Audit Committee Quality, Auditor Independence, and Internal Control Weaknesses

Transcription

1 Audit Committee Quality, Auditor Independence, and Internal Control Weaknesses Yan Zhang, Jian Zhou, and Nan Zhou * * All authors are from SUNY Binghamton. We thank two anonymous reviewers for detailed and insightful suggestions that have significantly improved the paper. We also thank workshop participants at the 2006 American Accounting Association Auditing Midyear Meeting and the 2006 American Accounting Association Annual Meeting for comments, and Raj Addepalli, Shanshan Chen, Yujing Pan, Gaurav Rastogi, Eric Romanoff, Grace Witte, and Meng Zhao for research assistance. Please address all correspondence to Jian Zhou, School of Management, SUNY Binghamton, Binghamton, NY ; jzhou@binghamton.edu; phone: (607)

2 Audit Committee Quality, Auditor Independence, and Internal Control Weaknesses Abstract In this paper we investigate the relation between audit committee quality, auditor independence, and the disclosure of internal control weaknesses after the enactment of the Sarbanes-Oxley Act. We begin with a sample of firms with internal control weaknesses and, based on industry, size, and performance, match these firms to a sample of control firms without internal control weaknesses. Our conditional logit analyses indicate that a relation exists between audit committee quality, auditor independence, and internal control weaknesses. Firms are more likely to be identified with an internal control weakness, if their audit committees have less financial expertise or, more specifically, have both less accounting financial expertise and non-accounting financial expertise. They are also more likely to be identified with an internal control weakness, if their auditors are more independent. In addition, firms with recent auditor changes are more likely to have internal control weaknesses.

3 Audit Committee Quality, Auditor Independence, and Internal Control Weaknesses 1. Introduction The Sarbanes-Oxley Act (hereafter SOX) of 2002 went into effect on July 30, 2002 to address the increasing concern of investors about the integrity of firms financial reporting, due to scandals involving once well-respected companies, such as Enron and WorldCom and auditors, such as Arthur Andersen. One important aspect of SOX is that it has two sections specifically focusing on internal control issues related to financial reporting. Under Section 302, management is required to disclose all material weaknesses in internal control, when they certify the periodic, annual, and quarterly statutory financial reports. Under Section 404, a firm is required to assess the effectiveness of its internal control structure and procedures for financial reporting and disclose such information in its annual reports. Furthermore, the firm s auditor is required to provide an opinion on the assessment made by the management in the same report. Because such mandatory disclosure under SOX provides us with more information on internal controls, we are interested in investigating the determinants of internal control weaknesses in the post-sox era. We begin with a sample of firms with internal control weaknesses, and, based on industry, size, and performance, match these firms to a sample of control firms without internal control weaknesses. Our conditional logit analyses indicate that a relation exists between audit committee quality, auditor independence, and internal control weaknesses. Firms are more likely to be identified with an internal control weakness, if their audit committees have less financial expertise or, more specifically, have less accounting financial expertise and non-accounting financial expertise. They are also more likely to 1

4 be identified with an internal control weakness, if their auditors are more independent. In addition, firms with recent auditor changes are more likely to have internal control weaknesses. Our paper is related to several recent papers on the determinants of internal control weaknesses. Krishnan (2005) examines the period prior to the enactment of SOX, when internal control problems are only disclosed in 8-Ks filed by firms when changing auditors. With information collected from 8-K filings, she finds that independent audit committees and audit committees with more financial expertise are significantly less likely to be associated with the incidence of internal control problems. Ge and McVay (2005) and Doyle et al. (2006a) find that material weaknesses in internal control are more likely for firms that are smaller, less profitable, more complex, growing rapidly, or undergoing restructuring. Ashbaugh-Skaife et al. (2006) find that firms with more complex operations, recent changes in organization structure, auditor resignation in the previous year, more accounting risk exposure, and less investment in internal control systems are more likely to disclose internal control deficiencies. We document that financial expertise in audit committees continues to be an important determinant of internal control weaknesses after the enactment of SOX. Our findings thus complement those in Krishnan (2005), who studies the pre-sox period. Focusing on the post-sox period enables us to take advantage of the wealth of information on internal control unleashed by SOX and to construct a sample of firms with internal control problems from both mandated disclosures in the firms 10-Q and 10-K filings under SOX and information disclosed in 8-K filings when firms change auditors. Consisting of only those firms that change auditors in the pre-sox period, the sample 2

5 firms in Krishnan (2005) tend to be smaller in size and are traded on smaller stock exchanges. We avoid this sample selection bias by focusing on the post-sox period, given that all firms are required to disclose material internal control weaknesses under SOX. In addition, we document that auditor independence is an important determinant of internal control weaknesses. This adds to the literature that supports the hypothesis that auditor independence matters, such as Frankel et al. (2002) and Krishnamurthy et al. (2006). Different from other researchers who also focus on the post-sox period, such as Ge and McVay (2005), Doyle et al. (2006a) and Ashbaugh-Skaife et al. (2006), we show that audit committee quality, characterized as having more financial expertise or, more specifically, having more accounting financial expertise and non-accounting financial expertise, is an important determinant of internal control weaknesses. In addition, we find that auditor independence, calculated as the ratio of audit fee to total fee, is also a determinant of internal control weaknesses. The rest of the paper is organized as follows. Section 2 introduces the background and proposes our hypotheses. Section 3 describes the sample selection procedures. Section 4 discusses the empirical findings, and Section 5 presents our conclusions. 2. Background and hypotheses 2.1. Background SOX emphasizes internal control, which is defined as a process, effected by an entity's board of directors, management and other personnel, designed to provide reasonable assurance regarding the achievement of objectives, according to the COSO 3

6 framework. 1 SOX Section 302 (hereafter SOX 302), which went into effect on August 29, 2002, requires management to disclose significant internal control deficiencies, when they certify annual or quarterly financial statements. Specifically, the signing officers, being responsible for internal controls, have evaluated the internal controls within the previous ninety days and reported in their findings: (1) a list of all deficiencies in the internal controls and information on any fraud that involves employees who are involved with internal control activities; (2) any significant changes in internal controls or related factors that could have a negative impact on the internal controls. Section 404 took this reporting a step further. It not only requires management to provide an assessment of internal controls, but also requires auditors to provide an opinion on management s assessment. Specifically, issuers are required to disclose information concerning the scope and adequacy of the internal control structure and procedures for financial reporting in their annual reports. This statement shall also assess the effectiveness of such internal controls and procedures. The registered auditing firm shall, in the same report, attest to and report on the effectiveness of the internal control structure and procedures for financial reporting. According to the rulings of the Securities Exchange Commission (SEC), a company that is an accelerated filer 2 must comply with SOX Section 404 (hereafter SOX 404) for its first fiscal year ending on or after November 15, A non-accelerated filer must begin to comply with these requirements for its first fiscal year ending on or after July 15, A foreign private 1 COSO stands for the Committee of Sponsoring Organizations of the Treadway Commission, who undertook an extensive study of internal control to establish a common definition that would serve the needs of companies, independent public accountants, legislators, and regulatory agencies and to provide a broad framework of criteria, against which companies could evaluate the effectiveness of their internal control systems. COSO published its Internal Control -- Integrated Framework in An accelerated filer is defined in Exchange Act Rule 12b-2. Generally, it refers to a U.S. company that has equity market capitalization over $75 million and has filed an annual report with the SEC. 4

7 issuer that files its annual report on Form 20-F or Form 40-F must begin to comply with the corresponding requirements in these forms for its first fiscal year ending on or after July 15, According to Compliance Week, most of the internal control weakness disclosures under SOX 302 and SOX 404 are related to financial systems and procedures. This group typically involves financial closing processes, account reconciliation, or inventory processes. For example, United Stationers disclosed problems with the design and effectiveness of internal controls relating to receivables from suppliers. Personnel issues rank as the second largest category of weakness disclosures. This category is related to the poor segregation of duties, inadequate staffing, or other related training or supervision problems. For example, Sanmina-SCI cited a lack of sufficient personnel with appropriate qualifications and training in certain key accounting roles. Other common types of weaknesses include revenue recognition, documentation, and IT system and controls (e.g. security and access controls, backup and recovery issues). In addition, issues related to international operations and mergers and acquisitions are sources of weakness disclosure, although they represent a relatively small percentage of all disclosures. For example, Masco cited internal control problems attributable to historical growth through acquisition and decentralized organizational structure, and GulfMark Offshore identified internal control deficiencies related to the complexity of their multi-national operations. Based on their severity, these internal control problems are classified into three types: material weakness, significant deficiency, and control deficiency. Auditing 3 The SOX compliance information is from and the SOX summaries are from 5

8 Standard (hereafter AS) No. 2 defines a material weakness as a significant deficiency, or combination of significant deficiencies, that results in more than a remote likelihood that a material misstatement of the annual or interim financial statements will not be prevented or detected. Under AS No.2, a significant deficiency is a control deficiency, or a combination of control deficiencies, that adversely affects the company s ability to initiate, authorize, record, process, or report external financial data reliably in accordance with genernally accepted accounting principles such that there is more than a remote likelihood that a misstatement of the company s annual or interim financial statements that is more than inconsequential will not be prevented or detected. A control deficiency occurs when the design or operation of a control does not allow management or employees, in the normal course of performing their assigned functions, to prevent or detect misstatement on a timely basis. Since only material weaknesses are required to be publicly disclosed under SOX 302 and SOX 404, we follow Doyle, Ge, and McVay (2006a; 2006b), and focus on firms that disclosed material weaknesses in our study. 4 For the sake of brevity, we will refer to material internal control weaknesses as internal control weaknesses hereafter Audit committee quality and internal control Since an entity s internal control is under the purview of its audit committee (Krishnan, 2005), we investigate the relation between audit committee quality and internal control weaknesses. The audit committee not only plays an important monitoring role to assure the quality of financial reporting and corporate accountability 4 This is also driven by the fact that Compliance Week lists only firms with material weaknesses starting March

9 (Carcello and Neal, 2000), but also serves as an important governance mechanism, because the potential litigation risk and reputation impairment faced by audit committee members ensure that these audit committee members discharge their responsibilities effectively. We thus expect that firms with high-quality audit committees are less likely to have internal control weaknesses than firms with low-quality audit committees. On measuring audit committee quality, we focus on the financial expertise in these committees. The Blue Ribbon Committee on Improving the Effectiveness of Corporate Audit Committees (BRC) s (1999) recommendation that each audit committee should have at least one financial expert highlights the importance of the financial literacy and expertise of audit committee members. 5 Section 407 of the SOX incorporates the above suggestion and requires firms to disclose in periodic reports, whether a financial expert serves on a firm s audit committee and, if not, why not. Such financial expertise of audit committee members has been shown to be important for dealing with the complexities of financial reporting (Kalbers and Fogarty, 1993) and for reducing the occurrence of financial restatements (Abbott et al., 2004). In addition, DeZoort and Salterio (2001) find that audit committee members with financial reporting and auditing knowledge are more likely to understand auditor judgments and support the auditor in auditor-management disputes than members without such knowledge. Moreover, financially knowledgeable members are more likely to address and detect material misstatements. Audit committee members with financial expertise can also 5 The Report of the BRC s recommendation related to Audit Committee Competence states that the audit committee should consist of at least three members, each of whom is "independent" (defined in the Report as having "no relationship to the corporation that may interfere with the exercise of their independence from management and the corporation") and "financially literate" (defined as "the ability to read and understand fundamental financial statements"). At least one member of the audit committee should have accounting or financial management expertise (defined as past employment or professional certification in accounting or finance, or comparable experience including service as a corporate officer with financial oversight responsibility). 7

10 perform their oversight roles in the financial reporting process more effectively, such as detecting material misstatements (Scarbrough et al., 1998; Raghunandan et al., 2001). Indeed, Abbott et al. (2004) find a significantly negative association between an audit committee having at least one member with financial expertise and the incidence of financial restatement. Krishnan (2005) presents evidence that audit committees with financial expertise are less likely to be associated with the incidence of internal control problems. Therefore, we have the following directional prediction. Hypothesis 1: Firms with greater audit committee financial expertise are less likely to have internal control weaknesses. DeFond et al. (2005) document significantly positive cumulative abnormal returns around the appointment of accounting financial experts to the audit committee, suggesting that audit committees with accounting financial expertise improve corporate governance. Therefore, we further separate audit committee financial expertise into accounting financial expertise and non-accounting financial expertise and test the relation between these two variables and internal control weaknesses. In measuring the financial expertise of an audit committee member, we follow the definition adopted in SOX Section 407, and, more specifically, modify the definition used in DeFond et al. (2005). An audit committee member is a financial expert if he or she can be classified into the following two categories: (a) an accounting financial expert who has experience as a public accountant, auditor, principal or chief financial officer, controller, or principal or chief accounting officer; or (b) a non-accounting financial 8

11 expert who has experience as the chief executive officer, president, or chairman of the board in a for-profit corporation, or who has experience as the managing director, partner or principal in venture financing, investment banking, or money management. With this definition, we measure audit committee financial expertise (ACFE) as the percentage of audit committee members who are financial experts. We further separate audit committee financial expertise into accounting financial expertise (ACCT_ACFE), measured as the percentage of audit committee members who are accounting financial experts, and non-accounting financial expertise (NONACCT_ACFE), the percentage of audit committee members who are non-accounting financial experts Auditor independence and internal control Auditor independence can be related to the disclosure of a firm s internal control problems. When there is a strong economic bond between an auditor and a client firm, the auditor has an incentive to ignore potential problems and issue a clean opinion on the client firm s internal controls. While some studies (DeFond et al., 2002; Ashbaugh et al., 2003; Chung and Kallapur, 2003; Reynolds et al., 2002; Francis and Ke, 2003) find no relation between non-audit fees and auditor independence and argue that an auditor s concern with maintaining its reputation for providing high quality audits could restrain it from undertaking activities that jeopardize independence, since the revenue from each client will be a small percentage of the auditor s total revenue, other studies suggest that the provision of non-audit services compromises auditor independence. For example, Frankel et al. (2002) find that non-audit services are associated with increased discretionary accruals and the achievement of certain earnings benchmarks and Krishnamurthy et al. 9

12 (2006) document that the abnormal returns for Andersen s clients around Andersen s indictment are significantly more negative, when the market perceived the auditor s independence to be compromised. Given these mixed empirical findings, we measure auditor independence (RATIO) as the ratio of the audit fee to the total fee, and propose the non-directional null hypothesis, as follows. Hypothesis 2: Auditor independence is not associated with the disclosure of internal control weaknesses Control variables Audit committee In addition to audit committee financial expertise, other attributes of an audit committee have been found to be important factors in effective monitoring. Specifically, we control for audit committee independence, since Krishnan (2005) finds that there is a positive relation between audit committee independence and the quality of internal control prior to the enactment of SOX. 6 While SOX requires that audit committees be composed of all independent directors for firms traded on an organized stock exchange (e.g., NYSE, AMEX) or a recognized dealer quotation system (e.g., NASDAQ), exemptions may be given by the SEC, if it determines that it is appropriate under certain circumstances. We thus still control for audit committee independence (ACIND), defined as the percentage of independent directors on the audit committee. Under SOX, an audit committee member is independent, if he or she is not affiliated with the firm and does not 6 Previous research has also found an association between audit committee independence and the quality of accounting information (e.g., Klein, 2002b; Abbott et. al., 2004). 10

13 accept any consulting fees. We next control for the natural logarithm of audit committee size (ACSZ), measured as the number of audit committee members, because research suggests that a large audit committee tends to enhance the audit committee s status and power within an organization (Kalbers and Fogarty, 1993), to receive more resources (Pincus et al., 1989), and to lower the cost of debt financing (Anderson et al., 2004). We thus expect that a large audit committee is more likely than a small one to improve the quality of internal controls, because increased resources and enhanced status will make the audit committee more effective in fulfilling its monitoring role. We also control for the natural logarithm of audit committee meetings (ACMEET), measured as the number of audit committee meetings held each year, because research shows that effective audit committees meet regularly (Menon and Williams, 1994; Xie et al., 2003). 7 Consistent with this hypothesis, McMullen and Raghunandan (1996) find that the audit committees of firms with SEC enforcement actions or earnings restatements are less likely to have frequent meetings than those without and Lennox (2002) finds that there is a significant increase in the number of audit committee meetings during an auditor dismissal year. However, it is also possible that an audit committee meets more frequently to discuss internal control issues, when there are significant problems associated with a firm s internal controls. Therefore, we make no prediction on the relation between the number of audit committee meetings and the quality of internal controls. 7 Hymowitz and Lublin (2003) report that many audit committees are spending far more time than they used to reviewing financial statements and overseeing auditors, meeting 10 or 11 times a year, up from three or four times. 11

14 Board of directors The quality of an entity s internal controls is a function of the quality of its control environment that includes the board of directors and the audit committee (Krishnan, 2005). First, we focus on board independence (BDIND), measured as the percentage of outside directors on the board, 8 because research suggests that board independence is negatively related to the likelihood of financial fraud and SEC enforcement actions (Beasley, 1996; Dechow et al., 1996). We also control for the natural logarithm of board size (BDSZ), measured as the number of directors on the board. While some researchers find that a large board has more expertise than a small one (Dalton et al., 1999), that it tends to be more effective in monitoring accruals (Xie et al., 2003), and that it leads to a lower cost of debt (Anderson et al., 2004). Others suggest that a small board is more effective in mitigating the agency costs associated with a large board (Yermack, 1996; Eisenberg et al., 1998; Hermalin and Weisbach, 1998, 2003). Given the mixed empirical evidence on board size, we expect that the relation between board size and the likelihood of internal control weaknesses is indeterminate. Finally, we control for the natural logarithm of board meetings (BDMEET), as measured by the number of board meetings held each year. While Conger et al. (1998) suggest that board meeting frequency is important to improve board effectiveness, Vafeas (1999) finds that it is inversely related to firm value, because of the increased board activities following share price declines. Since board independence, size, and meeting frequency all influence a board s effectiveness, they, in turn, are related to the quality of internal controls. 8 Outside directors are those who are not affiliated with the firm, other than serving on its board. We first exclude those directors who are the firm s officers and major shareholders, and then further exclude those who have consulting relationships or other related-party transactions with the firm. 12

15 Auditor Types We use a dummy variable (BIG4) to measure auditor type, 9 because a firm s decision to hire a Big 4 auditor is likely to be associated with internal controls for several reasons. Doyle et al. (2006a) find that smaller and less profitable firms are more likely to have internal control problems than larger or more profitable ones. On the one hand, such firms with internal control problems are less likely to hire a Big 4 auditor, because they are constrained by financial resources and cannot afford it. On the other hand, they might also be avoided by the Big 4 auditors, because they are perceived as being risky and may expose the Big 4 to potential litigations. Given that a firm shunned by a Big 4 auditor may signal that it has potential internal control problems, we introduce the dummy variable BIG4 to control for auditor quality Auditor Changes Ashbaugh-Skaife et al. (2006) find that firms with recent auditor changes are likely to have internal control problems. On the one hand, auditors may drop risky clients as part of their risk management strategies, since firms with material internal control weaknesses may represent high audit failure risk. On the other hand, firms may dismiss auditors for lack of performance, when the firms discover material internal control weaknesses. Therefore, we use a control variable AUDCHG, which is equal to one, if there is an auditor change in 2003 or 2004, and zero otherwise Other variables 9 The dummy variable (BIG4) takes a value of one, if a firm is a Big 4 client and zero otherwise. 13

16 We also control for firm characteristics that may be associated with internal control problems. Since Doyle et al. (2006a) show that small and high growth firms are likely to have internal control weaknesses, in our model, we control for size, measured as the natural logarithm of total assets (TA), and growth, measured as industry-median-adjusted sales growth (ADJSALEGR). It may take some time for a firm that recently engaged in mergers and acquisition to integrate different internal control systems; consequently, such a firm is more likely to have internal control problems. We thus introduce a dummy variable (ACQUISITION), which takes the value of one, if a firm engages in acquisitions during 2003, 2004 and from January to July of 2005, and zero otherwise. Since a firm experiencing restructuring is also likely to have internal control problems, because of the loss of experienced and valuable employees and because of the dramatic changes associated with such an event, we follow Ashbaugh-Skaife et al. (2006) and use a dummy variable (RESTRUCTURE), coded as one, if a firm has been involved in restructuring, and zero otherwise. 10 Because firms with greater complexity and scope of operations are more likely to have internal control problems than those without, we also include the natural logarithm of the number of business segments (BUS) and an indicator variable for foreign currency translation (FOREIGN) in our model (see Ashbaugh-Skaife et al., 2006; Ge and McVay, 2005). 3. Sample and control firms selection 3.1. Selection of sample firms 10 A firm is engaged in a restructuring, if it has non-zero values of COMPUSTAT data #376, #377, #378, or #

17 Table 1 provides the details for the sample selection. Our initial sample is from Compliance Week, an electronic newsletter that searchs through 8-Ks, 10-Qs, and 10-Ks for all public companies to identify any firms with internal control problems. While Compliance Week discloses firms with internal control problems on a monthly basis starting from November 2003, we examine the period from November 15, 2004 to July 31, 2005, so as to make it feasible to hand-collect most of the governance information required in our study. 11 For our sample period, there are 372 firms identified by Compliance Week as having various types of internal control problems under SOX 302 or SOX 404, 12 including material weakness, significant deficiency, reportable conditions, and control deficiency. 13 We exclude ten firms without Cusip numbers, nine firms not in COMPUSTAT, 13 foreign firms or subsidiaries, 57 firms with missing values for EBITDA profit margin, non-material weakness firms, 15 and 21 firms without proxy information. This leaves us with a final sample of 208 firms with material internal control weaknesses. We retrieve all financial information from 2004 COMPUSTAT, obtain the acquisition information from Securities Data Company, acquire the business segment information from COMPUSTAT Segment files, and hand-collect all audit and non-audit fee, audit committee, and board information from the firms proxy statements for the year of their material weakness disclosure in Compliance Week. 11 We start from November 15, 2004, given that there is increasing attention to internal control issues, since SOX 404 became effective for accelerated filers. 12 We exclude 31 duplicate appearances during the sample period. 13 Reportable conditions is an old term, which was defined by AICPA as a significant deficiency in the design or operation of the internal control structure that could adversely affect the company s ability to record, process, summarize, and report financial data consistent with the assertions of management in the financial statements. Compliance Week lists some of the early firms under this term. 14 We include this filter because our match firms are selected based on sales and EBITDA profit margin (EBITDA/sales). If a firm has a missing value for sales, it will also have a missing value for EBITDA profit margin. 15 These firms are identified as having significant deficiencies, control deficiencies, or reportable conditions. Starting from March 2005, Compliance Week lists only firms with material weaknesses. 15

18 3.2. Control firms selection To study the determinants of internal control weaknesses, we use the matchedpairs design. Although using matched samples when the number of treatment firms is not proportional to sample population may lead to biased parameters and probability estimates (Palepu, 1986), we adopt this approach to make it feasible for us to hand-collect audit committee and board information from the proxy statements. The 208 control firms without internal control weaknesses are matched to the 208 sample firms with internal control weaknesses, one-to-one, based on certain key characteristics. Specifically, we follow Purnanandam and Swaminathan (2004), who use selected publicly traded firms in the same industry as comparable firms. 16 Since their procedure balances between matching based on industry or sales, which can be too approximate, and matching based on a list of accounting variables, which can be so numerous that it becomes impossible to find match firms, we adapt their procedure to create our sample of match firms. The procedure is described as follows: (1) We select all firms in 2004 COMPUSTAT. From these firms, we eliminate subsidiaries and require firms to be incorporated in the U.S. We further require them to have information on CRSP, and be identified as common stocks of domestic U.S. firms (CRSP share code = 10 or 11). (2) We use COMPUSTAT SIC codes to group the remaining firms into 48 industries using the industry classification in Fama and French (1997). (3) The remaining firms in each industry are sorted into three portfolios by sales, and then each sales portfolio is sorted into three portfolios by EBITDA profit margin (EBITDA/sales), where EBITDA stands for earnings before interest, 16 Guo, Lev, and Zhou (2005) adapt this procedure to study the relative valuation of biotech IPOs. 16

19 taxes, depreciation, and amortization. As a result, we have 9 (3x3) portfolios of comparable firms in each industry. If there are not enough firms in an industry (fewer than 70 firms), we limit ourselves to a 2x2 classification, which leads to 4 portfolios of comparable firms in that industry. (4) We obtain sales and EBITDA margin for our sample firms from the 2004 COMPUSTAT and also classify them into different industries, according to the Fama-French industry classification. Each sample firm is matched with a portfolio of comparable firms based on industry, sales and EBITDA margin. In that portfolio, one firm with the closest total sales is selected as the match firm. If the match firm does not file a proxy or have sufficient information in proxy or has internal control weaknesses, we replace it with a non-weakness firm from the same portfolio that has the next closest total sales. 17 Following the same procedure as for the sample firms, we obtain all the information from COMPUSTAT and proxy statements for our control firms. Table 2 provides summary statistics for our sample firms and control firms. While the mean (median) sales for sample firms is $ million ($ million), the mean (median) sales for match firms is $ million ($ million). While the mean (median) EBITDA profit margin for sample firms is 0.14 (0.11), the mean (median) EBITDA profit margin for match firms is 0.45 (0.11). The large difference in the mean comparison of sales is driven by General Electric (GE), which has substantially larger sales than its closest match firm has. Without GE and its match firm, the mean (median) sales for the sample and control will be million ( million) and We check the Compliance Week list and 10-Ks to ensure that a match firm does not have internal control weaknesses. A match firm is replaced, if it appears on the Compliance Week list from November 2003 to July 2005 or is flagged with internal control weaknesses in its 10-K filed prior to July

20 million ( million), respectively. 18 If we exclude GE and its match firm, sample and control firms share similar characteristics in terms of sales and EBITDA profit margin, since our selection procedure for match firms is based on these two variables. 4. Empirical results 4.1. Univariate analyses Table 2 provides mean and median comparisons of the sample and control firms for our variables of interest. Because our sample firms are matched with control firms on a one-to-one basis, we use the paired t-test to test the difference in means and the Wilcoxon signed rank test to test the difference in medians. 19 There are several noticeable differences between these two groups of firms. On average, 75 percent of the audit committee members of the sample firms are financial experts, while 83 percent of the audit committee members of the control firms are financial experts. This difference, significant at the one-percent level, implies that firms with more audit committee financial expertise are less likely to have internal control problems, providing initial support for Hypothesis 1. We further separate audit committee financial experts into accounting financial experts and non-accounting financial experts. On average, accounting financial experts account for 22 percent of the sample firms and 23 percent of the control firms audit committee members, while non-accounting financial experts account for 53 percent of the sample firms and 59 percent of the control firms audit committee members. The difference in non-accounting financial expertise between the two groups is significant at the five-percent level. 18 We find that our main results in Table 4 remain unchanged, when we exclude GE and its match firm in our conditional logit regressions. 19 The Wilcoxon signed rank test is the nonparametric analog of the paired t-test. 18

21 We use the ratio of audit fee to total fee to measure auditor independence, where a low ratio indicates that the firm s auditor provides more non-audit services and thus lacks independence. The average audit fee ratios are 78 percent for the sample firms and 75 percent for the control firms. This significant difference indicates that independent auditors are more likely to uncover internal control problems, providing initial support for Hypothesis 2. In addition, the sample firms have more audit committee and board meetings, on average, probably in response to the sample firms internal control problems. Firms that have changed auditors or engaged in restructuring activities recently are also more likely to experience internal control weaknesses. Table 3 presents the correlation coefficients for the dependent and independent variables after we pool the sample and control firms together. We create a dummy for internal control weaknesses (ICW), which takes the value of one if a firm belongs to the sample firm group, and zero if it belongs to the control firm group. This dependent variable of interest is significantly negatively correlated with audit committee financial expertise, indicating that firms with greater audit committee financial expertise are less likely to have internal control weaknesses. Moreover, it is significantly positively correlated with the audit fee ratio, indicating that firms with more independent auditors are more likely to uncover internal control weaknesses. These results again provide preliminary support for Hypotheses 1 and 2. In addition, the internal control weakness dummy variable is positively correlated with the natural logarithms of audit committee meeting frequency and board meeting frequency. Thus, the audit committee and board of a firm with internal control weaknesses appear to hold additional meetings, dealing with the firm s internal control problems. Further, the internal control weakness dummy is 19

22 positively correlated to the variables for audit change and restructuring, suggesting that firms with recent auditor changes or restructuring activities are more likely to have internal control weaknesses Multivariate analyses Conditional logit regression models We use the conditional logit regression models to test our hypotheses that audit committee financial expertise and auditor independence are related to internal control weaknesses. Specifically, we express the internal control weakness variable as a function of audit committee quality, auditor independence, and a set of control variables. The conditional logistic regression is useful in investigating the relation between an outcome (whether the firm is a sample firm with internal control weaknesses or a control firm without such weaknesses) and a set of prognostic factors in a matched-pairs study. We match control firms to sample firms to minimize inherent variations in those factors. Because the traditional logistic regression cannot take into account the correlation structure of a matched design, we analyze our matched sample-control study using the conditional logistic regression that takes into account the non-random nature of the data. For each matched set consisting of one sample firm and one control firm, the conditional likelihood is as follows: where x i1 and xi0 i (1 + exp( β ( x ' i1 x i0 ))) 1 are vectors of the prognostic factors for the sample and control firm, respectively, of each ith matched set (Breslow, 1982; Hosmer and Lemeshow, 2000). 20

23 Conditional logit regression results Tables 4 and 5 present the regression results using conditional logit analyses. All variable definitions are provided in the Appendix. Table 4 presents four models with different measures of audit committee quality. Models 1 and 2 use audit committee financial expertise (ACFE) to measure audit committee quality. Klein (2002a) finds that the main determinant of audit committee independence is board independence, and thus audit committee characteristics and board characteristics are highly correlated. In order to avoid multicollinearity, we introduce only audit committee characteristics (ACIND, LOG(ACSZ), and LOG(ACMEET)), along with audit fee ratio (RATIO), in Model We further control for auditor type (BIG4), auditor change (AUDCHG), size (LOG(TA)), growth (ADJSALEGR), acquisition (ACQUISTION), restructuring (RESTRUCTURE), segments (LOG(BUS)), and foreign currency translation (FOREIGN). 21 The coefficient on ACFE is significant at the one-percent level, and the coefficient on RATIO is significant at the fivepercent level. This supports Hypothesis 1 and rejects the null of Hypothesis 2. Our evidence suggests that firms are more likely to be identified with an internal control weakness, if their audit committees have less financial expertise or their auditors are more independent. However, our result on auditor independence should be interpreted with caution, as there is an alternative explanation for this positive coefficient on RATIO. Clients that purchase fewer non-audit services may have fewer discretionary resources. This lack of discretionary resources may lead to a lack of investment in internal controls, resulting in internal control 20 Following Klein (2002a), we take the natural logarithms of ACSZ, ACMEET, BDSZ, and BDMEET. Our results remain unchanged, if we use the raw variables. 21 In an early version of Doyle et al. (2006a), restructuring is measured as special items (#17) divided by lagged total assets (#6). We replace RESTRUCTURE with this measure in Models 1 and 2 of Table 4 and find that our results remain unchanged. We also measure size as log of total assets (#6) and find that our results in Table 4 remain unaltered. 21

24 weaknesses. In addition, the coefficient on AUDCHG is significant at the one-percent level and the coefficient on RESTRUCTURE is significant at the ten-percent level. Consistent with Doyle et al. (2006a) and Ashbaugh-Skaife et al. (2006), our findings imply that firms with recent auditor changes or restructuring activities are more likely to have internal control weaknesses. Finally, it is worth noting that a few of the insignificant results, noticeably the one on size, might be due to the matched sample design used in this study. In Model 2, we add board characteristics (BDIND, LOG(BDSZ), and LOG(BDMEET)) to Model 1, and find that our results on ACFE, RATIO, AUDCHG and RESTRUCTURE remain unchanged. Moreover, firms with a large board are less likely to have internal control weaknesses. Klein (2002a) finds that board size is positively associated with audit committee independence, implying that firms with a large board are more likely to have effective audit committees and thus are more likely to demand high quality auditing services. Thus, our finding on board size is consistent with that in Klein (2002a). Finally, board meeting frequency is found to be positively related to internal control weaknesses. Therefore, firms with internal control weaknesses are more likely to hold additional meetings, dealing with their internal control problems. We do not find the relation between audit committee independence and internal control weaknesses in Models 1 and 2, as does Krishnan (2005), because SOX requires audit committees to be composed of all independent board members. Models 3 and 4 replicate Models 1 and 2 by replacing ACFE with two separate measures: accounting financial expertise (ACCT_ ACFE) and non-accounting financial expertise (NONACCT_ACFE). The coefficients on ACC_ ACFE and NONACCT_ ACFE are all significant at the one-percent level, suggesting that both accounting and non- 22

25 accounting financial experts are helpful in improving internal controls. Other results are similar to those reported for Models 1 and 2. Thus, our findings are robust to different ways of measuring audit committee financial expertise. Some board variables, namely board size and board meeting frequency, are found to be related to internal control weaknesses in Table 4. This suggests that firms with strong corporate governance may be less likely to have internal control problems. As a natural extension to Table 4, we control for corporate governance in Table 5. Specifically, we use the overall measure of corporate governance developed in DeFond et al. (2005, pp ). They capture the strength of the governance environment using a summary measure that combines the following six governance characteristics into a single dichotomous variable: board size, board independence, audit committee size, audit committee independence, shareholders rights as captured by the G index used in Gompers, Ishii, and Metrick (2003), and institutional ownership. Because the G index information is only available for 52 pairs of our sample and control firms, 22 we adapt the procedure in DeFond et al. (2005) and create our governance variable (GOVERN) based on the following dichotomous measures of the five governance characteristics for each firm. 1) Board size We code firms 1 (for strong governance), if the firm s board size is less than the sample median and 0, otherwise. 22 When we run regressions based on these 52 pairs for Models in Table 5, we have one-tailed significance at the ten-percent level for the coefficient on audit committee financial expertise in Model 1 and the coefficient on non-accounting financial expertise in Model 2, respectively. 23

26 2) Board independence We code firms 1 (for strong governance), if 60% or more of the directors are independent and 0, otherwise. 3) Audit committee size We code firms 1, if the proportion of the firm s audit committee size to its full board size is greater than the sample median and 0, otherwise. 4) Audit committee independence We code firms 1, if the committee is composed only of independent members and 0, if the committee includes at least one affiliated member. 5) Institutional ownership We code firms 1, if the firm s percentage of institutional ownership is greater than the sample median and 0, otherwise. 23 We first summarize the five dichotomous measures for each firm and then create a dichotomous variable based on the median of the summed values. This governance measure is equal to one, indicating strong governance, if it is equal to or greater than the median summed values and zero, otherwise. Note that the number of observations for Table 5 is 206 pairs or 412 firms, because the institutional ownership information is missing for two pairs. Table 5 presents the empirical results after controlling for the above summarized measure of corporate governance (GOVERN). Note that we no longer include ACIND, ACSZ, BDIND, and BDSZ in our models, because these variables are incorporated into GOVERN. The findings in Table 5 are very similar to those in Table 4. To measure 23 We retrieve the institutional ownership information from Compact D and supplement thirteen firms that have missing ownership data with information collected from Yahoo! Finance. Our results in Table 5 are unchanged, when these thirteen firms are excluded from our analyses. 24

27 audit committee quality, we use ACFE in Model 1, and ACCT_ACFE and NONACCT_ACFE in Model 2. The coefficients on these variables are all significant at the one-percent level, whereas the coefficient on GOVERN is not significant. After we control the influence of corporate governance, the relation between audit committee quality and internal control weaknesses still holds. In addition, the coefficients on RATIO and AUDCHG are significant at the five-percent level or better. Thus, auditor independence and auditor change continue to be positively associated with the disclosure of internal control weaknesses Robustness checks We perform the following additional tests to verify that our results in Tables 4 and 5 are robust. (1) We use the natural logarithm of sales or market value of equity instead of the natural logarithm of total assets. 24 (2) We use the acquisition value defined in Doyle et al. (2006a), instead of the acquisition dummy. (3) We use raw ACSZ, ACMEET, BDSZ, and ACMEET, instead of the natural logarithms of these variables. In all these cases, our results are robust to these alternative specifications, adding credence to our findings. 5. Conclusion 24 Because of missing information, there are only 206 pairs or 412 firms, when we use the market value of equity. 25

28 In this paper, we examine the relation between audit committee quality, auditor independence, and disclosure of internal control weakness after the enactment of the Sarbanes-Oxley Act. We begin with a sample of firms with internal control weaknesses and, based on industry, size, and performance, match these firms to a sample of control firms without internal control weaknesses. The results from our conditional logit analyses suggest that a relation exists between audit committee quality, auditor independence, and internal control weaknesses. Firms are more likely to be identified with an internal control weakness, if their audit committees have less financial expertise or, more specifically, have less accounting financial expertise and non-accounting financial expertise, as well. They are also more likely to be identified with an internal control weakness, if their auditors are more independent. In addition, firms with recent auditor changes are more likely to have internal control weaknesses. 26

29 Appendix Variable Definitions ICW: 1, if a firm is identified with a material internal control weakness; 0, otherwise ACFE: Percentage of audit committee members who are financial experts ACCT_ACFE: Percentage of audit committee members who are accounting financial experts NONACCT_ACFE: Percentage of audit committee members who are non-accounting financial experts RATIO: Ratio of audit fee to total fee ACIND: Percentage of outside directors on the audit committee ACSZ: Audit committee size ACMEET: Number of audit committee meetings BDIND: Percentage of outside directors on the board BDSZ: Size of the board of directors BDMEET: Number of board meetings BIG4: 1, if the auditor is a member of the Big 4; 0, otherwise AUDCHG: 1, if there is an auditor change in 2003 or 2004; 0, otherwise TOTAL ASSETS: Total assets (#6) ADJSALEGR: Two-digit industry median adjusted sales growth, measured as the percentage change in sales from the previous year minus the twodigit industry median sales growth ACQUISITION: 1, if a firm engages in acquisitions during 2003, 2004 and from January to July of 2005; 0, otherwise RESTRUCTURE: 1, if a firm engages in a restructuring (non-zero values of #376, #377, #378, or #379); 0, otherwise BUS: Number of business segments reported in 10-K FOREIGN: 1, if there is a foreign currency translation (#150); 0, otherwise SALES: Total sales (#12), in millions EBITDA/SALES: Ratio of EBITDA (#13) to sales (#12) GOVERN: Please see definition in the text. Note: COMPUSTAT item numbers are in parentheses. 27