The Price-Earnings Relation in Troubled Times: The Case of Arthur Andersen. January 12, 2003

Transcription

1 The Price-Earnings Relation in Troubled Times: The Case of Arthur Andersen January 12, 2003 By Paul K. Chaney* Owen Graduate School of Management Vanderbilt University (615) Kirk L. Philipich Fisher College of Business The Ohio State University * Contact Author Acknowledgement: We would like to thank Karl Hackenbrack, Debra Jeter, Craig Lewis, Ron Masulis, and Emad Mohd for helpful comments. We would also like to acknowledge the financial support from the Dean s Fund for Faculty Research.

2 1. Introduction In this study, we investigate the impact of a decline in auditor reputation on the price-earnings relation. Specifically, we examine both the market response to earnings announcements and the timeliness of losses for Arthur Andersen s clients and the clients of the remaining final four large auditors. During 2001, Arthur Andersen was fined or paid over $100 million to settle lawsuits for audit problems concerning two clients, Waste Management and Sunbeam. However the worst was yet to come, on November 8, 2001, a third company, Enron, announced that the company and its auditor, Andersen, determined that certain off-balance sheet entities (primarily a special purpose entity named Chewco) should have been consolidated in accordance with generally accepted accounting principles (GAAP). As a result, Enron stated that all earnings from 1997 through 2000 should not be relied upon and would be reduced by amounts ranging from a low of $96 million in 1997 to a high of $250 million in Also, Enron s debt had been understated by a high of $711 million in 1997 to a low of $628 million in Andersen had billed Enron $80,000 for a review of the Chewco transaction in Serious concerns were raised about Andersen s audit quality. On January 10, 2002, Andersen admitted to shredding a significant number of audit documents related to the Enron audit. On January 17, 2002, Enron fired Andersen and three days later the Powers report was released with more details concerning Andersen s involvement in certain questionable transactions. On March 15, 2002, the Department of Justice unsealed an indictment against Andersen and on June 15, 2002, the federal jury convicted Arthur Andersen of a single count of obstructing justice. Arthur Andersen was barred from conducting and reporting on the audits of SEC registered companies after

3 August This paper investigates whether the market s response to the earnings of Andersen s clients was affected by these issues. We investigate whether the market s response to Andersen s clients earnings has declined after news of Andersen s involvement in the Enron audit became public information. In addition, we examine the timeliness of bad news released by Andersen s clients. 1.1 VALUE OF AN AUDIT The Securities and Exchange Commission (SEC) requires that all firms that issue securities employ an independent auditor to examine its financial records. Auditors issue opinions attesting whether the firm s financial statements are prepared in accordance with generally accepted accounting principles (GAAP). The auditors issue an opinion stating whether the financial statements represent fairly the financial position of the company. The auditor must also state whether the company has a going concern issue. Because of the agency relationships among creditors, outside shareholders, and managers, Watts and Zimmerman [1983] argue that these relationships lead managers and entrepreneurs to desire high quality monitoring in order to reduce agency costs. Thus, auditor quality has important implications for investors, clients, and audit firms. Consider the incentives of managers of private firms considering an initial public offering (IPO). Titman and Trueman [1986] and Datar et al. [1991] develop models where the initial value for an IPO is an increasing function of audit quality. Feltham et al. [1991] in examining the U.S. IPO market provide weak empirical support for the Datar et al. [1991] proposition that the initial IPO value is dependent upon audit quality. Clarkson and Simunic [1994] reexamine this proposition within the context of the Canadian IPO market and find stronger support for the link between audit quality and the initial IPO value. 2

4 Balvers et al. [1988] provide a model in which IPO underpricing is a function of auditor quality. They purport that underpricing should be less severe when higher quality auditors are engaged. Both Balvers et al. [1988] and Beatty [1989] report less underpricing when a Big 8 auditor is engaged as compared to the underpricing for IPOs utilizing non-big 8 auditors. Thus higher audit quality benefits firms by leading to a greater value for the offering (and/or less underpricing). Large public auditors (Big 8, Big 6, or Big 5) are argued to have higher quality audits than smaller audit firms. Francis, Maydew, and Sparks [1999] find that firms are more likely to hire a Big 6 auditor (a proxy for audit quality) to enhance the credibility of their earnings. Dopuch and Simunic [1982] suggest that big auditors are more likely to invest in information technology and therefore have an increased ability to detect earnings management. Becker, DeFond, Jiambalvo, and Subramanyam [1998] find that clients of Big 6 auditors report lower discretionary accruals than non-big 6 auditor clients. An auditor s reputation, either in appearance or in fact, is crucial to the market in determining the relevance and reliability of its clients financial information for firm valuation. The audit report provides a level of assurance to the users of financial statements that the financial statements have been prepared in accordance with generally accepted accounting standards and that the financial statements reflect business reality. Choi and Jeter [1992] demonstrate a lessened response to earnings reports when a firm receives a qualified audit opinion. Teoh and Wong [1993] support a hypothesis that the earnings response coefficient for Big 8 auditor clients is greater than the earnings response coefficient for non-big 8 clients. Thus higher quality auditors have been argued to earn an audit fee premium, in part, because higher quality audits provide value to its client. 3

5 One argument addressed in the literature concerning the auditor-client relation is that high quality auditors have more to lose if their reputation is damaged. In order to maintain this reputation, these auditors are more likely to remain independent and are more likely to perform higher quality audits. Loss of auditor reputation has significant costs regardless of the size of the audit firm. Arthur Andersen, once one of the largest audit firms, is no longer conducting audits. Chaney and Philipich [2002] demonstrate that the clients of Arthur Andersen experienced a 2 percent decline in market value surrounding the dates on which information concerning Andersen s audit quality was questioned (specifically the shredding announcement date). Menon and Williams [1994] demonstrate that firms employing Laventhol and Horwath experienced a negative market reaction to the audit firm s bankruptcy. They attribute this cost to loss of insurance provided to future stakeholders. Thus the cost of the lost reputation is high for the client, shareholders, and the auditor. 2. Motivation and Hypotheses 2.1 EARNINGS REPSONSE COEFFICIENTS (ERC) After Enron, investors feared that more Enrons might be found. Suppose that auditors either ignored or allowed their clients to push the limits of GAAP or if the auditors allowed their independence to be violated; any firm s earnings quality might be questioned. In addition, numerous earnings restatements further eroded investors confidence in the integrity of the financial statements. The General Accounting Office (GAO) reported that earnings restatements due to accounting irregularities rose 145 percent between 1997 and June In 1997, less than one percent of firms restated earnings. During 2002, the number of firms restating earnings is likely to exceed three 4

6 percent. While Andersen did not lead the list of audit firms with the most restatements, they were involved in several of the largest (Enron, WorldCom, and Waste Management). Chaney and Philipich [2002] suggest that the lost reputation of Andersen directly impacted the valuation of its clients. In order for Andersen s client to regain any lost valuation due to auditor reputation, firms had several choices to improve the credibility of their financial statements. They could switch auditors or increase the quality of the information in earnings. Unfortunately, the speed of the indictment by the Department of Justice eventually resulted in all of Andersen s clients switching to other auditors (In other words, before the clients could make the switch decision for themselves, Andersen was indicted and not considered likely to survive). The shredding announcement date was January 10, 2002 and the indictment was unsealed on March 15. Most of Andersen s S&P 1500 clients remained with Andersen until after the indictment. Only thirteen clients of Andersen s S&P 1500 clients discharged Andersen before the indictment date. Approximately 71% of these S&P 1500 client report on a December year-end basis. It is unlikely that these firms would switch auditors in the middle of a year-end audit. Therefore, Andersen s clients needed alternative approaches to signal that their earnings were credible. We use the date Andersen announced that a significant number of audit documents related to the Enron audit were shredded as our cut-off date for increased uncertainty regarding the credibility and reliability of the financial statements. Because of the increased uncertainty by investors concerning earnings quality, given the heightened scrutiny of accounting by the business press, we expect that the market s response to earnings, released after the shredding announcement date, declined on average from the previous year s response. 5

7 Our first hypothesis relates to all firms and is not specific to Andersen s clients. This hypothesis is: H1: Earnings response coefficients (the mapping of earnings into prices) declined, on average, for firms after the decrease in the perceived credibility of financial statements (i.e. after the news that Andersen shredded documents related to the Enron audit). Because Andersen played a prominent role in the decreased confidence of accounting in general, we expect that Andersen s clients ERCs may decline to a greater extent. Alternatively, Andersen may attempt to curtail further erosion in their reputation by taking a harder line with audit clients. Andersen s auditors may require more detailed or timely disclosures. For instance, we might see an increase in write-offs. Andersen s clients may also wish to signal to its investors that the financial statements fairly represent the financial position of the firm. This argument would support an increase in ERCs for Andersen s clients. Our secondary hypothesis is: H2: Arthur Andersen s clients earnings response coefficients (the mapping of earnings into prices) decreased after the date Andersen announced that documents related to the Enron audit were shredded. 2.2 CONSERVATISM Basu [1997] tests the asymmetric timeliness of earnings. Using positive and negative returns as a proxy for good and bad news, he finds that the relation between earnings and returns is stronger for bad news versus good news. He attributes this finding to conservatism of accounting earnings, i.e. that bad news is reflected in earnings on a timely basis, while good news tends to be delayed (but more persistent over time). 6

8 Ball, Kothari, and Robin [2000] find that common-law accounting income exhibits significantly greater timeliness than code-law accounting income and suggest that this is due to conservatism. They suggest that conservative accounting facilitates the monitoring of managers. Auditing may play a role in examining the timeliness of earnings. In a heightened period of investor scrutiny of accounting numbers, managers or auditors may feel more pressure to recognize losses on a timely basis. Our third hypothesis is (stated in the null): H3: The asymmetric timeliness of earnings for Arthur Andersen s clients is not different from the asymmetric timeliness of earnings for other Big 5 audit firm clients. 3. Research Design 3.1 SAMPLE Our initial sample consists of 284 of the 287 Arthur Andersen clients included in the S&P Security prices were obtained from Datastream and CRSP. The most recently completed 10-Ks, 8-Ks, and/or annual reports were examined to identify the company s auditor. We include firms with annual earnings announcements in years 2001 and 2002 and eliminate any earnings announcement between September 7, 2001 and January 10, Descriptive data are provided in Tables 1 and 2. Arthur Andersen s clients represent 52 different industries using the Global Industry Classification Standard (GICS). Panel A of Table 1 reveals that the industries represented by the greatest number of Andersen clients are the Commercial Services and Supplies industry (28 firms), the Oil and Gas industry (17 firms), and the Machinery and Electrical Utilities industries (16 firms each). In panel B of Table 1, for the S&P 1500 firms we provide a comparison of the percentages of our sample of Arthur Andersen 7

9 clients within each segment of the S&P 1500 to the firms remaining with available data. Twenty-nine percent of Arthur Andersen s clients are in the S&P 500, 26% of the firms are in the Mid Cap index, and 45% of firms are found in the Small Cap index. The average asset size for our sample of Arthur Andersen s clients is $9,009 million and the mean stock market capitalization is $5,095 million. While the average size of Andersen s client is smaller than clients of other auditors ($13,585), the difference is not statistically different. Andersen s clients use more debt financing, i.e. have higher leverage, and on average earn less income before extraordinary items. The measure of growth opportunities, MVBV is also lower, but not significantly. Audit fees (not reported) are available for 208 of the 284 firms. The total audit fee consists of three components, the audit fee, the information systems fee (IS fee), and other fees (Other fee). Other fees include tax-related services, due diligence of potential acquisitions, SEC filings, and audit related services. The average total fee is $3.56 million with the audit component averaging $1.2 million. Information systems fees averaged $0.307 million and other fees averaged $2 million. The majority of the information systems fees and other fees resulted from services provided to Andersen s S&P 500 clients. In Figure 1, we plot the Russell 3000 index from January 2001 through June On average, the market declined over the entire period, with several periods of short recoveries. The largest decline occurred over the time period surrounding the September 11 attacks. All earnings announcements between September 7, 2001 and January 10, 2002 were eliminated from the sample. Since most of our earnings announcements occurred between January and June of 2001 and 2002, we also computed the buy and hold market return for this time period. From January to June of 2001 the buy and hold market return 8

10 was percent. During the same period in 2002, the return was percent. The standard deviation of the returns was approximately 30 percent larger in 2001 than While both time periods were similar in terms of the aggregate market performance, the returns in 2001 were slightly more volatile. 3.2 TEST OF OVERALL RESPONSE TO EARNINGS In order to test the hypotheses, we measure the market s reaction to annual earnings announcements. Determining the market reaction involves measuring daily abnormal returns (the difference between actual and expected returns). We use both market model and market adjusted abnormal returns. To control for the effects of market-wide fluctuations, the market model is used: where: R it = a i + ß i R mt + u it (1) R it = return for client i on day t, a i = intercept, R mt = return on the value-weighted Russell 3000 index on day t, ß i = beta for firm i, u it = error term. As presented below, the abnormal return on day t is the difference between the actual return and the expected return derived from the market model: AR it it ( α + β R ) = R ˆ ˆ (2) i i mt Market reaction is defined as the two-day cumulative abnormal return (CAR) measured beginning the day prior to the earnings announcement (t =-1) through the day of the earnings announcement (t=0): 9

11 CAR (3) = 0 AR it t= 1 We also compute market-adjusted abnormal returns, computed as the difference between the firm s return and a market return. Buy and hold raw returns are used as a third measure. We test the significance of the abnormal returns using a corrected version of the Z- test based upon Mikkelson and Partch [1988]. The corrected test accounts for the fact that within the window, the abnormal returns are serially correlated. Applications using the corrected test can be found in Cowan, Nayar and Singh [1990], Mann and Sicherman [1991], and Lee [1992]. In addition, we provide a generalized sign test that examines the null hypothesis that the fraction of positive returns is the same proportion as in the estimation period. The test applies a normal approximation of the binomial distribution, (Cowan [1992]). 1 Four alternative market return measures are used in estimating the market model. These include: 1) the Russell 3000 index, 2) the Russell 1000 index, 3) an equally weighted index of the S&P 1500 firms, and 4) a value-weighted S&P 500 index. 2 Since all market returns provide qualitatively similar results, we only present the results using the Russell 3000 index. We chose this index based on the relative sizes of the average market capitalization since the average (median) market capitalization for our sample was $5.4 billion ($1.3 billion) and the average (median) market capitalization for the Russell We also computed market-adjusted abnormal returns using various market returns. The results obtained using these procedures were similar to the results reported 2 The Russell 3000 Index includes the top 3000 NASDAQ, NYSE, and AMEX US domiciled firms ranked by market cap. The index is composed of 79.3% NYSE, 0.4% AMEX, and 20.3% NASDAQ firms. The Russell 1000 includes the top 1,000 Russell 3000 firms. The index is composed of 81.8% NYSE, 0.2% AMEX, and 17.9% NSDAQ firms. 10

12 was $4.6 million ($0.73 billion). The Russell 1000 has an average (median) market capitalization of $13 billion ($3.8 billion). 3.3 CROSS-SECTIONAL ANALYSIS Cross-sectional analysis is used to test the hypotheses. The following multivariate model is used to examine the change in the earnings response coefficient before and after our cut-off date. The model is estimated using OLS with heteroskedastic consistent standard errors for significance tests. Alternatively, the model is estimated using iteratively re-weighted least squares (also known as robust regression). In this procedure the weights used are scaled residuals. The scale is a median absolute deviation about a median residual divided by a constant. If the residual is small, the case weight is assigned a weight equal to one, but if the residual is large, the case weight is equal to a tuning constant divided by the absolute value of the scale. The model is re-estimated until it converges. The model used to test the first hypothesis is: CAR it = β 1 + β 2 Dum it +β 3 Unex it + β 4 Dum*unex it + β 5 MVBV it + β 6 Lev it + β 7 Persist it + µ it (4) where: CAR it = Cumulative mean abnormal return (two-day window), Dum it = 1 if the annual earnings announcement occurred after January 10, 2002, 0 otherwise, Unex it = unexpected earnings, measured using IBES forecasts, deflated by beginning of year price, Dum*unex it = interactive variable between Dum and Unex, MVBV it = Market value of equity plus book value of liabilities divided by total assets, Lev it = long-term debt plus debt in current liabilities divided by total assets, Persist = 1 if EPS/price ratio is in the middle 60% of positive EPS/price ratios, 0 otherwise. The ERC for the pre-shredding date is β 3 while the ERC for the post-shredding date is β 3 + β 4. 11

13 Kormendi and Lipe [1987] find that ERCs are increasing in earnings persistence. Our persistence measure follows the procedure used by Ali and Zarowin [1992] and Cheng, Liu, and Schaefer [1996]. They identify extreme values of the earnings to price ratio to indicate transitory earnings while non-extreme values indicate more permanent earnings. Collins and Kothari [1989] find that ERCs are increasing in growth opportunities. We use the ratio of market value of equity plus book value of long- and short-term debt divided by total assets as our growth variable. Collins and Kothari [1989] also find that ERCs are decreasing in the riskiness of earnings. We use leverage and the stock s beta to proxy for riskiness of earnings. Leverage is computed as total debt divided by total assets. The beta is computed using the market model and 200 days of returns. The second hypothesis is tested by including a dummy variable for the firm s auditor. The variable, Ander is set equal to one if the firm is audited by Andersen, 0 otherwise. This dummy variable is interacted with Unex and Dum*Unex to compute the ERC before and after the shredding date for Andersen and the remaining final four auditors. CAR it = β 1 + β 2 Dum it + β 3 Ander+ β 4 Ander*Dum it +β 5 Unex it + β 6 Dum*unex it + +β 7 Ander*Unex it + β 8 Ander*Dum*unex it + β 9 MVBV it + β 10 Lev it + β 11 Persist it + µ it (4) where: Ander it = 1 if the firm is audited by Arthur Andersen, 0 otherwise. The ERCs for clients of non-andersen auditors for the pre-shredding date is β 3 and β 3 + β 4 for the post-shredding date. The ERCs, for Andersen s clients, areβ 3 + β 7 and β 3 + β 6 +β 7 + β 8 for the pre- and post-shredding dates respectively. 12

14 4. Results 4.1 ERCs DURING TROUBLED TIMES The first hypothesis examines the change in ERCs between 2000 and 2001 for S&P 1500 firms with annual earnings announcements between January and June of the respective years. The results of estimating equation (4) are presented in Table 3, Panel A. In the first two columns, the coefficients are estimated using OLS with the market model CARs and market-adjusted CARs. In the last two columns, the regressions are repeated using iteratively re-weighted least squares for the same CARs. The coefficient on Unex is the ERC for 2000 and the sum of the coefficient on Unex and Dum*Unex is the ERC for The ERCs for 2000 are virtually identical among the various estimation procedures (0.732 for OLS market model CARs to for OLS market-adjusted CARs). The change in ERCs between 2000 and 2001 is reflected in the coefficient on Dum*Unex. ERCs declined in Using the market model CAR, the ERC is ( ). While the magnitude and sign of the coefficient is similar among the different estimation procedures (for the OLS estimation, the coefficients are for the market-adjusted CARs and for the market model CARs), the coefficients are only significant using the robust regression. The negative sign indicates that ERCs, on average, have declined providing support for the first hypothesis. We suggest that the increased skepticism of accounting in the press and the perceived lack of credibility of the financial statements led the market to a lessened response to earnings, in general, for earnings released after the Andersen shredding announcement date. As a robustness check, we computed a two-day return (-1,0) for each earnings announcement date. We replaced the abnormal return (CAR) with the two-day return (Ret). These results are reported in Panel B, of Table 3. Using raw returns, the ERC for firms 13

15 reporting earnings in 2000 was larger (0.924) than using CARs. Similarly, the change in ERCs was negative (-0.591) and statistically significant at the 0.05 level. Using raw returns also supports our hypothesis that ERCs have declined. The control variables for growth, persistence, and leverage are generally insignificant. We also interacted each of these variables with Unex, but the coefficients on these variables were statistically insignificant. Because Hayn [1995] reported lower ERCs for loss firms, we created interactive variables (for all variables) using a dummy variable for loss (loss=1 if income before extraordinary items was negative, 0 otherwise). The results of this regression (not reported) did not change any of the results. In general, loss firms had higher but insignificant ERCs in 2000 and lower but insignificant ERCs in Our second hypothesis examines whether the decreased ERCs are unique to clients of Andersen or if the decline in ERCs is universal to all clients of Big 5 auditors. The estimated coefficients from estimating equation (4) are reported in Table 4. In this regression, a dummy variable, Ander, is interacted with the unexpected earnings variables and the intercept. The ERCs for Andersen s clients for 2000 (the sum of the coefficients on Unex and Ander*Unex) are smaller than ERCs of clients of other auditors (although not significant). Using the market-model estimated coefficient, the ERCs for non-andersen clients for 2000 and 2001 are and ( ) respectively. This decline in ERCs is similar to the decline in ERCs reported in Table 3 for all clients. However, for Andersen s clients, the ERCs for 2000 and 2001 are ( ) and ( ). Relative to other clients, Andersen s clients ERCs increased from the previous years ERC, while the average non-andersen clients ERC declined. 14

16 Because auditors specialize in different industries, we estimated equation (4) replacing Unex with variables representing unexpected earnings for each individual Big 5 client. For instance, we define the following variables: Auditor = 1 if the auditor = {Ander, PM, PW, DT, EY}, 0 otherwise, where, Ander*Unex = Unex if the firm s auditor is Andersen, 0 otherwise, PM*Unex = Unex if the firm s auditor is KPMG, 0 otherwise, PW*Unex = Unex if the firm s auditor is PriceWaterhouseCoopers, 0 otherwise, DT*Unex = Unex if the firm s auditor is Deloitte and Touche, 0 otherwise, EY*Unex = Unex if the firm s auditor is Ernst and Young, 0 otherwise. Therefore, we compute the specific ERC for the clients of each Big 5 firm. The ERCs are reported in Table 5. ERCs are computed using the market model, the market-adjusted model, and using the returns model. All three models provide consistent coefficients; however, none of the ERCs are statistically significant. The ERCs for clients of Andersen and Ernst and Young increased, while the ERCs for the remaining Big 5 clients decreased. Because of the small sample size for each auditor, the power of the regressions is low. In general, no support is found for a decline in ERCs for Andersen s clients. If anything, the response to earnings increased over this time period. This suggests that clients of Andersen may have improved the quality of the information concerning earnings released to the market in an effort to demonstrate that the reliability of their financial statements were not affected by their auditor. Investors, aware of Andersen s audit problems, may have determined that Andersen in an effort to protect its remaining reputation was less likely to approve of earnings management. Andersen may have used a higher audit standards than in previous years in auditing these firms. 4.2 TIMELINESS OF BAD NEWS In the final hypothesis, we examine the asymmetric timeliness of earnings surrounding our cut-off date. If firms, due to conservatism, report more timely losses as a 15

17 result of increased scrutiny by investors or increased pressure by auditors, we might expect the asymmetric timeliness of earnings to increase from 2000 to If auditing provides a mechanism to facilitate monitoring of managers, auditors who are under fire for perceived lack of independence may require managers to recognize losses on a more timely basis. These results are presented in Table 6, panels A and B. For non-andersen clients, the timeliness of bad news relative to good news (the sum of the coefficients on positive and negative returns divided by the coefficient on positive news) declined from 10.4 times to 7.93 times in the post-period. For example, the coefficient on positive returns in the postperiod is and the coefficient on the negative returns is (the sum of positive coefficient of and the coefficient on interactive negative return of ). The timeliness measure is divided by , or 7.93 times. The intercepts are positive and significant, consistent with the current recognition of unrealized gains from earlier periods (that are uncorrelated with current news). The timeliness of Andersen s clients bad news was times in the pre-period and not significantly different from the good news in the post-period (i.e. not significantly different from one). This is a surprising result. Gigler and Hemmer (2001) argue that firms in less conservative financial reporting regimes (such as likely in our post period) are more likely to engage in timely preemptive disclosure than firms in more conservative regimes. To control for this, the returns window should exclude the market reaction to both the current period s and prior period s earnings announcements. These results have not yet been tabulated. In Table 7, we compute the average timeliness coefficient for the clients of each of the Big 5 auditors. The coefficient for bad news increased for three of the five audit firm clients; yet the timeliness of bad news relative to good news increased for clients of only 16

18 two of the five audit firms. The timeliness of bad news should be directly related to litigation risk. Given the increased criticism of accounting, it is surprising that. 17

19 5. Conclusion In this paper, we examine the information contained in the last earnings reported by clients of Arthur Andersen relative to clients of other audit firms. The perceived uncertainty of the quality of the financial statements began as a direct result of Andersen s Enron audit problems. We provide evidence that, on average, the earnings response coefficient declined for firms after the Andersen shredding announcement date. We interpret this finding to the lack of confidence investors placed on the information provided in unexpected earnings. If the accounting numbers cannot be taken as reality then unexpected earnings despite increased litigation risk, does not provide a reliable measure of information. Andersen s clients ERCs increased slightly (but not significantly) after the shredding announcement date suggesting that either the clients or the auditors provided increased disclosures concerning earnings. In addition, we are not able to demonstrate that firms increased the timeliness of bad news relative to good news in earnings. 18

20 REFERENCES ALI, A. and P. ZAROWIN The Role of Earnings Levels in Annual Earnings-Returns Studies. Journal of Accounting Research 30 (1992): BALL, R.; S.P. KOTHARI; And A. ROBIN. The Effect of International Institutional Factors on Properties of Accounting Earnings. Journal of Accounting and Economics 29 (2000): BALVERS, R.; B. MCDONALD; And R. MILLER. Underpricing of New Issues and the Choice of Auditor as a Signal of Investment Banker Reputation. The Accounting Review 63 (1988): BASU, S. The Coservatism Principle and the Asymmetric Timeliness of Earnings. Journal of Accounting and Economics 24 (1997): BEATTY, R. Auditor Reputation and the Pricing of Initial Public Offerings. The Accounting Review 64 (1989): BEATTY, R. The Economic Determinants of Auditor Compensation in the Initial Public Offerings Market. Journal of Accounting Research 31 (1993): CHANEY, P. and K. PHILIPICH Shredded Reputation: The Cost of Audit Failure. Journal of Accounting Research 40 (2002): CHENG, G., C-S. LIU; And T. SCHAEFER. Earnings Performance and the Incremental Information Content from Operations. Journal of Accounting Research 34 (1996): CHOI, S., AND D. JETER. The Effect of Qualified Audit Opinions on Earnings Response Coefficients. The Journal of Accounting and Economics (1992): CLARKSON, P., AND D. SIMUNIC. The Association Between Audit Quality, Retained Ownership, and Firm Risk in U.S. vs. Canadian IPO Markets. Journal of Accounting and Economics 17 (1994): COLLINS, D. and S.P. KOTHARI An Analysis of Intertemporal and Cross-sectional Determinants of Earnings Response Coefficients. Journal of Accounting and Economics 11 (1989): COWAN, A. Nonparametric Event Study Tests. Journal of Financial and Quantitative Analysis 25, 4 (1992): COWAN, A.; N. NAYAR; and A. SINGH. Stock Returns Before and After Calls of Convertible Bonds. Review of Quantitative Finance and Accounting 2, 4 (1990):

21 CRASWELL, A.; J. FRANCIS; and S. TAYLOR. Auditor Brand Name Reputations and Industry Specializations. Journal of Accounting and Economics 20 (1995): DATAR, S.; G. FELTHAM; and J. HUGHES. The Role of Audits and Audit Quality in Valuing New Issues. Journal of Accounting and Economics 14 (1991): DYE, R. Auditing Standards, Legal Liability, and Auditor Wealth. Journal of Political Economy 101 (1993): FELTHAM, G.; J. HUGHES; and D. SIMUNIC. Empirical Assessment of the Impact of Auditor Quality on the Valuation of New Issues. Journal of Accounting and Economics 14 (1991): FRANCIS, J. The Effect of Audit Firm Size on Audit Prices: A Study of the Australian Market. Journal of Accounting and Economics 6 (1984): FRANCIS, J.; E. MAYDEW; and H. SPARKS. The Role of Big 6 Auditors in the Credible Reporting of Accruals. Contemporary Accounting Research 16 (1999): FRANCIS, J., AND D. SIMON. A Test of Audit Pricing in the Small-Client Segment of the U.S. audit market. The Accounting Review 62 (1987): HAYN, C. The Information Content of Losses. Journal of Accounting and Economics 20, (1995): KORMENDI, R., AND B. LIPE. Earnings Innovations, Earnings Persistence and Stock Returns. Journal of Business 60 (1987): LEE, D. Management Buyout Proposals and Inside Information. Journal of Finance 47, 3 (1992): MANN, S., AND N. SICHERMAN. The Agency Costs of Free Cash Flow: Acquisition Activity and Equity Issues. Journal of Business 64, 2 (1991): MENON, K., AND D. WILLIAMS. The Insurance Hypothesis and Market Prices. The Accounting Review 69 (1994): MIKKELSON, W., and M. PARTCH. Withdrawn Security Offerings. Journal of Financial and Quantitative Analysis 23, 4 (1988): 487. SCHIPPER, K., AND R. THOMPSON. The Impact of Merger-Related Regulations on the Share-holders of Acquired Firms. Journal of Accounting Research (1983): TEOH, S., AND T. WONG. Perceived Auditor Quality and the Earnings Response Coefficient. The Accounting Review 68 (1993):

22 TITMAN, S., AND B. TRUEMAN. Information Quality and the Valuation of New Issues. Journal of Accounting and Economics 8 (1986): WATTS, R., AND J. ZIMMERMAN. Agency Problems, Auditing and the Theory of the Firm: Some Evidence. Journal of Law & Economics 26 (1983): WILLENBORG, M. Empirical Analysis of the Economic Demand for Auditing in the Initial Public Offerings Market. Journal of Accounting Research 37 (1999): WILSON, T., AND W. GRIMLUND. An Examination of the Importance of an Auditor's Reputation. Auditing: A Journal of Practice and Theory 9 (1990):

23 22

24 Table 1 Industry and Market Classifications Panel A: S&P 1500 Sample Industries*with 3 or more Arthur Andersen Clients Auditor** Industries AA E&Y D&T KPMG PWC Other Airlines Banks Biotechnology Building Products Chemicals Commercial Services & Supplies Communications Equipment Construction & Engineering Containers & Packaging Diversified Financials Divers. Telecommunication Services Electrical Utilities Electrical Equipment Electronic Equipment & Instruments Energy Equipment & Services Food & Drug Retailing Food Products Gas Utilities Health Care Equipment & Supplies Health Care Providers & Services Hotels Restaurants & Leisure Household Durables Leisure Equipment & Products Machinery Media Metals & Mining Multi-Utilities Multi-line Retail Oil & Gas Paper & Forest Products Pharmaceuticals Road & Rail Semiconductor Equipment & Products Software Specialty Retail Textiles & Apparel Others Total Firms * Using Global Industry Classification Standard (GICS) Sectors. Using 4-digit SIC codes, only two industries (1311 and 4911) contained more than 10 firms. Only three 4- digit SIC codes (2834, 4931, and 7372) contained between 5 and 10 firms. ** AA = Arthur Andersen, EY = Ernst & Young, DT = Deloitte & Touche, KPMG = Peat Marwick, and PW = Price Waterhouse 23

25 Table 1 Panel B: Market Index Classification Andersen s Clients Other Auditors Clients Totals Number Percentage Number Percentage Number Percentage S&P % % % Mid Cap 76 26% % % Small Cap % % % 600 Total 287 1,213 1,500 24

26 Table 2 Descriptive Statistics Panel A: All Observations Pooled (n=2,317) ($ Millions) Lower Upper Mean Quartile Median Quartile Total Assets 12, , , Market Value of Equity 8, , , Earnings before Extra Loss (=1 if NI<0) Leverage MVBV Beta EPS/Price CAR (%) Unex (%) Panel B: All Andersen Observations Pooled (n=378) ($ Millions) Lower Upper Mean Quartile Median Quartile Total Assets 9, , , Market Value of Equity 5, , , Earnings before Extra Loss (=1 if NI<0) Leverage MVBV Beta EPS/Price CAR (%) Unex (%) Panel C: All Non-Andersen Observations Pooled (n=1,939) ($ Millions) Lower Upper Mean Quartile Median Quartile Total Assets 13, , , Market Value of Equity 9, , , Earnings before Extra Loss (=1 if NI<0) Leverage (Lev) MVBV Beta EPS/Price CAR (%) Unex (%)

27 Table 2 definitions: Loss it = 1 if income before extraordinary items is less than 0, 0 otherwise, Lev it = long-term debt plus debt in current liabilities divided by total assets, MVBV it = Market value of equity plus book value of liabilities divided by total assets, Beta it = Computed using the market mode, EPS it /Price it = Earnings per share divided by price per share CAR it = Cumulative mean abnormal return (two-day window), Unex it = unexpected earnings, measured using IBES forecasts, deflated by beginning of year price, 26

28 Table 3 Cross-Sectional Regression Panel A: Abnormal Returns CAR it = β 1 + β 2 Dum it +β 3 Unex it + β 4 Dum*unex it + β 5 MVBV it + β 6 Lev it + β 7 Persist it + µ it OLS Regression IRLS Regression Market Market Market Model CAR Adjusted CAR Market Model CAR Adjusted CAR Estimate Estimate Estimate Estimate Variable (t-value) (t-value) (t-value) (t-value) Intercept * Dum ** 0.007** ** 0.008** Unex ** 0.753** ** 0.735** Dum*Unex * -434* MVBV ** * Lev Persist F Value ** 8.57** 9.84 ** 10.01** Adj R-Sq obs. 2,317 2,317 2,317 2,317 *, ** indicate significance at the 0.05 and 0.01 levels. White s heteroskedastic t-values are reported. where: CAR it = Cumulative mean abnormal return (two-day window), Dum it = 1 if the annual earnings announcement occurred after January 10, 2002, 0 otherwise, Unex it = unexpected earnings, measured using IBES forecasts, deflated by beginning of year price, Dum*unex it = interactive variable between Dum and Unex, MVBV it = Market value of equity plus book value of liabilities divided by total assets, Lev it = long-term debt plus debt in current liabilities divided by total assets, Persist = 1 if EPS/price ratio is in the middle 60% of positive earnings/price ratios, 0 otherwise. 27

29 Table 3 (continued) Cross-Sectional Regression Panel B: Raw Returns Ret it = β 1 + β 2 Dum it +β 3 Unex it + β 4 Dum*unex it + β 5 MVBV it + β 6 Lev it + β 7 Persist it + µ it IRLS OLS Regression Regression Estimate Estimate Variable (t-value) (t-value) Intercept Dum ** * Unex ** ** Dum*Unex * ** MVBV Lev * Persist F Value 9.41 ** ** Adj R-Sq obs. 2,317 2,317 *, ** indicate significance at the 0.05 and 0.01 levels. White s heteroskedastic t-values are reported. where: Ret it = The firm s return over days -1 and 0 (zero is the earnings announcement day) Dum it = 1 if the annual earnings announcement occurred after January 10, 2002, 0 otherwise, Unex it = unexpected earnings, measured using IBES forecasts, deflated by beginning of year price, Dum*unex it = interactive variable between Dum and Unex, MVBV it = Market value of equity plus book value of liabilities divided by total assets, Lev it = long-term debt plus debt in current liabilities divided by total assets, Persist = 1 if EPS/price ratio is in the middle 60% of positive earnings/price ratios, 0 otherwise. 28

30 Table 4 Cross-Sectional Regression CAR it (Ret it )= β 1 + β 2 Dum it + β 3 Ander+ β 4 Ander*Dum it +β 5 Unex it + β 6 Dum*unex it + +β 7 Ander*Unex it + β 8 Ander*Dum*unex it + β 9 MVBV it + β 10 Lev it + µ it it CAR Regressions Returns Reg Market Market Model CAR Adjusted CAR Ret Estimate Estimate Estimate Variable (t-value) (t-value) (t-value) Intercept Ander Dum ** ** * Ander*Dum Unex ** ** ** Ander*Unex Dum*Unex * Ander*Dum*Unex MVBV ** * Lev Persist F Value 6.34 ** 5.47 ** 5.75 ** Adj R-Sq obs. 2,317 2,317 2,317 29

31 *, ** indicate significance at the 0.05 and 0.01 levels. White s heteroscedasitic t-values are reported. where: CAR it = Cumulative mean abnormal return (two-day window), Dum it = 1 if the annual earnings announcement occurred after January 10, 2002, 0 otherwise, Unex it = unexpected earnings, measured using IBES forecasts, deflated by beginning of year price, Dum*unex it = interactive variable between Dum and Unex, MVBV it = Market value of equity plus book value of liabilities divided by total assets, Lev it = long-term debt plus debt in current liabilities divided by total assets, Persist = 1 if EPS/price ratio is in the middle 60% of positive earnings/price ratios, 0 otherwise. 30

32 Table 5 Average ERC for Big 5 Clients Fiscal Years 2000 and 2001 CAR it (Ret it )= β 1 + β 2 Dum it +β 3 Auditor*Unex it + β 4 Dum* Auditor*unex it + β 5 MVBV it + + β 6 Lev it + β 7 Persist it + µ it Big Five Firm Model 2000 ERC 2001 ERC Change in ERC Percentage Change Arthur Andersen MM CARs % MAR CAR % Ret % Ernst & Young MM CARs % MAR CAR % Ret % Deloitte and Touche MM CARs (1.1640) (81.6%) MAR CAR (73.3%) Ret (1.5560) (88.1%) KPMG MM CARs (0.2383) (33.5%) MAR CAR (27.7%) Ret (0.2518) (26.8%) PriceWaterhouseCoopers MM CARs (0.3863) (62.1%) MAR CAR (58.8%) Ret (0.5510) (77.2%) Where: Model MM CAR represents the market model, Model MAR CAR represents the market-adjusted model, RET represents the raw-returns model, Auditor =1 if {Andersen, E&Y, D&T, KPMG, or PW}, 0 otherwise. No statistical significance is found for the changes in ERCs for 31

33 Table 6 Timeliness of Earnings NI it /P it-1 = β 1 +β 2 Ander it + β 3 Dum it +β 4 Ander*Dum it + β 5 Ret it +β 6 Ander*Ret it + β 7 Dum*Ret it +β 8 Ander*Dum*Ret it + µ it Panel A: Interactive Andersen Timeliness Model Post Period Pre-Period (after 1/10/02) Coefficients (t-value) Coefficients (t-value) Intercept Ander Dum Ander*Dum Return Ander*Ret Dum*Ret Ander*Dum*Ret R-squared F value Pr. > F Number of obs. 1, Where: Ander Return Dum = 1 if the firm is audited by Andersen = The return is computed starting three months after the start of the year to three months after the firm s yearend. = 1 if the firm s return is negative, 0 otherwise. This dummy represents a proxy for bad/good news. 32

34 Table 6 (continued) Timeliness of Earnings Panel B: Timeliness Estimation Separated for Andersen NI it /P it-1 = β 1 + β 2 Dum it + β 3 Ret it + β 4 Dum*Ret it + µ it Andersen Clients Non-Andersen Clients Post Period Post Period Pre-Period (after 1/10/02) Pre-Period (after 1/10/02) Coefficients (t-value) Coefficients (t-value) Coefficients (t-value) Coefficients (t-value) Number of obs , Intercept Dum Return Dum*Return R-squared F-value Pr.>F

35 Table 7 Timeliness of Earnings Separated by Big 5 Auditor Coefficients on Positive and Negative Returns Andersen E&Y D&T KPMG PWC Positive Returns Pre-period Post-period Negative Returns Pre-period Post-period Timeliness of Negative News Pre-period Post-period The timeliness of negative news is computed by dividing the coefficient on negative returns by the coefficient on positive returns. 34

36 Figure 1. Russell 3000 Market Return (1/2/2001 to 6/25/2002) May 21, March Sept. 11 Jan Date 35