Auditor Tenure and Perceptions of Audit Quality

585

THE ACCOUNTING REVIEWVol. 80, No. 22005pp. 585–612

Auditor Tenure and Perceptionsof Audit Quality

Aloke GhoshU.S. Securities and Exchange Commission

andBaruch College—The City University of New York

Doocheol MoonState University of New York at Old Westbury

ABSTRACT: We analyze how investors and information intermediaries perceive auditortenure. Using earnings response coefficients from returns-earnings regressions as aproxy for investor perceptions of earnings quality, we document a positive associationbetween investor perceptions of earnings quality and tenure. Further, we find that theinfluence of reported earnings on stock rankings becomes larger with extended tenure,although the association between debt ratings and reported earnings does not varywith tenure. Finally, we find that the influence of past earnings on one-year-aheadearnings forecasts becomes greater as tenure increases. In general, our results areconsistent with the hypothesis that investors and information intermediaries perceiveauditor tenure as improving audit quality. One implication of our study is that imposingmandatory limits on the duration of the auditor-client relationship might impose unin-tended costs on capital market participants.

Keywords: auditor tenure; auditor independence; audit quality; earnings quality; man-datory auditor rotation; capital market perceptions.

Data Availability: All data used in this study are available from public sources.

We thank Ashiq Ali, Mark Beasley, Marti Benis, Donal Byard, Masako Darrough, Paquita Davis-Friday, AnitaDennis, Zeng Deng, Rajib Doogar, Amy Edwards, Mary Greenawalt, Zhaoyang Gu, Larry Harris, Prem Jain, JoeKerstein, Ying Li, Steve Lilien, Steve Lustgarten, Darius Miller, two anonymous referees, Srini Sankaraguruswamy,Bharat Sarath, Dave Smith, Jonathan Sokobin, Jan Sweeney, Tony Tinker, Joe Weintrop, and Jimmy Ye for theirhelpful comments. This paper was previously titled: ‘‘Does Auditor Tenure Impair Audit Quality?’’ We acknowledgefinancial support from the Robert Zicklin Center for Corporate Integrity of Baruch College.

The Securities and Exchange Commission disclaims responsibility for any private publication or statement by anySEC employee or commissioner. This paper expresses the authors’ views and does not necessarily reflect those ofthe Commission, the commissioners, or other members of the staff.

Editor’s note: This paper was accepted by Terry Shevlin, Senior Editor.Submitted August 2002Accepted August 2004

586 Ghosh and Moon

The Accounting Review, April 2005

I. INTRODUCTION

The recent rise in accounting irregularities has reopened questions about auditor ten-ure, independence, and audit quality (Bricker 2002).1 Recent studies provide valuableinsights into the debate surrounding auditor tenure by examining the association

between tenure and (1) accounting accruals, (2) analysts’ forecast errors, and (3) the costof debt. Myers et al. (2003) conclude that longer auditor tenure constrains managerialdiscretion with accounting accruals, which suggests high audit quality. Johnson et al. (2002)also find that accruals are larger and less persistent for firms with short auditor tenurerelative to those with medium or long tenure. Using credit spreads between bond yieldsand matched Treasury yields as the cost of debt, Mansi et al. (2004) find that the cost ofdebt declines with longer tenure, which suggests bondholders perceive audit quality asimproving with extended tenure. In contrast, Davis et al. (2002) conclude that audit qualitydeclines with extended tenure because, as tenure increases, client firms have greater re-porting flexibility and earnings forecast errors decline.

This study focuses on how investors and information intermediaries perceive auditortenure. Using earnings response coefficients from returns-earnings regressions as a proxyfor investor perceptions of earnings quality, we analyze whether investors perceive earningsquality as being affected by tenure. Given the importance of information intermediarieswho receive and process financial information for investors, we also analyze whether in-dependent rating agencies and financial analysts incorporate the potential effects of tenureon earnings quality. Specifically, we examine whether tenure affects the relationship be-tween reported earnings and (1) stock rankings, (2) debt ratings, and (3) analysts’ earningsforecasts.

A key distinction between our study and those of Davis et al. (2002) and Mansi et al.(2004) is the difference in the research design. We examine whether the extent to whichanalysts rely on past reported earnings to predict future earnings varies with tenure, whereasDavis et al. (2002) explore the association between forecast errors and tenure. Our researchdesign might be better suited because it is difficult to draw unambiguous inferences aboutanalysts’ perceptions from an association between forecast errors and tenure. Lower forecasterrors with longer tenure might suggest that earnings quality is perceived as improving withtenure because earnings are more predictable. However, it could also suggest lower earningsquality if managers increasingly guide earnings forecasts as auditor tenure lengthens. Fur-ther, Mansi et al. (2004) investigate the influence of tenure on the cost of debt, whereaswe examine whether the influence of reported earnings on debt ratings varies with tenure.Although Mansi et al. (2004) examine the effect of tenure on debt ratings, their primaryaim is to purge (orthogonalize) the information effects of tenure and other control variablesrelated to debt ratings and ultimately examine the insurance role of tenure on the cost ofdebt. In contrast, our emphasis is on the informational role of tenure. Our fundamentalobjective is to provide insights into any changes in the perceived credibility of reportedearnings with extended auditor tenure.

Emphasis on capital market perceptions of independence and audit quality is consistentwith the Financial Accounting Standards Board’s (FASB) conceptual framework for finan-cial reporting and principles of auditor independence (Carmichael 1999). According to the

1 Regulators express concerns that pressure to retain client firms and the ‘‘comfort level’’ created between auditorsand management over time impair auditor independence, which adversely affects audit quality (U.S. GovernmentAccounting Office [GAO] 2003). The accounting profession, on the other hand, claims that the likelihood ofaudit failures is greater during the initial period of an auditor-client relationship because of lack of informationabout client-specific risks (PricewaterhouseCoopers 2002).

Auditor Tenure and Perceptions of Audit Quality 587


Statement of Financial Accounting Concepts (SFAC No. 1, FASB 1978), ‘‘financial state-ments are often audited by independent accountants for the purpose of enhancing confidencein their reliability.’’ The AICPA (1994) also acknowledges the importance of consideringinvestor perceptions of auditor independence. A former chairman of AICPA, Elliott (2000)says ‘‘[The AICPA] believe[s] that appearances are very important and capital marketsrequire confidence in financial statements and audit reports, and the member firms of theAICPA are basing their business of auditing on their reputations, and that is heavily affectedby appearance.’’

Other things remaining constant, audited financial statements are less reliable (or oflower perceived quality) for investment and credit decisions if users of financial statementsview lengthy tenure as having an adverse effect on auditor independence and audit quality.Alternatively, investors and information intermediaries are more likely to rely on reportedaccounting numbers if they perceive that greater auditor expertise from longer tenure im-proves independence and audit quality. We assert that the relationship between perceptionsof earnings quality and auditor tenure provides insights into how capital market participantsview auditor tenure as affecting audit quality.

We use earnings response coefficients (ERCs) from contemporaneous returns-earningsregressions to measure investor perceptions of earnings quality. After controlling for theother determinants of ERC—such as the age of the firm, the quality of auditors, growth,earnings persistence, earnings volatility, systematic risk, firm size, financial leverage, andregulatory environment—we find that the magnitude of the ERC increases as the auditor-client relationship lengthens. One concern is that if the market anticipates a portion ofcurrent earnings more than one-year-ahead of the earnings release, then the estimated ERCmight be biased downward (Kothari 1992; Kothari and Sloan 1992). Additionally, if themarket is more likely to anticipate current earnings for firms with longer tenure, then thebias might be associated with tenure. Consistent with the premise that prices lead earnings,the estimated ERC for firms with extended tenure is larger when we increase the returnsmeasurement window or when we use non-market metrics, such as total assets, to deflateearnings. Thus, the ERC-based findings are consistent with the hypothesis that investorsperceive auditor tenure as enhancing earnings quality.

Further, controlling for the other determinants of stock rankings, we find that the influ-ence of reported earnings on Standard & Poor’s (S&P) common stock rankings becomeslarger with extended tenure. In contrast, the association between S&P debt ratings andreported earnings does not vary with tenure. Thus, our results provide modest evidence thatindependent rating agencies perceive reported earnings as being more reliable for firms withlonger tenure.

Finally, after controlling for factors that affect analysts’ earnings forecasts, we find thatthe influence of past earnings on one-year-ahead earnings forecasts becomes larger overextended auditor-client relationships. All else equal, analysts are more likely to rely onreported earnings to predict future earnings with longer tenure. Thus, the results fromanalyst earnings forecasts also suggest that analysts perceive earnings quality as improvingwith longer auditor tenure.

While our results are based on a large sample of firms spanning 11 years, there is oneconcern with our dataset. If auditor tenure is endogenous to audit quality, then the resultsare also consistent with an alternative hypothesis that auditor turnover is high for firmswith low earnings quality. In other words, high-quality auditors might terminate engage-ments with client firms that prefer low-quality financial statements (DeFond andSubramanyam 1998). We address potential concerns about frequent auditor turnover byconstructing a subsample for which the auditor-client relationship lasts for at least five

588 Ghosh and Moon


years, as in Myers et al. (2003). When we use this subsample, our inferences remainunchanged for all our tests.

In sum, our results are generally consistent with the hypothesis that reported earningsare perceived as being more reliable as auditor tenure increases. One implication of ourresults is that capital market participants view longer auditor tenure as having a favorableimpact on audit quality. Our results suggest that imposing mandatory limits on the durationof the auditor-client relationship might impose unintended costs on capital markets. How-ever, results from a regime without auditor term limits may not be applicable to a regulatedenvironment because of differences in economic incentives for both auditors and clients.

The rest of the paper is organized as follows. Section II establishes the links betweentenure, independence, and perceptions of audit quality. Section III discusses the researchdesign, and Section IV describes the sample selection procedure. Section V reports theresults of the association between tenure and perceptions of earnings quality. Section VIconcludes the paper.

II. TENURE, INDEPENDENCE, AND PERCEPTIONS OF AUDIT QUALITYIndependent auditors are considered the ‘‘gatekeepers’’ of the public securities markets

(SEC 2001, III.A). However, the recent rash of accounting irregularities has led many toquestion auditor independence (Wall Street Journal 2002a, 2002b). One perception is thatauditors are more likely to agree with managers on important reporting decisions as thelength of the audit engagement increases (Ryan et al. 2001; Farmer et al. 1987). Therefore,imposing mandatory limits on auditor tenure is expected to improve audit quality by re-ducing client firms’ influence over auditors (Turner 2002; Brody and Moscove 1998; SEC1994; AICPA 1978; U.S. Senate 1977; Mautz and Sharaf 1961).

An opposing viewpoint is that problem audits occur more frequently for newer clientsbecause auditors have less information about these firms (AICPA 1992). Client-specificknowledge of items such as operations, accounting system, and internal control structureis crucial for auditors to detect material errors and misstatements. In particular, Johnson etal. (2002) argue that lack of adequate client-specific knowledge during the early years ofengagement decreases the likelihood of detecting material errors and misstatements. As theauditor-client relationship lengthens, firm-specific expertise allows auditors to rely less onmanagerial estimates and become more independent of management (Solomon et al. 1999).

The crux of the debate rests on how tenure affects auditor independence. Proponentsof mandatory auditor rotation claim that lengthy auditor tenure erodes independence, whichin turn impairs audit quality. Others argue that independence and audit quality increasewith longer tenure because of improved auditor expertise from superior client-specificknowledge.2 Since independence is not observable, regulators, practitioners, and academicsoften rely on the appearance dimension to define auditor independence (Dopuch et al. 2003;Kinney 1999).

Our study investigates whether capital market participants perceive longer tenure asaffecting audit quality. Insights from a market-based approach are important along at leasttwo dimensions. First, academics often emphasize the need to understand capital marketperceptions of auditor independence and audit quality because ultimately the value of au-diting services depends on perceptions of independence (Dopuch et al. 2003; Shockley

2 Expert knowledge gained through years of on-the-job experience increases the likelihood that auditors will detecterrors in financial statements (Ashton 1991; Libby and Frederick 1990). In contrast, industry specialization,another frequently used proxy for auditor expertise, is based on training and practical experience gained fromauditing in a particular industry (Gramling and Stone 2001; Solomon et al. 1999; Hogan and Jeter 1999; Craswellet al. 1995).



1981). Second, regulatory institutions such as the FASB (SFAC No. 1) and SEC (2000)emphasize the importance of capital market perceptions of auditor independence. A per-ception that the auditors’ work is more objective and independent inspires greater confi-dence in auditor opinion, which increases the perceived reliability or quality of reportedaccounting numbers (Ryan et al. 2001; Elliott and Jacobson 1998). Independent auditorsincrease the reliability of financial statements because (1) they are more likely to preventor detect and correct material misstatements/omissions, and (2) they ensure that financialstatements comply with generally accepted accounting principles (Carmichael 1999). There-fore, to the extent that capital market participants view tenure as improving independenceand audit quality, financial statements are perceived as more reliable for financial decisionsas tenure lengthens.

In this study, we focus on the perceptions of investors because they are the principalusers of financial statements. In its Conceptual Framework, the FASB defines ‘‘quality’’ inrelation to the usefulness of financial statements to investors and links ‘‘usefulness’’ in turnto constructs such as relevance and reliability (SFACs No. 1 [FASB 1978] and No. 2 [FASB1980]). To draw inferences about investors’ perceptions of earnings quality, researchers tendto use stock-market-based metrics such as earnings response coefficients from regressionsof returns on earnings (Schipper and Vincent 2003; Warfield et al. 1995). Prior studiesdocument that investors pay a larger premium for ‘‘high-quality’’ earnings because high-quality earnings are viewed as sustainable (Schipper and Vincent 2003; Teoh and Wong1993). Thus, examining the influence of auditor tenure on the pricing of earnings is likelyto provide valuable insights into investors’ views of the association between the length ofthe auditor-client relationship and earnings quality.

Given that information intermediaries constitute an integral part of the capital marketby providing stock recommendations, debt ratings, and earnings forecasts (Lang andLundholm 1996), we also analyze how independent rating agencies and financial analystsview auditor tenure. Hunt (2002) states that independent rating agencies provide informationabout the creditworthiness of issuers and that credit ratings play a significant role in in-vestment decisions. Extant research finds links between earnings and debt ratings/stockrankings issued by independent rating agencies (Bhojraj and Sengupta 2003; Ziebart andReiter 1992; Van Horne 1992; Kaplan and Urwitz 1979), suggesting that perceptions ofearnings quality could be an important input in determining rankings/ratings. Similarly,financial analysts also play a prominent role as information intermediaries in capital marketsbecause of their ability to incorporate value-relevant information in their published reports,which impacts security prices (Francis and Soffer 1997; Schipper 1991; Lys and Sohn 1990;Brown et al. 1987). Prior research finds that analysts rely on earnings releases to estimatefuture earnings (Kasznik and McNichols 2002; Barron et al. 2002; Stickel 1989), whichsuggests that the extent to which analysts depend on reported earnings to make earningsforecasts might vary with perceptions of earnings quality.

Extending this line of research, we examine the association between auditor tenure and(1) stock rankings, (2) debt ratings, and (3) analysts’ forecasts of earnings per share. Ad-ditionally, we analyze how auditor tenure affects the association between reported earningsand rankings, ratings, and earnings forecasts. If auditor tenure is perceived as enhancingearnings quality, then, all else equal, the influence of reported earnings on rankings/ratingsand earnings forecasts is expected to become larger with longer auditor tenure becausereported earnings are viewed as more informative about future earnings. The converse istrue if information intermediaries perceive longer auditor tenure as eroding earnings quality.

Based on the association between our proxy measures for perceptions of earnings qual-ity and auditor tenure, we infer how investors and information intermediaries view auditor

590 Ghosh and Moon


tenure as affecting audit quality. A positive relationship between the proxies for perceptionsof earnings quality and auditor tenure is consistent with the hypothesis that longer tenureis perceived as improving independence and audit quality. In contrast, a negative relation-ship is consistent with the hypothesis that lengthy tenure is viewed as eroding independenceand audit quality.

III. RESEARCH DESIGNWe use the following basic regression framework to analyze whether investors, inde-

pendent rating agencies, and financial analysts perceive earnings quality as being affectedby auditor tenure:

Dependent variable � � � � E � � �E � � E*Tenure � � �E*Tenure1 2 3 4

9

� � Tenure � � E*Control variable�5 6�2( j�1) jj�1

9

� � �E*Control variable� 7�2( j�1) jj�1

9

� � Control variable � ε. (1)� 23�j jj�1

E and �E are reported earnings and changes in reported earnings, respectively. Tenure isthe duration of the auditor-client relationship (calculated using Compustat #149) in years,starting from 1982 (audit firm mergers are considered as a continuation of the prior auditor).Subscript j represents the number of control variables (1 to 9). Each of the control variablesis interacted with E/�E and is also included as a separate independent variable. Details onthe dependent and control variables are provided in subsequent subsections.

The sum of earnings levels and changes coefficients (�1 � �2) or the ‘‘earnings responsecoefficient’’ (ERC) is our proxy for capital markets’ perceptions of earnings quality. Ourinterest is in the sum of the E*Tenure and �E*Tenure coefficients (�3 � �4). If investors,rating agencies, and analysts perceive earnings quality as improving (declining) with longerauditor tenure, �3 � �4 is expected to differ from zero.

Perceptions of Investors and Auditor TenureWe measure investor perceptions of earnings quality using 12-month (ending three

months after the fiscal year-end) cumulative market-adjusted returns (CAR) as the dependentvariable in Equation (1). Market-adjusted returns are the difference between raw returnsand value-weighted CRSP market returns.

We include earnings changes and earnings levels in the same regression because in-cluding both increases the explanatory power and magnitude of earnings response coeffi-cients when earnings contain both transitory and permanent components (Easton and Harris1991; Ali and Zarowin 1992). E is income before extraordinary items (Compustat #18) and�E is the difference between income before extraordinary items for the current year andthat of last year. Both E and �E are deflated by market value of equity (Compustat #25� #199) at the beginning of the year.

We include various control variables because the ERC is associated with other firmcharacteristics (Warfield et al. 1995; Dhaliwal and Reynolds 1994; Collins and Kothari1989), which, in turn, might be associated with tenure. The control variables are definedas follows: FirmAge, computed using the beginning and end dates as reported in CRSP,



measures the number of years that the firm has been publicly traded as of the fiscal year-end; Big4 is an indicator variable that equals 1 when the client’s auditor is a large account-ing firm (Compustat #149); Growth is the sum of the market value of equity (Compustat#25 � #199) and the book value of debt (Compustat #9 � #34) scaled by the book valueof total assets (Compustat #6); Persistence (Volatility) is the first-order autocorrelation (stan-dard deviation) of income before extraordinary items per share (Compustat Quarterly #8/Quarterly #61) for the past 16 quarters;3 Beta is systematic risk computed using the past60 monthly stock returns; Size is the logarithmic transformation of the fiscal year-endmarket value of equity (Compustat #25 � #199) of the prior year; Leverage is the ratio oftotal debt (Compustat #9 � #34) to total assets (Compustat #6); and Regulation is anindicator variable that equals 1 for firms in a regulated industry with two-digit standardindustry classification codes between 40 and 49 or between 60 and 63.

We include FirmAge for two reasons: (1) older firms are more likely to be stable withless information asymmetry problems, which suggests higher ERCs; and (2) Tenure andFirmAge are positively correlated. We control for Big4 because large auditors are generallyassociated with high-quality audits (Becker et al. 1998; Teoh and Wong 1993). The inclu-sion of Growth, Persistence, Volatility, and Beta is primarily motivated by valuation con-siderations (Warfield et al. 1995). The inclusion of Size is motivated by the political costtheory: managers of large, politically sensitive firms are more likely to exploit the latitudein accounting to reduce political costs, which affects earnings quality. We include Leveragebecause of contracting considerations: firms with high leverage are more likely to use thelatitude in accounting to avoid possible debt-covenant violations (DeFond and Jiambalvo1994). Finally, the quality of earnings is affected in a regulated environment (Regulation)because managers have limited scope for opportunistic behavior (Warfield et al. 1995).

Perceptions of Information Intermediaries and Auditor TenureIndependent Rating Agencies

Our analysis of how tenure affects independent rating agencies’ perceptions of earningsquality is based on estimates from Equation (1) using Standard & Poor’s (S&P) commonstock rankings (Stock Rankings) and senior debt ratings (Debt Ratings) as dependent vari-ables. Stock Rankings represent numerical values of 1 to 7, corresponding to S&P commonstock rankings (Compustat #282). A value of 1 is assigned if a firm’s common stock rankingis rated as A� and as S&P common stock rankings decline, the numerical value increasesby 1.4 Similarly, Debt Ratings are assigned a value of 1 if a firm’s S&P senior debt (Com-pustat #280) is rated as AAA. As S&P debt ratings decline from AAA (the highest) to D(payment default), the numerical value increases by 1.

We include Growth, Volatility, Beta, Size, and Leverage to control for cross-sectionaldifferences in firm quality and riskiness (Bhojraj and Sengupta 2003; Van Horne 1992;Ziebart and Reiter 1992; Kaplan and Urwitz 1979). Growing firms with high earningsvolatility tend to be more risky, Beta controls for both operating and financial risk, largefirms tend to be less risky, and firms with higher leverage have greater financial risk. Wealso include FirmAge, Big4, Persistence, and Regulation because the association betweenreported earnings and rankings/ratings might depend on the firm’s life-cycle, whether firms

3 Since Persistence and Volatility measures are computed using time-series income per share numbers, we adjustfor stock-splits and stock dividends that occur subsequent to the end of a given period using the adjustmentfactor reported in Compustat (Quarterly #17).

4 For instance, values 2 through 7 correspond to the S&P common stock rankings of A, A–, B�, B, B–, and C,respectively. We do not include S&P common stock rankings of D (in reorganization) and LIQ (liquidation) forour sample because of going-concern issues.

592 Ghosh and Moon


use big audit firms, whether earnings are persistent, and whether firms are in a regulatedenvironment.

In a recent study, Mansi et al. (2004) find that longer tenure lowers the rate of returnrequired by bondholders and that the impact is larger for ‘‘information-sensitive’’ securitiessuch as non-investment grade bonds (debt ratings less than BBB�). Therefore, we alsoseparately estimate the effects of tenure on investment and non-investment grade debtratings.

Financial AnalystsFinally, our analysis of whether tenure influences financial analysts’ perceptions of

earnings quality is based on the estimates from Equation (1) using earnings forecasts(FEPSt) as the dependent variable. FEPSt is the consensus analyst forecasts of annualearnings per share from the Institutional Brokers Estimation System (I/B/E/S) database.Our proxy for consensus forecasts is the mean one-year-ahead forecast for year t issuedimmediately following the earnings announcement for year t�1 (earnings announcementdates are obtained from Compustat).5 We restrict the forecasts to those issued immediatelyfollowing earnings announcements, as in Kasznik and McNichols (2002) and Barron et al.(2002), because (1) any change in analysts’ perceptions of earnings quality is more likelyto be updated following earnings releases (Barron et al. 2002), (2) it excludes stale forecastssince forecast revisions are more common following earnings announcements (Stickel1989), and (3) it conditions the forecasts on the same set of publicly disclosed information.

Reported earnings per share (EPSt�1 and �EPSt�1) are used as measures for E and �Ein Equation (1). EPSt�1 is annual earnings per share reported for year t�1, and �EPSt�1 isthe absolute change in earnings per share for year t�1 defined as the difference in annualearnings per share in year t�1 and that in year t�2 (�EPSt�1 � EPSt�2�), both obtainedfrom the I/B/E/S database. As in Barron et al. (2002), we include the absolute value ofchanges in earnings because a larger earnings surprise tends to be temporary, which reducesthe potential usefulness of past earnings in predicting future earnings.

Since prior studies find that analysts’ incentives to acquire information about futureearnings are affected by firm characteristics (DeFond and Hung 2003; Barron et al. 2002;Lang and Lundholm 1996), we include the following control variables: FirmAge becauseolder firms are more likely to be stable and consequently earnings might be easier to predict;auditor type (Big4) because analysts might perceive earnings quality as improving for firmswith big auditors; Growth because high-growth firms tend to generate greater demand forprivate information, thereby reducing the reliance on reported earnings (Barron et al. 2002;Lang and Lundholm 1996); Volatility because analysts are more likely to attach lowerimportance on reported earnings for firms with higher earnings volatility (DeFond and Hung2003); risk using a market-based measure (Beta) and a balance-sheet-based measure (Lev-erage); Size because prior studies find that firm size is associated with risk and the infor-mation environment, which affects earnings predictability (DeFond and Hung 2003; Barronet al. 2002); Regulation because the predictability of earnings might vary between industries(O’Brien 1990); and the number of analysts (Analysts) providing annual earnings per share

5 We prefer consensus forecasts to individual forecasts because prior research finds that consensus forecasts aremore accurate than the individual forecasts underlying the consensus. One explanation for this result is that theforecast errors made by individual analysts are less than perfectly correlated and therefore in the process ofaggregation, the individual forecast errors offset each other (see Brown et al. [1985] for related papers). Theresults are similar when we use either the mean or the median forecast.



forecasts for a firm (Lang and Lundholm 1996). All the control variables including Tenureare measured at the end of year t�1.

In a recent study, Davis et al. (2002) find that forecast errors decline as tenure in-creases.6 However, it is difficult to draw unambiguous inferences about perceptions ofanalysts from an association between forecast errors and tenure. Lower forecast errors mightsuggest that analysts perceive earnings quality as improving with tenure because earningsare more predictable. Alternatively, lower forecast errors with longer tenure might suggestlower earnings quality if managers are more likely to guide earnings forecasts as auditortenure lengthens.7 Hence, we do not focus on earnings forecast attributes such as forecasterrors and dispersion.

IV. DATAWe construct a sample from the list of publicly traded firms in the 2001 Compustat

annual files (active and research). Since the 2001 Compustat files cover 20 years of data,and financial data are available from 1982, auditor tenure is 1 for the first year by construc-tion. The analysis begins with 1990 to provide some variation for auditor tenure and ex-cludes the year 2001 because of possible missing and incomplete data.8 Stock return andfirm age data are obtained from the 2001 CRSP files. We get earnings forecast data fromthe 2001 Institutional Brokers Estimation System (I/B/E/S) Summary Estimates.

We impose the following restrictions on the sample: (1) we delete the top and bottom1 percent of observations for the level of earnings (E), changes in earnings (�E), annualearnings per share (EPS), and the absolute change in annual earnings per share (�EPS);(2) we remove all observations with the absolute value of cumulative market-adjusted re-turns (CAR) greater than 100 percent; and (3) we also winsorize the top and bottom 1percent of observations for Growth, Persistence, Volatility, Beta, and Leverage.9 This sampleselection procedure results in a maximum of 38,794 observations for the ‘‘full’’ sampleover the years 1990 through 2000.

Some concerns remain with the full sample. Firms might frequently switch auditorseither because of ‘‘opinion shopping’’ (SEC 1988) or because of auditors’ preference forconservative accounting choices (DeFond and Subramanyam 1998). Alternatively, high-quality auditors might end engagements with clients that prefer low quality of financial

6 Researchers frequently use forecast errors and dispersion among forecasts as proxies for the uncertainty amonganalysts about the quality of accounting information (Hope 2003; Duru and Reeb 2002; Ashbaugh and Pincus2001; Barron et al. 1998; Lang and Lundholm 1996).

7 Managers have strong incentives to avoid negative earnings surprises because of the large stock-price reactionsto negative earnings (see Ghosh et al. 2005; Bartov et al. 2002; Myers and Skinner 2002; Skinner and Sloan2002; Barth et al. 1999). Firms can avoid earnings surprises by manipulating earnings to meet expectations orby guiding analyst expectations to avoid overly optimistic forecasts. Although existing studies suggest thatearnings management declines with longer auditor tenure (Myers et al. 2003; Johnson et al. 2002), we do notknow whether tenure affects managers’ propensity to guide analysts’ expectations.

8 Our computation of Tenure will understate the actual length of auditor tenure because Tenure is 1 for firms in1982 even if the auditor was the same for the prior years. However, when we replicate our analysis computingTenure from 1974 using information about the client’s auditor as reported in Compustat backfiles, similar toMansi et al. (2004), our conclusions remain unaffected.

9 The use of alternative cut-off points such as 0.5 percent also yields similar results, but it retains extremeobservations. For instance, the minimum of E is –541 percent and the maximum of �E is 629 percent whenwe delete the top and bottom 0.5 percent. We delete observations with absolute CARs greater than 100 percentto reduce the impact of major corporate restructuring activities such as mergers, bankruptcies, spin-offs, etc.However, when we delete the top and bottom 1 percent of CAR, we get similar results. The results on tenureare unaffected when we truncate the top and bottom 1 percent of observations for the control variables.

594 Ghosh and Moon


TABLE 1Descriptive Statistics

Variables Mean Min 25% 50% 75% Max n

Panel A: Full Sample

Tenure (years) 8.549 1.000 5.000 8.000 12.000 19.000 38,794FirmAge (years) 17.597 3.417 7.000 12.167 23.833 75.000 38,794CAR �0.060 �0.991 �0.343 �0.108 0.158 0.999 38,794E �0.007 �2.795 �0.022 0.048 0.083 0.523 38,794�E 0.006 �1.418 �0.032 0.006 0.034 2.815 38,794Stock Rankings 5.096 1.000 4.000 5.000 6.000 7.000 25,745Debt Ratings 9.147 1.000 6.000 9.000 12.000 23.000 9,462FEPS ($) 1.045 �4.140 0.450 0.890 1.490 7.250 16,417

Panel B: Restricted Sample

Tenure (years) 9.096 1.000 6.000 9.000 12.000 19.000 35,826FirmAge (years) 18.004 3.417 7.000 12.750 24.000 75.000 35,826CAR �0.054 �0.989 �0.332 �0.102 0.161 0.999 35,826E 0.000 �2.795 �0.013 0.050 0.084 0.523 35,826�E 0.006 �1.418 �0.030 0.006 0.033 2.804 35,826Stock Rankings 5.055 1.000 4.000 5.000 6.000 7.000 24,216Debt Ratings 9.087 1.000 6.000 9.000 12.000 23.000 9,060FEPS ($) 1.049 �4.140 0.450 0.890 1.500 7.250 15,699

The full sample includes Compustat firms with available data from 1990 to 2000. The restricted sample consistsof firms in the full sample with auditor-client relationships lasting for at least five years. Tenure is the durationof the auditor-client relationship in years starting from 1982. FirmAge measures the number of years that thefirm has been publicly traded as of the fiscal year-end. CAR is cumulative market-adjusted returns for the 12-month period ending three months after the fiscal year-end. Market-adjusted returns are the difference betweenraw returns and value-weighted CRSP market returns. E is income before extraordinary items deflated by marketvalue of equity at the beginning of the year. �E is the difference between income before extraordinary items forthe current year and that of last year deflated by market value of equity at the beginning of the year. StockRankings and Debt Ratings represent numerical values of S&P common stock rankings and S&P senior debtratings, respectively. Stock Rankings (Debt Ratings) are assigned a value of 1 if a firm’s S&P common stock(senior debt) is rated as A� (AAA), and as the S&P common stock rankings (debt ratings) decline, thenumerical value increases by 1. FEPS is the mean annual one-year-ahead earnings per share forecasts for year tissued immediately following the earnings announcement for year t�1.

reporting. We avoid some of these biases by constructing a subsample in which the auditor-client relationship lasts for at least five years. For auditor-client relationships lasting lessthan five years, we delete all of the firm-year observations. The maximum number ofobservations for the ‘‘restricted’’ sample is 35,826 firm-years.10

Table 1 reports summary statistics for both the full sample (Panel A) and the restrictedsample (Panel B). We report the mean, minimum, first quartile, median, third quartile, andmaximum for Tenure, FirmAge, CAR, E, �E, Stock Rankings, Debt Ratings, and FEPS.Auditor tenure is longer for the restricted sample than for the full sample; the mean/median

10 Although we use the Myers et al. (2003) methodology to construct the restricted sample, the percentage ofobservations deleted from the full sample is much smaller in our study. One primary reason for the differenceis that we include many more variables in our analyses, which reduces the size of our full sample. Thus, fewerobservations are lost from requiring that the auditor-client relationship lasts for at least five years in our studyrelative to those in Myers et al. (2003).



Tenure for the full sample is 8.5/8 years, while the corresponding number for the restrictedsample is 9.1/9 years. The mean FirmAge for the full sample is 17.6 years, compared to amean of 18.0 years for the restricted sample. FirmAge ranges between 3.4 years and 75years for the full and restricted samples.

The summary statistics for CAR, E, and �E indicate few differences across the twosamples. The mean/median CAR for the full sample is �0.060/�0.108, while that for therestricted sample is �0.054/�0.102.11 The mean/median E for the full sample is �0.007/0.048, compared to a mean/median of 0.000/0.050 for the restricted sample. The mean/median �E is identical for the full and restricted samples (0.006).

Similarly, there are almost no differences in Stock Rankings, Debt Ratings, and earningsforecasts (FEPS) between the two samples. The average firm in both samples has a B S&Pcommon stock ranking (a numerical score of 5), and a BBB S&P senior debt rating (anumerical score of about 9). The mean/median one-year-ahead analysts’ earnings per shareforecasts (FEPS) is $1.05/$0.89 and $1.05/$0.89 for the full and restricted samples,respectively.

Table 2 reports the Pearson correlation (�) matrix between CAR, E, �E, Stock Rankings,Debt Ratings, FEPS, Tenure, and FirmAge for the full sample. The high correlation betweenStock Rankings and Debt Ratings (� � 0.686) suggests that stock rankings and debt ratingsissued by independent rating agencies contain significant overlapping information. However,low correlations across variables that measure investor, rating agency, and financial analystperceptions indicate that analysis of each market participant’s perceptions provides distinctinsights into the overall perceptions of the quality of accounting information. Finally, thecorrelation between Tenure and FirmAge is positive and statistically significant (� � 0.337,p-value � 0.001), which underscores the need to control for FirmAge when analyzingTenure.

V. RESULTSAll the regression results reported in Tables 3 to 7 are based on yearly regressions from

1990 to 2000. The reported coefficients are the average of 11 yearly coefficients, and thecorresponding t-statistics are computed by comparing the average coefficient to its time-series standard error. We prefer this approach to a pooled time-series cross-sectional esti-mation because the standard errors in the latter approach are not adjusted for the correlationof regression residuals across firms, which lead to inflated t-statistics. The estimation pro-cedure of year-by-year cross-sectional regressions and the use of average slopes and theirtime-series standard errors to draw inferences allow for residual cross-correlation (for de-tails, see Fama and French 2000).

Perceptions of Investors and Auditor TenureIn this subsection, we report the findings of how auditor tenure affects the returns-

earnings association. Our interest is in the sign and magnitude of the sum of the coefficientson E*Tenure and �E*Tenure (�3 � �4). Consistent with prior research, we find that reportedearnings (E and �E) are significantly positively associated with returns (CAR) in Table 3.The sum of the coefficients on E and �E, or ERC, is 0.507 (t-statistic � 10.30) in the first

11 Prior studies also find negative mean CAR: (1) DeFond and Hung (2003) report that the mean CAR measuredover fifteen months for a sample of 21,908 observations from 1993 to 1999 is –0.05 (–0.04) for firms with(without) I /B /E /S cash flow forecasts; (2) Subramanyam and Wild (1996) find that the mean CAR over onequarter for a sample of 25,160 observations from 1981 to 1990 is –0.01; and (3) Ali and Zarowin (1992) reportthat the mean 12-month CAR from 1970 to 1985 is –0.05 (–0.04) for the transitory (permanent) group with4,355 (7,219) observations.

596 Ghosh and Moon


TABLE 2Pearson Correlation Matrix

Variables 1 2 3 4 5 6 7 8

1. CAR 1.0002. E 0.262

(0.001)1.000

3. �E 0.194(0.001)

0.368(0.001)

1.000

4. Stock Rankings �0.118(0.001)

�0.266(0.001)

0.068(0.001)

1.000

5. Debt Ratings �0.185(0.001)

�0.341(0.001)

�0.028(0.004)

0.686(0.001)

1.000

6. FEPS �0.002(0.802)

0.021(0.001)

�0.003(0.652)

�0.007(0.391)

�0.080(0.001)

1.000

7. Tenure 0.112(0.001)

0.122(0.001)

0.015(0.001)

�0.169(0.001)

�0.201(0.001)

0.024(0.001)

1.000

8. FirmAge 0.093(0.001)

0.115(0.001)

0.010(0.026)

�0.291(0.001)

�0.428(0.001)

0.025(0.001)

0.337(0.001)

1.000

Pearson correlation matrix for the full sample. CAR is cumulative market-adjusted returns for the 12-monthperiod ending three months after the fiscal year-end, where market-adjusted returns are the difference betweenraw returns and value-weighted CRSP market returns. E is income before extraordinary items deflated by marketvalue of equity at the beginning of the year. �E is the difference between income before extraordinary items forthe current year and that of last year deflated by market value of equity at the beginning of the year. StockRankings and Debt Ratings represent numerical values of S&P common stock rankings and S&P senior debtratings, respectively. Stock Rankings (Debt Ratings) are assigned a value of 1 if a firm’s S&P common stock(senior debt) is rated as A� (AAA), and as the S&P common stock rankings (debt ratings) decline, thenumerical value increases by 1. FEPS is the mean annual one-year-ahead earnings per share forecasts for year tissued immediately following the earnings announcement for year t�1. Tenure is the duration of the auditor-client relationship in years starting from 1982. FirmAge measures the number of years that the firm has beenpublicly traded as of the fiscal year-end.The p-values for correlation coefficients are reported in parentheses.

regression without the control variables using the full sample. More important, �3 � �4 ispositive and significant (0.012, t-statistic � 3.45). Regression estimates indicate that inves-tors on average pay a premium of 2.4 percent (� 0.012/0.507) for earnings as the auditor-client relationship increases by an additional year. The coefficient on Tenure (�5) is alsosignificant (0.005, t-statistic � 4.06), indicating that auditor tenure is positively associatedwith returns.

The ERC is 0.432 (t-statistic � 9.00) in the second regression when we include anumber of control variables. �3 � �4 remains positive and significant (0.008, t-statistic� 2.48). The parameter estimates suggest that, after controlling for the other determinantsof ERC, investors pay a premium of 2 percent (� 0.008/0.432) for earnings as the auditengagement increases by an additional year. Even though the magnitude of �3 � �4 issmaller in the second regression, we find it reassuring that the results continue to be sig-nificant even after including many control variables. When we estimate the augmentedmodel using the restricted sample for which the auditor-client relationship lasts for at leastfive years, we find similar results. Specifically, �3 � �4 is positive and significant at the 5percent level (0.008, t-statistic � 2.03) in the third regression.

The parameter estimates of the components of ERC provide added insights into investorperceptions of the time-series properties of earnings. The coefficients on E and �E in all



TABLE 3Earnings Response Coefficients and Perceptions of Investors

Variables (Coefficients)Full Sample

(1) (2) Restricted Sample

Intercept (�) �0.037 (�5.82)*** �0.096 (�6.40)*** �0.088 (�5.87)***E (�1) 0.356 (6.91)*** 0.261 (4.51)*** 0.195 (2.92)***�E (�2) 0.151 (8.09)*** 0.171 (5.04)*** 0.232 (5.60)***

(�1 � �2) 0.507 (10.30)*** 0.432 (9.00)*** 0.427 (7.09)***E*Tenure (�3) 0.007 (2.49)** 0.006 (1.80)* 0.009 (2.18)**�E*Tenure (�4) 0.005 (3.39)*** 0.002 (1.34) �0.001 (�0.78)

(�3 � �4) 0.012 (3.45)*** 0.008 (2.48)** 0.008 (2.03)**Tenure (�5) 0.005 (4.06)*** 0.003 (5.68)*** 0.003 (5.29)***Control Variables

E*FirmAge (�6) /�E*FirmAge (�7)(�6 � �7) �0.002 (�1.75) �0.002 (�1.71)

E*Big4 (�8) /�E*Big4 (�9)(�8 � �9) �0.091 (�2.25)** �0.117 (�2.11)**

E*Growth (�10) /�E*Growth (�11)(�10 � �11) 0.023 (1.35) 0.035 (1.92)*

E*Persistence (�12) /�E*Persistence (�13)(�12 � �13) 0.084 (2.42)** 0.079 (2.19)**

E*Volatility (�14) /�E*Volatility (�15)(�14 � �15) �0.083 (�7.40)*** �0.092 (�7.52)***

E*Beta (�16) /�E*Beta (�17)(�16 � �17) �0.061 (�1.87)* �0.070 (�1.99)*

E*Size (�18) /�E*Size (�19)(�18 � �19) 0.119 (9.27)*** 0.132 (8.79)***

E*Leverage (�20) /�E*Leverage (�21)(�20 � �21) �0.290 (�4.74)*** �0.296 (�4.21)***

E*Regulation (�22) /�E*Regulation (�23)(�22 � �23) �0.061 (�1.38) �0.051 (�1.27)

FirmAge (�24) 0.001 (4.58)*** 0.001 (4.47)***Big4 (�25) 0.047 (4.00)*** 0.041 (3.43)***Growth (�26) 0.023 (5.41)*** 0.024 (5.60)***Persistence (�27) 0.014 (1.93)* 0.015 (2.28)**Volatility (�28) �0.025 (�7.75)*** �0.025 (�8.53)***Beta (�29) �0.005 (�0.18) �0.003 (�0.10)Size (�30) �0.005 (�1.23) �0.005 (�1.39)Leverage (�31) �0.018 (�0.71) �0.014 (�0.53)Regulation (�32) 0.045 (4.01)*** 0.044 (3.72)***

Yearly Observations 3,044–4,122 3,044–4,122 2,897–3,803Adjusted R2 0.039–0.162 0.111–0.372 0.109–0.369

(continued on next page)

three regressions are positive and significant at the 1 percent level. Considering that earningslevels (earnings changes) are more important for pricing decisions when earnings containtransitory (permanent) components (Easton and Harris 1991; Ali and Zarowin 1992), theresults suggest that investors view reported earnings as containing both transitory and per-manent components. Individual coefficients on E*Tenure (�3) and �E*Tenure (�4) also

598 Ghosh and Moon


TABLE 3 (continued)

*, **, and *** denote statistical significance at the 0.10, 0.05, and 0.01 level, respectively, for a two-tailed test.The full sample includes Compustat firms with available data from 1990 to 2000. The restricted sample consistsof firms in the full sample with auditor-client relationships lasting for at least five years. The dependent variableis cumulative market-adjusted returns for the 12-month period ending three months after the fiscal year-end(CAR). Market-adjusted returns are the difference between raw returns and value-weighted CRSP market returns.E is income before extraordinary items deflated by market value of equity at the beginning of the year. �E is thedifference between income before extraordinary items for the current year and that of last year deflated bymarket value of equity at the beginning of the year. Tenure is the duration of the auditor-client relationship inyears starting from 1982. We suppress the individual coefficients on E*control variable and �E*control variable;instead, we report the sum of the two coefficients. The control variables are defined as follows: FirmAgemeasures the number of years that the firm is publicly traded as of the fiscal year-end; Big4 is an indicatorvariable that equals 1 when the client’s auditor is a large accounting firm; Growth is the sum of the marketvalue of equity and the book value of debt scaled by the book value of total assets; Persistence (Volatility) is thefirst-order autocorrelation (standard deviation) of income before extraordinary items per share for the past 16quarters; Beta is the systematic risk computed using the past 60 monthly stock returns; Size is the logarithmictransformation of the fiscal year-end market value of equity of the prior year; Leverage is the ratio of total debtto total assets; and Regulation is an indicator variable that equals 1 for firms in a regulated industry with two-digit SIC codes between 40 and 49 or between 60 and 63, otherwise it equals 0.The reported coefficients are the average of yearly coefficients from 1990 to 2000, and the corresponding t-statistics in parentheses are based on the distribution of the yearly coefficients.

provide some insights: (1) �3 is larger than �4 in all three regressions; and (2) �3 is statis-tically significant in all three regressions at the 10 percent level or below, while �4 issignificant in only one regression. The incremental importance of earnings levels in pricingdecisions suggests that, as the auditor-client relationship lengthens, investors perceive re-ported earnings as being more transitory.

We briefly discuss the results of the control variables because they have been exten-sively analyzed in prior studies (Subramanyam and Wild 1996; Warfield et al. 1995;Dhaliwal and Reynolds 1994; Collins and Kothari 1989). To conserve space, we do notreport the individual coefficients on the interactions between each control variable and E /�E. Instead, we report the sum of the two interaction coefficients. Controlling for otherdeterminants of ERC, our results based on the restricted sample suggest that the ERC doesnot vary with FirmAge (�6 � �7 is insignificant). Since �8 � �9 is negative and significant,our results indicate that the ERC is lower for client firms of large auditing firms (Big4),which is inconsistent with the findings of Teoh and Wong (1993). As in prior studies, wefind that the ERC is positively associated with Growth, Persistence, and Size, and is neg-atively associated with Volatility, Beta, Leverage, and Regulation. Also, the coefficients onFirmAge, Big4, Growth, Persistence, Volatility, and Regulation are all significant at the 5percent level or below.

Bias in the Estimated ERC from Prices Leading EarningsA potential concern in Table 3 is that the estimated ERC from a contemporaneous

returns-earnings regression might be biased downward if the market anticipates current yearearnings more than one year prior to the earnings release (Kothari 1992; Kothari and Sloan1992). The estimated ERC could be biased from using a 12-month contemporaneous period



to compute returns or from using market value of equity as a deflator.12 Further, the down-ward bias in the estimated ERC is expected to be larger for firms with longer tenure. Giventhat prior studies generally find that earnings quality improves with extended tenure, andconsidering that high-quality earnings are easier to predict, the market is more likely toanticipate current-year earnings for firms with longer tenure.

We mitigate potential biases in the estimated ERC in two ways. First, as in DeFondand Hung (2003) and Collins and Kothari (1989), we measure returns over a 15-monthperiod ending three months after the fiscal year-end. Increasing the returns measurementwindow enhances the probability of including the period in which the market anticipatescurrent earnings thereby reducing any measurement error. Second, as in Subramanyam(1996), we deflate earnings by total assets at the beginning of the year. Since total assetsare typically based on historical cost rather than market value, any bias in the estimatedcoefficients from prices leading earnings is likely to be reduced from using total assets asa deflator.

Panel A of Table 4 reports the results from using cumulative market-adjusted returnsmeasured over 15 months (CAR15-month) as the dependent variable. While the magnitude ofERC (�1 � �2) is smaller, we find that the size of the sum of the coefficients on E*Tenureand �E*Tenure (�3 � �4) is 25 to 38 percent larger than that reported in Table 3. Onereason why the ERC is smaller may be because of the large number of control variables.When we estimate the regressions in Panel A without the control variables, we find thatthe estimated ERC becomes larger. Specifically, unreported results show that the ERC forthe full sample without the control variables is 0.568, while the corresponding estimate is0.507 in Table 3.

We find similar results using total assets as a deflator for E and �E. The estimate of�3 � �4 using the full (restricted) sample is 0.027 (0.025) in Panel B of Table 4, while thecorresponding estimate in Table 3 is 0.008 (0.008). Thus, the magnitude of �3 � �4 is 213to 238 percent larger in Panel B compared to that reported in Table 3. The magnitude ofERC is also larger in Panel B than that in Table 3. Hence, the results in Table 4 suggestthat investors are more likely to anticipate current-year earnings more than one year aheadof the earnings release for firms with extended auditor-client relationships.

Overall, the results from Tables 3 and 4 are consistent with the hypothesis that investorsperceive earnings quality as improving with longer auditor tenure. Our results suggest thatinvestors view audit quality as improving with auditor tenure.

Perceptions of Information Intermediaries and Auditor TenureS&P Stock Rankings and Debt Ratings

Unlike stock returns, stock rankings and debt ratings are possibly correlated over time.Therefore, the residuals from yearly regressions may not be independent, which could leadto biased t-statistics. We compute an autocorrelation-corrected t-statistic (t-statisticAC) bycomparing the mean coefficients with the time-series-based standard error that accounts for

12 If prices lead earnings, current period earnings contain information that is ‘‘new’’ to the market and informationthat is ‘‘stale’’ or ‘‘old.’’ Given that stock prices change in response to new information, the independent variablecontains errors of measurement from using current period earnings as the regressor (also known as errors-in-variable problem). Measurement error on the independent variable biases the slope coefficient downward becauseof its correlation with the regression disturbance term (Greene 1993). Similarly, deflating earnings by marketvalue of equity might inject a downward bias in the estimated coefficients when the market anticipates currentperiod earnings at the beginning of the year.

600 Ghosh and Moon


TABLE 4Earnings Response Coefficients and Perceptions of Investors: Sensitivity Analysis

Variables (Coefficients) Full Sample Restricted Sample

Panel A: CAR15-month is the Dependent Variable

E (�1) 0.217 (2.62)** 0.165 (1.95)*�E (�2) 0.169 (3.80)*** 0.248 (4.07)***

(�1 � �2) 0.386 (5.50)*** 0.413 (5.21)***E*Tenure (�3) 0.008 (1.65) 0.011 (1.92)*�E*Tenure (�4) 0.003 (1.30) �0.001 (�0.91)

(�3 � �4) 0.011 (2.84)*** 0.010 (2.24)**Tenure (�5) 0.004 (5.99)*** 0.003 (5.37)***Control Variables (as in Table 3) Included IncludedYearly Observations 3,044–4,122 2,897–3,803Adjusted R2 0.145–0.322 0.154–0.327

Panel B: Earnings are Scaled by Total Assets

E (�1) 0.415 (5.06)*** 0.393 (5.09)***�E (�2) 0.379 (3.99)*** 0.468 (3.69)***

(�1 � �2) 0.794 (8.33)*** 0.861 (8.57)***E*Tenure (�3) 0.006 (0.84) 0.012 (1.16)�E*Tenure (�4) 0.021 (2.46)** 0.013 (1.58)

(�3 � �4) 0.027 (2.41)** 0.025 (2.38)**Tenure (�5) 0.002 (2.68)** 0.001 (0.94)Control Variables (as in Table 3) Included IncludedYearly Observations 3,218–4,207 3,063–3,888Adjusted R2 0.095–0.278 0.091–0.288

*, **, and *** denote statistical significance at the 0.10, 0.05, and 0.01 level, respectively, for a two-tailed test.The full sample includes Compustat firms with available data from 1990 to 2000. The restricted sample consistsof firms in the full sample with auditor-client relationships lasting for at least five years. In Panel A, thedependent variable is cumulative market-adjusted returns for the 15-month period ending three months after thefiscal year-end (CAR15-month). Market-adjusted returns are the difference between raw returns and value-weightedCRSP market returns. E is income before extraordinary items deflated by market value of equity at thebeginning of the year. �E is the difference between income before extraordinary items for the current year andthat of last year deflated by market value of equity at the beginning of the year. In Panel B, the dependentvariable is CAR measured over the 12-month period ending three months after the fiscal year-end. E is incomebefore extraordinary items deflated by total assets at the beginning of the year. �E is the difference betweenincome before extraordinary items for the current year and that of last year deflated by total assets at thebeginning of the year. The results of all the control variables in Panels A and B are not reported for brevity.The reported coefficients are the average of yearly coefficients from 1990 to 2000, and the correspondingt-statistics in parentheses are based on the distribution of the yearly coefficients.

any dependence in the estimated coefficients, as in Abarbanell and Bernard (2000).13 Since

13 We assume that serial correlation is second-order autoregressive, so that the correction involves multiplying the

yearly standard errors by the factor where n is the number of observations usedn(1 � r) 2*r*(1 � r )

�� 2� (1 � r) n*(1 � r)to compute the mean coefficient and r is the average of the first- and second-order autocorrelations estimatedfrom the series of annual coefficients. The correction factor is not applied when estimated autocorrelation is lessthan zero.



TABLE 5Stock Rankings and Perceptions of Information Intermediaries



Intercept (�) 5.746 (100.31)*** 6.610 (140.79)*** 6.614 (149.22)***E (�1) �1.969 (�5.25)*** �0.208 (�0.91) �0.366 (�1.01)�E (�2) 1.314 (4.77)*** �0.148 (�0.58) �0.030 (�0.08)

(�1 � �2) �0.655 (�5.25)*** �0.356 (�2.03)** �0.396 (�1.84)*E*Tenure (�3) �0.100 (�3.80)*** �0.061 (�5.16)*** �0.063 (�5.87)***�E*Tenure (�4) 0.046 (2.39)** 0.021 (1.93)* 0.023 (1.96)*

(�3 � �4) �0.054 (�3.76)*** �0.040 (�3.31)*** �0.040 (�2.97)***Tenure (�5) �0.068 (�6.05)*** �0.023 (�9.33)*** �0.024 (�9.79)***Control Variables

E*FirmAge (�6) /�E*FirmAge (�7)(�6 � �7) 0.017 (2.45)** 0.021 (2.72)**

E*Big4 (�8) /�E*Big4 (�9)(�8 � �9) 0.264 (2.54)** 0.262 (2.61)**

E*Growth (�10) /�E*Growth (�11)(�10 � �11) �0.134 (�1.66) �0.158 (�1.77)

E*Persistence (�12) /�E*Persistence (�13)(�12 � �13) 0.536 (3.40)*** 0.625 (4.23)***

E*Volatility (�14) /�E*Volatility (�15)(�14 � �15) 0.189 (2.60)** 0.193 (2.64)**

E*Beta (�16) /�E*Beta (�17)(�16 � �17) 0.265 (4.21)*** 0.320 (3.79)***

E*Size (�18) /�E*Size (�19)(�18 � �19) �0.208 (�4.40)*** �0.226 (�4.09)***

E*Leverage (�20) /�E*Leverage (�21)(�20 � �21) 0.489 (2.62)** 0.438 (1.83)*

E*Regulation (�22) /�E*Regulation (�23)(�22 � �23) 0.038 (0.28) 0.040 (0.24)

FirmAge (�24) 0.001 (0.88) 0.000 (0.46)Big4 (�25) 0.250 (11.51)*** 0.256 (14.51)***Growth (�26) 0.033 (2.66)** 0.030 (3.19)***Persistence (�27) �0.032 (�0.79) �0.023 (�0.53)Volatility (�28) 0.309 (15.93)*** 0.329 (15.88)***Beta (�29) 0.372 (8.71)*** 0.373 (9.44)***Size (�30) �0.375 (�26.10)*** �0.371 (�24.15)***Leverage (�31) 0.407 (5.77)*** 0.396 (5.63)***Regulation (�32) �0.207 (�6.00)*** �0.205 (�6.10)***



our focus is on auditor tenure, we compute t-statisticAC for coefficients that involve Tenureonly.

Table 5 reports the regression results of common stock rankings on auditor tenure. Thefirst column in Table 5 presents the results from the reduced model using the full sample.

602 Ghosh and Moon


TABLE 5 (continued)

*, **, and *** denote statistical significance at the 0.10, 0.05, and 0.01 level, respectively, for a two-tailed test.The full sample includes Compustat firms with available data from 1990 to 2000. The restricted sample consistsof firms in the full sample with auditor-client relationships lasting for at least five years. The dependent variableis S&P common stock rankings converted into numerical values (Stock Rankings). Stock Rankings take thevalues of 1 to 7 representing S&P common stock rankings of A�, A, A�, B�, B, B�, and C, respectively.Tenure is the duration of the auditor-client relationship in years starting from 1982. E is income beforeextraordinary items deflated by market value of equity at the beginning of the year. �E is defined as thedifference between income before extraordinary items for the current year and that of last year deflated bymarket value of equity at the beginning of the year. We suppress the individual coefficients on E*controlvariable and �E*control variable; instead, we report the sum of the two coefficients. The control variables aredefined as follows: FirmAge measures the number of years that the firm is publicly traded as of the fiscal year-end; Big4 is an indicator variable that equals 1 when the client’s auditor is a large accounting firm; Growth isthe sum of the market value of equity and the book value of debt scaled by the book value of total assets;Persistence (Volatility) is the first-order autocorrelation (standard deviation) of income before extraordinary itemsper share for the past 16 quarters; Beta is the systematic risk computed using the past 60 monthly stock returns;Size is the logarithmic transformation of the fiscal year-end market value of equity of the prior year; Leverage isthe ratio of total debt to total assets; and Regulation is an indicator variable that equals 1 for firms in a regulatedindustry with two-digit SIC codes between 40 and 49 or between 60 and 63, and 0 otherwise.The reported coefficients are the average of yearly coefficients from 1990 to 2000, and the correspondingt-statistics in parentheses are based on the distribution of the yearly coefficients.

As in Table 3, ERC, a proxy for earnings quality as perceived by ratings agencies, is definedas the sum of the coefficients on E and �E.14 The ERC is negative (�0.655) and statisticallysignificant at the 1 percent level (t-statistic � �5.25). �3 � �4, which measures the jointeffects of tenure and earnings on rankings, is also negative (�0.054) and significant at the1 percent level (t-statistic � �3.76; t-statisticAC � �3.14). All else equal, for a givenincrease in earnings, Stock Rankings improve by an additional 8.2 percent (� �0.054/�0.655) when the audit engagement increases by an added year.

The coefficient on Tenure (�5) is negative (�0.068) and significant at the 1 percentlevel (t-statistic � �6.05; t-statisticAC � �3.52). Other things remaining constant, if auditortenure for a firm is one standard deviation above the mean value of Tenure, the stock rankingis lower by 0.327 (� �0.068 � 4.807), compared to the average stock ranking of 5.096for the sample used in the regression.

The second column in Table 5 reports the results from the augmented model using thefull sample. �3 � �4 remains negative (�0.040) and significant (t-statistic � �3.31;t-statisticAC � �3.31). �5 is also negative (�0.023) and significant (t-statistic � �9.33;t-statisticAC � �6.59). The results are very similar when we use the restricted sample(reported in the last column). �3 � �4 is negative (�0.040) and significant (t-statistic ��2.97; t-statisticAC � �2.97), and �5 is highly significant (�0.024, t-statistic � �9.79;t-statisticAC � �6.78). Our results suggest that rating agencies place a greater weight onreported earnings as auditor tenure increases. Given that S&P incorporates the quality ofaccounting information as a factor in establishing its rankings/ratings (Standard & Poor’s1982), the importance of tenure suggests that, as tenure increases, independent rating agen-cies perceive audited financial statements as being more reliable.

We briefly discuss how various control variables affect Stock Rankings (based on thefull sample). The coefficients on the interactions between earnings and FirmAge, Persist-ence, Volatility, Beta, and Leverage are positive and significant at the 5 percent level or

14 Although the term ERC or earnings response coefficient is typically used for earnings coefficients in returns-earnings regressions, we use the term in a broader context to refer to earnings coefficients for the returns, stockrankings, debt ratings, and analyst forecasts regressions.



below. These results suggest that the effect of earnings on Stock Rankings is smaller formature/risky firms. The positive sign of the coefficient on the interactions between earningsand Big4 (�8 � �9) is puzzling. The coefficient on the interactions between earnings andSize (�18 � �19) is negative and significant at the 1 percent level, indicating that the effectof earnings on Stock Rankings is larger for bigger firms. The coefficients on the interactionsbetween earnings and Growth (�10 � �11) and those between earnings and Regulation (�22

� �23) are both insignificant.The control variables are also included as separate independent variables. The positive

sign of the coefficient on Big4 is again puzzling. The coefficients on Volatility, Beta, andLeverage are all positive and significant, which suggests that higher risk has an unfavorableimpact on stock rankings. The coefficient on Growth is also positive, which is inconsistentwith our expectations. If our model does not adequately control for risk, growth firms aremore likely to be associated with unfavorable Stock Rankings because higher growth couldproxy for larger risk. The coefficients on Size and Regulation are negative, which suggeststhat stock rankings are more favorable for larger firms and those in regulated industry. Thecoefficients on FirmAge and Persistence are insignificant.

Table 6 presents the regression results of the association between debt ratings andauditor tenure. The first column in Table 6 presents the results of estimating the reducedmodel using the full sample. Consistent with prior research, the ERC is negative and sig-nificant (�4.568, t-statistic � �5.87). Thus, earnings have a favorable influence on debtratings. The sum of the coefficients on E*Tenure and �E*Tenure (�3 � �4) is also negative,but insignificant (�0.093, t-statistic � �1.03; t-statisticAC � �0.98). In contrast, the coef-ficient on Tenure (�5) is negative and significant (�0.183, t-statistic � �6.33; t-statisticAC

� �3.42), consistent with the findings of Mansi et al. (2004).Once we include the control variables in Table 6, the ERC is insignificant in either the

second or the third regression. Although �3 � �4 is marginally significant (�0.216, t-statistic� �1.98; t-statisticAC � �1.85) for the full sample, it is insignificant (�0.240, t-statistic� �1.66; t-statisticAC � �1.62) for the restricted sample. On the other hand, �5 remainsnegative and significant for the full sample (�0.065, t-statistic � �4.78; t-statisticAC

� �2.70) and the restricted sample (�0.067, t-statistic � �4.51; t-statisticAC � �2.70).Thus, although longer tenure is associated with improved debt ratings, the association be-tween earnings and debt ratings does not appear to vary with tenure.15

In a recent study, Mansi et al. (2004) find that auditor tenure is negatively related tothe cost of debt and that the relation is more pronounced for non-investment-grade debt.This finding suggests that auditor tenure could influence debt ratings differently for invest-ment and non-investment-grade debt. Following Mansi et al. (2004), we separately estimatethe augmented model for investment and non-investment-grade debt to investigate whetherthe impact of auditor tenure on debt ratings depends on the quality of debt. When wepartition our sample into investment and non-investment-grade debt subsamples, the resultsare very similar to those reported in Table 6. For example, �3 � �4 is negative but notsignificant in either subsample, but �5 is negative and significant for both subsamples (forbrevity, the results are not tabulated).

In sum, the regression results from stock rankings and debt ratings provide modestevidence that independent rating agencies perceive firms with longer auditor tenure asproviding more reliable financial information.

15 Since the same control variables appear in the stock rankings and debt ratings regressions, and the results aresimilar across the two sets of regressions, we do not repeat our discussion of the control variables for the debtratings regressions.

604 Ghosh and Moon


TABLE 6Debt Ratings and Perceptions of Information Intermediaries



Intercept (�) 11.101 (64.35)*** 14.387 (42.74)*** 14.032 (31.54)***E (�1) �8.796 (�3.37)*** �1.378 (�0.21) 2.596 (0.40)�E (�2) 4.228 (2.13)** 14.567 (1.13) 11.871 (0.88)

(�1 � �2) �4.568 (�5.87)*** 13.189 (1.02) 14.467 (1.13)E*Tenure (�3) �0.169 (�0.84) 0.014 (0.18) 0.020 (0.22)�E*Tenure (�4) 0.076 (0.54) �0.230 (�2.06)** �0.260 (�1.89)*

(�3 � �4) �0.093 (�1.03) �0.216 (�1.98)* �0.240 (�1.66)Tenure (�5) �0.183 (�6.33)*** �0.065 (�4.78)*** �0.067 (�4.51)***Control Variables

E*FirmAge (�6) /�E*FirmAge (�7)(�6 � �7) 0.040 (3.11)*** 0.040 (2.63)**

E*Big4 (�8) /�E*Big4 (�9)(�8 � �9) �13.630 (�1.05) �15.302 (�1.17)

E*Growth (�10) /�E*Growth (�11)(�10 � �11) 1.333 (1.61) 1.641 (1.54)

E*Persistence (�12) /�E*Persistence (�13)(�12 � �13) 1.284 (3.17)*** 1.280 (2.97)***

E*Volatility (�14) /�E*Volatility (�15)(�14 � �15) 0.068 (0.57) 0.114 (0.67)

E*Beta (�16) /�E*Beta (�17)(�16 � �17) 1.011 (2.11)** 1.083 (2.26)**

E*Size (�18) /�E*Size (�19)(�18 � �19) �0.594 (�2.67)** �0.552 (�2.25)**

E*Leverage (�20) /�E*Leverage (�21)(�20 � �21) 0.449 (0.29) 0.512 (0.32)

E*Regulation (�22) /�E*Regulation (�23)(�22 � �23) �1.127 (�1.84)* �1.271 (�1.99)*

FirmAge (�24) �0.014 (�7.46)*** �0.014 (�7.13)***Big4 (�25) 1.473 (3.13)*** 1.914 (3.80)***Growth (�26) �0.102 (�1.94)* �0.108 (�1.98)*Persistence (�27) 0.306 (4.78)*** 0.305 (4.60)***Volatility (�28) 0.254 (7.55)*** 0.223 (5.36)***Beta (�29) 1.359 (12.12)*** 1.333 (11.66)***Size (�30) �1.087 (�31.65)*** �1.088 (�29.81)***Leverage (�31) 3.676 (13.00)*** 3.663 (12.30)***Regulation (�32) �1.025 (�10.18)*** �1.017 (�10.20)***

Yearly Observations 603–1,124 603–1,124 598–1,051Adjusted R2 0.132–0.245 0.602–0.699 0.596–0.697


Financial Analysts’ Earnings ForecastsTable 7 reports the results of how auditor tenure affects the association between one-

year-ahead consensus earnings per share forecasts and earnings per share reported just priorto the forecasts. As in the previous tables, our interest is in the sum of the coefficients onEPSt�1* Tenure and �EPSt�1*Tenure (�3 � �4). The first column in Table 7 reports the



TABLE 6 (continued)

*, **, and *** denote statistical significance at the 0.10, 0.05, and 0.01 level, respectively, for a two-tailed test.The full sample includes Compustat firms with available data from 1990 to 2000. The restricted sample consistsof firms in the full sample with auditor-client relationships lasting for at least five years. The dependent variableis S&P senior debt ratings converted into numerical values (Debt Ratings). Debt Ratings are assigned a value of1 if a firm’s S&P senior debt is rated as AAA. As the S&P debt ratings decline, the numerical value increasesby 1. Tenure is the duration of the auditor-client relationship in years starting from 1982. E is income beforeextraordinary items deflated by market value of equity at the beginning of the year. �E is defined as thedifference between income before extraordinary items for the current year and that of last year deflated bymarket value of equity at the beginning of the year. We suppress the individual coefficients on E*controlvariable and �E*control variable; instead, we report the sum of the two coefficients. The control variables aredefined as follows: FirmAge measures the number of years that the firm is publicly traded as of the fiscal year-end; Big4 is an indicator variable that equals 1 when the client’s auditor is a large accounting firm; Growth isthe sum of the market value of equity and the book value of debt scaled by the book value of total assets;Persistence (Volatility) is the first-order autocorrelation (standard deviation) of income before extraordinary itemsper share for the past 16 quarters; Beta is the systematic risk computed using the past 60 monthly stock returns;Size is the logarithmic transformation of the fiscal year-end market value of equity of the prior year; Leverage isthe ratio of total debt to total assets; and Regulation is an indicator variable that equals 1 for firms in a regulatedindustry with two-digit SIC codes between 40 and 49 or between 60 and 63, and 0 otherwise.The reported coefficients are the average of yearly coefficients from 1990 to 2000, and the correspondingt-statistics in parentheses are based on the distribution of the yearly coefficients.

results of the reduced model for the full sample. We find that reported earnings are posi-tively associated with analysts’ earnings forecasts. The ERC (�1 � �2) is positive and highlysignificant (1.008, t-statistic � 20.29). Moreover, �3 � �4 is also positive and significant(0.019, t-statistic � 2.68). Other things remaining constant, our results suggest that analystsappear to attach greater importance to the most recent reported earnings in forming theirexpectations about future earnings as auditor tenure grows longer.

When we estimate the augmented model that includes a number of control variables,�3 � �4 remains positive and significant (0.016, t-statistic � 2.90) in the second regression.The results are very similar when we estimate the augmented model using the restrictedsample. �3 � �4 continues to be positive and significant in the third regression (0.016,t-statistic � 2.51). On the other hand, �5 is insignificant in all three regressions.

The individual parameter estimates of ERC and �3 � �4 provide some added insights.The coefficients on EPSt�1 (�1) and �EPSt�1 (�2) are both significant in the first regression.However, in the second and third regressions that include the other determinants of earningsforecasts, �2 becomes insignificant. Large earnings surprises are more likely to be temporaryand, therefore, potentially less useful to analysts in predicting future earnings (Barron etal. 2002; Stickel 1989). In contrast, the coefficient on �EPSt�1*Tenure (�4) is significant inall three regressions at the 10 percent level or below. The results suggest that although pastearnings changes do not influence earnings forecasts, they are perceived as more relevantfor earnings forecasts as the auditor-client relationship increases.

Based on the results from the restricted sample, we briefly discuss how the variouscontrol variables affect analysts’ earnings forecasts (FEPSt). The coefficients on the inter-actions between reported earnings and FirmAge, Leverage, and Analysts are all negativeand significant at the 10 percent level or below. These results suggest that the effect ofreported earnings on FEPSt is smaller for mature/risky firms and for firms with largeranalyst following. The interaction between reported earnings and Size is positive and sig-nificant, which suggests that the impact of reported earnings on analysts’ earnings forecastsis greater for bigger firms. Also, the individual coefficients on FirmAge, Volatility, Leverage,and Analysts are all positive and significant at the 5 percent level or below, which suggests

606 Ghosh and Moon


TABLE 7Analyst Earnings Forecasts and Perceptions of Information Intermediaries



Intercept (�) 0.217 (8.84)*** 0.366 (15.47)*** 0.364 (14.87)***EPSt�1 (�1) 0.834 (22.34)*** 0.532 (15.28)*** 0.530 (15.64)***�EPSt�1 (�2) 0.174 (4.59)*** �0.068 (�0.66) �0.074 (�0.70)

(�1 � �2) 1.008 (20.29)*** 0.464 (4.62)*** 0.456 (4.49)***EPSt�1*Tenure (�3) 0.005 (1.75) 0.005 (1.79) 0.005 (1.54)�EPSt�1*Tenure (�4) 0.014 (2.00)** 0.011 (2.11)** 0.011 (1.95)*

(�3 � �4) 0.019 (2.68)** 0.016 (2.90)*** 0.016 (2.51)**Tenure (�5) �0.000 (�0.15) �0.003 (�1.16) �0.002 (�0.76)Control Variables

EPSt�1*FirmAge (�6) /�EPSt�1*FirmAge (�7)(�6 � �7) �0.004 (�2.83)*** �0.004 (�2.90)***

EPSt�1*Big4 (�8) /�EPSt�1*Big4 (�9)(�8 � �9) �0.060 (�1.29) �0.077 (�1.49)

EPSt�1*Growth (�10) /�EPSt�1*Growth (�11)(�10 � �11) �0.008 (�0.33) �0.010 (�0.41)

EPSt�1*Volatility (�12) /�EPSt�1*Volatility (�13)(�12 � �13) �0.032 (�0.67) �0.033 (�0.67)

EPSt�1*Beta (�14) /�EPS t�1*Beta (�15)(�14 � �15) �0.011 (�0.37) �0.012 (�0.38)

EPSt�1*Size (�16) /�EPSt�1*Size (�17)(�16 � �17) 0.140 (7.17)*** 0.147 (8.00)***

EPSt�1*Leverage (�18) /�EPSt�1*Leverage (�19)(�18 � �19) �0.235 (�2.18)** �0.225 (�1.99)*

EPSt�1*Regulation (�20) /�EPSt�1*Regulation (�21)(�20 � �21) �0.041 (�1.26) �0.034 (�1.04)

EPSt�1*Analysts (�22) /�EPSt�1*Analysts (�23)(�22 � �23) �0.013 (�3.85)*** �0.014 (�4.20)***

FirmAge (�24) 0.003 (4.92)*** 0.004 (4.95)***Big4 (�25) 0.028 (1.39) 0.034 (1.54)Growth (�26) �0.003 (�0.91) �0.003 (�1.00)Volatility (�27) 0.138 (3.04)*** 0.138 (2.86)***Beta (�28) �0.004 (�0.28) �0.002 (�0.18)Size (�29) �0.048 (�12.29)*** �0.050 (�12.99)***Leverage (�30) 0.174 (5.06)*** 0.168 (4.41)***Regulation (�31) 0.020 (0.89) 0.013 (0.55)Analysts (�32) 0.002 (1.82)* 0.002 (2.17)**



that analysts’ earnings forecasts are higher for mature firms, for firms with higher risk, andthose with larger analyst following. The coefficient on Size is negative, which suggests thatearnings per share forecasts are smaller for bigger firms. One explanation is that large,politically sensitive firms are more likely to report lower earnings to reduce their politicalcosts and, therefore, analysts’ earnings forecasts are also lower for larger firms.



TABLE 7 (continued)

*, **, and *** denote statistical significance at the 0.10, 0.05, and 0.01 level, respectively, for a two-tailed test.The full sample includes Compustat firms with available data from 1990 to 2000. The restricted sample consistsof firms in the full sample with auditor-client relationships lasting for at least five years. The dependent variableis the mean annual one-year-ahead earnings per share forecasts for year t issued immediately following theearnings announcement for year t�1 (FEPSt). EPSt�1 is annual earnings per share for year t�1, and �EPS t�1 isthe absolute change in earnings per share for year t�1 defined as the difference in annual earnings per share inyear t�1 and that in year t�2 (�EPSt�1 � EPSt�2�). All earnings per share variables (FEPSt, EPSt�1, and�EPSt�1) are obtained from I /B/E /S database. Tenure is the duration of the auditor-client relationship in yearsstarting from 1982. All the control variables including Tenure are measured as of the end of year t�1. Wesuppress the individual coefficients on E*control variable and �E*control variable; instead, we report the sum ofthe two coefficients. The control variables are defined as follows: FirmAge measures the number of years thatthe firm is publicly traded as of the fiscal year-end; Big4 is an indicator variable that equals 1 when the client’sauditor is a large accounting firm; Growth is the sum of the market value of equity and the book value of debtscaled by the book value of total assets; Volatility is the standard deviation of income before extraordinary itemsper share for the past 16 quarters; Beta is the systematic risk computed using the past 60 monthly stock returns;Size is the logarithmic transformation of the fiscal year-end market value of equity of the prior year; Leverage isthe ratio of total debt to total assets; Regulation is an indicator variable that equals 1 for firms in a regulatedindustry with two-digit SIC codes between 40 and 49 or between 60 and 63, and otherwise 0; and Analysts isthe number of analysts providing annual earnings forecasts.The reported coefficients are the average of yearly coefficients from 1990 to 2000, and the correspondingt-statistics in parentheses are based on the distribution of the yearly coefficients.

A key advantage of using consensus forecasts is that the aggregation process offsetsindividual forecast errors, or idiosyncratic errors, which results in the consensus forecastsbeing more accurate than the individual forecasts (Brown et al. 1985). However, one con-cern is that the idiosyncratic errors may not be fully diversified away when the consensusis based on a few individual forecasts (Barron et al. 1998). Although any measurementerror in the dependent variable does not bias the estimated coefficients, the estimated var-iances are larger and therefore the t-statistics are also lower (Gujarati 1995). One solutionto the potential measurement error problem is to impose a restriction on the number ofindividual forecasts required to compute the mean forecast. The diversification argumentsuggests that the measurement error declines with an increase in the number of individualforecasts.

Consistent with the diversification argument, we find that the magnitude of the t-statistics becomes larger when we impose restrictions on the number of individual forecasts.For instance, �3 � �4 is 0.01 and the associated t-statistic is 2.90 in the second regression(Table 7). However, as in Barron et al. (2002), when we require that the consensus forecastsbe based on a minimum of three individual forecasts, �3 � �4 is 0.01 but the t-statisticincreases to 3.40. The t-statistic jumps to 4.67 when the consensus forecast is based on aminimum of seven individual forecasts.

Thus, our results from analyst earnings forecasts are consistent with the hypothesis thatfinancial analysts perceive earnings quality as improving with longer auditor tenure.

Sensitivity AnalysisWe conduct a number of additional analyses to test the robustness of our results. Instead

of tabulating the results of the additional tests, we discuss them briefly in this subsection.First, given that raw returns are frequently used to estimate contemporaneous returns-earnings models (Lundholm and Myers 2002; Warfield et al. 1995; Collins and Kothari1989), we estimate the returns-earnings regression model using 12-month compounded rawreturns (ending three months after the fiscal year-end) as the dependent variable. The results

608 Ghosh and Moon


based on raw returns are very similar to those reported in Table 3 using cumulative market-adjusted returns. The sum of the coefficients on E*Tenure and �E*Tenure for the full andrestricted samples is 0.010 (t-statistic � 2.50) and 0.009 (t-statistic � 2.15), respectively.The coefficient on Tenure is also positive and significant for the full sample (0.004,t-statistic � 5.52) and the restricted sample (0.003, t-statistic � 4.84).

Second, considering that cash flow might be an important determinant of debt ratings,we estimate Debt Ratings regressions after including operating cash flow (deflated by mar-ket value of equity at the beginning of the year) in addition to all the other variables. Theinclusion of the cash flow variable does not affect the tenor of the results reported in Table6. The coefficient on Tenure continues to be negative and significant for the full sample(�0.066, t-statistic � �4.91) and the restricted sample (�0.067, t-statistic � �4.54). Thecoefficient on cash flow is not significant.

Third, we also analyze S&P subordinated debt ratings (Compustat #320). The samplesize is relatively small: the number of observations used in the yearly regressions variesbetween 251 and 372 for the full sample and between 245 and 358 for the restricted sample.The results from S&P subordinated debt ratings are very similar to those using senior debtratings. The coefficient on Tenure is negative and significant for the full sample (�0.047,t-statistic � �2.96) and for the restricted sample (�0.031, t-statistic � �1.89). The sumof the coefficients on E*Tenure and �E*Tenure is negative but insignificant for bothsamples.

Fourth, we estimate the effects of tenure on rankings/ratings using an ordered probitmodel. Since Stock Rankings and Debt Ratings (the dependent variables) are discrete out-comes, conventional regression methods could yield biased estimates (Greene 1993). Wefind that the results in Tables 5 and 6 are not affected when we use an ordered probitmodel. For example, the coefficient on Tenure is negative and highly significant for theStock Rankings regression (�0.036, �2 � 732.74) and the Debt ratings regression (�0.039,�2 � 473.14) for the full sample.

Fifth, the results reported in Table 7 are based on levels specification—earnings pershare forecasts (FEPS) and the two reported earnings per share measures (EPS and �EPS)are not deflated. One concern is that the ordinary least squares estimates might be inefficientand/or biased because of scale differences (Barth and Kallapur 1996; Dechow 1994). Ourconclusions remain unchanged when we estimate the analyst forecast model after (1) de-flating all the earnings per share variables (FEPS, EPS, and �EPS) by the stock price pershare at the beginning of the forecast period, and (2) deleting the top and bottom 1 percentof the observations for each variable. The sum of the coefficients on EPSt�1*Tenure and�EPSt�1*Tenure remains positive and significant for the full sample (0.015, t-statistic� 3.05) and the restricted sample (0.013, t-statistic � 2.08).

Finally, based on the specification used in prior studies, the analyst forecast model doesnot include earnings persistence. However, to maintain consistency with the returns, stockrankings and debt ratings specifications, we also examine whether the analyst forecastsresults are robust to the inclusion of Persistence. When we include EPSt�1*Persistence,�EPSt�1* Persistence, and Persistence as additional independent variables, our conclusionsremain unaffected. The sum of the coefficients on EPSt�1*Tenure and �EPSt�1*Tenure re-mains positive and significant for the full sample (0.017, t-statistic � 3.20) and the restrictedsample (0.016, t-statistic � 2.90).

VI. CONCLUSIONSThis study provides insights into the recent debate surrounding auditor tenure, inde-

pendence, and audit quality by analyzing the relationship between auditor tenure and audit



quality as perceived by capital market participants. We focus on a market-based approachbecause academics, regulatory institutions, and practitioners frequently emphasize the im-portance of incorporating capital market perceptions of independence and audit quality(Dopuch et al. 2003; Ryan et al. 2001; SEC 2000; Carmichael 1999; AICPA 1994). Basedon the association between perceived reliability of financial statements and auditor tenure,we infer how capital markets view tenure as affecting audit quality. All else equal, if capitalmarket participants perceive auditor tenure as improving (impairing) audit quality, thenreported financial statements are viewed as more (less) reliable for investment and creditdecisions.

We analyze the perceptions of three primary users of audited financial statements:investors, independent rating agencies, and financial analysts. Using earnings response co-efficients from returns-earnings regressions as a proxy for investor perceptions of earningsquality (Schipper and Vincent 2003), we find evidence consistent with the hypothesis thatinvestors perceive earnings quality as improving with auditor tenure. Our analysis of theperceptions of independent rating agencies is based on how tenure affects the associationbetween rankings/ratings and reported earnings. We find that the influence of reportedearnings on stock rankings becomes larger with extended tenure, although the influence ofreported earnings on debt ratings does not vary with tenure. Thus, our results providelimited evidence that independent rating agencies view auditor tenure as having a favorableimpact on earnings quality. Finally, using earnings coefficients from a regression of one-year-ahead consensus earnings forecasts on reported earnings as a proxy for analysts’ per-ceptions of earnings quality, we find evidence consistent with the hypothesis that financialanalysts perceive earnings quality as improving with extended auditor-client relationships.

In general, most of our results are consistent with the hypothesis that audited financialstatements, and in particular reported earnings, are perceived as more reliable for firms withlonger auditor tenure. One implication of our results is that many capital market participantsview longer tenure as having a favorable impact on audit quality. Overall, our study suggeststhat imposing mandatory limits on the duration of the auditor-client relationship mightimpose unintended costs on capital market participants. Our results are also subject to acaveat. Results from a regime without auditor term limits may not be applicable to a regimewith mandatory auditor rotation because the incentives for auditors and clients might differbetween the two regimes.

REFERENCESAbarbanell, J., and V. Bernard. 2000. Is the U.S. market myopic? Journal of Accounting Research

38: 221–242.Ali, A., and P. Zarowin. 1992. The role of earnings levels in annual earnings-returns studies. Journal

of Accounting Research 30: 286–296.American Institute of Certified Public Accountants (AICPA). 1978. The Commission on Auditors

Responsibilities: Report, Conclusions and Recommendations. New York, NY: AICPA.———. 1992. Statement of Position Regarding Mandatory Rotation of Audit Firms of Publicly Held

Companies. SEC Practice Section. New York, NY: AICPA.———. 1994. Professional Standards. New York, NY: Commerce Clearing House, Inc.Ashbaugh, H., and M. Pincus. 2001. Domestic accounting standards, international accounting stan-

dards, and the predictability of earnings. Journal of Accounting Research 39: 417–434.Ashton, A. H. 1991. Experience and error frequency knowledge as potential determinants of audit

experience. The Accounting Review 66: 218–239.Barron, O. E., O. Kim, S. C. Lim, and D. E. Steven. 1998. Using analysts’ forecasts to measure

properties of analysts’ information environment. The Accounting Review 73: 421–433.

610 Ghosh and Moon


———, D. Byard, C. Kile, and E. J. Reidl. 2002. High technology intangibles and analysts’ forecasts.Journal of Accounting Research 40: 289–312.

Barth, M. E., and S. Kallapur. 1996. The effects of cross-sectional scale differences on regressionresults in empirical accounting research. Contemporary Accounting Research 13: 527–567.

———, J. A. Elliott, and M. A. Finn. 1999. Market rewards associated with patterns of increasingearnings. Journal of Accounting Research 37: 387–413.

Bartov, E., D. Givoly, and C. Hayn. 2002. The rewards to meeting or beating earnings expectations.Journal of Accounting and Economics 33: 173–190.

Becker, C. L., M. L. DeFond, J. Jiambalvo, and K. R. Subramanyam. 1998. The effect of audit qualityon earnings management. Contemporary Accounting Research 15: 1–24.

Bhojraj, S., and P. Sengupta. 2003. Effect of corporate governance on bond ratings and yields: Therole of institutional investors and outside directors. Journal of Business 76: 455–475.

Bricker, R. 2002. Transparency, independence, technology and the CPA scope of services: New chal-lenges to the profession. The Ohio CPA Journal (April–June): 48–50.

Brody, R. G., and S. A. Moscove. 1998. Mandatory auditor rotation. National Public Accountant(May): 32–36.

Brown, L., G. Foster, and E. Noreen. 1985. Security Analyst Multi-Year Earnings Forecasts and theCapital Market. Studies in Accounting Research No. 21. Sarasota, FL: American AccountingAssociation.

———, R. Hagerman, P. Griffin, and M. Zmijewski. 1987. Security analyst superiority relative tounivariate time-series models in forecasting quarterly earnings. Journal of Accounting and Ec-onomics 9: 61–87.

Carmichael, D. R. 1999. In search of concepts of auditor independence. The CPA Journal (May): 39–43.

Collins, D. W., and S. P. Kothari. 1989. An analysis of intertemporal and cross-sectional determinantsof earnings response coefficient. Journal of Accounting and Economics 11: 143–181.

Craswell, A. T., J. R. Francis, and S. L. Taylor. 1995. Auditor brand name reputations and industryspecialization. Journal of Accounting and Economics 20: 297–322.

Davis, L. R., B. Soo, and G. Trompeter. 2002. Auditor tenure, auditor independence and earningsmanagement. Working paper, Boston College, Boston, MA.

Dechow, P. M. 1994. Accounting earnings and cash flows as measures of firm performance: The roleof accounting accruals. Journal of Accounting and Economics 18: 3–42.

DeFond, M., and J. Jiambalvo. 1994. Debt covenant violation and manipulation of accruals. Journalof Accounting and Economics 17: 145–176.

———, and K. R. Subramanyam. 1998. Auditor changes and discretionary accruals. Journal of Ac-counting and Economics 25: 35–67.

———, and M. Hung. 2003. An empirical analysis of analysts’ cash flow forecasts. Journal of Ac-counting and Economics 35: 73–100.

Dhaliwal, D. S., and S. S. Reynolds. 1994. The effect of default risk of debt on the earnings responsecoefficient. The Accounting Review 69: 412–419.

Dopuch, N., R. R. King, and R. Schwartz. 2003. Independence in appearance and in fact: An empiricalinvestigation. Contemporary Accounting Research 65: 83–113.

Duru, A., and D. M. Reeb. 2002. International diversification and analysts’ forecast accuracy and bias.The Accounting Review 77: 415–433.

Easton, P. D., and T. S. Harris. 1991. Earnings as an explanatory variable for returns. Journal ofAccounting Research 29: 19–36.

Elliott, R. K., and P. D. Jacobson. 1998. Audit independence. The CPA Journal (December): 30–37.———. 2000. Testimony on Auditor Independence before the Securities and Exchange Commission.

Available at: http: / /www.sec.gov/rules /proposed/s71300/ testimony/elliott1.htm.Fama, E. F., and K. French. 2000. Forecasting profitability and earnings. Journal of Business 73:

161–175.



Farmer, T., L. Rittenberg, and G. Trompeter. 1987. An investigation of the impact of economic andorganizational factors in auditor independence. Auditing: A Journal of Practice & Theory 7: 1–14.

Financial Accounting Standards Board (FASB). 1978. Objectives of Financial Reporting by BusinessEnterprise. Statement of Financial Accounting Concepts No. 1. Norwalk, CT: FASB.

———. 1980. Qualitative Characteristics of Accounting Information. Statement of Financial Ac-counting Concepts No. 2. Stamford, CT: FASB.

Francis, J., and L. Soffer. 1997. The relative informativeness of analysts’ recommendations and earn-ings forecast revisions. Journal of Accounting Research 34: 193–212.

Ghosh, A., Z. Gu, and P. C. Jain. 2005. Sustained earnings and revenue growth, earning quality, andearnings response coefficients. Review of Accounting Studies 10 (forthcoming).

Gramling, A. A., and D. N. Stone. 2001. Audit firm industry expertise: A review and synthesis of thearchival literature. Journal of Accounting Literature 20: 1–29.

Greene, W. H. 1993. Econometric Analysis. Englewood Cliffs, NJ: Prentice Hall.Gujarati, D. N. 1995. Basic Econometrics. New York, NY: McGraw-Hill.Hogan, C. E., and D. C. Jeter. 1999. Industry specialization by auditors. Auditing: A Journal of

Practice & Theory 18: 1–17.Hope, O. 2003. Accounting policy disclosures and analysts’ forecasts. Contemporary Accounting Re-

search 20: 295–322.Hunt, I. C. 2002. Testimony Concerning the Role of Credit Rating Agencies in the U.S. Securities

Markets before the Senate Committee on Governmental Affairs. Available at: http: / /www.sec.gov/news/ testimony/032002tsih.htm.

Johnson, V., I. K. Khurana, and J. K. Reynolds. 2002. Audit-firm tenure and the quality of financialreports. Contemporary Accounting Research 19: 637–660.

Kaplan, R. S., and G. Urwitz. 1979. Statistical models of bond ratings: A methodological inquiry.Journal of Business 52: 231–261.

Kasznik, R., and M. F. McNichols. 2002. Does meeting earnings expectations matter? Evidence fromanalyst forecast revisions and share prices. Journal of Accounting Research 40: 727–759.

Kinney, W. 1999. Auditor independence: A burdensome constraint or core value? Accounting Horizons13: 69–75.

Kothari, S. P. 1992. Price-earnings regressions in the presence of pricing leading earnings: Earningslevel versus changes specification and alternative deflators. Journal of Accounting and Econom-ics 15: 173–302.

———, and R. Sloan. 1992. Information in prices about future earnings: Implications for earningsresponse coefficients. Journal of Accounting and Economics 15: 143–171.

Lang, M., and R. Lundholm. 1996. Corporate disclosure policy and analyst behavior. The AccountingReview 71: 467–492.

Libby, R., and D. M. Frederick. 1990. Experience and the ability to explain audit findings. Journalof Accounting Research 22: 348–367.

Lundholm, R., and L. A. Myers. 2002. Bringing the future forward: The effect of disclosure on thereturns-earnings relation. Journal of Accounting Research 40: 809–839.

Lys, T., and S. Sohn. 1990. The association between revisions of financial analysts’ earnings forecastsand security price changes. Journal of Accounting and Economics 13: 341–364.

Mansi, S. A., W. F. Maxwell, and D. P. Miller. 2004. Does auditor quality and tenure matter toinvestors? Evidence from the bond market. Journal of Accounting Research 42: 755–793.

Mautz, R. K., and H. A. Sharaf. 1961. The Philosophy of Auditing. Monograph No. 6. Sarasota, FL:American Accounting Association.

Myers, J. N., L. A. Myers, and T. C. Omer. 2003. Exploring the term of the auditor-client relationshipand the quality of earnings: A case for mandatory auditor rotation? The Accounting Review 78:779–800.

Myers, L. A., and D. J. Skinner. 2002. Earnings momentum and earnings management. Workingpaper, University of Illinois at Urbana–Champaign and University of Michigan.

612 Ghosh and Moon


O’Brien, P. 1990. Forecast accuracy of individual analysts in nine industries. Journal of AccountingResearch 28: 286–304.

PricewaterhouseCoopers. 2002. Mandatory Rotation of Audit Firms: Will It Improve Audit Quality?New York, NY: PricewaterhouseCoopers LLP.

Ryan, S. G., R. H. Herz, T. E. Iannaconi, L. A. Maines, K. Palepu, C. M. Schrand, D. J. Skinner,and L. Vincent. 2001. Commentary. SEC auditor independence requirements: AAA FinancialAccounting Standards Committee. Accounting Horizons 15: 373–386.

Schipper, K. 1991. Analysts’ forecasts. Accounting Horizons 4: 105–121.———, and L. Vincent. 2003. Earnings quality. Accounting Horizons (Supplement) 17: 97–110.Securities and Exchange Commission (SEC). 1988. Financial Report. Release No. 31. Chicago, IL:

Commerce Clearing House.———. 1994. Staff Report on Auditor Independence. Chicago, IL: Commerce Clearing House.———. 2000. Hearing on Auditor Independence. July. Washington, D.C.: Government Printing

Office.———. 2001. Final Rule: Revision of the Commission’s Auditor Independence Requirements. File

No. S7-13-00. Washington, D.C.: Government Printing Office.Shockley, R. A. 1981. Perceptions of auditor independence: An empirical analysis. The Accounting

Review 56: 785–800.Skinner, D. J., and R. Sloan. 2002. Earnings surprises, growth expectations, and stock returns or don’t

let an earnings torpedo sink your portfolio. Review of Accounting Studies 7: 289–312.Solomon, I., M. D. Shields, and R. O. Whittington. 1999. What do industry specialist auditors know?

Journal of Accounting Research 37: 191–208.Standard & Poor’s. 1982. Credit Overview: Corporate and International Ratings. New York, NY:

Standard & Poor’s Corporation.Stickel, S. 1989. The timing of and incentives for annual earnings forecasts near interim financial

report releases. Journal of Accounting and Economics 11: 279–292.Subramanyam, K. R. 1996. The pricing of discretionary accruals. Journal of Accounting and Eco-

nomics 22: 250–265.———, and J. J. Wild. 1996. Going-concern status, earnings persistence, and informativeness of

earnings. Contemporary Accounting Research 13: 251–273.Teoh, S. H., and T. J. Wong. 1993. Auditor size and the earnings response coefficient. The Accounting

Review 68: 346–366.Turner, L. 2002. Testimony and Statements before the Committee on Banking, Housing, and Urban

Affairs United States Senate, Available at: http: / /banking.senate.gov/02 02hrg/022602/turner.htm.

U.S. General Accounting Office (GAO). 2003. Required Study on the Potential Effects of MandatoryAudit Firm Rotation. Report to the Senate Committee on Banking, Housing, and Urban Affairsand the House Committee on Financial Services. Available at: http: / /www.gao.gov/new.items/d04216.pdf.

U.S. Senate. 1977. Improving the Accountability of Publicly Owned Corporations and Their Auditors.Prepared by the Subcommittee on Reports, Accounting, and Management of the Committee onGovernmental Affairs. Washington, D.C.: Government Printing Office.

Van Horne, J. C. 1992. Financial Management and Policy. Englewood Cliffs, NJ: Prentice Hall.Wall Street Journal. 2002a. Questioning the books: SEC queries of global crossing likely to resurface.

(February 11): A8.———. 2002b. Questioning the books: Court documents show Andersen ties with Enron were grow-

ing in early 1990s. (February 26): A6.Warfield, T., J. Wild, and K. Wild. 1995. Managerial ownership, accounting choices, and informa-

tiveness of earnings. Journal of Accounting and Economics 20: 61–91.Ziebart, D. A., and S. A. Reiter. 1992. Bond ratings, bond yields and financial information. Contem-

porary Accounting Research 9: 252–282.

Documents

Auditor Tenure and Perceptions of Audit Quality