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 Mike Adams, Gary Biddle, Kevin Chen, Elisabeth Dedman, Jim Frederickson, Chul Park, TJ Wong and seminar participants at Bristol University, City University Business School (London), Edinburgh University, Lancaster University, Hong Kong University of Science 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 University of Science and Technology. I would also like to thank Angel Chu, Licardo Chan and Dammi Sze for their research assistance. Any remaining errors are my own. Tel: +852-23587571. E-mail: accl@ust.hk

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

DeFond and Subramanyam (1998) examine the association between earnings management and audit firm changes. They find discretionary 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. 2 Opinion shopping might be more prevalent in the UK where auditor change disclosure requirements are less strict compared to the US. SEC registrants are required to disclose audit firm changes within five business days of an incumbent auditors termination and

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

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. 3 Barry Vinocur (16 March, 1992), SEC Earthquake: Commission Delists Realty Pension Advisers, Barrons National Business and Financial Weekly (Boston). 4 There is no extant evidence on the role of audit committees in audit firm dismissals and little evidence regarding the hiring of new audit firms. Eichenseher and Shields (1985) find a positive association between audit committee existence and the hiring of large audit firms, but Cottell and Rankin (1988) find no significant relation. Abbott and Parker (2000) find companies with both active and independent audit committees are more likely to hire industry-specialist auditors, but the effects of activity and independence are

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

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.

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 (M it ) 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 (M it 1) . An auditor dismissal dummy ( Dit ) 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 (M it ) depends on the auditor dismissal decision ( Dit ) , the opinion received prior to the dismissal decision (M it 1) and a vector of other explanatory variables ( X it ) . The conditional probability that company i receives a modified report is Pr(M it = 1 | Dit , M it 1, X it )
D = 1) where the D and, for notational convenience, this probability is written as Pr( M it

superscript denotes the dismissal decision.

0 = 1) Reporting differences exist between retained and new auditors if Pr( M it 1 = 1) . Company i receives a modified report with probability Pr( M 0 = 1) if it retains its Pr( M it it 1 = 1) if it hires a new auditor. Company i uses the incumbent auditor and with probability Pr( M it 0 = 1) auditor dismissal decision to avoid a modified opinion if it dismisses its auditor when Pr( M it 1 = 1) and if it retains its auditor when Pr( M 0 = 1) Pr( M 1 = 1) . Estimation takes place > Pr( M it it it

in two stages. First, the modified opinion probabilities are predicted by estimating a probit model
0 = 1) - Pr( M 1 = 1) ) is of audit reporting. Next, the difference between opinion probabilities ( Pr( M it it

included in a probit model of auditor dismissal in order to test whether companies engage in opinion shopping. The audit reporting model is:

M it = 0 + 1M it 1 + 2 X it + 3Dit + 4 Dit M it 1 + 5Dit X it + vit

(1)

The 1 coefficient on prior opinions (M it 1) 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 X it 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 ( Dit ) and interaction terms ( Dit M it 1 and Dit X it ) 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 3 + 4 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.

0 = 1) - Pr( M 1 = 1) ) is The difference between modified opinion probabilities ( Pr( M it it

predicted from eq. (1) and included in an auditor dismissal model as follows: 0 = 1) - Pr(M 1 = 1) ) + Z + u Dit = 0 + 1(Pr(M 2 it it it it (2)

0 = 1) - Pr(M 1 = 1) ) captures the effect of reporting The opinion shopping variable ( Pr(M it it differences on auditor dismissal decisions. The 1 coefficient is positive if companies strategically dismiss auditors in order to avoid modified opinions. The Zit 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

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
The inclusion of interaction terms to capture reporting asymmetries is similar to the modelling approach of Basu (1997) who tests for asymmetry in the timeliness of accounting income. 6 The big five audit firms are Arthur Andersen, Ernst and Young, PricewaterhouseCoopers (Coopers and Lybrand or Price Waterhouse before 1st July 1998), Deloitte and Touche, and KPMG. The 18 other identified audit firms are: Altschuler, Melvoin and Glasser; BDO Seidman; Baird, Kurtz and Dobson; Cherry, Bekaert and Holland; Clarkson Gordon; Clifton Gunderson; Crowe Chizek; 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. 7 I identify other audit firm mergers by reading 8-K filings to see why the change occurs.
5

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.
8 9

A limitation of the Lennox (2000) study is its failure to distinguish between auditor dismissals and resignations. In general, 10-K filings are not collected for unqualified opinions with no explanatory language. An exception is that all opinions before 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 earnings management), so it is unclear how these opinions should be coded. In unreported results, the papers conclusions are found to be robust when these opinions are dropped from the sample.

10

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.
There 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 observations are dropped from the sample. 12 Some dismissal companies mistakenly refer to going concern opinions as audit qualifications in their 8-K filings, which is consistent with going concern opinions being viewed as negative.
11

11

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 ( M it = 0 ). 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 ( M it = 1 ). 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 no adverse opinions in the sample.

12

a significant limitation. Attention now turns to the control variables ( X it and Zit ) 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 ( PROFit ), liquidity ( LIQit ), leverage ( LEVit ) and corporate default ( DEFit ). Profitability ( PROFit ) is net income divided by total assets, liquidity ( LIQit ) is current assets divided by current liabilities, leverage ( LEVit ) is total liabilities divided by total assets and default ( DEFit ) is a dummy indicating the companys default status ( DEFit 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 ( SIZEit ) is measured using total assets and company growth ( GROWTHit ) 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
14

Moodys default data covers bond defaults for companies both rated and unrated by Moodys. Moodys compiles default histories from 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 contact with 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

13

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 ( GROWit ) and decline ( DECit ).15 After ranking the sample by growth, GROWit equals one for the top 10% and DECit 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 ( BM it ) 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 BM it and modified opinions. The book-to-market ratio ( BM it ) 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.
15

The non-monotonic relation between growth and auditor dismissal is also estimated using R (GROWTH it ) and R(GROWTH it ) 2 .

Consistent with the U-shape relation, the R (GROWTH it ) coefficient is significantly negative and the R(GROWTH it ) 2 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 BM it .

14

2.2.6 Functional form Table 2 provides descriptive statistics for the non-dummy control variables ( PROFit , LIQit ,
LEVit , SIZEit , GROWTHit , BM it ), 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 ( R( PROFit ) , R( LIQit ) , R(LEVit ) , R(SIZEit ) , R(GROWTHit ) , and
R( BM it ) ) are uniformly distributed between zero and one, as Fig. 1 illustrates for company

growth.

[INSERT TABLE 2 HERE]

Table 3 is a correlation matrix for the audit opinion ( M it ), auditor dismissal ( Dit ) 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.

[INSERT TABLE 3 HERE]

Kane and Meade (1998) find rank transformations contain information that is obfuscated by untransformed variables or alternative transformations. In addition, log and square root transformations are not viable for variables with non-positive observations ( PROFit , LIQit , GROWTH it ). Simulation studies indicate little loss of information even when rank transformations are applied to normally distributed variables (Conover and Iman, 1980).

17

15

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 ( 3 = 4 = 5 = 0 ). The positive coefficient on prior opinions ( M it 1 ) shows strong persistence in audit reporting. The negative coefficients on profitability ( R( PROFit ) ), liquidity ( R( LIQit ) ), size R(SIZEit ) and growth R(GROWTHit ) show modified opinions are issued more often to companies that are small, unprofitable and that have poor liquidity and growth. The positive coefficients on R( LEVit ) and DEFit show companies receive modified opinions more often if they are highly leveraged or in default of debt obligations. The positive coefficient on book-tomarket ( R( BM it ) ) 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 ( Dit ) 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 Dit M it 1 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 ( 3 + 4 = 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 ( 3 > 0, 4 < 0 ) imply that

16

0 = 1) Pr( M 1 = 1) ).19 An opinion shopping there exists scope for opinion shopping ( Pr( M it it

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 ( 3 + 4 < 0).

[INSERT TABLE 4 HERE]

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 0 = 1) - Pr(M 1 = 1) ). The predicted opinion probabilities are included in the new auditors ( Pr(M it it auditor dismissal model (eq. (2)) in order to test whether companies engage in opinion shopping.
0 = 1) - Pr( M 1 = 1) are significantly Columns (4)-(5) of Table 4 show the coefficients on Pr( M it it = 1.8, z = 6.6). This implies that incumbent auditors are dismissed more (less) often positive ( 1

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

Company observations may not be independent over time so unbiased standard errors are obtained using the robust cluster command in STATA. The usual assumption of independence between companies is retained. 19 Consistent with extant research (e.g., Monroe and Teh, 1993; Krishnan and Stephens, 1995), no significant reporting differences are 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. 20 The auditor dismissal model does not include all possible determinants of dismissal for example, it is not possible to control for audit fees as they are undisclosed during the sample period. It is investigated whether these omitted factors confound the results for opinion shopping by examining whether 8-K filings disclose the reasons for dismissal. There are 163 companies (19.7%) that disclose why auditors are dismissed and the most common reason is the need for lower fees (55 dismissals). When companies that disclose = 1.67, z = reasons are dropped from the sample, the opinion shopping coefficient and level of significance are similar to Table 4 ( 1 5.9), suggesting the opinion shopping coefficient is not biased by omitted variables. In addition, no significant correlation is found between opinion shopping and the companys propensity to disclose a reason for the dismissal.

18

17

 0 = 1) > It is predicted that an auditor dismissal is motivated by opinion shopping if Pr( M it 1 = 1) . There are 142 dismissals that are motivated by opinion shopping whereas 686 are Pr( M it

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 ( R( PROFit ) ) is low and leverage ( R( LEVit ) ) is high. The negative coefficient on company size ( R(SIZEit ) ) shows that large companies are less likely to dismiss than small companies. Companies that grow rapidly ( GROWit ) are more likely to dismiss but the decline dummy ( DECit ) is insignificant after controlling for financial health. Finally, the positive coefficient on the book-to-market ratio ( R( BM it ) ) 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
21

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

18

less likely to participate when opinion shopping motivates dismissals? Are audit committees that participate more likely to disapprove when dismissals are motivated by opinion shopping? If so, what happens to audit committee members when committees disapprove of opinion shopping?

3.1 Sample Accounting Series Release No. 247 requires that 8-K filings disclose whether audit committees approve audit firm changes (SEC, 1978).23 An important data limitation is that there is no disclosure of the audit committees role when an incumbent auditor is retained (there is no 8-K filing), so the following analysis is for auditor dismissals rather than retentions. This means it cannot be tested whether audit committee approval affects the likelihood of auditor dismissal. It is determined whether an audit committee exists by reading 10-K and proxy filings before and after auditor dismissals. Filings disclose that audit committees do not exist for 92 dismissals. Filings do not disclose whether audit committees exist for 52 dismissals and, in most cases, this probably reflects the absence of a committee. This leaves 684 (= 828 - 92 - 52) observations where audit committees exist during auditor dismissal years. The frequency of opinion shopping in the three sub-samples (92, 52 and 684 observations) is 30%, 31% and 14% respectively. Univariate results (unreported) reveal a significantly lower incidence of opinion shopping when companies have audit committees. However, multivariate results indicate the association is insignificant after controlling for company size, so it cannot be concluded that audit committee existence reduces opinion shopping. For the remainder of the paper, the focus is on companies that have audit committees (684 observations).

3.2 Audit committee participation and approval of auditor dismissals Fig. 2 illustrates an audit committees participation and approval decisions when an auditor is dismissed. The audit committee first decides whether to participate in the auditor dismissal
22

In unreported results, one digit SIC codes and year dummies are included in the audit reporting and dismissal models but their effects are jointly insignificant.

19

decision. A participation dummy ( PAit ) equals one if the committee participates and equals zero if it does not. If the audit committee does not participate ( PAit = 0 ) then, by definition, it cannot approve the auditor dismissal. If the committee does participate ( PAit = 1 ), an approval dummy ( APit ) equals one if the committee approves dismissal and equals zero if it disapproves.

[INSERT FIG. 2 HERE]

An 8-K filing discloses only the joint outcome of an audit committees participation and approval decisions ( PAit APit ). An auditor dismissal is disclosed as approved ( PAit APit = 1) if the audit committee both participates ( PAit = 1 ) and does not disapprove ( APit = 1 ). A dismissal is not approved ( PAit APit = 0) either because the committee does not participate ( PAit = 0 ) or because it participates and disapproves ( PAit = 1 , APit = 0 ).24 Since PAit and APit are only partially observed, an important task is to identify whether non-approval ( PAit APit = 0) is due to non-participation or disapproval. The distinction is important because nonparticipation 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 ( PAit APit ). Audit committee non-approval is 62% (= 61/98) when opinion 1 = 1) ) and 35% (= 207/586) when dismissals 0 = 1) > Pr(M shopping motivates dismissals ( Pr(M it it
23

Some 8-K filings use the word recommended rather than approved - recommendations and approvals are treated equivalently in this paper. 24 When PAit APit = 0 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).

20

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.

[INSERT TABLE 5 HERE]

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 ( MDit ) and a non-dismissal year ( MNDi. ). 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 ( MDit ) and the abnormal number of meetings ( AM it MDit MNDi. ). 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

Non-approval could be perceived to be a negative signal so it is tested whether the committees decision to approve affects the report of the incoming auditor. The (unreported) association between audit committee approval and modified opinions is found to be negative (as expected) but insignificant. 26 Proxy and 10-K filings disclose the number of audit committee meetings during the fiscal year but do not disclose meeting dates. It is straightforward to measure meeting activity during the auditor dismissal year if the dismissal occurs prior to the year-end reported on by the 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 19X2 and the incoming auditor issues a report for the year ending 31st December 19X1. In this example, meeting activity in the dismissal year is measured using the fiscal year 1st January 19X2 31st December 19X2 since this is the period in which the auditor is dismissed. This might introduce measurement error because the audit committee could meet to discuss the dismissal before 31st 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 no significant trend in meeting activity during the sample period (1996-98), so it makes no difference whether the non-dismissal year is the year before or the year after dismissal.

25

21

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 ( AM it = 0). Under the alternate hypothesis that some audit committees participate there are more meetings in dismissal years ( AM it > 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 ( PAit APit = 0) is always due to non-participation ( PAit = 0 ) or whether some audit committees participate and disapprove ( PAit = 1 , APit = 0 ). Under

22

the null hypothesis that non-approval is always due to non-participation (H2), there is no abnormal meeting activity when PAit APit = 0. Under the alternate hypothesis that some audit committees participate and disapprove, there is positive abnormal meeting activity when PAit APit = 0. Table 6 shows a significant increase of 0.26 meetings when PAit APit = 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 ( PAit APit = 0) is always due to disapproval ( PAit = 1 , APit = 0 ) or whether some audit committees do not participate ( PAit = 0 ). Under the null hypothesis that all committees participate (H3), abnormal meeting activity is equally high when
PAit APit = 0 compared to when PAit APit = 1. Under the alternate hypothesis that some audit

committees do not participate, abnormal meeting activity is lower when PAit APit = 0 compared to when PAit APit = 1. Table 6 shows an increase of 0.55 meetings when PAit APit = 1 compared to an increase of 0.26 meetings when PAit APit = 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 subsections I test whether opinion shopping explains the cross-sectional variation in participation and approval and I estimate the extent of audit committee non-participation.

[INSERT TABLE 6 HERE]

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
28

Menon 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 et al., (2001) find the average audit committee charter requires a minimum of 3.31 meetings (below the recommended number).

23

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 ( MDit ) since an active committee might discuss the dismissal as part of its regular meeting schedule, and (2) abnormal meeting activity ( AM it ) 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 ( MDit ) and the dependent variable in eq. (4) is abnormal meeting activity ( AM it ):
0 = 1) Pr(M 1 = 1)) + R(SIZE ) + MDit = 0 + 1(Pr(M it it 2 it it 0 1 AM it = 0 + 1(Pr(M it = 1) Pr(M it = 1)) + 2 R(SIZEit ) + it

(3) (4)

0 = 1) - Pr( M 1 = 1) ) and meeting activity A negative association between opinion shopping ( Pr( M it it

indicates that audit committees participate in auditor dismissal decisions less often when companies engage in opinion shopping ( 1 < 0 and 1 < 0 ). For example, an audit committee might leave the dismissal decision to the full board of directors in order to avoid conflict with senior management. On the other hand, a positive relation is possible ( 1 > 0 and 1 > 0 ) because an audit committee might meet in order to oppose opinion shopping. Company size ( R(SIZEit ) ) 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 ( 2 > 0 and 2 > 0 ). The results are reported in Columns (1)-(4) of Table 7. As expected, the association 2 = between company size and audit committee activity is positive and statistically significant ( 2 = 1.0, t = 3.3). Therefore, audit committees of large companies are more active 1.2, z = 10.4 and and are more likely to participate in auditor dismissal decisions.29 In contrast, there is no significant
1 = 0.3, z = 0.6 and 1 = 0.1, t = 0.1). It relation between opinion shopping and meeting activity (

24

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.

[INSERT TABLE 7 HERE]

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 ( PAit APit ) and later they are modelled separately using two dependent variables ( PAit and APit ). In eqs. (5)-(6) the dependent variable is the joint participation-approval outcome:
0 = 1) Pr(M 1 = 1)) + PAit APit = 0 + 1MDit + 2 (Pr(M it it it 0 = 1) (Pr(M 1 = 1)) + PAit APit = 0 + 1AM it + 2 (Pr(M it it it

(5) (6)

If active audit committees are more likely to participate in auditor dismissal decisions, the coefficients on the meeting variables ( MDit and AM it ) are positive ( 1 > 0 and 1 > 0). The
0 = 1) - Pr( M 1 = 1) ) captures the effect of companies dismissal opinion shopping variable ( Pr( M it it

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
1 = 0.2, z = relation exists between meeting activity and the joint participation-approval outcome ( 1 = 0.1, z = 2.2). This confirms that active audit committees are more likely to participate 4.8 and

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 found between meeting activity and audit committee size.

25

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

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

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
M ) are first predicted from Columns (2) and (4) of Table 7. The mean D and A variables ( M it it D ) is 2.35 when meeting activity is disclosed (N = 517) and predicted number of meetings ( M it

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.
M are included as predictors of the joint participation-approval D and A Next, M it it

outcome:
D + (Pr(M 0 = 1) Pr(M 1 = 1)) + PAit APit = 0 + 1M it it 2 it it M + (Pr(M 0 = 1) Pr(M 1 = 1)) + PAit APit = 0 + 1A it it 2 it it

(7) (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)-

26

(6)) are estimated for 517 observations using disclosed meeting activity. Columns (1)-(4) are consistent in showing that PAit APit 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.

[INSERT TABLE 8 HERE]

The third objection is addressed by modelling the participation and approval decisions separately:
D + PAit = 0 + 1M it it

(9) (10)

0 = 1) Pr(M 1 = 1)) + APit = 0 + 2 (Pr(M it it it

D ) explains audit committee participation ( PA ) and the The predicted number of meetings ( M it it 0 = 1) - Pr(M 1 = 1) ) explains approval ( AP ). Eqs. (9)predicted opinion shopping variable ( Pr(M it it it (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:
D )( + (Pr(M 0 = 1) Pr( M 1 = 1)))] [( 0 + 1M it it it 0 2 PAit APit =1 D )( + (Pr(M 0 = 1) Pr(M 1 = 1)))] [1 ( 0 + 1M it it it 0 2 PAit APit =0

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 M ) is used to explain participation: activity ( A it
M + PAit = 0 + 1A it it

(11)

27

 0 = 1) Pr(M 1 = 1)) + APit = 0 + 2 (Pr(M it it it

(12)

The results for eqs. (9)-(12) are shown in Columns (5)-(8) of Table 8. The positive M in Columns (5) and (7) show active audit committees are more D and A coefficients on M it it likely to participate in auditor dismissal decisions (z = 2.6). Columns (5) and (7) give mean
A = 1) ) of 85.9% and 85.4% respectively. It is predicted participation probabilities ( Pr( P it

therefore estimated that 15% of audit committees do not participate in auditor dismissal
0 = 1) - Pr( M 1 = 1) in Columns (6) and (8) confirm decisions. The negative coefficients on Pr( M it it

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 ( MEM it ) is 2.84 and the number of member departures during the auditor dismissal year ( MEMLit ) is 0.67. The mean departure rate (
MEMLit ) is 24.1% showing that nearly a quarter of audit committee MEM it

members leave during the auditor dismissal year. When auditor dismissals are approved
30

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

28

( PAit APit = 1) the mean departure rate is 19.2%, but when dismissals are not approved ( PAit APit = 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 ( PAit = 0 ) or with disapproval ( PAit = 1 , APit = 0 ). 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 0 = 1) > Pr(M 1 = 1) ) the mean departure rate ( MEMLit ) is 38.4%, but when shopping (( Pr(M it it MEM it 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:
A + A MEMLit = 0 + 1MEM it + 2 P it 3 Pit + it

(13)

The dependent variable is the number of members that leave the audit committee during the auditor dismissal year ( MEMLit ). Eq. (13) controls for the number of members prior to auditor dismissal ( MEM it ) because large audit committees have more members that might potentially
P ) are predicted from A ) and approval ( A leave ( 1 > 0 ). Audit committee participation ( P it it

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 ( 3 < 0 ). Since audit committee disapproval is predicted using the opinion shopping variable (eqs. (9) and (11)), 3 indirectly captures the effect of opinion shopping on member departure.

29

The results for eq. (13) are shown in Columns (1)-(2) of Table 9. The insignificant A ) show member departure is not associated with audit coefficients on participation ( P it 2 = -0.0, z = -0.2). The significant negative coefficients on approval committee inactivity ( P ) reveal that members are more likely to depart when audit committees disapprove of (A it 3 = -0.8, z = -2.8). Similar (unreported) findings emerge when the departure auditor dismissals ( 0 = 1) - Pr(M 1 = 1) ) and meeting model includes the predicted opinion shopping variable ( Pr(M it it M ). The coefficients for the meeting variables are insignificant, D and A activity ( M it it 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 ( R( PROFit ) , R( LIQit ) , R( LEVit ) and DEFit ) 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.

[INSERT TABLE 9 HERE]

30

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 ( AFFit ) or independent ( INDEPit ) 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

31

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 ( AFFLit ) as the dependent variable, and the dependent variable in eq. (15) is the departure of independent committee members ( INDEPLit ):
A + A AFFLit = 0 + 1AFFit + 2 P it 3 Pit + it A + A INDEPLit = 0 + 1INDEPit + 2 P it 3 Pit + it

(14) (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
3 = -0.6, z = -1.5), but significantly associated with the departure of committee members ( 3 = -1.1, z = -3.1). The difference in coefficient estimates is not independent members (

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

32

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.

33

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?

34

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 of Accounting 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 corporate governance. Gee, London, Carcello, J. V., Neal, T. L., 2000. Audit committee composition and audit reporting. The Accounting 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-concern opinions. 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. The Accounting 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.

35

Johnson, W. B., Lys, T., 1990. The market for audit services: evidence from voluntary auditor changes. 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 auditor switching 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 using public 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, Form 8-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.

36

Table 1 Audit opinion types

Panel A: a Unqualified opinions with no explanatory language Panel B: b Unqualified opinions with harmless explanatory language Panel C: c Going concern opinions Going concern + litigation uncertainty Going concern + related party transactions Going concern + correction of accounting errors Going concern + uncertainty about tax payable Going concern + uncertainty about environmental clean-up costs Going concern + litigation uncertainty + related party transactions Going concern + litigation uncertainty + development stage company Panel D: d Litigation uncertainty Uncertainty about foreign exchange losses Uncertainty about losses in Russia Uncertainty about value of mining assets Uncertainty about value of accounts receivable Uncertainty about consequences of opting out of Master Settlement Agreement Uncertainty about whether a distribution agreement will be re-negotiated Panel E: e Correction of accounting errors Sale of a significant part of the companys operations Revenues come from a limited number of sources Proposed merger or acquisition Death of CEO Possible acceleration of account payable Significant related party transactions Restructuring of long-term debt facilities Restructuring of short-term debt facilities Company is in development stage Intangibles comprise a significant part of total assets Results may not be indicative of those resulting from a stand alone company Stock repurchase Company is in breach of statutory insurance requirements Company has defaulted on its debt covenants Company is dependent upon its parent for financial support Company is not complying with SEC filing requirements Company is subject to a criminal investigation Company is subject to breaches of warranties and representations in connection with a merger agreement Panel F: f Except for reporting disagreement Except for limitation on audit scope + going concern Except for limitation on audit scope + going concern + litigation uncertainty 1,138 29 1 3 1 1 1 1 1,175 30 2 1 3 2 1 1 40 5 15 5 10 1 3 10 3 1 12 2 4 2 2 1 1 1 3 1 82 6 3 2 11 0.06 0.43 0.21 6.10 Observations 16,844 1,108 % 87.40 5.75

37

Table 1 (contd.) Audit opinion types

Panel G: g Opinion disclaimer Opinion disclaimer + going concern Opinion disclaimer + going concern + litigation uncertainty + significant related party transactions Total

2 8 3 13 19,273 0.07 100.00

Notes: Table 1 provides information on the different types of audit opinions issued. Panel A consists of unqualified opinions with no explanatory language. Panels B-E consist of unqualified opinions with explanatory language. Opinions are grouped into Panels B-E according to the type of explanatory language used. Explanatory language is harmless (Panel B) 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 (d) the current year financial statements are not comparable with those of the previous year. Panel C consists of going concern opinions. Panel D consists of fundamental uncertainties. Panel E consists of other emphases of matter. Panel F consists of qualified (except for) opinions. Panel G consists of opinion disclaimers. Opinions with multiple bad news disclosures are shown separately in Panels C-G. Audit opinions in Panels A and B are viewed as favorable and so are coded as unmodified ( M it = 0). Audit opinions in Panels C, D, E, F and G are viewed as unfavorable and so are coded as modified ( M it = 1).
a b, c, d, e, f, g

Data source is COMPUSTAT. Data source is COMPUSTAT and companies 10-K filings.

38

Fig. 1. Company growth for auditor dismissals and retentions (a) Auditor dismissals (N = 828)

.17

Fraction

0 0 R(GROWTHit) 1

(b) Auditor retentions (N = 18,445)

.10

Fraction

0 0 R(GROWTHit) 1

Notes: Fig. 1 provides histograms for company growth when auditors are dismissed and retained. Fig. 1a shows a U-shape relation for growth when auditors are dismissed. Figs. 1a and 1b show the rank-transformed growth variable ( R (GROWTH it ) ) is uniformly distributed between zero and one for the sample as a whole. R (GROWTH it ) = Rank transformation of GROWTH it , where GROWTH it = Percentage annual growth in total assets.

39

Table 2 Descriptive statistics for the untransformed variables Mean PROFit LIQit
LEVit SIZEit
-0.14 3.47 0.59 1124.82 7.26 35.83

Median
0.03 1.99 0.49 87.76 0.10 0.94

Min
-452.25 0 0.0005 0.001 -0.99 0.0003

Max
286.67 716.67 496 153498 97579 613909

GROWTH it
BM it

Notes: Table 2 provides descriptive statistics for the non-dummy control variables, which are highly skewed and contain outliers. Skewness and outlier problems are avoided by applying rank transformations (see Table 3). PROFit = Net income/Total assets. LIQit = Current assets/Current liabilities. LEVit = Total liabilities/Total assets. SIZE it = Total assets ($ million). GROWTH it = Percentage annual growth in total assets. BM it = Book value of total assets/Market value. Observations = 19,273.

40

Table 3 Correlation matrix

Dit M it M it 1 R( PROFit )
.08** .11** -.09** -.04** .05** .02* -.10** -.01 .04** .05** .03**

M it
1 .61** -.31** -.24** .20** .07** -.31** -.20** -.02* .29** .02**

M it 1

R( PROFit ) R( LIQit ) R( LEVit )

DEFit

R( SIZEit )

R(GROWTH it ) GROWit DECit

1 -.20** -.18** .16** .02* -.27** -.12** .01 .18** -.01 1 .14** -.18** -.05** .34** .30** -.08** -.38** -.16** 1 -.70** -.03** -.16** .15** .13** -.08** -.33** 1 .06** .24** -.13** -.10** .08** .46** 1 .01 -.04** -.01 .07** .06** 1 .14** -.03** -.28** .15** 1 .52** -.52** -.25** 1 -.11** -.16** 1 -.01

R( LIQit )
R( LEVit ) DEFit R(SIZEit ) R(GROWTHit ) GROWit DECit R( BM it )
Notes:

Table 3 provides a correlation matrix for the dummy variables and the rank-transformed variables. M it = 1 if company i receives a modified audit opinion; = 0 otherwise (see Table 1). Dit = 1 if company i dismisses its incumbent auditor; = 0 if the incumbent auditor is retained. R ( PROFit ) = Rank transformation of PROFit , where PROFit = Net income/Total assets. R ( LIQit ) = Rank transformation of LIQit , where LIQit = Current assets/Current liabilities. R ( LEVit ) = Rank transformation of LEVit , where LEVit = Total liabilities/Total assets. DEFit = 1 if company i is in default; 0 otherwise. R ( SIZE it ) = Rank transformation of SIZE it , where SIZE it = Total assets ($ million). R (GROWTH it ) = Rank transformation of GROWTH it , where GROWTH it = Percentage annual growth in total assets. GROWit = 1 if company is percentage annual growth is in the top 10%; = 0 otherwise. DEC it = 1 if company is percentage annual growth is in the bottom 10%; = 0 otherwise. R ( BM it ) = Rank transformation of BM it , where BM it = Book value of total assets/Market value. * = Statistically significant (5% level, 2-tailed). ** = Statistically significant (1% level, 2 tailed). Observations = 19,273.

Table 4 Audit reporting and auditor dismissal models M it = 0 + 1M it 1 + 2 X it + 3Dit + 4 Dit M it 1 + 5Dit X it + vit
0 = 1) - Pr(M 1 = 1) ) + Z + u Dit = 0 + 1(Pr(M 2 it it it it Eq. (1) Eq. (1) (1) (2) CONSTANT M it 1
-0.39 (-4.00)** 1.95 (32.25)** . . -1.95 (-15.74)** -1.06 (-10.58)** 0.36 (3.59)** 1.04 (3.92)** -1.59 (-16.44)** -0.36 (-4.90)** . . . . 0.30 (3.88)** . . . . . . . . . . . . . . . . . . -0.47 (-4.58)** 2.07 (30.73)** . . -1.98 (-14.86)** -1.03 (-9.69)** 0.41 (3.94)** 1.06 (3.79)** -1.61 (-15.99)** -0.31 (-4.06)** . . . . 0.34 (4.05)** 0.83 (2.61)** -0.81 (-4.85)** 0.07 (0.17) -0.30 (-0.84) -0.53 (-1.60) -0.24 (-0.28) 0.13 (0.33) -0.34 (-1.37) -0.27 (-1.08)

Eq. (1) Eq. (2)

Eq. (1) (3)

-0.41 (-4.21)** 2.07 (30.80)** . . -1.98 (-15.63)** -1.06 (-10.51)** 0.36 (3.66)** 1.03 (3.85)** -1.60 (-16.34)** -0.34 (-4.71)** . . . . 0.31 (3.94)** 0.17 (1.96)* -0.85 (-5.32)** . . . . . . . . . . . . . .

Eq. (2) (4)

-1.59 (-17.70)** . . 1.77 (6.69)** -0.25 (-3.58)** -0.06 (-0.72) 0.30 (3.55)** 0.40 (1.58) -0.80 (-12.25)** . . 0.32 (6.33)** 0.04 (0.69) 0.29 (4.59)** . . . . . . . . . . . . . . . . . .

Eq. (2) (5)

-1.62 (-31.15)** . . 1.75 (6.62)** -0.27 (-4.15)** . . 0.36 (5.55)** . . -0.81 (-12.61)** . . 0.31 (6.30)** . . 0.29 (4.59)** . . . . . . . . . . . . . . . . . .

0 = 1) - Pr(M 1 = 1) Pr(M it it R( PROFit ) R( LIQit )

R( LEVit )

DEFit
R( SIZEit ) R(GROWTH it ) GROWit DECit

R( BM it )
Dit Dit M it 1 Dit R( PROFit ) Dit R( LIQit ) Dit R( LEVit ) Dit DEFit

Dit R( SIZEit )
Dit R(GROWTH it )

Dit R( BM it )

Notes: Eq. (1) tests for reporting differences between new and retained audit firms. Eq. (2) tests whether companies engage in opinion shopping ( 1 > 0 ). Unbiased standard errors are obtained using the robust cluster command in STATA because company observations may not be D = 1) = independent over time (z-statistics are reported in parentheses). All columns are estimated using probit. Observations = 19,273. Pr( M it Probability that company i receives a modified audit opinion conditional on its dismissal decision (D = 1 if company i appoints a new auditor, D = 0 if company i retains its incumbent auditor). The modified opinion probabilities are predicted using the results from Column (3). See Table 3 for other variable definitions. * = Statistically significant (5% level, 2-tailed). ** = Statistically significant (1% level, 2-tailed).

Fig. 2. Audit committee participation and approval of auditor dismissal decisions

Does the audit committee participate in the auditor dismissal decision?

Yes ( PAit = 1 )

Does the audit committee approve of auditor dismissal?

No ( PAit = 0 )

No ( APit = 0 )

Yes ( APit = 1 )

Notes: Fig. 2 shows the sequential nature of the audit committees decision to participate and approve of the auditor dismissal. PAit = 1 if the audit committee participates in the auditor dismissal decision; 0 otherwise. APit = 1 if a participating audit committee approves the auditor dismissal decision; 0 otherwise. 8-K filings disclose only the joint outcome of an audit committees participation and approval decisions ( PAit APit ). PAit APit = 1 if the auditor dismissal is approved by the audit committee or PAit APit = 0 if it is not approved.

43

Table 5 The association between opinion shopping and audit committee participation-approval
0 = 1) Pr( M 1 = 1) Pr( M it it 0 = 1) > Pr( M 1 = 1) Pr( M it it
61 37 98

Total
268 416 684

PAit APit = 0 PAit APit = 1 Total

207 379 586

Notes: Table 5 reports the association between opinion shopping and audit committee participation and approval of auditor dismissal decisions. PAit = 1 if the audit committee participates in the auditor dismissal decision; 0 otherwise. APit = 1 if a participating audit committee approves the auditor dismissal decision; 0 otherwise. 8-K filings disclose only the joint outcome of an audit D = 1) = Probability that company i receives a committees participation and approval decisions ( PAit APit ) (see Fig. 2). Pr( M it modified audit opinion (D = 0 if company i retains its incumbent auditor, D = 1 if company i appoints a new auditor). A 0 = 1) > Pr( M 1 = 1) or by other reasons if Pr( M 0 = 1) Pr( M 1 = 1) . dismissal is motivated by opinion shopping if Pr( M it it it it The modified opinion probabilities are predicted from Column (3) of Table 4.

44

Table 6 Audit committee meeting activity Audit committee meetings in auditor dismissal years ( MDit )
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Observations (%)

Audit committee meetings in nondismissal years ( MNDi. )

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Observations (%)

28 (5.4%) 169 (32.7%) 130 (25.2%) 84 (16.2%) 54 (10.4%) 31 (6.0%) 11 (2.1%) 4 (0.8%) 2 (0.4%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 3 (0.6%) 1 (0.2%) 517

53 (10.2%) 195 (37.7%) 127 (24.6%) 68 (13.2%) 51 (9.9%) 12 (2.3%) 9 (1.7%) 1 (0.2%) 1 (0.2%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 0 (0.0%) 517

Total Mean ( MDit ) = 2.35

Hypotheses tests:

Total Mean ( MNDi. ) = 1.90

H1: Mean abnormal meeting activity ( AM it ) = 0.45. Reject null hypothesis that AM it = 0 (t = 6.71)**. H2: When auditor dismissals are not approved ( PAit APit = 0, N = 189), mean abnormal meeting activity ( AM it ) = 0.26. Reject null hypothesis that AM it = 0 when PAit APit = 0 (t = 3.34)**. H3: When auditor dismissals are approved ( PAit APit = 1, N = 328), mean abnormal meeting activity ( AM it ) = 0.55. Reject null hypothesis that AM it is equally high when PAit APit = 1 compared to PAit APit = 0 (t = 2.30)*.
Notes: Table 6 shows the level and change in audit committee meeting activity during the auditor dismissal year. MDit = Number of audit committee meetings in auditor dismissal years. MNDi. = Number of audit committee meetings in non-dismissal years. AM it = Abnormal meeting activity in auditor dismissal years ( AM it MDit MND i. ). PAit = 1 if the audit committee participates in the auditor dismissal decision; 0 otherwise. APit = 1 if a participating audit committee approves the auditor dismissal decision; 0 otherwise. H1 tests whether some audit committees participate in auditor dismissal decisions (null is that committees never participate). H2 tests whether non-approval ( PAit APit = 0) is sometimes due to disapproval (null is that non-approval is always due to non-participation ( PAit = 0)). H3 tests whether non-approval ( PAit APit = 0) is sometimes due to non-participation (null is that non-approval is always due to disapproval ( PAit = 1, APit = 0)). The results from H1-H3 show that some but not all audit committees participate in auditor dismissal decisions, and of committees that participate, some but not all approve of dismissal decisions. * = Statistically significant (5% level, 2-tailed). ** = Statistically significant (1% level, 2-tailed).

45

Table 7 Audit committee meeting models 0 = 1) Pr(M 1 = 1)) + R(SIZE ) + MDit = 0 + 1(Pr(M it it 2 it it 0 = 1) Pr(M 1 = 1)) + R(SIZE ) + AM it = 0 + 1(Pr(M it it 2 it it Eq. (3) (1) Eq. (3) (2)
0.29 (4.42)** . . 1.20 (10.37)**

Eq. (3) Eq. (4) Eq. (4) (3)

0.01 (0.09) 0.10 (0.10) 1.00 (3.19)**

Eq. (4) (4)

0.01 (0.07) . . 1.01 (3.30)**

CONSTANT
0 = 1) - Pr( M 1 = 1) Pr( M it it

R( SIZE it )

0.29 (4.23)** 0.34 (0.57) 1.18 (9.68)**

Notes: Eqs. (3)-(4) test the effects of company size and opinion shopping on audit committee meeting activity. Robust standard errors are calculated (z-statistics are reported in parentheses). MDit = Number of audit committee meetings in auditor dismissal years. MNDi. = Number of audit committee meetings in non-dismissal years. AM it = Abnormal meeting activity in auditor dismissal years ( AM it MD it MND i. ). Columns (1) and (2) are estimated using Poisson because MDit takes discrete non-negative values. Columns (3) and (4) are estimated using OLS regression. Observations = 517. R ( SIZE it ) = Rank transformation of SIZE it , where D = 1) = Probability that company i receives a modified audit opinion (D = 0 if company i SIZE it = Total assets ($ million). Pr( M it retains its incumbent auditor, D = 1 if company i appoints a new auditor). A dismissal is motivated by opinion shopping if 0 = 1) > Pr( M 1 = 1) or by other reasons if Pr( M 0 = 1) Pr( M 1 = 1) . The modified opinion probabilities are predicted Pr( M it it it it from Column (3) of Table 4. * = Statistically significant (5% level, 2-tailed). ** = Statistically significant (1% level, 2-tailed).

46

Table 8 Audit committee participation and approval 0 = 1) Pr(M 1 = 1)) + PAit APit = 0 + 1MDit + 2 (Pr(M it it it
0 = 1) (Pr(M 1 = 1)) + PAit APit = 0 + 1AM it + 2 (Pr(M it it it D + (Pr(M 0 = 1) Pr(M 1 = 1)) + PAit APit = 0 + 1M it it 2 it it M + (Pr(M 0 = 1) Pr(M 1 = 1)) + PAit APit = 0 + 1A it it 2 it it

Eq. (5) Eq. (6) Eq. (7) Eq. (8) Eq. (7) (3)
-0.44 (-2.71)** . . . . 0.33 (4.77)** . . -1.88 (-2.92)**

D + PAit = 0 + 1M it it 0 1 = 1)) + APit = 0 + 2 (Pr(M it = 1) Pr(M it it M + PAit = 0 + 1A it it 0 = 1) Pr(M 1 = 1)) + APit = 0 + 2 (Pr(M it it it

Eq. (9) Eq. (10) Eq. (11) Eq. (12) Eq. (12) (8)
0.59 (6.07)** . . . . . . . . -2.29 (-2.72)**

Eq. (5) (1)

Eq. (6) (2)

0.33 (5.59)** . . 0.08 (2.21)* . . . . -1.99 (-2.55)*

Eq. (8) (4)

-0.10 (-1.14) . . . . . . 1.01 (5.26)** -1.79 (-2.79)**

Eq. (9) (5)

-2.91 (-2.30)* . . . . 2.18 (2.61)** . . . .

Eq. (10) (6)

0.58 (6.66)** . . . . . . . . -2.28 (-2.74)**

Eq. (11) (7)

-0.10 (-0.46) . . . . . . 4.22 (2.65)** . .

CONSTANT

MDit
AM it

D M it
M A it
0 = 1) - Pr( M 1 = 1) Pr( M it it

-0.07 (-0.64) 0.19 (4.81)** . . . . . . -1.69 (-2.18)*

Notes: Eqs. (5)-(12) test the effects of meeting activity and opinion shopping on audit committee participation and approval. Robust standard errors are calculated (z-statistics are reported in parentheses). Eqs. (5)-(8) are estimated using standard probit models. Eqs. (9)-(12) are estimated using sequential probit models of partial observability (Abowd and Farber, 1982). Columns (1)-(2) are estimated using disclosed meeting activity (N =517). Columns (3)-(8) are estimated using predicted meeting activity (N = 684). PAit = 1 if the audit committee participates in the auditor dismissal decision; 0 otherwise. APit = 1 if a participating audit committee approves the auditor dismissal decision; 0 otherwise. MDit = Number of audit committee meetings in auditor dismissal years. MNDi. = Number of audit committee meetings in non-dismissal years. AM it = Abnormal meeting activity in M are predicted from Columns (2) and (4) of Table 7. Pr( M D = 1) = Probability that company i receives a modified audit opinion (D = 0 if company i D and A auditor dismissal years ( AM it MDit MND i. ). M it it it 0 = 1) > Pr( M 1 = 1) or by other reasons if Pr( M 0 = 1) Pr( M 1 = 1) . The modified retains its incumbent auditor, D = 1 if company i appoints a new auditor). A dismissal is motivated by opinion shopping if Pr( M it it it it opinion probabilities are predicted from Column (3) of Table 4. * = Statistically significant (5% level, 2-tailed). ** = Statistically significant (1% level, 2-tailed).

Table 9 Audit committee member departure A + A MEMLit = 0 + 1MEM it + 2 P it 3 Pit + it

A + A AFFLit = 0 + 1AFFit + 2 P it 3 Pit + it A + A INDEPLit = 0 + 1INDEPit + 2 P it 3 Pit + it

Eq. (13) Eq. (14) Eq. (15) Eq. (14) (4)

-2.47 (-10.44)** . . 1.16 (16.37)** . . -0.07 (-0.77) -0.59 (-1.52)

Eq. (13) (1)

Eq. (13) (2)

-0.67 (-2.92)** 0.26 (3.41)** . . . . -0.02 (-0.37) -0.82 (-2.81)**

Eq. (14) (3)

-2.49 (-10.76)** . . 1.17 (16.34)** . . -0.05 (-0.84) -0.60 (-1.54)

Eq. (15) (5)

-1.02 (-4.63)** . . . . 0.34 (4.61)** 0.01 (0.06) -1.11 (-3.17)**

Eq. (15) (6)

-1.01 (-4.54)** . . . . 0.35 (4.90)** -0.02 (-0.22) -1.08 (-3.14)**

CONSTANT MEM it
AFFit
INDEPit A P it P A it

-0.68 (-2.97)** 0.26 (3.32)** . . . . -0.01 (-0.16) -0.84 (-2.84)**

Notes: Eq. (13) tests the effect of audit committee participation and approval on the departure of audit committee members. Eq. (14) tests the effect of audit committee participation and approval on the departure of affiliated audit committee members. Eq. (15) tests the effect of audit committee participation and approval on the departure of independent audit committee members. Robust standard errors are calculated (z-statistics are reported in parentheses). A dismissal is in the sample if 10K or proxy filings disclose audit committee membership before and after auditor dismissal (N = 457). Eqs. (13)-(15) are estimated using Poisson because the dependent variables take discrete non-negative values. MEMLit = Number of audit committee members that leave during the auditor dismissal year. MEM it = Number of audit committee members prior to auditor dismissal. AFFLit = Number of affiliated audit committee members that leave during the auditor dismissal year. AFFit = Number of affiliated audit committee members prior to auditor dismissal. An audit committee member is affiliated if he/she is an employee of the company, a former employee, a relative of senior management or has other business relations with the company. INDEPLit = Number of non-affiliated (independent) audit committee members that leave during the auditor dismissal year. INDEPit = Number of non-affiliated audit committee members prior to auditor dismissal. PAit = 1 if the audit committee participates in the auditor dismissal decision; 0 otherwise. APit = 1 if a P are A and A participating audit committee approves the auditor dismissal decision; 0 otherwise. In Columns (1), (3) and (5), P it it predicted from Columns (5) and (6) of Table 8. In Columns (2), (4) and (6), PAit and APit are predicted from Columns (7) and (8) of Table 8. ** = statistically significant (1% level, 2-tailed).