the opinions in Panels A and B are not unfavorable, they are coded as unmodified ( 0=itM ).

Going concern opinions warn about bankruptcy, fundamental uncertainties indicate high risk, and

qualifications and opinion disclaimers are obviously unfavorable, so the opinions in Panels C, D,

F and G are modified ( 1=itM ). The Panel E opinions are more difficult to code as they are not

obviously unfavorable. The reported results are obtained after coding the Panel E opinions as

modified. This results in a sample of 17,952 unmodified opinions and 1,321 (6.8%)

modifications, most of which mention going concern problems.

There are three objections to this mapping of the audit opinion variable. First, the coding

of the Panel E opinions is inappropriate if they are favorable. In unreported results, the papers

conclusions are found to be robust when: (1) the Panel E opinions are dropped from the sample,

(2) the Panel E opinions are unmodified, or (3) some Panel E opinions are modified (e.g.,

accounting errors) and others are unmodified (e.g., death of CEO). The robustness is

unsurprising since Panel E accounts for only 0.43% of opinions. Second, it might be argued that

disclaimers (Panel G) are more negative than qualifications (Panel F), which in turn are more

negative than unqualified opinions. However, a separate analysis of disclaimers and

qualifications is not meaningful since Panels F and G contain only 11 and 13 opinions

respectively. Third, multiple bad news disclosures might be worse than single disclosures. For

example, a going concern opinion that also mentions accounting errors might be more negative

than a going concern opinion alone. In practice, any attempt to rank audit opinions is fraught with

difficulties. For example, is a going concern opinion plus litigation uncertainty worse than a

going concern opinion plus accounting error? Are these multiple unqualified opinions worse

than a single qualification? Without answers to such questions, a ranking is too subjective and so

is avoided. In any case, only 53 (0.28%) reports have multiple negative disclosures, so this is not

13 An adverse opinion should be issued when an auditor believes that the overall financial statements are not fair. There are noadverse opinions in the sample.


13/48

13

a significant limitation. Attention now turns to the control variables ( itX and itZ ) in the audit

reporting (eq. (1)) and auditor dismissal (eq. (2)) models.

2.2.3 Financial health

Financially distressed companies are more likely to receive modified opinions compared to

healthy companies (Levitan and Knoblett, 1985; Chen and Church, 1992; Monroe and Teh, 1993;

Krishnan, 1994; Carcello et al., 1995) and they are more likely to change auditors (Menon and

Schwartz, 1985; Krishnan, 1994; DeFond and Subramanyam, 1998). It is therefore important to

control for financial health in the audit reporting and auditor dismissal models. The financial

health variables are profitability (

it

PROF ), liquidity (

it

LIQ ), leverage (

it

LEV ) and corporate

default ( itDEF ). Profitability ( itPROF ) is net income divided by total assets, liquidity ( itLIQ )

is current assets divided by current liabilities, leverage ( itLEV ) is total liabilities divided by total

assets and default ( itDEF ) is a dummy indicating the companys default status ( itDEF equals

one if the company is in default in the period -365 to +90 days around the year-end). Financial

ratios are obtained from COMPUSTAT and Moodys provide the default data.14

2.2.4 Company size and growth

Small companies receive modified opinions more often than large companies (Monroe and Teh,

1993; Krishnan, 1994; Carcello et al., 1995) and small companies are more likely to change

auditors (Krishnan et al., 1996). Company size ( itSIZE ) is measured using total assets and

company growth ( itGROWTH ) is the annual percentage change in total assets. Both the audit

reporting and auditor dismissal models control for company size and growth. The association

between company growth and modified opinions is expected to be negative since low growth

14Moodys default data covers bond defaults for companies both rated and unrated by Moodys. Moodys compiles default historiesfrom a variety of sources including: Industrial, Railroad, and Public Utilities Manuals; reports of the National Quotation Service; The

Commercial and Financial Chronicle; financial reports; press releases; press clippings; internal memoranda; records of analyst contactwith rated issuers; and documents from the Securities and Exchange Commission, Dun & Bradstreet, the New York Stock Exchange,and the American Stock Exchange. A bond default is defined as any missed or delayed disbursement of interest and/or principal,bankruptcy, receivership, or distressed exchange where (i) the issuer offers bondholders a new security or package of securities that


14/48

14

could be a symptom of poor performance. The relation between growth and auditor dismissals is

expected to be non-monotonic because companies that grow or decline rapidly are more likely to

change auditors compared to companies that grow at a steady rate (Williams, 1988; Johnson and

Lys, 1990; Beattie and Fearnley, 1995). Rapid growth and decline reflect structural changes in

clients and are associated with changes in the type of audit firm demanded. Growing companies

tend to switch to larger audit firms whereas declining companies switch to smaller firms.

Fig. 1 confirms a U-shape relation between company growth and auditor dismissals, with

dismissals more frequent in the top and bottom 10% of the growth ranking. The auditor dismissal

model captures this non-monotonic relation by including dummy variables for rapid growth

( itGROW ) and decline ( itDEC ).15 After ranking the sample by growth, itGROW equals one for

the top 10% and itDEC equals one for the bottom 10%.

[INSERT FIG. 1 HERE]

2.2.5 Investment opportunities

It is expected that companies with poor investment opportunities are more likely to receive

modified audit opinions. The book-to-market ratio ( itBM ) is often used as a proxy for

investment opportunities but, to my knowledge, has not been used in prior research on audit

opinions.16 Since investment opportunities are inversely related to the book-to-market ratio, a

positive association is expected between itBM and modified opinions. The book-to-market ratio

( itBM ) is also included in the auditor dismissal model.

amount to a diminished financial obligation (such as preferred or common stock, or debt with a lower coupon or par amount) or (ii)the exchange has the apparent purpose of helping the borrower avoid default.15The non-monotonic relation between growth and auditor dismissal is also estimated using )( itGROWTHR and

2)( itGROWTHR .

Consistent with the U-shape relation, the )( itGROWTHR coefficient is significantly negative and the2)( itGROWTHR coefficient is

significantly positive but this (unreported) model gives a slightly worse fit compared to the one reported using growth dummies.16

In unreported results, the papers conclusions are not found to be sensitive to inclusion of itBM .


15/48

15

2.2.6 Functional form

Table 2 provides descriptive statistics for the non-dummy control variables ( itPROF , itLIQ ,

itLEV , itSIZE , itGROWTH , itBM ), and shows the variables are highly skewed and contain

outliers. Sample trimming and log or square root transformations are traditional methods for

dealing with skewness and outliers, but these procedures are deemed less effective than the rank

transformations adopted in this paper (e.g., see Kane and Meade, 1998).17

For a variable with N

observations in year t, each observation is replaced with its corresponding rank (from i = 1, . . . ,

N in ascending order) and the rank assigned to observation i is divided by N+1. The rank

transformed variables ( )( itPROFR , )( itLIQR , )( itLEVR , )( itSIZER , )( itGROWTHR , and

)( itBMR ) are uniformly distributed between zero and one, as Fig. 1 illustrates for company

growth.


Table 3 is a correlation matrix for the audit opinion ( itM ), auditor dismissal ( itD ) and

control variables. Most of the correlations are statistically significant but the coefficients are low

enough to suggest that multicollinearity is not a problem. The signs of all correlation coefficients

are in the direction expected. For example, companies are more likely to default if profitability

and liquidity are low and leverage is high. Companies receive modified opinions and dismiss

auditors more often if they are small, unprofitable, highly leveraged, have poor liquidity and poor

investment opportunities.


17Kane and Meade (1998) find rank transformations contain information that is obfuscated by untransformed variables or alternativetransformations. In addition, log and square root transformations are not viable for variables with non-positive observations

( itPROF , itLIQ , itGROWTH ). Simulation studies indicate little loss of information even when rank transformations are applied to

normally distributed variables (Conover and Iman, 1980).


16/48

16

2.3 Are there significant reporting differences between retained and new audit firms?

This sub-section investigates whether there exists scope for opinion shopping by testing whether

significant reporting differences exist between retained and new audit firms. Columns (1)-(3) of

Table 4 provide results for three models of audit reporting (eq. (1)).

18

Column (1) is a benchmark

model, which makes no distinction between the opinions of new and retained auditors

( 0543 === ). The positive coefficient on prior opinions ( 1itM ) shows strong persistence

in audit reporting. The negative coefficients on profitability ( )( itPROFR ), liquidity ( )( itLIQR ),

size )( itSIZER and growth )( itGROWTHR show modified opinions are issued more often to

companies that are small, unprofitable and that have poor liquidity and growth. The positive

coefficients on )( itLEVR and itDEF show companies receive modified opinions more often if

they are highly leveraged or in default of debt obligations. The positive coefficient on book-to-

market ( )( itBMR ) reveals companies receive modified opinions more often when investment

opportunities are poor or when assets are not being efficiently utilised.

Column (2) adds the auditor dismissal dummy ( itD ) and interaction terms in order to

identify reporting differences between new and retained auditors. The positive coefficient on the

dismissal dummy shows that new auditors issue modified opinions more often than retained

incumbents when prior opinions are unmodified. The negative coefficient on 1 itMitD

captures the incremental sensitivity of audit opinions to dismissal when prior opinions are

modified. The other interaction terms have insignificant coefficients ( 5 = 0) and are omitted

from Column (3). The total sensitivity of audit opinions to dismissal is negative when prior

opinions are modified since the incremental effect ( 4 ) more than offsets the positive coefficient

on the dismissal dummy ( 4

3 + = 0.2 - 0.8 = -0.6). This implies that new auditors issue

modified opinions less often than retained incumbents when prior opinions are modified. The

significant reporting differences between retained and new auditors ( 04,03

) imply that


17/48

17

there exists scope for opinion shopping ( )10Pr( =it

M )11Pr( =itM ).

19 An opinion shopping

company would retain an incumbent whose prior opinion is unmodified because dismissal

increases the likelihood of a modified opinion ( 3 > 0). An opinion shopping company would

dismiss an incumbent whose prior opinion is modified because dismissal reduces the likelihood

of another modified opinion ( 4

3 + < 0).


2.4 Do companies successfully engage in opinion shopping?

This sub-section tests whether companies exploit significant reporting differences between

retained and new audit firms in order to avoid modified opinions. The results in Column (3) of

Table 4 are used to predict the difference between modified opinion probabilities for retained and

new auditors ( )10Pr( =it

M - )11Pr( =it

M ). The predicted opinion probabilities are included in the

auditor dismissal model (eq. (2)) in order to test whether companies engage in opinion shopping.

Columns (4)-(5) of Table 4 show the coefficients on )10Pr( =it

M - )11Pr( =it

M are significantly

positive ( 1 = 1.8, z = 6.6). This implies that incumbent auditors are dismissed more (less) often

when they are more (less) likely to give modified opinions compared to new auditors. It is

therefore concluded that companies successfully engage in opinion shopping.20

18

Company observations may not be independent over time so unbiased standard errors are obtained using the robust clustercommand in STATA. The usual assumption of independence between companies is retained.19Consistent with extant research (e.g., Monroe and Teh, 1993; Krishnan and Stephens, 1995), no significant reporting differencesare found between large (big five) and small (non-big five) audit firms. Therefore, the scope for opinion shopping does not depend on

audit firm size.20The auditor dismissal model does not include all possible determinants of dismissal for example, it is not possible to control foraudit fees as they are undisclosed during the sample period. It is investigated whether these omitted factors confound the results foropinion shopping by examining whether 8-K filings disclose the reasons for dismissal. There are 163 companies (19.7%) that disclosewhy auditors are dismissed and the most common reason is the need for lower fees (55 dismissals). When companies that disclose

reasons are dropped from the sample, the opinion shopping coefficient and level of significance are similar to Table 4 ( 1 = 1.67, z =

5.9), suggesting the opinion shopping coefficient is not biased by omitted variables. In addition, no significant correlation is foundbetween opinion shopping and the companys propensity to disclose a reason for the dismissal.


18/48

18

It is predicted that an auditor dismissal is motivated by opinion shopping if )10Pr( =it

M >

)11Pr( =it

M . There are 142 dismissals that are motivated by opinion shopping whereas 686 are

motivated by other reasons. This demonstrates that opinion shopping motivates only a minority of

dismissals (17.1%). It is interesting to note the opinion shopping dismissals occur significantly later

in the fiscal year compared to other dismissals. The mean (median) number of days from the date of

dismissal to the year-end reported on by the new auditor is 49 (28) for the opinion shopping

dismissals and 105 (85) for the other types of dismissals.21 The difference in mean (median) days is

statistically significant (t = -2.71, t = -4.92 respectively). The incoming auditor is probably under

greatest time pressure when appointed after the fiscal year-end. It is predicted that opinion shopping

motivates 29.2% of dismissals after the year-end but only 13.8% of dismissals before the year-end.

The later dismissal date for opinion shopping companies is consistent with companies dismissing

when they believe incumbent auditors are likely to give modified opinions and when information

asymmetries are high for incoming auditors.

Table 4 also shows financial health is a statistically significant predictor of auditor

dismissals. As expected, companies are more likely to dismiss when profitability ( )( itPROFR ) is

low and leverage ( )( itLEVR ) is high. The negative coefficient on company size ( )( itSIZER )

shows that large companies are less likely to dismiss than small companies. Companies that grow

rapidly ( itGROW ) are more likely to dismiss but the decline dummy ( itDEC ) is insignificant

after controlling for financial health. Finally, the positive coefficient on the book-to-market ratio

( )( itBMR ) reveals dismissal companies have poor investment opportunities.22

3. Opinion shopping and the role of audit committees

This section considers the role of audit committees and addresses the following questions: To what

extent do audit committees participate in auditor dismissal decisions? Are audit committees more or

21Care is taken to use the date of dismissal rather than when the dismissal is announced.


19/48


20/48

20

decision. A participation dummy ( itPA ) equals one if the committee participates and equals zero if

it does not. If the audit committee does not participate ( 0=itPA ) then, by definition, it cannot

approve the auditor dismissal. If the committee does participate ( 1=itPA ), an approval dummy

( itAP ) equals one if the committee approves dismissal and equals zero if it disapproves.

[INSERT FIG. 2 HERE]

An 8-K filing discloses only thejointoutcome of an audit committees participation and

approval decisions ( itAPitPA ). An auditor dismissal is disclosed as approved ( itAPitPA = 1)

if the audit committee both participates ( 1=itPA ) and does not disapprove ( 1=itAP ). A

dismissal is not approved ( itAPitPA = 0) either because the committee does not participate

( 0=itPA ) or because it participates and disapproves ( 1=itPA , 0=itAP ).24

Since itPA and itAP

are only partially observed, an important task is to identify whether non-approval ( itAPitPA =

0) is due to non-participation or disapproval. The distinction is important because non-

participation could indicate laziness whereas disapproval might be a sign of audit committee

integrity. Audit committee non-participation is expected to be important for two reasons. First,

prior research indicates that many voluntary committees are inactive and are formed for the sake

of appearance (Menon and Williams, 1994). Second, most sample companies are listed on stock

exchanges that require the formation of audit committees, and compulsory committees might be

less active than committees that are formed voluntarily.

Table 5 provides descriptive statistics for opinion shopping and audit committee

participation-approval ( itAPitPA ). Audit committee non-approval is 62% (= 61/98) when opinion

shopping motivates dismissals ( )10Pr( =it

M > )11Pr( =it

M ) and 35% (= 207/586) when dismissals

23Some 8-K filings use the word recommended rather than approved - recommendations and approvals are treated equivalently in

this paper.24When 0= itit APPA the 8-K filing either does not disclose who approves the dismissal or it discloses approval by individuals

other than the audit committee (usually the full board of directors).


21/48

21

are motivated by other factors and the difference is significant at the 1% level. This implies that

audit committees are either less likely to participate or they are more likely to disapprove when

opinion shopping motivates dismissals.25 The next three sub-sections investigate whether opinion

shopping is associated with non-participation or with disapproval.


3.3 Audit committee participation and meeting activity

This sub-section identifies audit committee participation by examining meeting activity. The

number of audit committee meetings is obtained from 10-K and proxy filings for the auditor

dismissal year ( itMD ) and a non-dismissal year ( .iMND ). The non-dismissal year is the year prior

to auditor dismissal or, if meeting data for that year are unavailable, the year after dismissal is

used.26 Audit committee activity during the dismissal year is measured using the number of

meetings ( itMD ) and the abnormal number of meetings ( .iMNDitMDitAM ).

Meeting activity is undisclosed for 167 dismissals thereby reducing the sample to 517 (=

684 167).27 Non-disclosure is a potential cause for concern because a company might not

disclose the number of meetings if the audit committee does not meet (i.e., it may not disclose

zero meetings). In this case, the disclosed number of meetings overstates the true level of audit

committee activity. It is shown later (sub-section 3.5) the evidence supports this conjecture. The

predicted level of meeting activity is found to be significantly lower for companies that do not

25Non-approval could be perceived to be a negative signal so it is tested whether the committees decision to approve affects thereport of the incoming auditor. The (unreported) association between audit committee approval and modified opinions is found to be

negative (as expected) but insignificant.26Proxy and 10-K filings disclose the number of audit committee meetings during the fiscal year but do not disclose meeting dates. It isstraightforward to measure meeting activity during the auditor dismissal year if the dismissal occurs prior to the year-end reported on bythe new auditor. However, if the dismissal occurs after the year-end it is unclear which year should be used to measure meeting activity.For example, suppose the fiscal year is 1st January 19X1 31st December 19X1, the outgoing auditor is dismissed on 1st February 19X2and the incoming auditor issues a report for the year ending 31st December 19X1. In this example, meeting activity in the dismissal year ismeasured using the fiscal year 1st January 19X2 31st December 19X2 since this is the period in which the auditor is dismissed. Thismight introduce measurement error because the audit committee could meet to discuss the dismissal before 31 st December 19X1.Fortunately only a minority of dismissals (22.9%) occur between the year-end and the incoming auditors report and it is found(unreported) that dropping these from the sample does not affect the following results.27

One year is used to measure meeting activity in the non-dismissal period in order to avoid losing more observations. There is nosignificant trend in meeting activity during the sample period (1996-98), so it makes no difference whether the non-dismissal year isthe year before or the year after dismissal.


22/48

22

disclose compared to companies that do disclose. However, the results for audit committee

participation and approval are robust whether the analysis uses predicted meeting activity (684

observations) or disclosed meeting activity (517 observations). Therefore, the lack of meeting

disclosure is not a significant limitation.

The American Bar Association (1978) recommends that audit committees hold a minimum

of two meetings per year one during the planning stage of the audit and one to discuss the audited

financial statements. More recently, the Blue Ribbon Committee (1999) recommends that audit

committees meet at least four times a year. Table 6 shows nearly half of audit committees hold

fewer than two meetings and 85% hold fewer than four meetings. In the non-dismissal year, 10.2%

of committees have no meetings, 37.7% meet once, 24.6% meet twice, 13.2% meet three times,

9.9% meet four times and 4.4% meet five or more times. These meeting frequencies are lower than

when committees are formed voluntarily (Menon and Williams, 1994).28

Meeting activity in the non-dismissal year is compared to the dismissal year in order to test

whether committees participate in auditor dismissal decisions. The rationale is that a committee

might arrange one or more extra meetings in order to discuss the dismissal. A limitation of this test

is that audit committees might participate in auditor dismissal decisions even if they do not arrange

additional meetings - this possibility is addressed in sub-section 3.4.

Under the null hypothesis that audit committees do not participate (H1), there is no

abnormal meeting activity in the dismissal year ( itAM = 0). Under the alternate hypothesis that

some audit committees participate there are more meetings in dismissal years ( itAM > 0). Table 6

shows an average increase of 0.45 meetings in dismissal years. This abnormal meeting activity is

significantly positive (t = 6.71), indicating that some audit committees participate in auditor

dismissal decisions.

Next it is tested whether non-approval ( itAPitPA = 0) is always due to non-participation

( 0=itPA ) or whether some audit committees participate and disapprove ( 1=itPA , 0=itAP ). Under


23/48

23

the null hypothesis that non-approval is always due to non-participation (H2), there is no abnormal

meeting activity when itAPitPA = 0. Under the alternate hypothesis that some audit committees

participate and disapprove, there is positive abnormal meeting activity when itAPitPA = 0. Table

6 shows a significant increase of 0.26 meetings when itAPitPA = 0 (t = 3.34). It is therefore

concluded that some committees participate and disapprove of auditor dismissal decisions.

Finally, it is tested whether non-approval ( itAPitPA = 0) is always due to disapproval

( 1=itPA , 0=itAP ) or whether some audit committees do not participate ( 0=itPA ). Under the null

hypothesis that all committees participate (H3), abnormal meeting activity is equally high when

itAPitPA = 0 compared to when itAPitPA = 1. Under the alternate hypothesis that some audit

committees do not participate, abnormal meeting activity is lower when itAPitPA = 0 compared

to when itAPitPA = 1. Table 6 shows an increase of 0.55 meetings when itAPitPA = 1

compared to an increase of 0.26 meetings when itAPitPA = 0. The difference is statistically

significant (t = 2.30), so it is concluded that some audit committees do not participate in auditor

dismissal decisions.

To summarize, some but not all audit committees participate in auditor dismissal decisions,

and, of committees that participate, some but not all approve of dismissals. In the next two sub-

sections I test whether opinion shopping explains the cross-sectional variation in participation and

approval and I estimate the extent of audit committee non-participation.


3.4 The association between opinion shopping and audit committee participation

A limitation of the previous sub-section is that an audit committee might participate in the auditor

dismissal decision without arranging extra meetings. This is particularly likely if the audit

28Menon and Williams (1994) find 6% of voluntary committees do not meet during the year, 32% meet once, 28% meet twice, 12%

meet three times, 13% four times and 9% five or more times. Following the report of the Blue Ribbon Committee (1999), Carcello etal., (2001) find the average audit committee charter requires a minimum of 3.31 meetings (below the recommended number).


24/48

24

committee is already active since it can discuss the dismissal as part of its regular meeting schedule.

This possibility is addressed by using two meeting variables to capture audit committee

participation: (1) the level of meeting activity ( itMD ) since an active committee might discuss the

dismissal as part of its regular meeting schedule, and (2) abnormal meeting activity ( itAM ) since a

committee might hold extra meetings to discuss the dismissal. As will be shown, the conclusions are

robust to using either variable as a predictor of audit committee participation.

Attention now turns to the association between opinion shopping and audit committee

participation (meeting activity). The dependent variable in eq. (3) is the number of meetings

( itMD ) and the dependent variable in eq. (4) is abnormal meeting activity ( itAM ):

ititSIZERitMitMitMD ++==+= )(2))11Pr()10(Pr(10 (3)

ititSIZERitM

itMitAM ++==+= )(2))1

1Pr()10(Pr(10 (4)

A negative association between opinion shopping ( )10Pr( =it

M - )11Pr( =it

M ) and meeting activity

indicates that audit committees participate in auditor dismissal decisions less often when companies

engage in opinion shopping ( 01 ) because an audit committee

might meet in order to oppose opinion shopping. Company size ( )( itSIZER ) is included as a control

variable because Menon and Williams (1994) find audit committees of large companies are more

active than those of small companies ( 02 > and 02 > ).

The results are reported in Columns (1)-(4) of Table 7. As expected, the association

between company size and audit committee activity is positive and statistically significant ( 2 =

1.2, z = 10.4 and 2 = 1.0, t = 3.3). Therefore, audit committees of large companies are more active

and are more likely to participate in auditor dismissal decisions.29 In contrast, there is no significant

relation between opinion shopping and meeting activity ( 1 = 0.3, z = 0.6 and 1 = 0.1, t = 0.1). It


25/48

25

is therefore concluded that opinion shopping does not affect audit committee participation in

dismissal decisions. The next sub-section tests whether audit committees are more likely to

disapprove when opinion shopping motivates auditor dismissals.


3.5 The association between opinion shopping and audit committee approval

This sub-section tests the relation between opinion shopping and audit committee approval of

dismissal decisions. The participation and approval decisions are first modelled jointly using a

single dependent variable ( itAPitPA ) and later they are modelled separately using two

dependent variables ( itPA and itAP ). In eqs. (5)-(6) the dependent variable is the joint

participation-approval outcome:

ititM

itMitMDitAPitPA +==++= ))1

1Pr()10(Pr(210 (5)

ititM

itMitAMitAPitPA +==++= ))1

1(Pr()10(Pr(210 (6)

If active audit committees are more likely to participate in auditor dismissal decisions, the

coefficients on the meeting variables (it

MD andit

AM ) are positive (1 > 0 and

1 > 0). The

opinion shopping variable ( )10Pr( =it

M - )11Pr( =it

M ) captures the effect of companies dismissal

motives. The coefficients 2 and 2 are negative if audit committees disapprove of dismissals that

are motivated by opinion shopping.

The results for eqs. (5)-(6) are shown in Columns (1)-(2) of Table 8. A significant positive

relation exists between meeting activity and the joint participation-approval outcome ( 1 = 0.2, z =

4.8 and 1 = 0.1, z = 2.2). This confirms that active audit committees are more likely to participate

in auditor dismissal decisions. The coefficients on the opinion shopping variable are significantly

29 In unreported results, meeting activity shows no significant trend over the sample period and no significant relation is foundbetween meeting activity and audit committee size.


26/48

26

negative ( 2 = -1.7, z = -2.2 and 2 = -2.0, z = -2.5). Therefore, audit committees are more likely

to disapprove when opinion shopping motivates auditor dismissals.

Three objections to these results are now considered. First, meeting activity is undisclosed

for 167 observations. Non-disclosure causes meeting activity to be biased upwards if companies are

less likely to disclose when committees do not meet. Second, the estimates of 1 and 1 could be

biased because meeting activity is endogenous. In other words, meeting activity might be

relatively high in the auditor dismissal year for reasons not directly associated with dismissal or

opinion shopping. For example, accounting issues might independently increase the likelihood of

auditor dismissal and increase the number of audit committee meetings, causing 1 and 1 to be

biased upwards. Third, audit committee participation and approval are modelled jointly, whereas

Fig. 2 depicts them as separate sequential decisions.

The first two objections are addressed using company size as an instrumental variable for

meeting activity. Company size data are available for all dismissals so estimation results can be

obtained using predicted rather than disclosed meeting activity. Company size is also a strong

predictor of meeting activity, so it is an ideal instrument to control for endogeneity. The meeting

variables ( itDM and itMA ) are first predicted from Columns (2) and (4) of Table 7. The mean

predicted number of meetings ( itDM ) is 2.35 when meeting activity is disclosed (N = 517) and

2.00 when it is not disclosed (N = 167). This difference is statistically significant (t = 5.15)

implying that companies are less likely to disclose meeting activity when activity is low.

Next, itDM and itMA are included as predictors of the joint participation-approval

outcome:

ititM

itMitDMitAPitPA +==++= ))1

1Pr()10(Pr(2

10 (7)

ititM

itMitMAitAPitPA +==++= ))1

1Pr()10(Pr(2

10 (8)

The results for eqs. (7)-(8) are shown in Columns (3)-(4) of Table 8. Note that Columns (3)-(4) are

estimated for 684 observations using predicted meeting activity, whereas Columns (1)-(2) (eqs. (5)-


27/48

27

(6)) are estimated for 517 observations using disclosed meeting activity. Columns (1)-(4) are

consistent in showing that itAPitPA is positively associated with meeting activity and negatively

associated with opinion shopping. Therefore, the results are robust to the non-disclosure and

endogeneity of meeting activity.


The third objection is addressed by modelling the participation and approval decisions

separately:

ititDMitPA ++=

10 (9)

ititM

itMitAP +==+= ))1

1Pr()10(Pr(20 (10)

The predicted number of meetings ( itDM ) explains audit committee participation ( itPA ) and the

predicted opinion shopping variable ( )10Pr( =it

M - )11Pr( =it

M ) explains approval ( itAP ). Eqs. (9)-

(10) are estimated taking into account the sequential decision-making process illustrated in Fig. 2

and the partial observability of participation and approval. Abowd and Farber (1982) developed the

estimation procedure for sequential decisions and partially observed dependent variables in labour

economics.30 The likelihood function for eqs. (9)-(10) is:

=

==++

=

==++

0)))]11Pr()10(Pr(20()

10(1[

1

)))]11Pr()10(Pr(20()

10([

itAPitPAitMitMitDM

itAPitPAitMitMitDM

The likelihood function is programmed into STATA and maximized numerically using the

softwares algorithms. Eqs. (11)-(12) are the same as eqs. (9)-(10) except that abnormal meeting

activity ( itMA ) is used to explain participation:

ititMAitPA ++=

10 (11)


28/48

28

ititM

itMitAP +==+= ))1

1Pr()10(Pr(20 (12)

The results for eqs. (9)-(12) are shown in Columns (5)-(8) of Table 8. The positive

coefficients on itDM and itMA in Columns (5) and (7) show active audit committees are more

likely to participate in auditor dismissal decisions (z = 2.6). Columns (5) and (7) give mean

predicted participation probabilities ( )1Pr( =itAP ) of 85.9% and 85.4% respectively. It is

therefore estimated that 15% of audit committees do not participate in auditor dismissal

decisions. The negative coefficients on )10Pr( =it

M - )11Pr( =it

M in Columns (6) and (8) confirm

that audit committees are more likely to disapprove when opinion shopping motivates auditor

dismissals (z = -2.7).

3.6 What happens to audit committee members when committees disapprove of opinion

shopping?

This section tests whether audit committee disapproval of opinion shopping is associated with a

higher frequency of member departure. It is hypothesized that senior management might remove

audit committee members that disapprove of opinion shopping or members might leave voluntarily

if they disapprove. It is not possible to discriminate between these two explanations because

companies do not disclose whether member departures are resignations or dismissals. The names of

audit committee members before and after auditor dismissals are obtained from 10-K and proxy

filings. Members names are unavailable for 227 dismissals due to missing filings and lack of

disclosure so the sample size is 457 (= 684 - 227).

The mean number of audit committee members prior to auditor dismissal ( itMEM ) is 2.84

and the number of member departures during the auditor dismissal year ( itMEML ) is 0.67. The

mean departure rate (itMEMitMEML ) is 24.1% showing that nearly a quarter of audit committee

members leave during the auditor dismissal year. When auditor dismissals are approved

30Maddala (1983) (p. 278-280) provides a detailed description of the sequential probit model with partial observability.


29/48

29

( itAPitPA = 1) the mean departure rate is 19.2%, but when dismissals are not approved

( itAPitPA = 0) the mean departure rate is 32.1%, which is significantly higher (t = 3.85). It is

important to determine whether member departure is positively associated with non-participation

( 0=itPA ) or with disapproval ( 1=itPA , 0=itAP ). Committee members might have less incentive

to participate if they know they are going to depart or they might be removed because they are

inactive. Alternatively, members might leave voluntarily if they disapprove or managers might

remove members who disapprove.

The predicted opinion shopping variable provides a clue as to whether departure is

associated with non-participation or with disapproval. When dismissals are motivated by opinion

shopping (( )10Pr( =itM > )11Pr( =it

M ) the mean departure rate (itMEMitMEML ) is 38.4%, but when

dismissals are not due to opinion shopping the mean departure rate is 22.1%, which is

significantly lower (t = 2.57). Since audit committee members are more likely to depart when

companies engage in opinion shopping, departure appears to be associated with disapproval

rather than non-participation.

Eq. (13) estimates the effect of audit committee participation and approval on member

departure:

ititPAitAPitMEMitMEML ++++=

3

210 (13)

The dependent variable is the number of members that leave the audit committee during the

auditor dismissal year ( itMEML ). Eq. (13) controls for the number of members prior to auditor

dismissal ( itMEM ) because large audit committees have more members that might potentially

leave ( 01> ). Audit committee participation (

itAP ) and approval (

itPA ) are predicted from

Columns (5)-(8) of Table 8. It is expected that audit committee members are more likely to

depart when audit committees disapprove of auditor dismissals ( 03


30/48

30

The results for eq. (13) are shown in Columns (1)-(2) of Table 9. The insignificant

coefficients on participation ( itAP ) show member departure is not associated with audit

committee inactivity ( 2 = -0.0, z = -0.2). The significant negative coefficients on approval

( itPA ) reveal that members are more likely to depart when audit committees disapprove of

auditor dismissals ( 3 = -0.8, z = -2.8). Similar (unreported) findings emerge when the departure

model includes the predicted opinion shopping variable ( )10Pr( =it

M - )11Pr( =it

M ) and meeting

activity ( itDM and itMA ). The coefficients for the meeting variables are insignificant,

confirming that member departure is not associated with activity. The coefficient on opinion

shopping is positive and significant, showing that members are more likely to depart when

opinion shopping motivates auditor dismissals.

Additional robustness checks are performed in order to test whether the results reflect the

confounding effects of omitted variables. First, financial health variables ( )( itPROFR ,

)( itLIQR , )( itLEVR and itDEF ) are included as controls because member departure could be

caused by financial distress. The financial health variables are found to be jointly insignificant

and the association between audit committee disapproval and member departure remains

significant. Second, the departure of board members who do not sit on the audit committee is

examined in order to test whether the results are unique to the audit committee. This addresses an

alternative explanation, which is that investors change board members (including audit

committee members) when companies engage in opinion shopping. In unreported results, no

statistically significant association is found between opinion shopping and the departure of board

members who do not sit on the audit committee. Therefore, the association between opinion

shopping and member departure is unique to the audit committee.



31/48

31

In summary, audit committees that disapprove of opinion shopping experience a higher

rate of member departure and this is consistent with two explanations. First, members might

leave committees voluntarily when they disapprove of opinion shopping. Second, senior

management might remove audit committee members who disapprove of opinion shopping.

3.7 Audit committee independence and opinion shopping

A number of studies (e.g., Beasley, 1996; Carcello and Neal, 2000) posit that audit committees are

more effective if members are independent of senior management. Audit committee members are

assumed to be affiliated if they are relatives of senior management, current or former employees of

the company, advisors to the company, officers of significant customers or suppliers, or interlocking

directors.

There are two measurement problems in identifying whether audit committee members lack

independence. First, companies are not required to disclose all affiliations with audit committee

members. For example, an audit committee member could be a personal friend of the CEO but this

would not typically be disclosed. Certain financial ties also need not be disclosed. For example,

Enrons proxy filing does not disclose contributions made to affiliated organizations of two audit

committee members (Business Week (Asian Edition), January 21, 2002, pages 38-39). Second, as

shown in sub-section 3.6, there is high turnover of audit committee members during the auditor

dismissal year. Since the dates of member departures and appointments are not disclosed, the high

turnover rate means that audit committee composition at the auditor dismissal date is unknown.

Therefore, it is not possible to correlate opinion shopping with audit committee independence at the

auditor dismissal date. However, I do test whether audit committee disapproval is associated with

the departure of independent members during the auditor dismissal year.

Using the same classification as previous studies (e.g., Beasley, 1996; Carcello and Neal,

2000), I identify whether audit committee members prior to auditor dismissal are affiliated ( itAFF )

or independent ( itINDEP ) using proxy and 10-K filings. I find 8.0% of audit committee members

are insiders (employed by the company) and 19.6% have some non-employment affiliation with the


32/48

32

company (gray directors), so 27.6% of members are affiliated. These percentages are very similar

to those reported in previous studies (e.g., see Table 2 of Carcello and Neal, 2000).

Two variants of eq. (13) are estimated in order to test whether audit committee disapproval

of opinion shopping is associated with the departure of independent or affiliated members. Eq. (14)

has the departure of affiliated audit committee members ( itAFFL ) as the dependent variable, and

the dependent variable in eq. (15) is the departure of independent committee members ( itINDEPL ):

ititPAitAPitAFFitAFFL ++++=

3

210 (14)

ititPAitAPitINDEPitINDEPL ++++=

3

210 (15)

Columns (3)-(6) of Table 9 provide the results for eqs. (14)-15). I find audit committee

disapproval of opinion shopping is insignificantly associated with the departure of affiliated

committee members ( 3 = -0.6, z = -1.5), but significantly associated with the departure of

independent members ( 3 = -1.1, z = -3.1). The difference in coefficient estimates is not

statistically significant (t = 0.94), so it cannot be concluded that audit committee disapproval has

a significantly bigger impact on the departure of independent members. However, it is concluded

that audit committees are significantly more likely to lose independent audit committee members

when companies engage in opinion shopping.

4. Conclusion

This paper tests whether companies engage in opinion shopping and examines the role of audit

committees when incumbent auditors are dismissed. There are three key findings. First,

companies successfully engage in opinion shopping. This conclusion differs from studies that

focus on the opinions of outgoing and incoming auditors (Krishnan, 1994; Krishnan and

Stephens, 1995), but is consistent with Lennox (2000) who shows it is necessary to identify

reporting differences between new and retained auditors. It is found that a new auditor is more

likely than a retained incumbent to issue an opinion that differs from the previous year and this

reporting difference provides scope for opinion shopping. Opinion shopping companies take


33/48

33

advantage of the reporting difference by dismissing incumbents who are likely to give modified

opinions and by retaining incumbents who are likely to give unmodified opinions. The results are

therefore consistent with those reported for the UK (Lennox, 2000).

Second, the paper examines whether audit committees participate in and approve auditor

dismissal decisions. It is found that nearly half of audit committees hold fewer than the

recommended minimum number of meetings (two per year) and approximately 15% do not

participate in auditor dismissal decisions. It is therefore concluded that senior management has

considerable influence over auditor dismissal decisions. Audit committees are more active and

are more likely to participate in dismissal decisions when companies are large. No significant

association is found between opinion shopping and audit committee participation in auditor

dismissal decisions. However, a committee that participates is more likely to disapprove if

opinion shopping motivates the auditor dismissal. This is consistent with evidence that active

audit committees help maintain the integrity of the financial reporting process (DeFond and

Jiambalvo, 1991; Dechow et al., 1996).

Finally, independent audit committee members are more likely to depart during the

auditor dismissal year when committees disapprove of opinion shopping. This is consistent with

two complementary explanations. First, committee members might resign because they do not

wish to be associated with opinion shopping dismissals. Second, senior management might

remove disapproving members in order to exert greater control over the committee. In either

case, audit committees are undermined when companies engage in opinion shopping.

The papers primary limitation is I am unable to estimate the effect of audit committee

disapproval on auditor dismissal decisions, because companies do not disclose whether audit

committees approve auditor retentions. The audit committees in this sample are ineffective since

auditors are dismissed even if committees participate in auditor dismissal decisions and

disapprove. Of course, it cannot be inferred from this that audit committees are generally

ineffective because it is not possible to identify instances in which auditors are retained when

audit committees disapprove of dismissals.


34/48

34

Several questions arise from the results and could be investigated in future research.

Why do reporting differences exist between new and retained audit firms? This question is of

interest to regulators who might wish to remove reporting differences in order to reduce the

scope for opinion shopping. There are two possible explanations. First, audit firms might have

different views about the circumstances in which modified opinions should be issued. In this

case, regulators might issue more explicit reporting standards in order to reduce subjectivity.

Second, auditors might report differently in response to client pressure. In this case, regulators

might try to prevent clients from discussing reporting matters when they solicit new audit firms.

It is found that opinion shopping dismissals occur significantly later in the fiscal year compared

to other types of dismissals. Therefore, regulators might deter opinion shopping by restricting the

period in which companies are allowed to dismiss their incumbent auditors (e.g., not after the

next fiscal year-end).

The recent audit fee disclosure requirement will enable researchers to examine the

association between audit pricing and opinion shopping. Questions that might be addressed

include: How important is the fee motive compared to the opinion shopping motive as a

determinant of auditor dismissal? Do incoming audit firms charge higher fees when opinion

shopping motivates auditor dismissals? If so, does the impact of opinion shopping on fees vary

between large and small audit firms?


35/48

35

References

Abbott, L. J., Parker, S., 2000. Auditor selection and audit committee characteristics. Auditing: A

Journal of Practice and Theory 19, 47-66.

Abowd, J. M., Farber, H. S., 1982. Job queues and the union status of workers. Industrial and

Labor Relations Review 35, 354-67.

American Bar Association, 1978. Corporate Directors Guidebook, Chicago, IL: ABA.Basu, S., 1997. The conservatism principle and the asymmetric timeliness of earnings. Journal ofAccounting and Economics 24, 3-37.

Beasley, M. S., 1996. An empirical analysis of the relation between the board of director

composition and financial statement fraud. The Accounting Review 71, 443-65.Beasley, M. S., Petroni, K. R., 2001. Board independence and audit firm type. Auditing: A

Journal of Practice and Theory 20, 97-114.

Beattie V., Fearnley, S., 1995. The importance of audit firm characteristics and the drivers of

auditor change in UK listed companies. Accounting and Business Research 25, 227-239.Blue Ribbon Committee, 1999. Report and recommendations of the Blue Ribbon committee on

improving the effectiveness of corporate audit committees. Stamford, CT.

Cadbury Committee, 1992. Report of the committee on the financial aspects of corporategovernance. Gee, London,

Carcello, J. V., Neal, T. L., 2000. Audit committee composition and audit reporting. TheAccounting Review 75, 453-67.

Carcello, J. V., Hermanson, D. R., Huss, H. F., 1995. Temporal changes in bankruptcy-related

reporting. Auditing: A Journal of Practice and Theory 14, 133-43.

Carcello, J. V., Hermanson, D. R., Neal, T. L., 2001. Disclosures in audit committee reports and

charters. Working Paper.Carcello, J. V., Palmrose, Z-V. 1994. Auditor litigation and modified reporting on bankrupt

clients. Journal of Accounting Research 32 (Supplement), 1-30.

Chen, K. C., Church, B. K., 1992. Default on debt obligations and the issuance of going-concernopinions. Auditing: A Journal of Practice and Theory 11, 30-49.

Chen, K. C., Church, B. K., 1996. Going concern opinions and the markets reaction to

bankruptcy filings. The Accounting Review 71, 117-128.Chtourou, S. M., Bdard, J., Courteau, L., 2001. Corporate governance and earnings

management. Working Paper.

Conover, W. J., Iman, R. L., 1980. The rank transformation as a method of discrimination with

some examples. Communication in Statistics - Theory and Methods 9, 465-487.

Cottell, P. G., Rankin, L. J., 1988. Do audit committees bias auditor selection? Akron Business

and Economic Review 19, 87-103.

Dechow, P., Sloan, R., Sweeney, A., 1996. Causes and consequences of earnings manipulation:An analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting

Research 13, 1-36.

DeFond, M. L., Jiambalvo, J., 1991. Incidence and circumstances of accounting errors. TheAccounting Review 66, 643-655.

DeFond, M. L., Subramanyam, K. R., 1998. Auditor changes and discretionary accruals. Journal

of Accounting and Economics 25, 35-67.

Eichenseher, J. W., Shields, D., 1985. Corporate director liability and monitoring preferences.Journal of Accounting and Public Policy 4, 13-31.

Firth, M., 1980. A note on the impact of audit qualifications on lending and credit decisions.

Journal of Banking and Finance 4, 257-67.Fleak, S. K., Wilson, E. R., 1994. The incremental information content of the going concern

audit opinion. Journal of Accounting, Auditing and Finance 9, 149-66.

Frost, C. 1994. Uncertainty-modified audit reports and future earnings. Auditing: A Journal of

Practice and Theory 13, 22-35.


36/48

36

Johnson, W. B., Lys, T., 1990. The market for audit services: evidence from voluntary auditorchanges. Journal of Accounting and Economics 12, 281-308.

Kane, G. D., Meade, N. L., 1998. Ratio analysis using rank transformation. Review of

Quantitative Finance and Accounting 10, 59-74.

Klein, A., 2000. Audit committee, board of director characteristics, and earnings management.

Working paper.

Krishnan, J., 1994. Auditor switching and conservatism. The Accounting Review 69, 200-15.Krishnan, J., Krishnan, J., Stephens, R., 1996. The simultaneous relation between auditorswitching and audit opinion: An empirical analysis. Accounting and Business Research 26,

224-236.

Krishnan, J., Stephens, R., 1995. Evidence on opinion shopping from audit opinion conservatism.Journal of Accounting and Public Policy 14, 179-201.

Lennox, C., 2000. Do companies successfully engage in opinion shopping: Evidence from the

UK? Journal of Accounting and Economics 29, 321-37.

Levitan, A. S., Knoblett, J. A., 1985. Indicators of exceptions to the going concern assumption.Auditing: A Journal of Practice and Theory 5, 26-39.

Maddala, G. S., 1983. Limited-Dependent and Qualitative Variables in Econometrics.

Econometric Society Monographs No. 3, Cambridge University Press.Magee, R. P., Tseng, M., 1990. Audit pricing and independence. The Accounting Review 65,

315-336.Menon, K., Schwartz, K., 1985. Auditor switches by failing firms. The Accounting Review

60, 248-61.

Menon, K., Williams, J., 1994. The use of audit committees for monitoring. Journal of

Accounting and Public Policy 13, 121-39.

Monroe, G. S., Teh, S. T., 1993. Predicting uncertainty audit qualifications in Australia usingpublic available information. Accounting and Finance 33, 79-106.

Public Oversight Board, 1993. In the public interest: A special report by the public oversight

board of the SEC Practice Section, AICPA. Stamford, CT.Securities and Exchange Commission, 1978. Accounting Series Release No. 247, Auditor

changes: Final rules.

Securities and Exchange Commission, 1988. Disclosure Amendments to Regulation S-K, Form8-K and Schedule 14A regarding changes in accountants and potential opinion shopping

situations. SEC Financial Reporting Release No. 31, Washington, D.C.

Williams, D. 1988. The potential determinants of auditor change. Journal of Business

Finance and Accounting 15, 243-261.


37/48

37

Table 1

Audit opinion typesObservations %

Panel A:a

Unqualified opinions with no explanatory language 16,844 87.40

Panel B: bUnqualified opinions with harmless explanatory language 1,108 5.75

Panel C:c

Going concern opinions 1,138

Going concern + litigation uncertainty 29

Going concern + related party transactio

Documents

SSRN-Id299843Opinion Shopping, Audit Firm Dismissals, And Audit Committees