International Research Journal of Applied Finance ISSN Vol. V Issue 10 October, Contents. Article Title & Author (s)

Transcription

1 Contents Article Title & Author (s) Page No. Audit Fee Premiums: Pre and Post the Andersen Scandal and SOX Dr. Jiuzhou Wang, and Nancy Manzhen Fan The Association among Auditor Litigation, Earnings Management, and IPO Underpricing: A Theoretical Perspective Yan Xiong, and Caixing Liu On Maximizing Annualized Option Returns Charles J. Higgins

2 Call for Papers / Case Studies International Research Journal of Applied Finance (IRJAF) is a double blind peer-reviewed open access online journal. Every month a copy of IRJAF reaches faculty members in the areas of accounting, finance, and economics working in 80% of AACSB accredited Business Schools across the world. The journal provides a dedicated forum for Academicians, practitioners, policy makers and researchers working in the areas of finance, investment, accounting, and economics. The Editor of the Journal invites papers with theoretical research/conceptual work or applied research/applications on topics related to research, practice, and teaching in all subject areas of Finance, Accounting, Investments, Money, Banking and Economics. The original research papers and articles (not currently under review or published in other publications) will be considered for publication in International Research Journal of Applied Finance. All paper will be double blind reviewed on a continual basis. Normally the review process will be completed in about 4 weeks after the submission and the author(s) will be informed of the result of the review process. If the paper is accepted, it will be published in the next month issue. By the final submission for publications, the authors assign all the copyrights to the Kaizen Publications. The Editorial Board reserves the right to change/alter the final submissions for IRJAF for editorial purposes. Copyright: Articles, papers or cases submitted for publication should be original contributions and should not be under consideration for any other publication at the same time. Authors submitting articles/papers/cases for publication warrant that the work is not an infringement of any existing copyright, infringement of proprietary right, invasion of privacy, or libel and will indemnify, defend, and hold IRJAF or sponsor(s) harmless from any damages, expenses, and costs against any breach of such warranty. For ease of dissemination and to ensure proper policing of use, papers/articles/cases and contributions become the legal copyright of the IRJAF unless otherwise agreed in writing. Click here to submit your paper: Click here to submit your case study:

3 Audit Fee Premiums: Pre and Post the Andersen Scandal and SOX Dr. Jiuzhou Wang Nancy Manzhen Fan Abstract Purpose This paper aims to examine audit fee premiums along the timeline from 2000 through 2005, a turbulent period of time in audit market with the demise of Author Anderson and passage of Sarbanes-Oxley Act (SOX). Design/methodology/approach The paper employs the treatment effects model. Audit firms are classified into three segments: Big 5 (4) auditors, second-tier auditors (Grant Thornton and BDO Seidman), and third-tier auditors (others). Findings -- Before the demise of Andersen and the passage of SOX, there were no audit fee premiums, but there was a reverse trend afterward. While Big 4 auditors do not charge fee premiums over the second-tier auditors, the second-tier auditors charges audit fee premiums over the third-tier auditors. In addition, our findings show a selectivity bias between Big 5 (4) auditors and the second-tier auditors but no selectivity bias between the second- and third-tier auditors. Research limitations/implications The paper focuses on the cost efficiency of auditor selection only and future research should explore the effects of auditor reputation, monopoly power and product differentiation of Big 4 auditors on audit fee premiums. Originality/value The paper contributes to the extant audit premiums literature by utilizing treatment effect model in examining three tiers of auditors. Keywords Audit fee premium, Treatment effects model, Selectivity bias, Three tiers of auditors I. Introduction The demise of Arthur Andersen and the enactment of the Sarbanes-Oxley Act (SOX) have dramatically changed the U.S. audit market (Landsman et al. 2006). The strict regulatory environment has raised audit risk, and a considerable increase in audit demand has caused a shortage of auditors for all public accounting firms. In light of this changing environment, Big 4 auditors are significantly increasing audit fees (Asthana et al. 2004) and carefully screening both existing and potential clients (Schloetzer 2006) to reduce litigation risk. Prior studies report mixed findings on audit fee premiums (e.g., Simunic 1980; Taffler and Ramalinggam 1982; Francis 1984; Francis and Stokes 1986; Palmrose 1986a; Francis and Simon 1987; Chung and Lindsay 1988; Pong and Whittington 1994; Firth 1997; Ferguson and Stokes 2002). This study uses a treatment effects model to examine how the change in the audit market and the regulation affect audits fee premiums. Almost all the previous studies have examined whether the audit market is classified into two categories, namely, Big N[1] and non-big N auditors (e.g., Simunic 1980; Palmrose 1986a; Francis 1984; Francis and Stokes 1986; Rubin 1988; Chan et al. 1993; Pong and Whittington 1994; Chaney et al. 2004; etc.). However, among non-big N auditors there are national, regional/local firms that possess different resources and employ different strategies (e.g., pricing, investment in technology and training etc.) that subsequently affect the market structure[2]. Francis and Simon (1987) classify auditors into three categories, i.e. Big Eight, national (second-tier) and local/regional auditors, and they find that Big Eight auditors charge audit fee premiums over both the national and local/regional auditors. However, no fee difference is identified between the national and the local/regional audit firms. The divergence among non-big N audit firms becomes more apparent after the recent change in regulations. Therefore, we examine whether auditor selection and audit pricing follow a 1235

4 three-tier classification of audit firms: Big N, the second-tier (Grant Thornton and BDO Seidman)[3], and the third-tier (others). Our study contributes to the extant literature on audit fees in three respects. First, our study supports previous arguments that there is lack of audit fee premiums in 2000 and 2001 (Chaney et al. 2005). However, our results show that, after the demise of Andersen and the passage of SOX, there are audit fee premiums charged by Big 4 auditors over non-big 4 auditors. These results illustrate the impact of the changing audit market and regulations on audit fee premiums. The lack of audit fee premiums in 2000 and 2001 suggests that both Big 5 and non-big 5 clients select auditors based on a cost-efficiency consideration. And, the presence of audit fee premiums after 2001 indicates that Big 4 clients would have paid less audit fees had they chosen non-big 4 auditors. Second, this study further classifies non-big 4 audit firms into the second and third tiers to examine audit fee premiums between different tiers of auditors. Using a three-tier audit firm classification, we find that a selectivity bias exists between Big N and the second-tier auditors but not between the second- and third-tier auditors. This indicates that client firms do not distinguish between the second- and third-tier audit firms and still use the two-tier classification (Big N vs. non-big N) in selecting auditors. On the other hand, we do not find any audit fee premiums between Big N and the second-tier auditors. However, the secondtier auditors charge audit fee premiums over the third-tier auditors across all years, implying that the second-tier auditors consider their reputation and audit quality to be comparable to those of Big N auditors and consequently charge rates similar to those of Big N auditors. Third, we also contribute to the research methodology in the literature. Some studies argue that the mixed findings on audit fee premiums may be partly attributed to the assumption of the exogeneity of the auditor selection (Copley et al. 1995; Chaney et al. 2004; 2005). In this study, we use the treatment effects model to examine audit fee premiums, and the model treats the auditor selection as endogenous. There are two reasons why we use the treatment effects model instead of the Heckman model as in Chaney et al. (2004; 2005). First, the Heckman model is mostly used in labor economics because the wages of non-participants of labor force are not observable. However, the audit fees of both Big N and non-big N clients are available. Second, the Heckman model may generate unreasonable and illogical results. Chaney et al. (2004) emphasize the importance of using different slope coefficients of the audit fee determination equations between Big 5 and non-big 5 segments to capture audit firms different investments in training, technologies, and facilities. However, the Heckman model truncates samples and only keeps one type of auditor selection to run the regression, which may generate unreasonable and illogical results due to extrapolation beyond the scope from which the model is derived (Hogan 1997)[4] when using counterfactual effects to examine audit fee premiums. In contrast, the treatment effects model pools all observations of the whole sample to run the second-stage regression and can avoid such problems although it may not perfectly fit both the Big N and non-big N subsamples. Furthermore, we demonstrate that using a nonlinear model to capture the relation between audit fees and client size can avoid sensitivity to minor changes in different model specifications (Francis and Lennox 2006)[5]. The remainder of this paper is organized as follows. Section 2 discusses prior studies and the hypothesis development. Section 3 describes the method, including our sample data and models, followed by empirical results in Section 4. Section 5 presents our robustness checks and the final section provides a summary of conclusions. 1236

5 II. Literature on audit fee premiums and hypothesis development Auditing theories suggest that Big N audit firms may charge audit fee premiums for three reasons. First, Big N audit firms build their brand name reputation by providing distinguished quality services[6] to their clients (e.g., DeAngelo 1981b; Simunic and Stein 1987; Francis and Wilson 1988; Brinn et al. 1994; Lee 1996; Simon 1997; DeFond et al. 2000; Peel and Roberts 2003). To secure and protect client-specific quasi-rents, Big N auditors need to develop and maintain their brand name reputations by conducting sufficient audit work and reporting independently (DeAngelo 1981b; Francis and Wilson 1988). Otherwise, they may lose client-specific quasi-rents and also decrease their market shares. Second, Big N auditors help client firms to reduce both agency costs (Jensen and Meckling 1976; Simunic and Stein 1987; and DeFond 1992) and the costs of capital (Beatty 1989). When a company hires a Big N auditor, it signals higher audit quality and thereby reduces its agency cost. The credibility demand arises in the circumstance of moral hazard as an important relationship to be addressed with the external agency. As suggested by the demand-based model of product differentiation, high quality audit services can both lower a company s internal agency conflict (DeFond 1992) and enhance the external agency relationship (Simunic and Stein 1987). Thirdly, the Big N auditors have the market (monopolistic) power. Balachandran and Ramakrishnan (1987) argue that audit fees are jointly determined by auditors ability to monitor their clients and by the clients ability to write contracts. The latter is affected by the auditors relative performance with other clients and their industry-specific expertise. Large audit firms normally possess more negotiation power than small audit firms, and therefore can charge higher fees. Just as there are conflicting arguments for audit fee premiums, empirical studies report mixed findings. In the U.S. audit market, Palmrose (1986a) finds Big Eight audit fee premiums for public and non-public companies. Her conclusions are consistent with product differentiation of Big Eight auditors in that audit efforts are also positively associated with the brand name of the Big Eight auditors. Deis and Giroux (1996) study school district audits in Taxes and reach the same conclusion as Palmrose (1986a). Audit fee premiums are also identified by Simon and Francis (1988) for Big Eight auditors as a group. However, the pricing rates of the Big Eight auditors are shown to be different (Simunic 1980; Balachandran and Simon 1993; and Simon 1997). Simunic (1980), Rubin (1988), and Chaney et al. (2005) did not find audit fee premiums for Big Eight auditors in the U.S. market. Audit fee premiums are also identified in U.K (Taffler and Ramalinggam 1982; Chan et al. 1993; Pong and Whittington 1994; and Ireland and Lennox 2002), Australian (Francis 1984; Francis and Stokes 1986 for small companies; Craswell et al. 1995; Ferguson and Stokes 2002; and Hamilton 2005), Canada (Anderson and Zeghal 1994) and other audit markets (Lee 1996; Simon et al. 1992; Firth 1993; Johnson et al. 1995; Simon and Taylor 2002; Willekens and Achmadi 2003; etc.). As in the U.S., evidence is mixed elsewhere as well. Audit fee premiums are not found in U.K. (Copley et al (mixed evidence in the study); Chaney et al. 2004), Canada (Chung and Lindsay 1988), New Zealand (Firth 1985), Norway (Firth 1997) and other audit markets (Simon 1995; Karim and Moizer 1996; Langendijk 1997; etc.) An important point to bear in mind is that most of the prior studies assume the selection of Big N and non-big N auditors to be exogenous, i.e., client firms are randomly assigned to audit firms. Therefore, they examine audit fee premiums by including a dummy variable of Big N auditors in an ordinary least squares (OLS) regression. However, as pointed out by Copley, Gaver, and Gaver (1995), the selection of Big N auditors is endogenous, and the use of OLS to investigate audit fee premiums is inappropriate. Their argument is supported by Chaney et al. (2004; 2005), who find that, using a two-stage Heckman model (Heckman 1979; 1237

6 Lee 1979), audit fee premiums for small UK firms as well as for U.S. firms (in 2001) revealed by the OLS regression, disappear. Given the mixed findings reported by previous studies and the recent finding of a lack of audit fee premiums using U.S. market data (2001) by Chanel et al. (2005), we expect no audit fee premiums prior to the Andersen dissolution if the selectivity bias is considered. Furthermore, Chan et al. (2001) argue that the equilibrium audit fees charged to a prospective client are affected by the production costs of the next closest audit firm in a competitive market. This suggests that audit firms must respond to market competition. The dissolution of Andersen significantly increased the concentration of the top-tier audit market, which could reduce competition. Also, accelerated filers (more than $75M market capitalization) were the first group that was required to file SOX Section 404 reports. Since the great majority of the accelerated filers selected Big 4 auditors, this significantly increased the demand for Big 4 audit services. Taken together, the lowered competition and the increased demand for audit services give Big 4 auditors higher negotiation power and allow them to consider the opportunity costs of their resources in setting higher price for a given client (Sullivan 2006). As such, we expect that Big 4 auditors charge audit fee premiums over non-big 4 auditors after The above discussion leads to the following hypothesis: H1: No Big N audit fee premium exists before the demise of Arthur Andersen and the passage of the Sarbanes-Oxley Act, but it does emerge afterwards. The business community has long classified public accounting firms into Big N and non-big N. Some studies have used a three-tier classification to examine audit fee premiums (Francis and Simon 1987; Gist 1995), economy of scale in producing audit services (Danos and Eichenseher 1982), the relation between audit quality and agency conflicts (DeFond 1992), and competition between middle-tier and top-tier auditors (Doogar et al. 2004). Francis and Simon (1987) report that while audit fee premiums exist between Big 8 and national (secondtier) auditors, they do not exist between national and local/regional (third-tier) audit firms. In addition to Francis and Simon (1987), Gist (1995) identifies fee premiums between the second-tier and third-tier auditors, but not between Big Eight and the second tier auditors. However, his conclusions are based on a small segment of the audit market. Over the past two decades, the audit market has changed dramatically due to the mergers and the failure of Andersen. The number of top-tier accounting firms is down from eight to four, which may have reduced competition among them. Over the past years, the second-tier audit firms have improved their service and audit quality by making investment in technologies and employees training. The rapid development of the second-tier audit firms has helped them to earn a reputation that is close to Big N audit firms. Gist (1995) attributes the non-existence of audit fee premiums between Big N and the second-tier auditors to their equivalent value in providing audit services to small auditees. Furthermore, in the audit market reshuffling process, there was quite a large overlap in the clients that Big 4 and the second-tier audit firms compete for. Evidence has shown that the second-tier audit firms are competing with Big 4 audit firms for larger clients after the demise of Andersen. For instance, Grant Thornton has experienced more than 30 percent growth each year since Consequently, there has been a growing voice favoring the inclusion of Grant Thornton in the Big N group due to its comparable audit quality and service and the small number of Big 4 audit firms. On this basis, we predict that audit fee premiums do not exist between Big N and the second-tier audit firms. 1238

7 H2: There is no audit fee premium between Big 4 and the second-tier auditors even after the demise of Arthur Andersen and the enactment of SOX. Before the demise of Arthur Andersen, there were five big audit service suppliers, so less attention has been paid to non-big 5 auditors, and the gap between the second and the third tier auditors were not prominent. Thus we expect that there was no audit fee premium charged by the second-tier auditors over the third tier auditors before After 2002, with the collapse of Arthur Andersen and the passage of SOX, the number of big audit service suppliers has been reduced, and the demand for audit services has increased. The above changes have resulted in a quick growth of the second tier auditors and enabled them to possess the ability to charge similar rate of audit fees to that of Big 4 auditors, which leads us to predict the existence of audit fee premiums between the second- and third-tier audit firms. So our third hypothesis is described as follows. H3: There was no audit fee premium of the second-tier auditors over the third tier auditors before 2002, but it would emerge after that. III. Sample selection and research design III.1 Sample Selection To estimate audit fee premiums, we obtain audit fees and auditor information from Audit Analytics, firms financial information from Compustat, and daily stock returns from CRSP[7]. During our sample period between 2000 and 2005, we obtain 68,790 observations from Audit Analytics. When we merge the data from Audit Analytics with the data from Compustat and CRSP, we delete 27,684 observations without matching firm/years in these two databases (see Table 1). Following prior studies [8], we exclude financial firms (SIC code between 6,021 and 6,799) because the pricing structure of financial firms is different from other industries (Simunic 1980) and the financial ratios of financial firms are quite different from other firms[9]. As a result, we delete another 7,627 observations. Table 1: Sample Selection Procedure Total observations from Audit Analytics 68,790 Less: Observations without matching firms in Compustat and CRSP (27,684) Financial firms (7,627) Subsidiaries (743) Observations with missing values in financial information in Compustat (7,726) Duplicated observations in audit fees (1,385) Observations with missing values in CRSP (6,805) Final Sample 17,820 Three databases are used in this study: Audit Analytics, Compustat and CRSP. Following prior studies and based on the purpose of this study, the above sample selection procedure is employed. From Audit Analytics, 68, 790 observations are retrieved from 2000 to After merging with Compustat and CRSP, 27, 684 OBS without matching firm/years are lost from the sample. Following prior studies, 7,627 financial firm/years are excluded. 743 subsidiaries are further deleted because they are usually not independent decision makers of auditor selection. Observations with missing values in Compustat (7,726) and CRSP (6,805) are excluded because they can not be used in the regressions. Besides, because there are duplicate observations in audit fees, especially for the years of auditor switches, after special manipulation, 1,385 duplicate observations are excluded from the sample. After the above selection procedure, there are 17,820 observations left in the final sample in the six-year research period. 1239

8 Furthermore, we delete 743 subsidiaries that normally follow their parent firms selection of auditors and are not the decision-makers in auditor selection. Also, due to missing values from Compustat, we lose 7,726 observations. When firms switch auditors, they are listed more than one time in that year. To avoid duplication, we further delete 1,385 observations and keep only the engaged auditors. After excluding 6,805 observations with missing values from CRSP, our final sample consists of 17,820 observations. III.2 Treatment Effects Model As discussed earlier, to examine audit fee premiums by adding a dummy variable of auditor size in the traditional OLS, the dummy variable should be exogenous, namely, clients are randomly assigned across Big N and non-big N auditors. However, if the dummy variable is endogenous, the estimates of the OLS model will be biased, and the OLS model is not appropriate to examine audit fee premiums. Therefore, we use a two-step treatment effects model (Wooldridge 2002; Green 2003) to correct the selectivity bias. Detailed descriptions of the treatment effects model are as follows. To describe the treatment effects model, we start from the traditional OLS model with a dummy variable of Big N. y = x β + δbig _ N + v (1) i i i Here y i is the dependent variable, i.e., audit fees in this study, xi is a set of audit fee determinants, BIG_N is the indicator variable of Big N auditors, β and δ are the coefficients of x i and BIG_N respectively. Because auditor selection is potentially endogenous, we have the following auditor selection equation: * AUD = γ + ε (2) i z i i * * Where AUD i is a latent variable and BIG_N =1 if AUD i >0, and BIG_N is equal to 0 otherwise; z i is the set of auditor selection determinants and γ is the set of their coefficients. We assume that v i and ε i are normally distributed. Combining the two equations and taking the expectation of equation (1) conditional on all independent variables and BIG_N, for the Big N group, we get E y BIG _ N 1, x, z = x β + δ + E[ v z γ + ε > 0 [ ] ] i i = i i i i i i φ( ziγ ) = xiβ + δ + ρσ v, (3) Φ( ziγ ) and for the non-big N group, φ ( ziγ ) E[ yi BIG_ Ni = 0, xi, zi ] = xiβ + ρσv. (4) 1 Φ( z γ ) φ( ziγ ) φ( ziγ ) Where and are the selectivity terms for the Big N and non-big N groups, Φ( ziγ ) 1 Φ( ziγ ) respectively. They are from the truncated normal distribution. Where φ ( ) and Φ ( ) are the probability density function and distribution function for normal distribution respectively. Equation (3) and (4) illustrate how to run the two-step regression of the treatment effects model. In details, we run a probit regression of the auditor selection equation to get the estimates and then calculate the selectivity terms which are included in the second step. Then we run the OLS regression including a selectivity term (lambda) in the model to correct the selectivity bias in the audit fee determination equation. i 1240

9 The difference in the expected audit fees between selecting Big N and non-big N firms is E [ y i BIG_ N i = 1, x i, z i ]- E [ y i BIG_ N i = 0, x i, z i ] = φ( γ ) δ + zi ρσ v Φ( ziγ )(1 Φ( ziγ )) (5) As seen in equation (5), the audit fee difference between selecting Big N and non-big N firms is decided not only by the coefficient of the dummy variable BIG_N, but also by the selectivity term and its coefficient. Also from equation (3) and (4), if the error term ε in the selection equation is independent of the error term in the audit fee regression equation, then the selectivity term will be zero and the estimated β will be unbiased. In this regard, we can conclude that the client firms are randomly assigned to audit firms, and the traditional OLS method is appropriate for examining Big N audit fee premium. However, if the error term ε in the selection model is correlated with the error term in the audit fee regression equation, there is a selectivity bias. In this study, the first step of the treatment effects model (i.e., the selection equation) is to use the following probit regression to determine the likelihood of selecting a Big N auditor. Also, to see the trend of audit fee premiums along the time line of the changing audit environment, we run the regression models by years: _ N = γ + γ LOGASSET + γ ASSET TURN + γ ROA + γ DA + γ QUICK BIG _ γ 6 INVENTORY + γ 7 RECEIVABLE + γ 8 NO _ EXPERTISE + γ 9SEGMENTS + γ 10 LOSS + γ 11FOREIGN _ SALES + γ 12 NET _ LIABILITY + γ 13OPINION + ε All the estimates of γ s are acquired by running the above regression and then the selectivity terms (lambda) for Big N and non-big N clients can be calculated, respectively. In the second step, we run the following audit fee determination equation: LOGFEE = β 0 + β1logasset+ β2sq _ LOGASSET+ β3asset _ TURN + β4roa+ β5da + β QUICK + β INVENTORY + β RECEIVABLE + β BUSY _ SEASON 6 + β SEGMENTS + β LOSS + β FOREIGN _ SALES + β NET _ LIABILITY 10 + β OPINION + β SWITCH + β BIG _ N + β LAMBDA + v where: LOGFEE=Natural logarithm of audit fees LOGASSET=Natural logarithm of total assets SQ_LOGASSET=The square of LOGASSET ASSET_TURN=Asset turnover; sales divided by total assets ROA=Return on assets DA=Long-term debts to total assets ratio QUICK=Quick ratio INVENTORY=Inventory to total assets ratio RECEIVABLE=Receivables to total assets ratio BUSY_SEASON=1 if a company s fiscal year end falls between December 1st and March 31st, which is the normal busy season for auditors, and 0 otherwise SEGMENTS=The number of industry segments of a firm LOSS=1 if net income before extraordinary items is less than zero, and 0 otherwise FOREIGN_SALES=Foreign sales as a percentage of total sales NET_LIABILITY=1 if a company s total liabilities are bigger than its total assets, and 0 otherwise OPINION=1 if a company receives a qualified audit opinion, and 0 otherwise NO_EXPERTISE=1 if a company s audit committee has no financial expert, and 0 otherwise SWITCH=1 if a firm changes its auditor in a year, and 0 otherwise BIGN=1 if a firm selects one of Big N auditors, and 0 otherwise LAMBDA=Inverse mills ratios calculated from the auditor selection equation

10 We first discuss the explanatory variables used in the audit fee determination equation, followed by a discussion of the additional variables in the auditor selection equation. Consistent with prior studies, we expect that client firm size, complexity and risk will affect both auditor choices and audit fees. To control for audit effort, we include the natural logarithm of total assets (LOGASSET) and asset turnover (ASSET_TURN) in our model. A nonlinear relationship may exist between audit fee and client size (Francis and Simon 1987; Chaney et al. 2005), and therefore, we add the quadratic term of LOGASSET (SQ_LOGASSET) to the model[10]. We will explain why this variable is added later. In addition, we control for audit risk by including variables for profitability (ROA), long-term (DA and NET_LIABILITY) and short-term financial structure (QUICK), and highly risky assets (RECEIVABLE, INVENTORY). Other control variables that capture the effects of firm and auditor characteristics on audit fees include SEGMENT, FOREIGN_SALES, OPINION, SWITCH, and BIG_N. In Chaney et al. (2004), the independent variables of the auditor selection equation are a strict subset of the independent variables in the audit fee determination equation. Although the audit fee determination equation can be identified because of the nonlinear function of the selectivity terms, for better identification, the model should include at least one variable in the auditor selection equation different from the variables in the audit fee determination equation (Maddala 1983). Ireland and Lennox (2002) include two variables which are the determinants of auditor choice but not of audit fees in the auditor selection model in order to provide power for the selectivity test. The variables they use are the proportion of board members who are non-executives and board members former affiliations with audit firms. As such, we include NO_EXPERTISE (company s audit committee has no financial expert) in the auditor selection equation since it may affect the auditor selection but not the magnitude of audit fees. IV. Empirical Results IV.1. Descriptive Statistics Table 2 shows the means and standard deviations for the whole sample, Big N and non-big N clients. As expected, ASSET and FEE of Big N clients are significantly greater than those of non-big N clients. Compared to non-big N clients, Big N clients have higher DA, more business segments and higher FOREIGN_SALES, as well as lower ASSET_TURN, INVENTORY, RECEIVABLE, LOSS, NET_LIABILITY, and SWITCH. Furthermore, on average, Big N clients have higher BUSY_SEASON and OPINION than do non-big N clients. However, there are no significant differences for ROA and QUICK between Big N and non-big N clients. There is a low percentage of audit committees without experts (NO_EXPERTISE) for Big N clients than non-big N clients. 1242

11 Table 2: Descriptive Statistics of the Sample Whole sample Big N Non-Big N No. of obs 17,820 14,876 2,944 Mean S.D. Mean S.D. Mean S.D. t-value FEE a LOGFEE ASSET a LOGASSET ASSET_TURN ROA DA QUICK INVENTORY RECEIVABLE SEGMENTS LOSS FOREIGN_SALES NET_LIABILITY OPINION BUSY_SEASON NO_EXPERTISE b SWITCH ACC_FILER c Means and stardard deviations of variables are separately calculated for the whole sample, Big N and non-big N sub-samples. a Fees and assets are in million U.S. dollars. b NO_EXPERTISE is excluded in 2005 because it is not available in Audit Analysis. There are 15,336 observations from 2000 to 2004, where 12,955 for the Big N group and 2,381 for the non-big N group. c The statistics of ACC_FILER are only for year 2004 and And there are 6,344 observations (4,954 for the Big N group and 1,390 for the non-big N group). Variable definitions: FEE=Audit fees a firm pays to its auditor; LOGFEE=Natural logarithm of audit fees; ASSET=Total assets at a fiscal year end; LOGASSET=Natural logarithm of total assets; ASSET_TURN=Asset turnover; sales divided by total assets; ROA=Return on assets; DA=Long-term debts to total assets ratio; QUICK=Quick ratio; INVENTORY=Inventory to total assets ratio; RECEIVABLE=Receivables to total assets ratio; SEGMENTS=The number of industry segments in which a firm operates; LOSS=1 if net income before extraordinary items is less than zero, and 0 otherwise; FOREIGN_SALES=Foreign sales as a percentage of total sales; NET_LIABILITY=1 if a company s total liabilities are bigger than its total assets, and 0 otherwise; OPINION=1 if a company receives a qualified audit opinion, and 0 otherwise; BUSY_SEASON=1 if a company s fiscal year end falls between December 1st and March 31st, which is the normal busy season for auditors, and 0 otherwise; NO_EXPERTISE=1 if a company s audit committee has no financial expert, and 0 otherwise; SWITCH=1 if a firm changes its auditor in a year, and 0 otherwise; ACC_FILER=1 if a firm is an accelerated filer (market capitalization is greater than 75 million U.S. dollars), and 0 otherwise. 1243

12 Table 3: Pearson Correlations B C D E F G H I J K L M N O P Q ACC_FILER b LOGFEE A a LOGASSET B BIG_N C ASSET_TURN D ROA E DA F QUCIK G INVENTORY H RECEIVABLE I SEGMENTS J LOSS K FOREIGN_SALES L NET_LIABILITY M OPINION N BUSY_SEASON O NO_EXPERTISE P SWITCH Q a The letters in the first row of the table represent the same variables in the first column as the corresponding letters specified in the second column b The correlation coefficients between ACC_FILER and other variables are only for year 2004 and Variable definitions: LOGFEE=Natural logarithm of audit fees; LOGASSET=Natural logarithm of total assets; BIG_N=1 if a firm selects one of Big N as its auditor, and 0 otherwise; ASSET_TURN=Asset turnover; sales divided by total assets; ROA=Return on assets; DA=Long-term debts to total assets ratio; QUICK=Quick ratio; INVENTORY=Inventory to total assets ratio; RECEIVABLE=Receivables to total assets ratio; SEGMENTS=The number of industry segments in which a firm operates; LOSS=1 if net income before extraordinary items is less than zero, and 0 otherwise; FOREIGN_SALES=Foreign sales as a percentage of total sales; NET_LIABILITY=1 if a company s total liabilities are bigger than its total assets, and 0 otherwise; OPINION=1 if a company receives a qualified audit opinion, and 0 otherwise; BUSY_SEASON=1 if a company s fiscal year end falls between December 1st and March 31st, which is the normal busy season for auditors, and 0 otherwise; NO_EXPERTISE=1 if a company s audit committee has no financial expert, and 0 otherwise; SWITCH=1 if a firm changes its auditor in a year, and 0 otherwise; ACC_FILER=1 if a firm is an accelerated filer (market capitalization is greater than $75 million), and 0 otherwise. 1244

13 IV.2. Correlations Table 3 presents correlation coefficients among variables used in the regressions. The correlation coefficient between LOGFEE and LOGASSET (0.799) is the most notable, indicating that firm size is the most significant determinant of audit fees. Of further interest are the positive correlations between LOGFEE and SEGMENTS (0.381) and FOREIGN_SALES (0.322), reflecting the complexity of auditing a firm. Also, the high positive correlation between LOGASSET and BIG_N (0.422) indicates the inclination of large firms to select Big N as their auditors. Overall, firm size is positively correlated with DA, SEGMENTS and FOREIGN_SALES and negatively related with QUICK and LOSS. Furthermore, ASSET_TURN is positively correlated with INVENTORY and RECEIVABLE. IV.3. Audit Fee Premiums IV.3.1. H1 Big 4 vs. non-big 4 auditors Following most of the prior studies on audit fee premiums, we also run OLS regressions with BIG_N as a dummy variable. The regression results are shown in Table 4. Our results show that the coefficients of BIG_N are positive and highly significant across all six years. Furthermore, using Simon and Francis (1988) s economic interpretation for the coefficients on BIG_N dummy, the results suggest audit fee premiums of 11.84%, 21.45%, 32.84%, 34.74%, 67.59%, and 59.49% from 2000 to 2005, respectively. Comparable to most of the prior studies on audit pricing, our model explains more than 75% of variations in audit fees across the six years. Our OLS results reveal a nonlinear relation between log audit fees and log firm size from 2000 to 2003 (firm size and its quadratic term are negative and positively significant respectively), which is different from Francis and Simon (1987). 1245

14 The OLS Model: Table 4: Traditional OLS regression LOGFEE= β 0 + β1logasset+ β2sq_ LOGASSET+ β3asset_ TURN + β4roa+ β5da+ β6quick+ β7inventory+ β8receivable+ β9busy_ SEASON + β 10SEGMENTS + β11loss + β12foreign _ SALES + β13net _ LIABILITY + β14opinion + β15switch + β16big _ N + v Year 2000 Year 2001 Year 2002 Year 2003 Year 2004 Year 2005 Exp. Sign Coeff. p-value Coefficient p-value Coeff. p-value Coeff. p-value Coeff. p-value Coeff. p-value INTERCEPT < < < < < LOGASSET? < < < < < <.0001 SQ_LOGASSET? < < < < ASSET_TURN ROA < < < < <.0001 DA? <.0001 QUICK < < < < INVENTORY RECEIVABLE < < < < < <.0001 BUSY_SEASON < SEGMENTS < < < < < <.0001 LOSS < < < < <.0001 FOREIGN_SALES < < < < < <.0001 NET_LIABILITY < < < <.0001 OPINION < < < SWITCH? < < <.0001 BIG_N < < < < <.0001 Adj-R No. of Observations Variable definitions: LOGFEE=Natural logarithm of audit fees; LOGASSET=Natural logarithm of total assets; BIG_N=1 if a firm selects one of Big N as its auditor, and 0 otherwise; ASSET_TURN=Asset turnover; sales divided by total assets; ROA=Return on assets; DA=Long-term debts to total assets ratio; QUICK=Quick ratio; INVENTORY=Inventory to total assets ratio; RECEIVABLE=Receivables to total assets ratio; SEGMENTS=The number of industry segments in which a firm operates; LOSS=1 if net income before extraordinary items is less than zero, and 0 otherwise; FOREIGN_SALES=Foreign sales as a percentage of total sales; NET_LIABILITY=1 if a company s total liabilities are bigger than its total assets, and 0 otherwise; OPINION=1 if a company receives a qualified audit opinion, and 0 otherwise; BUSY_SEASON=1 if a company s fiscal year ends between December 1st and March 31st, which is the normal busy season for auditors, and 0 otherwise; SWITCH=1 if a firm changes its auditor in a year, and 0 otherwise. 1246

15 We use the treatment effects model to identify if there are audit fee premiums charged by Big 4 over non-big 4 auditors and the results are summarized in Table 5. As seen from Panel A, LOGASSET is the most important determinant of auditor selection across the years. This suggests that the larger the firm is, the more likely it is to select Big N auditors, which is intuitively appealing and consistent with Chaney et al. (2004; 2005). It is also consistent with the high correlation between Big N and LOGASSET (0.422) in Table 3. In addition, the higher QUICK is, the more likely it is for firms to select Big N auditors (from 2002 to 2005). As expected, we observe that firms with a higher percentage of foreign sales tend to select Big N auditors (except for 2004) and firms with higher DA are less likely to select Big N auditors (except for 2002). It is likely that debt holders are able to monitor management closely, which reduces the agency cost. Moreover, firms without experts on the audit committees tend not to select Big N auditors. Following Chaney et al. (2004), we use a cutoff level of 50 percent to assess the accuracy of our auditor selection prediction. As shown in Panel A, the accuracy rate of our model prediction of auditor selection is very high (at least of 93%). Furthermore, Big N audit firms have declining market share (from 89.15% in year 2000 to 76.24% in year 2005)[11]. Regarding the second step of the treatment effects model (audit fee determination model), Panel B shows that the coefficients of the selectivity terms (lambda) are statistically significant for all the years (except for year 2002 at 10% level). These results suggest that there is a selectivity bias between Big N and non-big N auditors, and that the OLS model does not hold since it assumes the dummy variable BIG_N is an exogenous variable. Note that we add a quadratic term of LOGASSET (SQ_LOGASSET) in the audit fee determination regressions[12]. As seen in Table 5, the coefficient of LOGASSET is negative but the coefficient of SQ_LOGASSET is positive. Ceteris paribus, the relation between LOGFEE and LOGASSET is a convex curve. Also, the curve is quite flat when LOGASSET is small, and LOGFEE marginally decreases when the firm size increases and it will decrease until it passes the lowest point of the curve. The non-linear relation between LOGFEE and LOGASSET suggests that when firm size increases to a certain level, audit fees will increase more sharply. Basically our test shows that the maximum total assets do not reach the steeply increasing stage of the curve. It is possible that when the firm size increases to a certain level, its complexity enlarges quickly, especially for firms with foreign operations. Consequently, large firms demand more audit efforts and advanced techniques. As predicted, we find that LOGFEE has positive and significant associations with RECEIVABLE, SEGMENTS, LOSS, FOREIGN_SALES, NET_LIABILITY (except for 2004), and OPINION (except for 2001), and a negative and significant association with QUICK across all six years. The magnitudes of the coefficients of ROA are too small to attribute a meaningful effect on audit fees. Furthermore, the associations between LOGFEE and ASSET_TURN, DA, INVENTORY and BUSY_SEASON are not consistently significant across years. 1247

16 Panel A Step 1: Auditor selection (Dependent variable---bign) Table 5: Treatment Effects Model of the Determination of Audit Fees The Probit Model: BIG N = γ + γ LOGASSET+ γ ASSET_ TURN+ γ ROA+ γ DA+ γ QUICK+ γ INVENTORY+ γ RECEIVABLE+ _ γ 8 + γ 9 SEGMENTS+ γ 10LOSS + γ 11FOREIGN_ SALES+ γ 12NET _ LIABILITY+ γ 13OPINION+ ε NO_ EXPERTISE Year 2000 Year 2001 Year 2002 Year 2003 Year 2004 Year 2005 Exp. Sign Coeff. p-value Coeff. p-value Coeff. p-value Coeff. p-value Coeff. p-value Coeff. p-value LOGASSET ASSET_TURN ROA? DA? QUICK? INVENTORY RECEIVABLE NO_EXPERTISE? SEGMENTS LOSS FOREIGN_SALES NET_LIABILITY? OPINION? INTERCEPT % Correctly Classified % Selecting Big N (continued on next page) 1248

17 Table 5 (Continued) Panel B Step 2: Audit fee determination (Dependent variable--logfee) The OLS Model: LOGFEE= β 0 + β1logasset+ β2sq_ LOGASSET+ β3asset_ TURN + β4roa+ β5da+ β6quick+ β7inventory+ β8receivable+ β9busy_ SEASON + β 10 SEGMENTS + β11loss + β12foreign _ SALES + β13net _ LIABILITY + β14opinion + β15switch + β16big _ N + β17 LAMBDA + v Year 2000 Year 2001 Year 2002 Year 2003 Year 2004 Year 2005 Exp. Sign Coeff. p-value Coefficient p-value Coeff. p-value Coeff. p-value Coeff. p-value Coeff. p-value INTERCEPT LOGASSET? SQ_LOGASSET? ASSET_TURN ROA DA? QUICK INVENTORY RECEIVABLE BUSY_SEASON SEGMENTS LOSS FOREIGN_SALES NET_LIABILITY OPINION SWITCH? BIG_N? LAMBDA? (continued on next page) 1249

18 Table 5 (continued) Panel C: Counterfactual effects between Big N and non-big N clients E (LOGFEE-LOGALT_FEE) for Big N clients E (LOGFEE-LOGALT_FEE) for non-big N clients Year 2000 Year 2001 Year 2002 Year 2003 Year 2004 Year 2005 Mean Diff. t-value Mean Diff. t-value Mean Diff. t-value Mean Diff. t-value Mean Diff. t-value Mean Diff. t-value Panel D: Counterfactual effects between Big N and SECOND clients E (LOGFEE-LOGALT_FEE) for Big N clients E (LOGFEE-LOGALT_FEE) for SECOND clients Year 2000 Year 2001 Year 2002 Year 2003 Year 2004 Year 2005 Mean Diff. t-value Mean Diff. t-value Mean Diff. t-value Mean Diff. t-value Mean Diff. t-value Mean Diff. t-value Variable definitions: BIG_N=1 if a firm selects one of Big N as its auditor, and 0 otherwise; LOGFEE=Natural logarithm of audit fees; LOGASSET=Natural logarithm of total assets; SQ_LOGASSET=The quadratic term of LOGASSET; ASSET_TURN=Asset turnover; sales divided by total assets; ROA=Return on assets; DA=Long-term debts to total assets ratio; QUICK=Quick ratio; INVENTORY=Inventory to total assets ratio; RECEIVABLE=Receivables to total assets ratio; NO_EXPERTISE=1 if a company s audit committee has no financial expert, and 0 otherwise; SEGMENTS=The number of industry segments in which a firm operates; LOSS=1 if net income before extraordinary items is less than zero, and 0 otherwise; FOREIGN_SALES=Foreign sales as a percentage of total sales; NET_LIABILITY=1 if a company s total liabilities are bigger than its total assets, and 0 otherwise; OPINION=1 if a company receives a qualified audit opinion, and 0 otherwise; BUSY_SEASON=1 if a company s fiscal year end falls between December 1st and March 31st, which is the normal busy season for auditors, and 0 otherwise; SWITCH=1 if a firm changes its auditor in a year, and 0 otherwise; LAMBDA=Selectivity term, the inverse mill ratios derived from the probit regression. 1250

19 In the treatment effects model, we cannot simply look at the coefficient of BIG_N to infer whether there is a Big N premium since the audit fee premium is jointly determined by both the coefficient of BIG_N and the selectivity term. In this regard, we follow Chaney et al. (2004) to use the counterfactual estimation method (Maddala 1983) by comparing the actual audit fees firms pay (LOGFEE) to Big N (non-big N) auditors and the fees they would have paid had they chosen non-big N (Big N) auditors (LOGALT_FEE)[13]. A positive mean difference suggests that Big N (non-big N) auditors charge higher audit fees than those hypothetically charged by non-big N (Big N) auditors had Big N (non-big N) clients made an alternative choice of auditors. In Table 5, Panel C summarizes the counterfactual effect results between Big N and non-big N clients. As seen in Panel C, in 2000 and 2001, the mean differences between the LOGFEE and LOGALT_FEE for both Big N and non-big N clients are negative and significant, suggesting that there are no audit fee premiums in 2000 and 2001, and that firms select auditors cost-efficiently. This finding is consistent with Chaney et al. (2005) using only the year 2001 data. Furthermore, we observe a reverse trend of audit fee premiums for Big 4 clients from year 2002 to 2005, i.e., positive mean differences between LOGFEE and LOGALT_FEE. Noticeably, the signs of mean difference for non-big N clients are negative and significant over the six years, suggesting that non-big N clients are cost conscious by staying with their non-big N auditors. These results support our first hypothesis[14]. IV.3.2. H2, H3 Audit fee premiums along three tiers of auditors We further partition non-big N auditors into the second-tier (hereinafter SECOND)[15] and third-tier (hereinafter THIRD) groups and examine if there is a selectivity bias between Big N and SECOND auditors and between SECOND and THIRD auditors. We run a treatment effects model by using the subsample of Big 4 and SECOND client firms and find a selectivity bias[16]. As seen in Panel D of Table 5, the results of the counterfactual effects estimation reveal negative mean fee differences between LOGFEE and LOGALT_FEE for Big N and SECOND clients. Also, all the mean differences are significant except for SECOND clients in 2000 and 2003, and Big 4 clients in These results suggest that Big N auditors do not charge audit fee premiums and that both Big N and SECOND clients select their auditors cost-efficiently, or at the very least, not cost-inefficiently. This finding is especially apparent for SECOND clients in both 2004 and 2005, i.e., after the implementation of Section 404 of SOX. As such, our second hypothesis is supported. We also run a treatment effects model for SECOND and THIRD client firms but find no selectivity bias[17]. Therefore, we include a dummy variable SECOND (equal to 1 if a firm s auditor is one of the SECOND auditors, and 0 otherwise) in the OLS regressions to examine whether the second-tier auditors charge audit fee premiums over the third tier auditors. The regression results are shown in Table 6. We find that the coefficients of SECOND are positive and significant across the years, indicating that SECOND auditors charge fee premiums over THIRD auditors. Also, these coefficients increase from (in 2000) to (in 2004), which suggests that SECOND auditors charge increasing fee premiums over THIRD auditors over the years. The higher fees charged by the second-tier auditors may partly explain why a number of former Big 4 clients switched to the third-tier auditors. Therefore, our results partially support the third hypothesis [18]. 1251

20 Table 6: Regression of Audit Fees of the Two-tier non-big N Auditors and Auditor Switching Effects (Dependent variable--logfee) Year 2000 Year 2001 Year 2002 Year 2003 Year 2004 Year 2005 Coeff. p-value Coeff. p-value Coeff. p-value Coeff. p-value Coeff. p-value Coeff. p-value INTERCEPT < < < LOGASSET SQ_LOGASSET < < < ASSET_TURN ROA DA QUICK INVENTORY RECEIVABLE SEGMENTS LOSS < < <.0001 FOREIGN_SALES < <.0001 NET_LIABILITY OPINION BUSY_SEASON SECOND < < < < <.0001 BIGN_TO_ SECOND <.0001 BIGN_TO_THIRD < < <.0001 SECOND_TO_THIRD <.0001 THIRD_TO_SECOND SECOND_TO_SECOND THIRD_TO_THIRD AA_TO_SECOND AA_TO_THIRD No. of OBS Adj R (continued on next page) 1252

21 Table 6 (continued) Variable definitions: LOGFEE=Natural logarithm of audit fees; LOGASSET=Natural logarithm of total assets; SQ_LOGASSET=The quadratic term of LOGASSET; ASSET_TURN=Asset turnover; sales divided by total assets; ROA=Return on assets; DA=Long-term debts to total assets ratio; QUICK=Quick ratio; INVENTORY=Inventory to total assets ratio; RECEIVABLE=Receivables to total assets ratio; SEGMENTS=The number of industry segments in which a firm operates; LOSS=1 if net income before extraordinary items is less than zero, and 0 otherwise; FOREIGN_SALES=Foreign sales as a percentage of total sales; NET_LIABILITY=1 if a company s total liabilities are bigger than its total assets, and 0 otherwise; OPINION=1 if a company receives a qualified audit opinion, and 0 otherwise; BUSY_SEASON=1 if a company s fiscal year end falls between December 1st and March 31st, which is the normal busy season for auditors, and 0 otherwise; SECOND=1 if a firm s auditor is one of the second-tier auditors, and 0 otherwise; BIGN_TO_SECOND=1 if a firm transferring from a Big N auditor to a second-tier auditor, and 0 otherwise; BIGN_TO_THIRD=1 if a firm transferring from a Big N auditor to a third-tier auditor, and 0 otherwise; SECOND_TO_THIRD=1 if a firm transferring from a second-tier auditor to a third-tier auditor, and 0 otherwise; THIRD_TO_SECOND=1 if a firm transferring from a third-tier auditor to a second-tier auditor, and 0 otherwise; SECOND_TO_SECOND=1 if a firm transferring from one second-tier auditor to another second-tier auditor, and 0 otherwise; THIRD_TO_THIRD=1 if a firm transferring from one third-tier auditor to another third-tier auditor, and 0 otherwise; AA_TO_SECOND=1 if a former Arthur Andersen client transferring to a second-tier auditor, and 0 otherwise in 2002; AA_TO_THIRD=1 if a former Arthur Andersen client transferring to a third-tier auditor, and 0 otherwise in

22 V. Robustness Check V.1. Results Sensitive to the Model Specifications Francis and Lennox (2006) show that the Heckman model used by Chaney et al. (2004) is not only sensitive to multicollinearity, but also sensitive to minor changes in model specifications. They change the model specifications by using different proxies of client size in four model specifications: (1) log total assets (LTA) in the selection equation and log total sales (LTS) in the audit fee equation, (2) LTS in the selection equation and LTA in the audit fee equation, (3) LTA in the selection and both LTA and LTS in audit fee equations, (4) including interaction terms between LTA and all the independent variables in both the selection equation and the audit fee equation. In turn, they argue that OLS regressions are independent of the model specifications and produce consistent results of audit fee premiums. We conduct robustness tests to check if our results are sensitive to model specifications, as pointed out by Francis and Lennox. In model specification 1 we reach nearly the same conclusions on audit fee premiums as our main results in Section 4. In model specification 2, the mean differences between actual fees and alternative fees are negative and insignificant in 2000 but positive and significant (0.039; t=3.81, p=0.0001) in 2001 for Big 5 clients. Other results remain unchanged. In model specification 3 and 4, we obtain insignificant coefficients of the selectivity terms in 2002 and negative and insignificant mean differences of fees for model specification 3 in 2000 and Other results are qualitatively similar to our main conclusions. Therefore, in these two model specifications, we conclude that before 2001, both Big N and non- Big N clients select their auditors cost-efficiently, or at least not cost-inefficiently. All in all, these results show that our main conclusions are basically robust and consistent across the different model specifications. IV.2. The Effect of Accelerated Filers on Auditor Switching Effects In response to mounting complaints and pressure by small companies and foreign private issuers, the SEC requires that only accelerated filers (ACC_FILER), excluding registered investment companies, with fiscal year ending on or after November 15, 2004, file Section 404 reports in the first year[19]. We find a high correlation between LOGFEE and ACC_FILER (0.549 in Table 3). It is likely that accelerated filers correlate with firm size and that these companies are required to have their internal control reports assessed by auditors, which apparently increases audit fees. To examine whether the obligation to produce internal control report explain on audit fees premiums in 2004 and 2005, we include the dummy variable of ACC_FILER in the regression model to re-examine the results. As expected, we find that coefficients of ACC_FILER are positive and significant for all the models. The results show that after including the variable of ACC_FILER in the models, our conclusions on audit fee premiums remain unchanged. To avoid multicollinearity problems caused by the high correlation between LOGASSET and ACC_FILER (the correlation coefficient is 0.610), we do not include this variable in our formal models. Besides, we also test the above influence by only keeping accelerate filers in the sample in 2004 and 2005, and we reach the same conclusion as in the total sample, indicating that our results on audit fee premiums are not driven by the implementation of SOX section

23 VI. Discussion and Conclusions In light of the prominent changes of the market structure and regulations in the U.S. audit market, this study employs the treatment effects model to examine audit fee premiums in the timeline from 2000 through After correcting for the selectivity bias and using counterfactual estimation, we find a lack of Big 5 audit fee premiums over non-big 5 auditors in 2000 and 2001, indicating that both Big 5 clients and non-big 5 clients choose their auditors cost-efficiently. This implies that when companies choose auditors, they consider the fee structures of audit firms as well as their own firm-specific characteristics. As predicted, we observe the presence of Big 4 audit fee premiums after the Andersen failure and the enactment of SOX. Non-Big 4 clients make a cost-efficient decision by staying with their auditors. Our study also further classifies non-big 4 audit firms into the second and third-tiers to examine audit fee premiums. We do not find Big N audit fee premiums over the second-tier auditors across all six years. This finding is opposite to Francis and Simon (1987) while consistent with Gist (1995), which suggests that the second-tier auditors have grown significantly over the years and are able to compete with Big N auditors at similar audit rates. Also, contrary to Francis and Simon (1987) our results show audit fee premiums between the second-tier and third-tier auditors during our sample period. This suggests that the divergence between the second- and third-tier auditors may have increased in recent years. These results suggest that future research on audit pricing should consider the apparent divergence between the second- and third-tier audit firms and cannot assume the presence of homogeneity between these two groups. Similarly, future studies that examine issues concerning brand name reputation of audit firms should consider the similarity between Big 4 and the second-tier audit firms by including the second-tier auditors in the top-tier group. Our robustness checks show that our models do not suffer from sensitivity to minor changes in the model specifications pointed out by Francis and Lennox (2006). We also check whether the missing variable of ACC_FILER drives our results in 2004 and 2005, and find that the conclusions remain the same. It is noteworthy to mention the limitations of this study in interpreting our findings. First, this study mainly focuses on the costs side of audit services, i.e., the cost efficiency of auditor selection. However, from the counterfactual part, namely, from the results that Big N clients would like to make cost-inefficient choice of auditors and pay higher fees, we can infer that these clients need the high quality services from Big 4 auditors. Second, our study does not distinguish whether the audit fee premiums following the demise of Andersen and the enactment of SOX are attributable to monopoly power or to product differentiation of Big 4 auditors. Future research should explore the effects of auditor reputation, monopoly power and product differentiation of Big 4 auditors on audit fee premiums. 1255

24 Notes 1. Big N is a general concept for big audit service providers and refers to Big 8, Big 6, Big 5 or Big 4 in different periods. In this study, the research period covers the change of Big N auditors from 5 to 4. We consistently use Big N to represent either Big 5 or Big 4 before and after the demise of Arthur Andersen in the study. 2. Based on the number of clients, we observe a declining market share of Big 4 audit firms in U.S from 2003 to 2005, but the market share did not change much based on audit fees. However, there has been a concern by regulators and investors in the UK that Big 4 audit firms build an excessive market share (i.e., audit all but one of the companies in the FTSE 100 and 97% of the FTSE 250) (Japson 2006). 3. According to rankings by revenue, Grant Thornton (GT) and BDO Seidman (BDO) are next to Big 4 auditors. GT received the most benefit due to Andersen s demise. Some argue that the U.S. audit market includes Big 4, GT, and then everyone else. Furthermore, Ashbaugh-Skaife, Collins and Kinney (2006) classify GT, BDO and Big 4 audit firms as dominant suppliers. 4. For example, using the Heckman model and the U.S. data of 2001, Chaney et al. (2005) report a mean difference between natural logarithm of actual fees and alternative fees (if hiring the other group of audit firms) of -2.58, which implies that the actual audit fees are only 8 percent of the alternative fees. 5. They criticize that Chaney et al. s (2004) model is sensitive to model specifications (e.g., different proxies of firm size) and results in different conclusions on audit fee premiums. 6. DeAngelo (1981b) defines the quality of audit service as the market-assessed joint probability that a given auditor will both discover a breach in a client s accounting system and report the breach. 7. CRSP is merged into other data for factor analysis in a later study. Our results are not sensitive to the sample selection procedure of whether CRSP is included or not. 8. Simunic (1980) finds that the pricing structure in banking industry is different from other industries. Prior studies exclude this industry from their samples based on Siminic s finding and the argument that the financial ratios in banking are different from other industries (Maher et al. 1992; Lee 1996; Firth 1997; Chaney et al. 2004). 9. In a robustness test, we include financial firms in our models and find qualitatively similar results. 10. Francis and Simon (1987) include a quadratic term of LOGASSETS in their audit fee model to check for a nonlinear relation between LOGFEE and LOGASSETS. 11. However, the market share of Big 4 audit firms (exclude Andersen) has increased from 2000 (69.91%) to 2005 (76.24%). 12. In the treatment effects model, the selectivity term (lambda) is nonlinear (see Equation (3) and (4)). If there is nonlinear relationship between audit fees and other independent variables and we neglect the relationship, the selectivity term is likely to pick up all the nonlinear effects in the second step regression (Maddala 1983, P. 269). Therefore, if we exclude SQ_LOGASSET in the regression, the coefficient estimates (p-values) of lambda are (0.000), (0.000), (0.000), (0.000), (0.017), and (0.010) for 2000 to 2005, respectively. The signs of the coefficients of lambda from 2000 to 2003 are opposite to those in the regression results in Panel B of Table 5, suggesting that there is a significant nonlinear relation between LOGFEE and LOGASSET captured by the selectivity term from 2000 to We also use counterfactual effects to estimate 1256

25 audit fee premiums without including the quadratic term for 2004 and We find qualitatively similar results, suggesting that there may be a model structure change after The LOGALT_FEE is calculated as follows. For Big N clients, we get LOGALT_FEE by substituting the parameters of Big N clients into equation (4), namely, the equation of Non-Big N clients. For non-big N clients, we predict LOGALT_FEE by substituting the parameters of non-big N clients into equation (3), namely, the equation of Big N clients. 14. We also use maximum likelihood estimation to test audit fee premiums and we find qualitatively similar results. 15. We further use another classification of the second-tier auditors by including only Grant Thornton. Our main conclusions do not change. 16. The coefficients (significance levels) of the selectivity term for the Big N and SECOND subsample are (0.003), (0.000), (0.015), (0.036), (0.000), and (0.000) for years 2000 to 2005, respectively. 17. The coefficients (significance levels) of the selectivity term for the SECOND and THIRD subsample are (0.352), (0.575), (0.639), (0.466), (0.177), and (0.257) for year 2000 to 2005, respectively. 18. As expected, our additional analysis shows that Big N auditors charge audit fee premiums over third-tier auditors. 19. Foreign private issuers received a one-year extension until the first fiscal year ending on or after July 15, 2006, to comply with the regulation. Due to the complexity and substantial compliance costs associated with implementation, in September 2005 the SEC further extended the compliance date for non-accelerated filers until References Anderson, T., and Zéghal, D The Pricing of Audit Services: Further Evidence from the Canadian Market. Accounting & Business Research, Summer94, Vol. 24 Issue 95, p Ashbaugh-Skaife, H., D. W. Collins, and W. R. Kinney Jr The discovery and reporting of internal control deficiencies prior to SOX-mandated audits. Research Paper, McCombs School of Business, University of Texas at Austin (ACC-02-05). Asthana, S., S. Balsam, and S. Kim The effect of Enron, Anderson, and Sarbanes-Oxley on the market for audit services. Working Paper, Temple University. Balachandran, B., and R. T. S. Ramakrishnan A theory of audit partnerships: Audit firm size and fees. Journal of Accounting Research 25 (1): Balachandranc B. V., and Simon, D Audit Services and Fees of Large Accounting Firms. Journal of Economics and Management Strategy, Vol. 2, No. 3, Fall 1993, pp Beatty, R. P Auditor reputation and the pricing of initial public offerings. The Accounting Review 64 (4): Chan, D. K., A. Ferguson, D. A. Simunic, and D. Stokes A spatial analysis and test of oligopolistic competition in the market for audit services. Working Paper, Hong Kong University of Science and Technology. Chan, Philip; Ezzamel, Mahmoud; Gwilliam, David., Determinants of Audit Fees For Quoted UK Companies. Journal of Business Finance & Accounting, Nov93, Vol. 20 Issue 6, p Chaney, P. K., D. C. Jeter, and L. Shivakumar Self-selection of auditors and audit pricing in private firms. The Accounting Review 79 (1):

26 Self-selection of auditors and size nonlinearities in audit pricing. Working Paper, Vanderbilt University. Chung, D. Y., and W. D. Lindsay The pricing of audit services: The Canadian perspective. Contemporary Accounting Research 5 (1): Copley, P. A., J. J. Gaver, and K. Gaver Simultaneous estimation of the supply and demand of differentiated audits: Evidence from the municipal audit market. Journal of Accounting Research 33 (1): Craswell, A.T., Francis, J.R., and Taylor, S.L Auditor Brand Name Reputations and Industry Specializations. Journal of Accounting & Economics 20 (Dec.): Danos, P., and J. W. Eichenseher Audit industry dynamics: Factors affecting changes in client-industry market shares. Journal of Accounting Research 20 (2): DeAngelo, L. E. 1981a. Auditor independence, 'Low Balling', and disclosure regulation. Journal of Accounting and Economics 3 (2): b. Auditor size and audit quality. Journal of Accounting and Economics 3 (3): DeFond, M. L The association between changes in client firm agency costs and auditor switching. Auditing: A Journal of Practice and Theory 11 (1): Deis Jr, D. R., and G. Giroux The effect of auditor changes on audit fees, audit hours, and audit quality. Journal of Accounting and Public Policy 15 (1): Doogar, R., N. Fargher, and K. Hong Leveling the playing field, or crumbs from the table? The contestability of the audit market to middle tier firms. Mimeo (December) Ferguson, A., and D. Stokes Brand name audit pricing, industry specialization, and leadership premiums post-big 8 and Big 6 mergers. Contemporary Accounting Research 19 (1): Firth, M Price Setting and the Value of a Strong Brand Name. Intern. J. of Research in Marketing 10 (1993) North-Holland Firth, M An Analysis of Audit Fees and Their Determinants in New Zealand. Auditing: A Journal of Practice and Theory, Vol. 4, No. 2, Spring 1985 Firth, M The provision of nonaudit services by accounting firms to their audit clients. Contemporary Accounting Research 14 (2):1-21. Francis, J. R The effect of firm size on audit prices. Journal of Accounting and Economics 6 (2): Francis, J. R., and C. S. Lennox Pick a model, any model: Assessing the robustness of the self-selection model. Working Paper, University of Missouri-Columbia. Francis, J. R., and D. T. Simon A test of audit pricing in the small-client segment of the U.S. audit market. The Accounting Review 62 (1): Francis, J. R., and D. J. Stokes Audit prices, product differentiation and scale economies: Further evidence from the Australian market. Journal of Accounting Research 24 (2): Francis, J. R., and E. R. Wilson Auditor changes: A joint test of theories relating to agency cost and auditor differentiation. The Accounting Review 63 (4): Gist, W A Test of Audit Pricing in the Small-Client Segment: A Comment. Journal of Accounting, Auditing and Finance. Spring95, Vol. 10 Issue 2, p Greene, W. H Econometric Analysis. 5th ed: Prentice Hall. Hamilton J., Y. Li, and D. Stokes Listed Company Auditor Self-selection Bias and Audit Fee Premiums. Working Paper Heckman, J Sample selection bias as a specification error. Econometrica 47 (1):

27 Hogan, C. E Costs and benefits of audit quality in the IPO market: A self-selection analysis. The Accounting Review 72 (1): Ireland, J.C., and C.S. Lennox The Large Audit Firms Fee Premium: A Case of Selectivity Bias? Journal of Accounting, Auditing and Finance 17 (Winter): Japson, B Dominance of big four audit firms challenged. Financial Times, August 7. Jensen, M. C., and W. H. Meckling Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics 3 (4): Johnson, E. N., Walker, K. B., and Westergaard, E Supplier Concentration and Pricing of Audit Services in New Zealand. Auditing: A Journal of Practice and Theory, Autumn95, Vol. 14 Issue 2, p74-89 Karim, W. and Moizer, P Determinants of Audit Fees in Bangladesh. The International Journal of Accounting, Vol. 31, No 4 pp Langendijk; H The Market for Audit Services in the Netherlands, European Accounting Review, 6(2): Landsman, W. R., K. K. Nelson, and B. R. Rountree An empirical analysis of Big N auditor switches: Evidence from the pre- and post-enron eras. Working Paper, Rice University. Lee, D. S.-y Auditor market share, product differentiation and audit fees. Accounting and Business Research 26 (4): Lee, L.-F Identification and estimation in binary choice models with limited (censored) dependent variable. Econometrica 47 (4): Maddala, G. S Limited-Dependent and Qualitative Variables in Econometrics. New York: Cambridge University Press. Palmrose, Z.-V. 1986a. Audit fees and auditor size: Further evidence. Journal of Accounting Research 24 (1): Pong, C. M., and G. Whittington The determinants of audit fees: Some empirical models. Journal of Business Finance and Accounting 21 (8): Rubin, M. A., Municipal Audit Fee Determinants. The Accounting Review, Apr88, Vol. 63 Issue 2, p Schloetzer, J. D Arthur Anderson, SOX Section 404 and auditor switching: Theory and evidence. Working Paper, University of Pittsburgh. Simon D. T The Market for Audit Services in South Africa. The International Journal of Accounting, (1995), 30: Simon, D. T Additional evidence on the large audit-firm fee premium as an indication of auditor quality. Journal of Applied Business Research 13 (4): Simon, D. T., and J. R. Francis The effects of auditor change on audit fees: Tests of price cutting and price recovery. Accounting Review 63 (2): Simon, D.T. and Taylor M.H A Survey of Audit Pricing in Ireland, International Journal of Auditing, 6: Simon D. T., Teo, S., and Trompeter, G A Comparative Study of the Market for Audit Services in Hong Kong, Malaysia and Singapore, The International Journal of Accounting, (1992) 27: Simunic, D. A The pricing of audit services: Theory and evidence. Journal of Accounting Research 18 (1):

28 Simunic, D. A., and M. T. Stein Product Differentiation in Auditing: Auditor Choice in the Market for Unseasoned New Issues. Research Monograph, 13 ed. Vancouver: The Canadian Certified General Accounts' Research Foundation. Sullivan, M. W Great Migration: How recent events changed that switching behavior of to-tier audit clients. In American Accounting Association, 2007 Midyear Conference. Charleston, SC. Taffler, R. J., and K. S. Ramalinggam The determinants of audit fees in the U.K.: An exploratory study. Working Paper 37, City University Business School, London. Willekens, M., and C. Achmadi Pricing and Supplier Concentration in the Private Client Segment of the Audit Market: Market Power or Competition? International Journal of Accounting 38 (4): Wooldridge, J. M Econometric Analysis of Cross Section and Panel Data. 1st ed. Cambridge MA: MIT Press. Authors Dr. Jiuzhou Wang Norwegian School of Economics, Bergen, Norway Nancy Manzhen Fan Accounting Department, California State Polytechnic University, Pomona, Pomona, CA 91768, USA, 1260

29 The Association among Auditor Litigation, Earnings Management, and IPO Underpricing: A Theoretical Perspective Yan Xiong Caixing Liu Abstract This paper constructs models to examine the relationship between the probability of auditor litigation, earnings management, and IPO underpricing anomaly in the initial public offering context. This paper contributes to the literature in the three ways. First, this paper examines the relationship between auditor litigation and earnings management in the IPO market. The IPO context provides an ideal setting for testing both earnings management and auditor litigation because several unique features related to the IPO market provides exceptional incentives and opportunities for earnings management. Second, this paper proposes three different earnings management measures to test the relationship between auditor litigation and earnings management. In the absence of a perfect earnings management measures, the employment of three different measures may explain the inconclusive results from prior studies. Third, this paper makes the first attempt to propose that there is a positive relationship between auditor litigation and IPO underpricing. Key Words: Auditor litigation; Earnings Management; IPO Underpricing Article Classification: M4 I. Introduction This paper takes the first step to construct models to examine the relationship between the probability of auditor litigation, earnings management, and IPO underpricing anomaly in the initial public offering (hereafter IPO) context. Prior research examines the relationship between auditor litigation and earnings management in the regular market (Stice 1991, Lys and Watts 1994, Heninger 2001, Cahan and Zhang 2006). However, these studies provide conflicting evidence on the relationship. This paper improves prior research in two aspects. First, this paper constructs models to examine the relationship between auditor litigation and earnings management in the IPO market. The IPO context provides an ideal setting for testing both earnings management and auditor litigation because several unique features related to the IPO market provides exceptional incentives and opportunities for earnings management. Second, this paper proposes to test the relationship between auditor litigation and earnings management in the IPO market by employing three different earnings management measure to test the robustness of the results. Although a number of methods have been developed in the literature to obtain proxies for earnings management, no existing method is perfect in measuring earnings management. The inconclusive results from prior studies may result from the use of different measures of earnings management in these studies. This paper proposes that there is a positive relationship between auditor litigation and earnings management in the IPO market. Specifically, the paper proposes that the probability of auditor litigation increases as IPOs report more positive (income-increasing) abnormal accruals. Furthermore, this paper makes first attempt to propose the relationship between auditor litigation and the IPO underpricing anomaly. IPO underpricing is a documented phenomenon in which IPO offer prices are, on average, less than the market prices at the end of the first day of trading. Smith (1986) provides empirical evidence that the returns from initial offer price to the first 1261

30 market closing price have exceeded 15 percent on average. Prior research claims that high quality auditors, although more expensive than low-quality auditors, reduce the total cost of the IPO, by reducing underpricing (Balvers et al. 1988). Prior research also shows that there is a positive relationship between the level of IPO underpricing and the level of earnings management (Xiong et al. 2012). Therefore, this paper further proposes that the level of IPO underpricing is positively related to the probability of auditor litigation. II. Literature Review Public perception of the accounting profession after Enron was low, and was just starting to recover when the subprime and credit crisis hit. Now congressional hearings have been held on how audits can better prevent the next financial crisis. The recent Lehman report raises new questions about the quality of auditing and financial reporting. Litigation against auditors has increased dramatically in recent years. Auditors are usually among the first to be blamed when investors and creditors suffer financial losses (Stice 1991). The effects of litigation include financial damages, reputation losses and even bankruptcy (Palmrose 1988; Dalton et al. 1997).The risk to auditors of being named defendants in securities class actions has been of great concern to both auditors and academia. There is considerable evidence analyzing different characteristics of auditor litigation. Palmrose and Scholz (2000) provide evidence on the restatement circumstances associated with lawsuits against auditors. They classify the accounting issues resulting to restatements as either economic or technical. Economic restatements involve transactions and accounts related to core (recurring) earnings while all other restatements are technical. They examine 416 US restatements of financial statements issued between 1995 and They find that auditors are significantly more likely to be sued over economic restatements than technical ones. Accounting accruals are another subject related to auditor litigation that prior research has focused on. Accounting accruals are subjective estimates of future outcomes and cannot be objectively verified by auditors prior to occurrence. Francis and Krishnan (1999) argue that the difficulty of verifying accounting accruals causes audits of high-accrual firms to pose more uncertainty than audits of low accrual firms because of potential estimation error and a greater chance that high-accrual firms have undetected asset realization and /or going concern problems which are related to the high level of accruals. They investigate whether auditors have a tendency to qualify more frequently in the presence of higher accruals compared with lower accruals. They demonstrate that differential qualification tendencies in the face of accruals would be consistent with the view that auditors consider accruals may be used to distort the earnings figure. Becker et al. (1998) find that clients of non Big 6 firms report discretionary accruals that increase income relatively more than the discretionary accruals reported by client of Big 6 auditors but there is no evidence of differential qualification rates resulting from income increasing as opposed to income decreasing discretionary accruals. This paper is going to concentrate on auditor litigation in IPOs. Auditor litigation in IPOs is more prominent. Section 11 of the Securities Act of 1933 mandates that auditors, underwriters, issuers, and other experts who prepare any part of the registration statement are jointly and severally liable for damages resulting from false or misleading information presented in an initial public offering registration statement. Joint and several liability places the auditors at risk of 1262

31 suffering damages from all activities of all participants involved in the preparation of the IPO registration statement. All IPO participants in the coalition are jointly and severally liable for each others actions. In practice, this means that they were routinely sued for various omissions in the IPO prospectus when the public market valuation fell below the IPO offering price. Different characteristics of auditor litigation in IPOs have been analyzed. Prior research argues that there is a demand for high-quality auditors in the context of IPOs to provide credible information to investors (Balvers et al. 1988). Prior research also indicates that managers have a greater incentive to mislead auditors by manipulating earnings if the company is in financial distress (Lys and Watts 1994) or if managers face the threat of dismissal (DeAngelo 1988). Initial public offerings are primary candidates for earnings management. Several features particular to the IPO process make earnings vulnerable to management. First, there is an information asymmetry problem between investors and issuers at the time of a firm's initial public offering since very little information is disclosed about most IPO firms prior to their going public. Second, accounting regulation APB 20 (1971) gives a firm undertaking an initial public offering special permission to change any or all accounting principles via retroactive restatement of the financial statements presented in the offering prospectus. These features give IPO firms exceptional opportunities to manage their earnings. However, prior research does not provide conclusive evidence on the relationship between earning management and auditor litigation risk (Stice 1991, Lys and Watts 1994, Heninger 2001). Stice (1991) identifies several client and auditor characteristic as being associated with lawsuits against lawsuits. He finds that the asset structures of clients, firms financial conditions, firms market values, and variability of firms returns each influence the likelihood of litigation against auditors. One important specific finding of this paper is that higher levels of accounts receivable and inventory account balance are associated with increased litigation risk, likely because earnings management may affect the levels of these accounts. However, the total levels of these account balances should depend primarily on business fundamentals. Lys and Watts (1994) investigate the relation between audit litigation and total accruals. They report conflicting results in their univariate and multiple regression analyses. Based on univaraite analysis, they report that litigation firms have more income-decreasing accruals than do control firms. However, based on multiple regression analysis, they conclude that litigation is positively associated with income-increasing accruals. Finally, they conclude no relation between auditor litigation and total accruals for their most recent subsample of lawsuits that allege wrongdoing after Heninger (2001) extended Stice (1991) and Lys and Watts (1994) by investigating whether abnormal accruals, estimated using the Modified Jones model (Dechow et al. 1995), are associated with auditor litigation. He found that the risk of auditor litigation increases with more income-increasing abnormal accruals, even after controlling for other variables that affect litigation. This paper extends prior research in two aspects. First, this paper examines the relationship between auditor litigation and earning management in IPO firms. Second, the paper examines the relationship by employing three different earnings management measures. This paper hypothesizes that the higher the level of earnings management the higher the probability of auditor litigation. In addition, IPO underpricing is one of two major anomalies in the IPO market. Many theories have been proposed to explain this anomaly. However, none of these theories have been universally accepted. This paper investigates the relationship between the level of IPO 1263

32 underpricing and the probability of auditor litigation. Prior research claims that high quality auditors, although more expensive than low-quality auditors, reduce the total cost of the IPO, by reducing underpricing (Balvers et al. 1988). Prior research also shows that there is a positive relationship between the level of IPO underpricing and the level of earnings management (Xiong et al. 2008). Therefore, this paper proposes that the level of IPO underpricing is positively related to the probability of auditor litigation. III. Hypotheses Development Prior studies have examined the relationship between auditor litigation and earnings management. Stice (1991) identifies several client and auditor characteristic as being associated with lawsuits against lawsuits. He finds that the asset structures of clients, firms financial conditions, firms market values, and variability of firms returns each influence the likelihood of litigation against auditors. One important specific finding of this paper is that higher levels of accounts receivable and inventory account balance are associated with increased litigation risk, likely because earnings management may affect the levels of these accounts. However, the total levels of these account balances should depend primarily on business fundamentals. Lys and Watts (1994) investigate the relation between audit litigation and total accruals. They report conflicting results in their univariate and multiple regression analyses. Based on univaraite analysis, they report that litigation firms have more income-decreasing accruals than do control firms. However, based on multiple regression analysis, they conclude that litigation is positively associated with income-increasing accruals. Finally, they conclude no relation between auditor litigation and total accruals for their most recent subsample of lawsuits that allege wrongdoing after Heninger (2001) extended Stice (1991) and Lys and Watts (1994) by investigating whether abnormal accruals, estimated using the Modified Jones model (Dechow et al. 1995), are associated with auditor litigation. He found that the risk of auditor litigation increases with more income-increasing abnormal accruals, even after controlling for other variables that affect litigation. This paper extends prior research in two aspects. First, this paper examines the relationship between auditor litigation and earning management in IPO firms. Second, the paper takes the first step to investigate the relationship between auditor litigation and IPO underpricing anomaly. This paper hypothesizes that the higher the level of earnings management and IPO underpricing, the higher the probability of auditor litigation. H1: The probability of auditor litigation in IPO firm s increases as IPO firms report more positive (income-increasing) abnormal accruals. H2: The higher the level of IPO underpricing, the higher of the probability of auditor litigation. IV. A Theoretical Development and Identification of Research Design and Models Two regression models are proposed to investigate the relationship between the dependent variable-the probability of auditor litigation and the independent variables-discretionary accruals (earnings management measures) and IPO underpricing after controlling for other variables. The control variables are identified from prior studies. It is expected that the probability of auditor litigation increases as managers in IPO firms overstate more of their earnings, and also the probability of auditor litigation increases as IPO underpricing increases. 1264

33 Measures of Earnings Management Earnings management cannot be directly measured. However, a number of methods have been used in the literature to obtain proxies for earnings management. These methods include the discretionary accruals method, the total accruals method, the single accrual method, the accounting change method, and the distribution method. Since no existing method is perfect in measuring earnings management, three methods are employed in this study to test the robustness of the earnings management measures. The three methods are the discretionary total accruals method, the discretionary working capital accruals method and the total accruals method. Each of these methods is described in detail below. The discretionary total accruals method has been widely employed in tests of the earnings management hypothesis. The major difficulty involved with using this method is the need to identify and separate total accruals into unmanaged and managed components. The most frequently used models for separating expected and discretionary accruals are the Jones (1991) and modified Jones (Dechow et al. 1995) models. The current study adopts the modified Jones model as the means of decomposing total accruals into their unmanaged and managed components. This method is consistent with Teoh et al. (1998). Each sample IPO firm is first matched with all firms on the Research Insight database with the same three-digit SIC code. Total accruals are regressed on gross property, plant, and equipment and the adjusted change in revenues for each group of control firms matched with a given sample firm. Data for the regression is taken from the fiscal year prior to the sample firm s initial offering and all variables are scaled by beginning of the period total assets. TAC it / TA it-1 = a 0j (1/TA it-1 ) + a 1j ( REV it - REC it )/ TA it-1 + a 2j ( PPE it / TA it-1 ) + e it Where: TAC it = the total accruals (net income before extraordinary items minus cash flow from operations) in year t for the i th control firm; TA it = the total assets in year t for the i th control firm; REV it = the change in revenues from year t-1 to year t for the i th control firm; REC it = the change in receivables from year t-1 to year t for the i th control firm; PPE it = the gross property, plant, and equipment in year t for the i th control firm; and e it = the regression error terms, assumed cross-sectionally uncorrelated and normally distributed with mean zero. The estimated coefficients from the control-firm regressions are then used to estimate the level of managed accruals for each sample IPO firm by subtracting the estimate of unmanaged accruals from total accruals as follows: TAEM j,t = TAC jt / TA jt-1 - a 0j (1/TA jt-1 ) - a 1j ( REV jt - REC jt )/ TA jt-1 - a 2j ( PPE jt / TA jt-1 ) Where: TAEM j,t = the managed component of total accruals for IPO sample firm j in year t, which is equal to discretionary total accruals, and all other variables are as previously determined. There has been considerable discussion of the efficiency of the modified Jones model in detecting earnings management (Kang et al. 1995; Guay et al. 1996; Peasnell et al. 1998). 1265

34 Peasnell et al. (1998) demonstrate that the modified Jones model controls for only a small amount of normal working capital accrual activities. Therefore, a working capital accrual model (Teoh et al. 1998) is also considered in this study. To maintain consistency between the discretionary total accruals model and the working capital model, the change in revenue is adjusted for the change in receivables. The coefficients for unmanaged accruals are obtained by regressing working capital accruals on changes in revenues, cross-sectionally, for each group of control firms matched with a given sample firm. The control firm regression is: WAC i,t / TA i,t-1 = b 0j (1/TA i, t-1 ) + b 1j ( REV it - REC it )/ TA i, t-1 +e i,t Where: WAC i,t = the working capital accruals (sum of changes in inventory, accounts receivable, and other current assets less the sum of changes in accounts payable, income taxes payable, and other current liabilities) in year t for the i th control firm; and the other variables are as described previously. The estimated coefficients from the control-firm regressions are then used to estimate the level of managed working capital accruals for each sample IPO firm by subtracting the estimate of unmanaged working capital accruals from total working capital accruals as follows: WCEM j,t = WAC j,t / TA j,t-1 b 0j (1/TA j,t-1 ) b 1j ( REV jt - REC jt )/ TA j,t-1 Where: WCEM j,t = the managed component of working capital accruals for IPO sample firm j in year t, which is equal to discretionary working capital accruals, and all other variables are as previously described. In the absence of a perfect model for detecting earnings management, a third measure is also employed. Following Healy s theoretical proposal (1985), this third measure is defined as the difference between net income and cash flow from operations. As Warren Buffett, the legendary investor has cautioned investors who are obsessed with earnings, generally accepted accounting principles (GAAP) give managers great discretion in determining reported earnings and managers are likely to utilize this discretion to produce earnings numbers they prefer (Individual Investor Group 2001). Buffett recommends that individual investors use cash flow from operations as a check on the quality of a company s earnings. In situations where there is a large discrepancy between a firm s reported earnings and its cash flow from operations, the firm is likely engaging in a certain degree of earnings management. Thus, an easy way for ordinary investors to identify earnings management is to examine the difference between cash flows from operations and reported earnings. To be consistent with the prior measures, both reported earnings and cash flows from operation are scaled by beginning-of-period total assets. There are some differences between total accruals as measured in this study and the measure used by Healy (1985). Since the statement of cash flows was not available until 1987, Healy (1985) estimated cash flow from operations as working capital from operations (reported in the funds statement) less changes in inventory and receivables and plus changes in payables and income tax payable. Cash flow from operations in this study is obtained directly from the statement of cash flows. The earnings management for firm i during period t is measured as: CFOEM j,t = NI j,t / TA j,t-1 - CFO ijt / TA j,t

35 Where: CFOEM j,t = the managed component of earnings for IPO sample firm j s during period t, which is equal to total accruals; NI j,t = the reported net income for IPO sample firm j during period t; CFO j,t = the cash flow from operations for IPO sample firm j during period t; and total assets (TA j,t-1 ) is as described previously. Although this measure is noisy in detecting the magnitude of earnings management, and can not identify the specific measures used for managing earnings, this measure can provide an indication of the existence of earnings management. Moreover, this measure provides an easy way for investors to evaluate the likelihood that a firm is engaging in earnings management without the use of sophisticated statistical methods. Measure of IPO Underpricing The first day IPO price change is measured as the difference between the market price of the IPO at the close of its first day of trading and its initial offer price. PRC i, OFFER Where, PRCH i = the first day price change of IPO firm i, PRC i,offer = offer price of IPO firm i, and PRC i,close = Market price of IPO firm i at the close of the first day of trading. PRCHi = PRCi, CLOSE - Control Variables The purpose of this study is to address whether earnings management and IPO underpricing are associated to auditor litigation. Therefore, other factors that may influence the association between earnings management and auditor litigation and between IPO underpricing and Auditor Litigation be controlled. The control variables used in this paper are those suggested by prior research, including audit firm size, the proportion of audit fee received from a particular client, and the length of the auditor-client relationship as well as client characteristics such as the client s financial health, size, and growth. A Theoretical Development and Identification of Models Two models were proposed to test hypotheses 1 and 2. The models use auditor litigation, a dummy variable, as dependent variable and different earnings management measures and IPO underpricing and the control variables as independent variables. Hypothesis 1 predicts that the probability of auditor litigation increases as managers in IPO firms overstate more of their earnings. To test this hypothesis, I proposed a model to regress the probability of auditor litigation on three earnings management measures and the previously discussed control variables. If the coefficients on the earnings management measures are positively significant, the results should indicate that the probability of auditor litigation is positively associated with earnings management. LIT j =b 1 + b 2 EM jt + b 3 RO j + b 4 AR j + b 5 UR j + b 6 VC j +e jt (1) Where: LIT j, = a dummy variable where it equals to 1 if the IPO sample firm j is named in a lawsuit and 0 otherwise; EM jt = earnings management measures for sample firm j at time t; AR j = a dummy variable for auditor reputation, equal to 1 if the auditor of sample firm j is a Big 5 public accounting firm and 0 otherwise; 1267

36 CI j = the ratio of a particular client s sales to the sales of all client firms audited by a given auditor; ST j = a dummy variable for the tenure of the audit firm, equal to 1 if the client has engaged the auditor continuously for three years or less and 0 otherwise; FC j = Zmijewaski s (1984) financial condition index estimates the probability of client failure, so the greater the index the weaker the client s financial condition; SIZE j = the natural log of total client assets; GROWTH j = the percentage change in the client sales from year t-1 to year t; and e jt is the error term for this model. Hypothesis 2 predicts that the probability of auditor litigation increases as the level of IPO underpricing increases. To test this hypothesis, I proposed a model to regress the probability of auditor litigation on IPO underpricing and the previously discussed control variables. If the coefficients on the earnings management measures are positively significant, the results should indicate that the probability of auditor litigation is positively associated with the level of IPO underpricing. LIT j =b 1 + b 2 PRCH jt + b 3 RO j + b 4 AR j + b 5 UR j + b 6 VC j +e jt (2) Where: LIT j, = a dummy variable where it equals to 1 if the IPO sample firm j is named in a lawsuit and 0 otherwise; PRCH jt = the first day price change of IPO firm j at time t; AR j = a dummy variable for auditor reputation, equal to 1 if the auditor of sample firm j is a Big 5 public accounting firm and 0 otherwise; CI j = the ratio of a particular client s sales to the sales of all client firms audited by a given auditor; ST j = a dummy variable for the tenure of the audit firm, equal to 1 if the client has engaged the auditor continuously for three years or less and 0 otherwise; FC j = Zmijewaski s (1984) financial condition index estimates the probability of client failure, so the greater the index the weaker the client s financial condition; SIZE j = the natural log of total client assets; GROWTH j = the percentage change in the client sales from year t-1 to year t; and e jt is the error term for this model. Due to the limited availability of the auditor litigation data in the IPO market, this paper could not perform an empirical test of the models developed at this point of time. However, this paper presents the two models that examine earnings management and auditor litigation in the IPO context, an ideal context to examine the aforementioned relationship. Future research may conduct empirical tests to evaluate the validity of the two models constructed in this paper and examine the relationships between auditor litigation, earnings management and IPO Underpricing. IV. Conclusion and Implication This paper presents models to examine the relationship between the probability of auditor litigation, earnings management, and IPO underpricing anomaly in the initial public offering context. This paper contributes to the literature in the three ways. First, this paper presents a 1268

37 model to examine the relationship between auditor litigation and earnings management in the IPO market. The IPO context provides an ideal setting for testing both earnings management and auditor litigation because several unique features related to the IPO market provides exceptional incentives and opportunities for earnings management. Second, this paper proposes three different earnings management measures to test the relationship between auditor litigation and earnings management. In the absence of a perfect earnings management measures, the employment of three different measures may explain the inconclusive results from prior studies. Third, this paper makes the first attempt to propose that there is a positive relationship between auditor litigation and IPO underpricing. Future research may conduct empirical tests to evaluate the validity of the two models constructed in this paper. References Accounting Principles Board (APB) Balvers, R. J., B. McDonald, and R. E. Miller Underpricing of new issues and the choice of auditor as a signal of investment of banker reputation. The Accounting Review (63): Becker, C.M., J.J. DeFond, and K. Subramanyan The effect of audit quality on earnings management. Contemporary Accounting Research (15): Cahan, S. F, and W. Zhang After Enron: Auditor conservatism and ex-andersen clients. The Accounting Review (81): Dalton, D., J. Hill, and R. Ramsey The threat of litigation and voluntary partner/manager turnover in Big Six firms. Journal of Accounting and Public Policy (16): DeAngelo, L.E Auditor Size and auditor quality. Journal of Accounting and Economics 3: Dechow, P.M., R.G. Sloan, and A.P. Sweeney Detecting earnings management. The Accounting Review (70): Francis, J.R. and J. Krishnan Accounting Accruals and Auditor Reporting Conservatism Contemporary Accounting Research (16 ): Guay, W.A., S.P. Kothari, and R.L. Watts A market-based evaluation of discretionaryaccrual models. Journal of Accounting Research 34(Supplement): Healy, P.M., and J.M. Wahlen A review of the earnings management literature and its implications for standard setting. Accounting Horizon (13): Heninger, W. G The association between auditor litigation and abnormal accruals. The Accounting Review (76): Individual Investor Group, Inc. How to read statement of cash flow. Jones, J.J Earnings management during import relief investigations. Journal of Accounting Research (29): Kang, S., ad K. Sivaramakrishnan Issues in testing earnings management and an instrumental variables approach. Journal of Accounting Research (33): Lys, T., and R. L. Watts Lawsuits again auditors. Journal of Accounting Research 32 (Supplement): Palmrose, Z.V An Analysis of Auditor Litigation and Audit Service Quality. The Accounting Review (63):

38 Palmrose, Z.V. ands. Scholz Restated Financial Statements and Auditor Litigation. Working Paper. University of Southern Californai. Peasnell, K.V., P.F. Pope, and S. Young Detecting earnings management using crosssectional abnormal accrual models. Accounting and Business Research (30): Smith, C.W., Jr., Investment banking and the capital acquisition process. Journal of Financial Economics (15): Stice, J. D Using financial and market information to identify pre-engagement factors associated with lawsuits against auditors. The Accounting Review 59 (April): Teoh, S.H., I. Welch, and T.J. Wong Earnings management and the post-issue performance of seasoned equity offerings. Journal of Financial Economics (50): Xiong, Y Earnings Management and IPO Underpricing. Working paper. Authors Yan Xiong, PhD Professor of Accounting, College of Business Administration, California State University Sacramento, Caixing Liu Associate Professor of Accounting, College of Business Administration, California State University Sacramento 1270

39 On Maximizing Annualized Option Returns Charles J. Higgins, PhD Abstract While options do generally demonstrate an increase in prices as time increases, an annualized return of their excess premiums exhibit other characteristics including a lower return on options farther out of the money, that as the exercise price is farther out of the money that the expiration with the greatest annualized return is longer in time, and more interestingly that for underlying securities having larger standard deviations the greatest annualized option returns are found with options having shorter expirations. I. Introduction A call option is a contract to buy and a put option is a contract to sell an underlying security. Call and put options can likewise be bought or sold. As Aswath Damodaran in Option Pricing Basics noted: A call option gives the buyer of the option the right to buy the underlying asset at a fixed price (strike price or K) at any time prior to the expiration date of the option. The buyer pays a price for this right. At expiration, if the value of the underlying asset (S) > Strike Price (K) [the] buyer makes the difference: S K; if the value of the underlying asset (S) < Strike Price (K) [the] buyer does not exercise. More generally, the value of a call increases as the value of the underlying asset increases [and] the value of a call decreases as the value of the underlying asset decreases. If the nominal intrinsic value is negative it is normally zero in that, unlike a futures contract, an option owner can walk away from the contract (some slight exceptions are observed near expiration dates and often reflect transaction costs in terms of the amount of the negative excess premium). The option owner is on the positive side of the intrinsic value and the option seller is on the negative side of this same valuation with the option owner in control of whether the option may be exercised. There are European and American options where the former may be exercised upon expiration and the latter may be exercised at any time prior to expiration. Generally options have excess premiums above the intrinsic values and are a function of interest rates, volatility of the underlying security (standard deviation), time to expiration, expectations with particular attention to dividend distributions, and the relationship between the exercise price and security price with the greatest excess premiums usually associated with exercise prices closest to the underlying security s price. There are various methods for modeling options; they include: the Black-Scholes options pricing model which particularly describes European call options, the binomial options pricing model, and a Monte Carlo simulations model among others. While subject to academic disdain, Wikepedia descriptions of each model provide concise summaries of each: One of the attractive features of the Black-Scholes model is that the parameters in the model other than the volatility (the time to maturity, the strike, the risk-free interest rate, and the current underlying price) are unequivocally observable. All other things being equal, an option s theoretical value is a monotonic increasing function of implied volatility. 1271

40 And: And: For options with several sources of uncertainty (e.g., real options) and for options with complicated features (e.g., Asian options), binomial methods are less practical due to several difficulties, and Monte Carlo option models are commonly used instead. Although computationally slower than the Black Scholes formula, it is more accurate, particularly for longer-dated options on securities with dividend payments. For these reasons, various versions of the binomial model are widely used by practitioners in the options markets. In terms of theory, Monte Carlo valuation relies on risk neutral valuation. Here the price of the option is its discounted expected value; see risk neutrality and rational pricing. The technique applied then, is (1) to generate a large number of possible (but random) price paths for the underlying (or underlying) via simulation, and (2) to then calculate the associated exercise value (i.e. "payoff") of the option for each path. (3) These payoffs are then averaged and (4) discounted to today. This result is the value of the option. This approach, although relatively straightforward, allows for increasing complexity II. Pricing Simulations versus Real Data One can simulate a security s sequential price distribution and thus the option value at each moment by P t = P t-1 (1+k) where k is N(µ,σ) and σ is derived from (-2log(ř 1 )) 1/2 sin(2πř 2 ) with each ř distributed as U(0,1) noting that some Excel computations using its random normal number generator have been shown to be sometimes problematic. The daily security standard deviation creates an annual standard deviation and approximates the square root of time which here is 16 times from 256^.5 which closely equals the number of trading days per year. A graphic of a frequency distribution of simulated security prices plotted against various days up to a year was created by a GWBASIC program (see Figure 1). Figure 1. Simulated one-year security price frequency distribution 1272

41 Now consider an at-the-money option formed from a simulation of a security price initially set at 100 with an exercise price of 100 (see Figure 2). Figure 2. Simulated at-the-money daily one-year call option pricing In contrast would be an out-of-the-money option exhibiting a different pricing graphic. Consider an otherwise similar simulated call option with a security price of 100 but an exercise of 110 (see Figure 3a). Figure 3a. A simulated daily one-year out-of-the-money call option pricing 1273

42 and unlike the graphic with a strike price of 100, one could create a tangency associated with the greatest annualized return (see Figure 3b). Figure 3b. A simulated daily one-year out-of-the-money call option pricing In consideration of other simulated daily exercise strike prices during a year, see Figures 4a and 4b. Figure 4a. Simulated daily in-the-money puts & out-of-the-money calls 1274

43 Figure 4b. Simulated daily in-the-money calls & out-of-the-money puts However, an observation of real option prices presents some interesting differences when the option price excess premiums are annualized with an eye toward maximizing a continuing portfolio return. In an examination of currently traded options as well as some option trades five years ago (when there were fewer exercise strike prices and there were no weekly or quarterly options) one sees some difficulties with real option trading data. If one uses closing prices they are in fact last prices. An option s last price may not be contemporaneous with the closing underlying security price. Consider the call options of Boeing Aircraft which closed at $52.53 on November 17, 2009 (see Table 1 where the prices are presented, then the in-the-money call options were adjusted for excess premium [Call Security + Exercise], then the excess premium was annualized by dividing by time to expiration). Likewise consider the call options for General Electric which closed at $16.00 on November 16, 2009 (see Table 2 for the option prices then the annualized return but without the computation for excess premiums). Now consider currently traded options where there are newer additional exercise strike prices and also weekly and quarterly options. In an examination of International Business Machines with a price of $ on October 29, 2014 closing or last prices were used. See Table 3 and Figure 5 the IBM call option prices and near-the-money exercise strike prices. The options were then adjusted for excess premium then the excess premium was annualized by dividing by time to expiration; see Figure

44 Figure 5. IBM Option Prices October 29, 2014 Figure 6. IBM Annualized Excess Option Premiums Likewise consider the SPDR DIA index ETF with a price of $ October 29, 2014 (see Table 4 using frequently traded option prices with near-the-money exercise five dollar multiple strike prices and adjusted for excess premiums, then annualized by expiration). The occasional negative excess premium is likely indicative of the non contemporaneous pricing of the option and/or that explained by transactions near expiration dates. Another examination was made of Proctor & Gamble (see Table 5) where bid and ask prices were averaged together in an attempt to provide a more contemporaneous pricing to the underlying security s closing price. I do note that bid and ask prices may change or expire the close of the market and in my experience that some options may execute at either the upper or lower range of the bid ask spread depending 1276

45 upon the security in question thus diminishing the transparency of using the average of last bid and ask prices for options. Be that as it may, see Figure 7 for the Proctor & Gamble closing options prices sorted by expiration then by exercise strike prices for November 3, Figure 7. Proctor & Gamble call option prices Nov. 4, 2014 The Proctor & Gamble call option prices were then annualized then again for a second presentation after subtracting the intrinsic value for an excess premium (see Table 5). The excess premium annualized option returns are presented in Figure 8 now sorted by exercise strike prices then expiration dates. Figure 8. Proctor & Gamble call options by strikes then expiration 1277

46 One can see that unsurprisingly greater annualized returns for near-the-money options. Further for out-of-the-money options as the exercise strike price increases that the annualized return decreases but with maximums associated with greater expiration dates. A need for a simulation derived graphic description of each exercise strike price and expiration date now becomes apparent. What follows is a bar for each out-of-the-money dollar by dollar exercise strike price with a simulated month by month expiration therein (see Figure 9). Note that the near-the-money options had annualized excess premiums with the shortest expirations and vice versa. Moreover there occurred a maximum annualized excess premiums with an expiration of one year when the exercise strike price was somewhere in between. Figure 9. Monthly annualized call option premiums by strike prices The daily security standard deviation creates an annual standard deviation and approximates the square root of time which here is about 16 times from the number of trading days per year which in fact is a few days shy of 256. A reconfigured graphic, now arranged by major strike prices then daily expirations, makes clearer the maximum annualized excess premium computations (see Figure 10a). 1278

47 Figure 10a. Out-of-the-money call option simulated pricing If one were to draw a tangent from the origin to each of these simulated option prices, it would provide the highest annualized return (see Figure 10b). Figure 10b. Out-of-the-money call option simulated pricing A computation of simulated annualized option returns was performed for 1, 2, and 3 percent daily standard deviations for some simulated 256 trading days for the underlying security starting price of 100 and reporting the exercise strikes prices of 100, 105, 110, 115, and 120 (see Figures 11a, 11b, and 11c all having the same vertical scale as Figures 10a and 10b). 1279

48 Figure 11a. Annualized out-of-the-money calls,.16/year standard deviation Figure 11b. Annualized out-of-the-money calls,.32/year standard deviation Figure 11c. Annualized out-of-the-money calls,.48/year standard deviation 1280

49 III. Conclusion A call writer or a put seller may consider the various option expirations and exercise strike prices noting that for out of the money options that annualized premium returns decrease as the exercise strike price rises but that maximum annualized returns will be associated with greater expiration dates. Likewise for securities which have a higher standard deviation they will have a larger annualized return but with a shorter expiration for maximum annualized returns. References Black. Fischer and Myron Scholes The Pricing of Options and Corporate Liabilities Journal of Political Economy, 1973 Bliss, Robert R., On the Monotonicity of the Option-Value/Risk Relation Research Department Federal Reserve Bank of Chicago [2003] Bodie, Zvi, Alex Kane, & Alan J. Marcus Essentials of Investments 9th ed., McGraw-Hill Irwin, 2010 Damodaran, Aswath Option Pricing Basics New York University Stern School of Business DrCinvests, Annualizing Option Premiums ExcelUser, An Introduction to Excel's Normal Distribution Functions Great Option Trading Strategies, How to Calculate Annualized Returns on Option Trades Jackwerth, Jens Carsten and Mark Rubinstein, Recovering Probability Distributions from Option Prices The Journal of Finance 51 (5) [December 1996], pp Jones, Christopher A Nonlinear Factor Analysis of S&P 500 Index Option Returns The Journal of Finance 61 (5) [October 2006], pp Longstaff, Francis A. and Eduardo S. Schwartz Interest Rate Volatility and the Term Structure: A Two-Factor General Equilibrium Model Journal of Finance 47 (4) [September 1992], pp Longstaff, Francis A. and Eduardo S. Schwartz, Valuing American Options by Simulation: A Simple Least-Squares Approach The Review of Financial Studies, Vol. 14, No. 1. [Spring 2001], pp Mathematics Stack Exchange, Why doesn't NORMSINV(RAND()) in Excel work as a standard normal random number generator? Author Charles J. Higgins, PhD Dept. Finance/CIS, Loyola Marymount University, One LMU Dr., Los Angeles, CA , chiggins@lmu.edu See my related video Annualizing Option Returns on YouTube/DrCinvests:

50 Appendix Table 1. Boeing Aircraft call options adjusted for excess premiums Table 2. General Electric call options not adjusted for excess premiums 1282

51 Table 3. International Business Machines major adjusted call options Table 4. SPDR Dow Jones Industrial ETF major adjusted call options 1283