MSc Accountancy & Control, specialization Accountancy Faculty of Economic and Business, University of Amsterdam
The effect of the Sarbanes‐Oxley Act,
regarding audit committees, on audit
quality
What is the effect of the Sarbanes‐Oxley Act, regarding audit committees, on audit quality? Master Thesis, 19th June 2015 By: Bob Groen 10688277 Supervisor: Dr. R. Felleg Master thesis Accountancy & Control, 2014/2015, 6314M0044; word count: 19,1051
Statement of Originality
This document is written by student Bob Groen who declares to take full
responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and
that no sources other than those mentioned in the text and its references have
been used in creating it.
The Faculty of Economics and Business is responsible solely for the
supervision of completion of the work, not for the contents.
2
Abstract
After several huge accounting scandals the U.S. Senate and the House of Representatives enforced the Sarbanes‐Oxley Act (SOX). The Securities Exchange Commission stated that the SOX was implemented to increase audit quality and regain investors’ confidence. My study focuses on whether or not the implementation of the SOX led to an increase in audit quality for companies who (partially) complied with the Blue Ribbon Committee’s recommendations before the implementation of the Sarbanes‐Oxley Act. The purpose of this paper is twofold. This paper provides additional evidence, with regards to the effectiveness of audit committees, for prior studies. In addition, this paper studies the effectiveness of the Sarbanes‐Oxley Act on audit quality for companies that (partially) complied with the BRC guidelines. For this study I created three samples. One sample to answer each hypotheses. Each sample was created from a merger with the Compustat and ISS database. For each sample I calculated the absolute discretionary accruals, the negative discretionary accruals and the positive discretionary accruals. For each sample and type of accrual I created a correlation matrix, a multicollinearity test and an ordinary least square regression. I did not find any evidence that companies who complied with all BRC/SOX guidelines pre‐SOX would not see a significant increase in audit quality if they complied with all BRC/SOX guidelines post‐SOX. I also did not find any evidence that companies who did not comply with all BRC/SOX guidelines pre‐SOX would not see an increase in audit quality if they complied with all BRC/SOX guidelines post‐SOX. This research contributes to the knowledge of the implementation of new/updated rules and directives and whether these rules are based on what is socially acceptable or if these rules are intended to actually improve the audit quality. In addition, this study not only explains the statistically significance of the results but also the economic significance of the results. The main limitation of this paper is the high amount of noise in the samples. In addition, in my study I only used one method to describe audit quality. Keywords: Audit committee; audit quality; SOX; BRC; accruals.3
Table of Contents
1 Introduction ... 5 2 Literature review and hypotheses ... 7 2.1 Audit committee ... 7 2.1.1 Independence of the audit committee ... 9 2.1.2 Composition of the audit committee... 10 2.1.3. Size of the audit committee ... 11 2.1.4 The frequency of audit committee meetings ... 11 2.2 Audit quality ... 12 2.2.1 Measuring audit quality ... 12 2.2.2 Audit quality indicators ... 13 2.2.2.1 Input‐based audit quality indicators ... 13 2.2.2.2 Output‐based audit quality indicators ... 14 2.3 Hypotheses ... 17 3 Research methodology ... 20 3.1 Sample ... 20 3.2 Measures of audit quality ... 21 3.3 Measures for complying with SOX/BRC requirements ... 22 3.4 Control variables ... 22 3.5 Empirical model ... 244 4 Results ... 26 4.1 Descriptive statistics ... 26 4.2 Results of hypothesis tests ... 29 4.2.1 The effect of having an audit committee on audit quality ... 30 4.2.1.1 Correlation hypotheses 1 ... 30 4.2.1.2 OLS regression hypotheses 1 ... 33 4.2.1.3 Conclusion hypotheses 1 ... 34 4.2.2 The effect of complying to all BRC/SOX guidelines ... 37 4.2.2.1 Correlation hypotheses 2A ... 37 4.2.2.2 OLS regression hypotheses 2A ... 40 4.2.2.3 Conclusion hypotheses 2A ... 42 4.2.3 The effect of not complying to all BRC/SOX guidelines ... 46 4.2.3.1 Correlation hypotheses 2B ... 46 4.2.3.2 OLS regression hypotheses 2B ... 49 4.2.3.3 Conclusion hypotheses 2B ... 51 5 Conclusion ... 55 References ... 59
5
1 Introduction
After a series of corporate (accounting) scandals the U.S. Senate and the House of Representatives enforced the Sarbanes‐Oxley Act (SOX). This meant that all companies, listed on the U.S. stock exchange, were forced to implement an audit committee which complied with the guidelines stated in the SOX. The U.S. senate and the House of Representative stated that the SOX was implemented to regain investors’ confidence (SEC, 2003). However, the majority of the guidelines stated in the SOX were based on a report by the Blue Ribbon Committee (BRC)(Myers and Ziegenfuss, 2006). Because there were no clear guidelines regarding audit committees prior to 1999, the BRC introduced, in 1999, recommendations regarding the audit committee. These recommendations were adopted by over 90% of the companies listed on the U.S. stock exchange (Myers and Ziegenfuss, 2006). My research objective is to study the relationship between the recommended guidelines by the BRC and the mandatory guidelines by the SOX with regards to the audit committee. Like I stated before, the SOX is based, for the most part, on the BRC guidelines. The main difference between the BRC recommendations and the SOX guidelines is that the BRC was voluntary and the SOX is mandatory. However, over 90% of the companies complied with the BRC recommendations. My motivation for this research is whether or not a mandatory audit committee will lead to a higher quality audit or that the implementation of a mandatory audit committee is more for ‘’show’’ (There are scandals so we introduce new rules). Based on my motivation I formulated the following research question: ‘‘What is the effect of mandatory audit committees, for US listed companies, on audit quality?’’ I formulate three hypotheses based on my research question. For each hypotheses I will calculate the absolute accruals (ABSDA), the negative accruals (NEGDA) and the positive accruals (POSDA). I first research whether or not having an audit committee would lead to an increase in audit quality. For this research I will use a total sample of 10,089 firm‐years. For hypothesis 2A I will use a sample of 5,112 firm‐years. This sample consists of companies that complied with all BRC/SOX guidelines pre‐SOX and companies that6 complied with all BRC/SOX guidelines post‐SOX. This hypothesis will answer the question as to whether or not a company that has an audit committee that complied with all BRC/SOX guidelines pre‐SOX, would see an increase in audit quality when they complied with all BRC/SOX guidelines post‐SOX. My third and final sample consists of 6,630 firm‐ years. This sample consists of companies that did not comply with all BRC/SOX guidelines pre‐SOX and companies that did comply with all BRC/SOX guidelines post‐SOX. The sample will be used to answer the question about whether or not companies that did not comply with all BRC/SOX guidelines pre‐SOX will see an increase in audit quality if they complied with all BRC/SOX guidelines post‐SOX. I found significant statistical evidence that the audit quality for companies with an audit committee is higher than for companies without an audit committee. However, I did not find any evidence that companies who complied with all BRC/SOX guidelines pre‐SOX would not see a significant increase in audit quality if they complied with all BRC/SOX guidelines post‐SOX. I also did not find any evidence that companies who did not comply with all BRC/SOX guidelines pre‐SOX would not see an increase in audit quality if they complied with all BRC/SOX guidelines post‐SOX. From a social point of view this research is interesting because it investigates whether or not the implemented rules are used to improve the audit quality or to satisfy the users of the annual report. This research contributes to the knowledge of the implementation of new/updated rules and directives and whether these rules are based on what is socially acceptable (‘’do the right thing’’) or if these rules are intended to actually improve audit quality. In addition, this study not only explains the statistically significance of the results but also the economic significance of the results. Also, this study corroborates earlier research by Abbott et al. (2004). In the second chapter of this research I will review earlier literature. In addition, the second chapter will also state the hypotheses of this research. The third chapter will explain the research methodology and the fourth chapter will discuss the results. Finally, in chapter five, I will conclude my findings and answer my research question. In the fifth chapter I will also state the limitations of my research and I will do some suggestions about possible future research.
7
2 Literature review and hypotheses
2.1 Audit committee
In 1940 the SEC made its first recommendation, towards companies, to establish an audit committee (Williams, 1977). This recommendation was issued based on the aftermath of the McKesson & Robbins Inc. fraud in the late 1930s and was introduced to enhance the quality of the audit. Due to this fraud the New York Stock Exchange (NYSE) and the Securities and Exchange Commission (SEC) recommended that the audit committee, composed of non‐executive board members, should be appointed to select the external auditor of a company (Williams, 1977; Ghafran and O’Sullivan, 2013). Until the late sixties not many companies acted on the recommendations of the SEC and the NYSE, even though the Journal of Accountancy expressed, in 1953, support for the concept. In the late sixties the concept of an audit committee began to attract more attention. This led the American Institute of Certified Public Accountants (AICPA) to recommend that companies, that were publicly owned, should appoint independent directors to nominate the independent auditor (Williams, 1977, p. 71). But, research done by Mautz and Neumann (1970), shows that out of the 385 surveyed companies, only 121 reported that they had implemented an audit committee. It was not until the late 1970s that companies started to act on the recommendations of the SEC and the AICPA. In the late 1970s the corporate failures of Franklin National Bank and Penn Central Transportation Co. motivated regulators to recommend that listed firms in the U.S. should create an audit committee (Zeff 2003; Ghafran and O’Sullivan, 2013). The recommendation of regulators to implement an audit committee led the NYSE, in 1977, to require all listed companies on the NYSE to implement an audit committee. However, the rules about the size and composition of an audit committee were considered to be vague at best (Bronson et al. 2009). In the late 1990s the SEC formed the Blue Ribbon Committee on Improving the Effectiveness of Corporate Audit Committees (BRC). The formation of the BRC was part of Chairman Levitt’s (Chairman of the SEC) action plan to increase accounting integrity and to reinforce the audit committees’ oversight (Backman and Salan, 1999). The main goal of the BRC was to develop guidelines to improve the effectiveness of the audit committee and thereby improve audit quality (BRC, 1999). According to the BRC the size of the audit8 committee, the independence of the audit committee, the financial expertise and literacy of the audit committee members and the frequency of meetings the audit committee has, have got an effect on audit quality (BRC, 1999). This is corroborated by Abbott et al. (2004) who found that implementing these recommendations affect audit quality. In their research they argue that the BRCs report represents an organized set of best practices which should result in a more effective oversight by the audit committee on the financial reporting process and hence, increase audit quality (Abbott et al., 2004). These recommendations led the NYSE and the NASD to modify the listing requirements for companies. This meant that companies that were listed on these stock exchanges were required to create independent audit committees or if they did not comply, explain why they would not comply (Bronson et al. 2009). Research shows that one year after the BRC recommendations, in 2000, more than 75 percent of the researched companies (232/296) had taken steps to address all the BRC effectiveness recommendations and that only 5 percent of the sample did not make any effort to address the recommendations (Myers and Ziegenfuss, 2006). In 2001 the high profile corporate and accounting scandals of Enron, WorldCom and Arthur Andersen led to a decline in investors’ confidence in financial statements (Giroux, 2008). This decrease in investors’ confidence was the main motivation for the passage of the Sarbanes‐Oxley Act (SEC, 2003). In 2002 the US Senate and the House of Representatives passed the Sarbanes‐Oxley Act (SOX). One of the key components of the SOX is to strengthen audit committees and corporate governance to enhance audit quality (EY, 2012). The adoption of the SOX means that every listed company in the United States is required to implement an audit committee independent of management (SOX section 301). These companies are also required to disclose whether at least one ‘’financial expert’’ is on the audit committee (SOX section 407) and they are required to have the independent audit committee to be directly responsible for the appointment, compensation and oversight of the external auditor (SOX section 202). The SOX does not set any other requirements regarding the audit committee. This means that companies can freely decide on the amount of audit committee meetings, the size of the audit committee, the composition of the board, et cetera. With the passing of SOX the rules regarding the audit committee were changed from a principle‐based structure to a rule‐based structure. This
9 meant that every listed company was forced to implement an audit committee and that the ‘’comply or explain’’ approach no longer sufficed (Hoeven, 2005). 2.1.1 Independence of the audit committee Research shows that one of the characteristics that affects audit quality is the independence of the audit committee board members (e.g. Carcello et al. 2011; Abbott et al. 2004; Archambeault et al. 2008). Carcello et al. (2011) states that independent audit committees are associated with less restatements, and therefore higher audit quality, for several reasons. First, they argue, based on research done by Abbott et al. (2003), that an independent audit committee requires more extensive external audit procedures. These more extensive external audit procedures should increase the likelihood that material misstatement will be discovered by the auditor before the financial statements are issued. Secondly, the effectiveness of the internal audit, and therefore the quality of the external audit, will improve when the function of the internal audit would report to an entire independent audit committee (e.g. Deli and Gillan, 2000; Raghunandan et al. 2001). Finally, Carcello et al. (2011) argue that an independent audit committee increases audit quality because the audit committee would, in its own attempt of overseeing management, review the financial statements before the financial statements are issued. Moreover, researchers also found that audit committee independence is associated with higher quality reporting in case of a client’s financial distress (Carcello and Neal, 2000), less dismissals of auditors after a going‐concern opinion is issued (Carcello and Neal, 2003), and research shows that companies with an independent audit committee face less cases of fraudulent financial reporting (Beasley et al. 2000). The BRC report states that independence can be achieved when the audit committee member does not have a relationship to the company which could interfere with the independence of the audit committee member. They advise to exclude directors, controlling shareholders, or executive officers who are involved with other companies from which the entity received significant payments (Abbott et al. 2004). The BRC recommends this exclusion because an audit committee member without any financial, family or other
10 ties to the company is better able to evaluate the internal control and reporting practices (BRC, 1999). Also, according to the BRC, research shows that there is a correlation between the degree of audit committee independence and audit quality (BRC, 1999). The correlation between the degree of audit committee independence and audit quality means that the audit quality increases when the number of independent audit committee members increases. However, research done by Bronson et al. (2009) argues that the benefits related to the independent audit committee, are only consistently achieved, when the audit committee is completely independent (every member of the audit committee is independent). Also, Carcello et al. (2011) argue that the definition of independence, as stated by the BRC and the SOX, can be more clearly defined. Even though the stricter regulation of the SOX limits the possibility of hiring dependent audit committee members, it is not possible to consider all the countless personal connections between directors and audit committee members. Therefore Carcello et al. (2011) argue that there are two types of independence. These are ‘’independence in fact’’ and ‘’independence in appearance’’. Carcello et al. (2011) argues that if an audit committee member is independent in appearance, which means that the audit committee member has personal ties to a director or management, the independence of the audit committee member is negatively affected. Hence, Carcello et al. (2011) argue that independence can only be achieved if the audit committee member is independent in fact (which means the audit committee member has no personal ties to any director or someone in management). So according to Carcello et al. (2011), if an audit member is not ‘’independent in fact’’, the company does not comply with the rules regarding audit committee independence. This means that the member that is not ‘independent in fact’ needs to be dismissed. 2.1.2 Composition of the audit committee One of the guidelines of the BRC and the SOX is about the composition and size of the audit committee. Both the SOX and the BRC state that at least one member of the audit committee has to be financially literate (SOX section 407, 2002; BRC recommendation 3, 1999). In addition, the BRC also recommends that out of all the financially literate audit committee members at least one of them has to be an expert in financial management or accounting. The BRC defines financial literacy as ‘’the ability to read and understand
11 fundamental financials statements’’ and the BRC states that whether or not an audit committee member possesses accounting and/or financial management skills can be based on his or her employment experience or CPA certification (Abbott et al. 2004). These guidelines are based on research done in the past. This earlier research shows that having financial expertise on the audit committee is associated with a lower likelihood of restatements, thus associated with a higher degree of audit quality (e.g. Kalbers and Fogarty 1993; McMullen and Raghunandan 1996; Abbott et al. 2004; Agrawal and Chadha 2005). Several researchers have studied the relation between the financial expertise of the audit committee and earnings management. The results of these studies show that financial experts in an audit committee is an important factor to constrain managers in engaging in earnings management. This means that having a financial expert in the audit committee leads to an increase in audit quality (e.g. Xie et al. 2003; Bédard et al. 2004). 2.1.3. Size of the audit committee Neither the SOX nor the BRC include guidelines towards the size of the audit committee (besides the number of financial literate members). Jensen (1993) and Yermack (1996) state that smaller audit committees are more effective in monitoring whereas Beasley (1996) and Salterio (2001) found that larger audit committees are more effective in monitoring because a larger audit committee has more knowledge to their disposal. The study done by Abbott et al. (2004) researched the optimal size for an audit committee. They found a significant statistical relation between audit quality and the size of the audit committee but did not find an optimal number of audit committee members. They do state that an audit committee with too many members might lead to an ineffective audit committee (Abbott et al. 2004) which ultimately leads to a decline in audit quality. However, they do not mention what number of audit committee members is too many (Abbott et al. 2004). 2.1.4 The frequency of audit committee meetings Neither the BRC nor the SOX give guidelines towards the number of audit committee meetings. Because of the wide variety of types of companies, the BRC argues that it is not possible to mandate a specific number of audit committee meetings for each type of company (BRC, 1999). Even though there are no guidelines towards the frequency of audit
12 committee meetings, researchers did study the effect that the number of audit committee meetings had on audit quality (Abbott et al, 2004; Xie et al. 2003; Kent et al. 2010). Researchers found a significant negative association with the frequency of audit committee meetings and audit quality (e.g. Abbott et al. 2004; Xie et al. 2003; Kent et al. 2010). This means that the chance of a financial restatement is higher for companies that have an audit committee that meets less than four times a year. Research that was done by DeZoort et al. (2002) suggests that when audit committees meet frequently the extent of financial misreporting can be reduced. But, Ghafran and O’Sullivan (2013) mention that other studies that link the frequency of audit committee to the impact of earnings management show mixed results. This is supported by other research which did not find evidence for an association between the number of audit committee meetings and audit quality (e.g. Bédard et al. 2004; Baxter and Cotter 2009).
2.2 Audit quality
Through the years the term audit quality has been thoroughly researched. But, after decades of research, there has been little agreement on how to define audit quality (Dickins et al. 2014). Researchers like Abbott et al. (2004) define the term audit quality as the ability that auditors are able to discover and report misstatements, while others claim that audit quality is the ability that auditors meet investors’ needs (PCAOB, 2013). 2.2.1 Measuring audit quality Just like the definition of audit quality, there is also little agreement on how audit quality should be measured. For instance, Abbott et al. (2004) uses restatements to measure audit quality while others use discretionary accruals to measure audit quality (e.g., Cahan et al. 2011; De Vlaminck and Sarens, 2015; Ghosh et al. 2010). Dickins et al. (2014) states that the PCAOB categorizes audit quality measures as either input‐audit quality measures or as output‐audit quality measures. The PCAOB makes this distinction between input‐ and output‐audit quality indicators in order to be better able to more clearly identify audit quality indicators. The objective for separating audit quality indicators in input‐ and output‐quality indicators is so that these indicators can help the decision making process of audit committees, investors and managers in positively influencing the audit firms and to be better able to inform regulators about audit quality (Dickins et al. 2014).13 2.2.2 Audit quality indicators The United States Department of the Treasury’s Advisory Committee on the Auditing Profession (TACAP) describes two types of audit quality indicators (AQI’s). These are the AQI’s that are output‐based, also referred to as output‐AQI’s and the input‐based, also referred to as input‐AQI’s. The TACAP states that output‐based indicators are: ‘’Indicators determined by what the auditing firm has produced in terms of its audit work, such as number of frauds discovered and the nature and reason for financial restatements related to time periods when the underlying reason for restatement occurred during the auditing firm’s tenure as auditor for the client.’’ The TACAP formulates input‐based indicators as: ‘’Indicators of what the auditing firm puts into its audit work to achieve a certain result, such as the auditing firm’s processes and procedures used for detecting fraud, the average experience level of auditing firm staff on individual engagements, the average ratio of auditing firm professional staff to auditing firm partners on individual engagements, and annual staff retention’’ (Advisory Committee on the Auditing Profession, (2008), pp. 116). The PCAOB (2013) states several examples of input‐AQI’s and output‐AQI’s. Because the concept of separating audit quality measures into input‐AQI’s and output‐AQI’s is relatively new, the PCAOB states that the examples that they give might not be complete or might not include the most relevant indicators of audit quality (PCAOB, 2013). 2.2.2.1 Input‐based audit quality indicators The PCAOB (2013) divides the input‐AQI’s into operational inputs and process inputs. Examples of operational inputs are; the ratio of partners to staff, the chargeable hours per professional, the training hours per professional, the percentage of work outsourced to service centres and the specialist hours as a percentage of the overall engagement hours. A few of the examples that the PCAOB (2013) gives regarding process inputs are; the number and nature of internal quality review findings, the leverage ratio of audit staff to partners, the number and size of auditor resignations and the percentage of clients assessed as high risk. The information, which is needed for input‐based AQI’s, comes from audit firms and/or regulators. The information required for input‐based AQI’s is usually confidential. Dickins et al. (2014) state that it might be possible to predict audit quality based on input indicators. But, because most of the information that is needed to use input‐based AQI’s is
14 not publicly available, either due to confidentiality or because there are no databases that record the required information, it would be a hard task to use input‐based AQI’s to measure audit quality. 2.2.2.2 Output‐based audit quality indicators The second type of audit quality indicators are the output‐based AQI’s. The PCAOB (2013) states several examples of output‐AQI’s. A few examples of these output‐AQI’s are the frequency and market impact of financial restatements, the calculation of discretionary accruals to measure audit quality, the number of audit reports including a going concern opinion which did not have a subsequent bankruptcy and the number of audit reports lacking a going concern opinion which had a subsequent bankruptcy. The first measure, the frequency and market impact of financial restatements, is a relatively solid signal of audit quality (PCAOB, 2013). AuditAnalytics has created a database that covers all SEC registrants, since 1 January 2001, which disclosed a financial restatement (AuditAnalytics, 2015). The database created by AuditAnalytics is used by many academia to research, for example, audit quality (e.g., Drake et al. 2015; Francis et al. 2013; Ettredge et al. 2013). The ‘’Restatements’’ section of the AuditAnalytics database makes a distinction between restatements due to fraud, irregularities and misrepresentations, restatements due to accounting and clerical errors and restatements due to application failures of accounting rules (GAAP/FASB). Francis et al. (2013) argue that an audit is of high quality when the auditor correctly enforces the General Auditing Accounting Principles (GAAP). Francis et al. (2013) refers in their research to the SAS guideline, which state that auditors have ‘’a responsibility to plan and perform the audit to obtain reasonable assurance about whether the financial statements are free of material misstatement, whether caused by error or fraud’’ (SAS No.1 1972). Francis et al. (2013) state that it depends on the auditor’s own competence to give the right statement. Therefore they imply that a restatement, based on the application failure of the accounting rules, is the outcome of a low‐quality audit. They argue that restatements due to an internal company error or fraud cannot be fully attributed to the competence of the auditor and therefore cannot be used by academics to research audit quality (Francis et al. 2013). The PCAOB (2013) argues that the frequency of restatements may also be a good indicator for a
15 low‐quality audit. However, the PCAOB (2013) states that the users of financial statements have to keep in mind that there are other factors besides poor audit quality, like the complexity of the financial statements, that could lead to restatements. The second measure that is given by the PCAOB (2013) is the use of discretionary accruals. Although, the PCAOB (2013) has not reached a consensus about which earnings management model should be used to measure audit quality, in academic literature the use of discretionary accruals as a measure of audit quality is prevalent (e.g. Cahan et al. 2011; De Vlaminck and Sarens, 2015; Ghosh et al. 2010). The most common way to measure discretionary accruals is by using the Jones model (Jones, 1991) or its modifications (e.g. Dechow et al. 1995; Kothari et al. 2005; Cahan et al. 2011). Jones (1991) used the model of Healy (1985) and DeAngelo (1986) to develop a way in which she could separate the discretionary accruals from the non‐discretionary accruals. Accruals can be divided in non‐discretionary accruals and discretionary accruals. The difference between these types of accruals is that management can affect the discretionary accruals and they cannot affect the non‐discretionary accruals. Francis and Krishnan (1999) state that discretionary accruals are important to accountants because discretionary accruals have an effect on the inherent risk of a client. Because discretionary accruals are based on future events they are susceptible to the judgement of management. These estimation errors can be used by managers that have an incentive to manage earnings. The incentives managers have to influence earnings are created by earnings based contracts or targets that managers need to reach (Becker et al. 1998). These incentives lead to information asymmetry. Becker et al. (1998) states that auditing reduces information asymmetry between the managers and the stakeholders of a firm. They argue that the information asymmetry is reduced because there is an outside party to verify the validity of the financial statements. Becker et al. (1998) argue that factors that affect management’s ability to manage earnings, and therefore create information asymmetry, are the quality of a firm’s internal governance and previously made accounting decisions that restrict choices in discretionary accruals in the future. Becker et al. (1998) states that the effectiveness of an audit, and the ability that the auditor can constrain management to take part in earnings management, can vary with the quality of the auditor. They argue that
16 a high quality auditor is better able to detect accounting practices that are questionable, like discretionary accruals, and to object to their use. Hence, low discretionary accruals are positively related to high audit quality (e.g. Becker et al. 1998; Francis et al. 1999). However, the downside to the use of discretionary accruals are estimation errors. That is why Dechow et al. (1995) modified the original Jones’ model to reduce the amount of estimation errors in discretionary accruals. Dechow et al. (1995) tested several discretionary models, by artificially adding a fixed and known amount of accruals to each firm‐year to test each model. They found that a modified version of the Jones model, that is developed to detect revenue‐based earnings management, is the best model to reduce estimation errors and therefore the best way to measure audit quality if a researcher uses discretionary accruals. The third measure is based on the fact that whether or not an entity went bankrupt after receiving a going concern opinion. This means that if there was a going concern opinion issued and the entity did not go bankrupt, according to the PCAOB (2013), this can be seen as a low‐quality audit. The fourth measure, which is the opposite of the third measure, is when an entity goes bankrupt and the auditor did not give a going concern opinion, according to the PCAOB, this can be interpreted as a low‐quality audit (PCAOB 2013). The downside to using the going concern opinion as a measure for audit quality is the fact that accountants are conservative (Basu, 1997). Basu (1997) states that accountants need a higher degree of verification for good news than for bad news. This means that the accountant will be inclined to issue a going concern opinion when there is a slight chance that the entity might go bankrupt. This means that a company can receive a going‐concern opinion even though the chances for bankruptcy are small. Hence, the assumption made by the PCAOB that companies that are not going bankrupt after receiving a going‐concern opinion is a result of bad auditing and therefore low audit quality, could be argued to be invalid. In addition, Blay et al. (2011) found that if a company receives a going concern opinion the market immediately increases the risk of bankruptcy for that company.
17
2.3 Hypotheses
Prior to the recommendations of the BRC and the implementation of the SOX, regulatory bodies and other official organizations stated that the role of the audit committee is important in improving financial reporting (Kalbers and Fogarty, 1993). However, until 1993 there has been little empirical research on the effectiveness of the audit committee and on what characteristics of the audit committee affect audit quality (Kalbers and Fogarty, 1993). One of the first empirical studies, done by Kalbers and Fogarty (1993), show that companies that implement an audit committee will see an increase in audit quality compared to companies that do not implement an audit committee. This is corroborated by several other studies that researched the characteristics of the audit committee and how these characteristics affect audit quality (e.g. Abbott et al. 2004; Archambeault et al. 2008; De Vlaminck and Sarens, 2013). The first hypothesis I formulate is to check whether or not my findings correspond to the findings of earlier research. Based on the aforementioned I formulate my hypothesis as follows: H1: Audit quality increases for companies that did not have an audit committee pre‐SOX. The ‘’comply and explain’’ approach meant that companies could opt to not implement an audit committee if, for example, implementing an audit committee did not lead to a reduction in agency costs. (e.g. Pincus et al. 1989; Ghafran and O’Sullivan, 2013; De Vlaminck and Sarens, 2015). Agency theory suggests that the principal (owner) and the agent (management) have different incentives. These differences can lead to agency problems and costs. For example, management tries to generate as much revenue as possible because the size of their bonus depends on the size of revenue. However, this could mean that management takes risks which the principal is not willing to take. To reduce these agency costs the principal can force management to implement an audit committee (Jensen and Meckling, 1976). The audit committee is viewed as a monitoring mechanism which improves the flow of information between the principal and the agent (Pincus et al., 1989). Myers and Ziegenfuss (2006) researched the differences between the ‘’comply or explain’’ approach (recommendations of the BRC) and the mandatory approach (SOX legislation). They researched how independence is defined, whether or not an audit18 committee should be independent, the composition and financial literacy of the audit committee members and the responsibilities of the audit committee. They found that all recommendations of the BRC are similar to the guidelines of the SOX. Myers and Ziegenfuss (2006) also state that although the BRC recommendations did not have the same weight as the SOX legislation, the BRC recommendations were well‐ publicized and companies were able to implement these recommendations before the SOX legislation came into effect (Myers and Ziegenfuss, 2006). This is corroborated by the fact that the manufacturing industry saw an increase in the percentage of independent audit committee members of 18 percent (from 74 percent to 92 percent) during the period 1999‐ 2000 (before the introduction of the BRC until one year after the introduction). Overall, when all 11 industry groups are examined, the percentage of companies that reported independent financially literate audit committee members increased from 90.9 percent to 93.2 percent during the period 1999‐2000 (Myers and Ziegenfuss, 2006). Based upon the above, I argue that the implementation of the SOX did not result in a significant increase in audit quality for companies that already (partly) complied with the recommendations of the BRC. Firstly, based on the agency theory, the rules regarding the audit committee should be based on a ‘’comply or explain’’ approach. However, the U.S. implemented, quickly after the introduction of the BRC recommendations, mandatory guidelines. Furthermore, Sharma et al. (2009) made an overview of several countries, which represent different geographic regions, and how these countries regulate the implementation and structure of the audit committee. The countries that Sharma et al. (2009) use are the U.S., the U.K., Spain, Australia, New Zealand, Singapore and China. All these countries, except for the U.S. and China, have a voluntary audit committee where they use the ‘’comply or explain’’ approach. In China the audit committee is voluntary and does not use the ‘’comply or explain’’ approach. In the U.S. the audit committee is mandatory (Sharma et al. 2009). This research shows that most developed countries use a ‘’comply or explain’’ approach. I argue that if mandatory audit committees would lead to a significant increase in audit quality, compared to the audit committees based on the ‘’comply or explain’’ approach, more
19 countries would have adopted the mandatory audit committees. Therefore, I reject the idea that making the audit committee mandatory has a significant effect on audit quality. Secondly, because the SOX legislation is similar to the recommendations of the BRC, I argue that the implementation of the SOX would not lead to a significant increase in audit quality for companies that had audit committees that complied with all the recommendations introduced by the BRC. H2a: Ceteris paribus, companies with an audit committee that fully complied with the BRC recommendations pre‐SOX will not see a significant increase in audit quality post‐SOX. Thirdly, Myers and Ziegenfuss (2006) found that over 93 percent of the companies complied with the BRC recommendations. This means that only 7 percent of the companies did not comply with the BRC recommendations. The agency theory predicts that companies with poor governance do not want a strong audit committee. This means that companies with poor governance, hence companies that do not comply to all BRC recommendations, would see a lower audit quality than companies that do comply with all BRC recommendations. Therefore I argue that the implementation of mandatory guidelines (SOX) leads to an increase in audit quality for companies that did have an audit committee pre‐SOX but did not comply with all BRC recommendations. Based on the aforementioned, I formulate my hypothesis as follows: H2b: Ceteris paribus, companies that did have an audit committee pre‐SOX but did not (fully) comply with the BRC recommendations regarding the audit committee pre‐SOX will see an increase in audit quality post‐SOX.
20
3 Research methodology
3.1 Sample
For my research I created three different samples. My main sample consists of firms that are included in the ISS Directors Legacy database from 1996 through 2006. My sample starts at 1996 and ends at 2006 because this is the range of the ISS Directors Legacy database. After creating the necessary variables I was left with 17,409 firm‐years. After merging the data with data gotten from the Compustat North America Fundamentals Annual database I was left with 11,631 firm‐years. After merging the databases I excluded 299 firm‐years for companies in the financial services industry (SIC codes 6000‐6900). I excluded the financial services industry because the accruals of these companies are expected to diverge from the accruals of companies in other industries. Besides that I also excluded companies that missed data on governance and companies that missed financial data. After creating the variables for my hypotheses’ I was left with a total sample of 10,089 firm‐years to estimate ACORNOT, which is the variable that answers my first hypotheses. In my research I classify the firm‐years prior, or equal to 2002 as the years pre‐SOX and I classify the firm‐years equal or after 2003 as post‐SOX. I also state that companies comply with all BRC/SOX requirements when: (1); there is an audit committee. (2); all members of the audit committee are independent. (3); at least one member of the audit committee is financially literate. For the estimation of the second sample I used the main sample of 10,089 firm years. I then dropped all companies that did not have an audit committee. I dropped all companies without an audit committee because I am only comparing companies that do have an audit committee. This left me with a sample of 8,897 firm‐years. After dropping all firm that did not comply with all BRC/SOX guidelines pre‐SOX I was left with a sample of 5,112 firm‐ years. I use this sample to answer hypotheses 2A. For the estimation of the third sample I again used the main sample of 10,089 firm‐years. I then dropped, again, all companies that did not have an audit committee. This left me with a sample of 8,897 firm years. After dropping all the firms that did comply with all BRC/SOX guidelines pre‐SOX I was left with a sample of 6,630 firm‐years. I dropped all companies21 that did comply with the BRC/SOX guidelines because, for hypotheses 2B, I am comparing companies that did not comply with all BRC/SOX guidelines pre‐SOX with companies that did comply with all BRC/SOX post‐SOX. That means that I do not want any companies that did comply with all BRC/SOX guidelines in the sample for hypotheses 2B.
3.2 Measures of audit quality
Based on earlier research the two most common ways to research audit quality is by using discretionary accruals and/or restatements. I have chosen to only use discretionary accruals to measure audit quality. I use discretionary accruals due to the limited number of available restatements for my sample size. To measure discretionary accruals I use the model created by Cahan et al. (2011). This model estimates the modified Jones model (1991) cross‐sectionally and includes a correction for performance (Kothari et al., 2005). The Cahan model also includes, as suggested by Ashbaugh et al. (2003) and Reichelt and Wang (2010), lagged return on assets. In the Cahan model the lagged return on assets is used to estimate the non‐ discretionary accruals. The following model is estimated for all industries with two‐digit SIC codes, with sufficient data and with at least ten observations in year t: ß ß 1 ß ∆ ∆ ß ß 1 , TAC stands for the total accruals that are computed from the cash flow statement, which is equal to income before extraordinary items minus operating cash flows and adjusted for discontinued operations and extraordinary items. ∆SALES stands for the change in sales revenues in year t and ∆AR is the change in accounts receivable in year t. PPE stands for the gross plant, property and equipment in year t and LAGINC stands for the lagged income before extraordinary items. ‘A’ stands for the lagged total assets. To mitigate the effects of extreme observations I winsorized the accrual‐based measure to the top and bottom 1% of the sample. Consistent with prior research by Kothari et al. (2005) I use the residual of (1) as an estimate for discretionary accruals. Based on research by Cahan et al. (2011) I use the absolute value of discretionary accruals (ABSDA), negative discretionary accruals (NEGDA)22 and positive discretionary accruals (POSDA) to measure audit quality. Research done by DeFond and Subramanyam (1998) and Kim et al. (2003) state that managers have incentives to overstate earnings and that the risk, that the auditor gets sued, is higher if earnings are overstated. Therefore auditors are more likely to monitor positive discretionary accruals than negative discretionary accruals (Cahan et al. 2011). Based on this assumption I am especially interested in positive discretionary accruals (POSDA).
3.3 Measures for complying with SOX/BRC requirements
I created the proxies ACORNOT, ACCOMPLYALL and ACNOTCOMPLYALL to test my hypotheses. However, the database does not have a variable that shows whether or not an audit committee complies with all BRC/SOX guidelines. So, to answer the second hypotheses I needed to create a variable to show whether or not a company complied with the BRC/SOX requirements (complySOXBRC). I used six different variables from the ISS Directors Legacy database to code the variable complySOXBRC. I used the variable classification to check for independence and the variable audit_membership to check whether or not a company had an audit committee. For the last requirement, the financial literacy of at least one audit committee member, I created the variable aclit. I had to create this variable because there is no financial literacy variable available in the ISS Directors Legacy database. Earlier research by Abbott et al. (2004), Carcello et al. (2011) and Archambeault et al. (2008) show that an audit committee member can be classified as financially literate if the member is either an accountant, CFO, CEO, treasurer, investor, offers real estate services or a small business owner. Based on this research I used the variables empl_category, employment_ceo, employment_cfo and employment_treasurer to create the new variable aclit. I also used the variable empl_category because this variable includes accountants, investors and employees that offer real estate services or are small business owners.3.4 Control variables
I use five control variables in my study. The first control variable I use is ALL_ATT_75pct. The variable ALL_ATT_75pct is a dummy variable that measures whether or not all audit committee members attended at least 7523 percent of the meetings, with a minimum of four meetings. For this dummy variable the value ‘0’ means that at least one audit committee member did not attend at least 75 percent of the audit committee meetings and the value ‘1’ means that all audit committee members attended at least 75 percent of the audit committee meetings. This dummy variable is based on research done by Abbott et al. (2004) who argue that all publicly held firms should have their audit committee meet at least four times a year. The threshold of at least four audit committee meetings is based on the fact that, at the minimum, the audit committee should review the quarterly results of the company. Secondly, I included the dummy variable BIG5 to control for companies with a big 5 auditor and companies without a big 5 auditor (Arthur Andersen, KPMG, EY, Deloitte, PWC). This dummy variable is based on research done by Michas (2011) who found that total and abnormal accruals are lower for companies that are audited by a big 4 auditor. This later research is based on the big 4 because Arthur Anderson, the auditor I include in my sample, was bought by Deloitte. The third control variable I use in my research is the control variable firm size (natural log of average total assets). I use this control variable because earlier research shows that the control variable firm size is a common control variable to control for audit quality (Michas, 2011; Abbott et al., 2004; Archeambeault, 2008). The fourth control measure I use is the control variable leverage (total long‐term debt divided by average total assets). I use this control variable because research done by Watts (2003) found that companies with a higher level of leverage often have greater conflicts with debt‐holders and shareholders. This means that these companies are at higher risk for using discretionary accruals to meet debt‐ and shareholders’ needs. The fifth control measure I use is the control variable litigation. According to Francis et al. (2009) companies with the SIC codes 2833‐2836, 3570‐3577, 3600‐3674, 5200‐5961 and 7370‐7374 are companies with a high litigation risk. I use the control variable litigation based on the fact that companies that overstate their net assets have higher expected litigation costs than companies who do not overstate their net assets (Watts, 2003).
24
3.5 Empirical model
To examine my hypotheses I estimated the following empirical models: _ _75 5 (1) NEG _ _75 5 (2) POS _ _75 5 (3) ABSDA = Absolute discretionary accruals used to measure audit quality POSDA = Positive discretionary accruals used to measure audit quality NEGDA = Negative discretionary accruals used to measure audit quality DESCRIPTOR = Descriptor is either hypotheses 1 (ACORNOT), hypotheses 2A (ACCOMPLYALL) or hypotheses 2B (ACNOTCOMPLYALL) ACORNOT = Dummy variable; 0 for a company that does not have an audit committee and 1 for a company that does have an audit committee. ACCOMPLYALL = Dummy variable; 0 for companies that comply with all SOX/BRC recommendations <=2002; 1 for companies that comply with all SOX/BRC recommendations >=2003. ACNOTCOMPYALL = Dummy variable; 0 for companies that do not comply with all SOX/BRC recommendations <=2002; 1 for companies that comply with all SOX/BRC recommendations >=2003. ALL_ATT_75pct = Dummy variable; 0 for companies of which at least one audit member did not attend at least 75 percent of the audit25 committee meetings; 1 if all audit committee members attended at least 75 percent of the audit committee meetings. BIG5 = Dummy variable; 0 for companies that are not audited by a big 5 auditor; 1 for companies that are audited by a big 5 auditor. SIZE = Natural log of total average assets. LEV = Total long‐term debt divided by average total assets. LIT = Dummy variable; 0 for companies with a low litigation risk; 1 for companies with a high litigation risk.
If is the descriptor ACORNOT and if has a positive coefficient in the ABSDA (1) and POSDA (3) model than it is consistent with my hypothesis that the audit quality is higher for companies with an audit committee than for companies that do not have an audit committee. If , has a negative coefficient in the NEGDA (2) model than it is consistent with my hypothesis that the audit quality is higher for companies with an audit committee than for companies that do not have an audit committee.
If is the descriptor ACCOMPLYALL and has a positive coefficient in the ABSDA (1) and POSDA (3) model than it is consistent with my hypothesis that companies who already complied with the BRC/SOX recommendations pre‐SOX do not see a significant increase in audit quality post‐SOX. If has a negative coefficient in the NEGDA (2) model than it is consistent with my hypothesis that companies who already complied with the BRC/SOX recommendations pre‐SOX do not see a significant increase in audit quality post‐SOX. If is the descriptor ACNOTCOMPLYALL and has a positive coefficient in the ABSDA (1) and POSDA (3) model than it is consistent with my hypothesis that companies who did not comply with the BRC/SOX recommendations pre‐SOX would see an increase in audit quality post‐SOX. If has a negative coefficient in the NEGDA (2) model than it is consistent with my hypothesis that companies who did not comply with the BRC/SOX recommendations pre‐SOX would see an increase in audit quality post‐SOX.
26
4 Results
4.1 Descriptive statistics
In table 1 the industry distribution of the total sample is shown. Table 1 shows that 53% of the sample’s firm‐years represent the manufacturing industry. Also, the industries transportation and public utilities, wholesale and retail; and services are, with approximately 14%, evenly divided over the sample. Table 2 shows the distribution of the means of the total sample for the audit quality measures. I expect the distribution of the means to decrease starting in the year 2001, one year after the introduction of the BRC recommendations. I also expect the distribution of the means to further decrease in the 2003, one year after the introduction of the SOX legislation. Table 2 shows that the distributions of the mean for all accrual measurements decrease after 2001. However, when I look at the distribution of the mean for 2003 I see that for the accrual measure ABSDA, the distribution of the mean increases after 2002, which is not in line with my expectation. But, when I look at the distribution of the mean for the accruals measure POSDA, which is, according to Cahan et al. (2011), the most important accruals measurement, I see that after the introduction of the BRC recommendations, in which the distribution of the mean went down, the distribution of the mean stays fairly consistent.27 Table 3 shows the descriptive statistics for the pooled sample. I found that the mean values for the accrual measures ABSDA and POSDA are positive and that the mean value for NEGDA is negative. The fact that ABSDA is positive means that the average of the absolute discretionary value is positive. This is expected because the majority of the sample has positive discretionary accruals (Table 3). These findings are consistent with the findings of Cahan et al. (2011) who also found positive mean values for ABSDA and POSDA and a negative mean value for NEGDA. About 88% of the sample has an audit committee and about 94% of these audit committees have members that attend at least 75% of all audit committee meetings. These findings are roughly the same as the findings of Ziegenfuss and Myers (2006). The variable ACCOMPLYALL consists for 44% of companies that complied with SOX guidelines post‐SOX and for 56% of companies that complied with BRC/SOX guidelines before the implementation of SOX. The variable ACNOTCOMPLYALL consists for 57% of companies that did not comply with all BRC/SOX guidelines pre‐SOX and for 43% of companies that complied with SOX guidelines post‐SOX. Also, 98% of the sample is audited by a big 5 firm and about 25% of the companies is qualified as a company with a high litigation risk. Other research shows that my sample has a high amount of companies that use a big 5 auditor. Other research shows samples in which approximately 80% of the sample is audited by a big 5 auditor (Francis et al., 1999A; Lawrence et al., 2011). The litigation risk of 25% is considered to be high because Abbott et al. (2007) shows a litigation risk of only 5%. However, his sample size is much smaller (less than 100) which could affect the outcome.
28
29
4.2 Results of hypothesis tests
I have used the correlation test of Spearman to compute the correlations in table 4, table 6 and table 8. In this test all variables are first ranked, and then assigned a new value based on the rank. I use the Spearman test to make sure that the correlation is not driven by some extreme observations or any other non‐linearities. Based on the descriptive statistics I expect the variables ABSDA and POSDA (NEGDA) to show a positive (negative) coefficient. To get the least amount of noise I want the correlation matrix to be highly correlated with my outcome variable and I want the predictor variables to be minimally correlated with each other. I consider the control variables to have a high, medium, low correlation at a coefficient level of respectively > 0.7, 0.3 – 0.7 and <0.3. . I also checked for multicollinearity (not tabulated). If the VIF, in my test for multicollinearity, is lower than 5 I conclude that the sample is not affected for multicollinearity. Another way to check for multicollinearity is to check the tolerance level. If the tolerance level is higher than 0.2 I conclude that the sample is not affected by multicollinearity. I created an OLS regression in table 5, table 7 and table 9. I created an OLS regression for the ABSDA model, the NEGDA model and the POSDA model. The first column (I) of each OLS regression represents the dependent variable without any control variables. The second column (II) represents an OLS regression with the dependent variable and the control variables. This regression has revised t‐statistics for heteroskedasticity. The third column (III) in the OLS regression represents the dependent variable with the control variables. In the third column the OLS regression includes two‐way clustered standard errors. The first column was added to check for the effects of the control variables on the dependent variable. I added the second column to test the robustness of the sample. The third column was added based on research by Gow et al. (2010). They state that cross‐sectional and time‐series dependence can negatively affect test statistics. Also, the risk of creating wrong test statistics could be reduced by using a regression analysis that includes two‐way clustered standard errors (Gow et al. 2010). Therefore I added the third column that includes a regression analysis with two‐way clustered standard errors.30 4.2.1 The effect of having an audit committee on audit quality In this paragraph I will discuss the results for the correlation matrix and the OLS regression that I ran on the full sample (10,089 firm‐years). I will discuss the variable of interest (dependent variable) and I will discuss the control variables. The variable of interest in this paragraph is the variable ACORNOT. 4.2.1.1 Correlation hypotheses 1 The ABSDA model in table 4A, with a sample size of 10,089 firm years, shows low correlations between the outcome variable and the control variables. This implies that the sample includes a lot of noise. However, the control variables do not show a high correlation with each other except for the correlation between the control variable SIZE and LEV. Which implies that the control variables that are used do have explanatory power. The results of the multicollinearity test shows that the mean of the VIF and the tolerance level of most variables lie around respectively 1.07 and 0.98. There are however two exceptions. In line with my earlier findings I find a higher VIF (1.16, 1.21) and a lower tolerance level (0.86, 0.83) for respectively SIZE and LEV. This means that these variables are more correlated than the other variables. To summarize, based on the ABSDA model, the outcome variable is not highly correlated with the control variables. This implies that the sample is very noisy and might affect the outcome of the study. On the other hand, the control variables are not (highly) correlated with each other, which means that the control variables used, have explanatory power. The NEGDA model in table 4B, with a sample size of 4,346, shows low correlations between the dependent variable and the control variables. The dependent variable of the NEGDA model shows, overall, similar correlations as the dependent variable in the ABSDA model. But, the dependent variable in the NEGDA model increases the level of noise as the correlation coefficient, and also the significance level, drops. This implies that the sample includes more noise than the ABSDA sample. Again, the control variables are not highly correlated with each other except for the variable SIZE and LEV. However, compared to the ABSDA sample both variables are less correlated with each other. This means they (slightly) increase in explanatory power compared to the ABSDA sample.
31 The results of the multicollinearity test shows that the mean of the VIF and the tolerance level of most variables lie around respectively 1.05 and 0.98. There are however two exceptions. In line with my earlier findings I find a higher VIF (1.12, 1.15) and a lower tolerance level (0.89, 0.87) for respectively SIZE and LEV. This means that these variables are higher correlated than the other variables. Compared to the ABSDA model the tolerance (VIF) level increases (decreases) which is in line with my earlier explanation that the explanatory power of the variables SIZE and LEV decreases. To summarize, based on the NEGDA model, the outcome variable is not highly correlated with the control variables. This implies that the sample is very noisy and might affect the outcome of the study. On the other hand, the control variables are not (highly) correlated with each other, which means that the control variables used, have explanatory power. Compared to the ABSDA model, the control variables have a better explanatory power. However, because the correlation coefficient, and the significance level drops for NEGDA, and thereby the noise increases, I conclude that the ABSDA model is better suited for my research. The POSDA model in table 4C, with a sample size of 5,743, shows low correlations between the outcome variable and the control variables. The POSDA model shows, overall, slightly higher correlations with the ABSDA model. However, the dependent variable (POSDA) decreases the level of noise, compared to the ABSDA model, as the correlation coefficient, and also the significance level, increases. In addition, the correlation coefficient of the variables LEV and LIT tripled compared to the ABSDA model. This implies that both these control variables reduce noise. However, even though the correlation coefficient tripled the correlation coefficient still shows a low correlation between the dependent and independent variables. The implication of these results is that the POSDA sample includes less noise than the ABSDA sample. Just like the earlier findings I find that the control variables are not highly correlated with each other except for the variable SIZE and LEV. However, compared to the ABSDA sample both variables are more correlated with each other. This means they (slightly) decrease in explanatory power compared to the ABSDA sample.
32 The results of the multicollinearity test shows that the mean of the VIF and the tolerance level of most variables lie around respectively 1.09 and 0.97. There are however two exceptions. In line with my earlier findings I find a higher VIF (1.20, 1.25) and a lower tolerance level (0.83, 0.80) for respectively SIZE and LEV. This means that these variables are higher correlated than the other variables. Compared to the ABSDA model the tolerance (VIF) level increases (decreases) which is in line with my earlier explanation that the explanatory power of the variables SIZE and LEV increases. To summarize, based on the POSDA model, the outcome variable is not highly correlated with the control variables. This implies that the sample is very noisy and might affect the outcome of the study. On the other hand, the control variables are not (highly) correlated with each other, which means that the control variables used, have explanatory power. Compared to the ABSDA model, the explanatory power of the control variables in the POSDA model is lower. However, because the correlation coefficient, and the significance level increase for POSDA, and thereby the noise decreases, I conclude that the POSDA model is best suited for my research. This is corroborated by earlier research of Cahan et al. (2011) who also found that positive discretionary accruals can be best used to explain audit quality. Overall, the correlation coefficients in table 4A show that ABSDA and ACORNOT are positively and significantly correlated. In addition, the correlation in table 4C also shows that the correlation between the variables POSDA and ACORNOT is positively and significantly correlated. Also, table 4B shows that NEGDA and ACORNOT are negatively and significantly correlated. However, NEGDA and ACORNOT are less significant correlated than the correlation between ABSDA and ACORNOT and the correlation between POSDA and ACORNOT. The results of the correlation matrixes are in agreement with my first hypotheses, as NEGDA moves contrary to POSDA. To summarize, the correlation in table 4A and 4C (4B) show that the accruals significantly increase (decrease) when a company does not have an audit committee. Hence, an increase (decrease) in ACORNOT in table 4A and 4C (4B) means that the audit quality decreases. Therefore, companies without an audit committee will see a significant increase in audit quality when they implement an audit committee. These findings are
33 corroborated by Abbott et al. (2004) who found that, based on restatements, the audit quality is higher for companies with an audit committee than for companies without an audit committee. However, as stated earlier the sample has a lot of noise. The downside of the big amount of noise means that the results of my study can be bias. 4.2.1.2 OLS regression hypotheses 1 In table 5 I created an OLS regression to analyze the effect of having an audit committee. I ran an OLS regression on the ABSDA model (10,089 firm‐years), the NEGDA model (4,346 firm‐years) and the POSDA model (5,743 firm‐years). The ABSDA model shows in all three columns a significant relation with the dependent variable ACORNOT. This implies that the coefficients are accurately estimated. However, the coefficients for ACORNOT are so low that they are not economically significant. For example, when I look at the first column (I) I see a coefficient of .0143. When I 1, which is if a company wants to go from no audit committee (0) to an audit committee (1), by the coefficient I get approximately 70. This implies that having an audit committee only leads to an increase in audit quality when the absolute discretionary accruals increase by 70. Hence, the dependent variable is economically insignificant. In addition, the range of the accruals in the ABSDA model lies between ‐6.30 and 1.39 (table 3A). Hence, it is very unlikely that companies reach 70 in the ABSDA accruals model. Also, I see that the p‐value in the first column of the ABSDA model is only a little higher compared to columns two and three. In addition, the significance level stays the same. This implies that the control variables have no impact on the dependent variable. This is corroborated by the fact that all coefficients are very small. Which implies that if they have any effect on the dependent variable, the effect is small. The NEGDA model also shows in all three columns a significant relation with the dependent variable ACORNOT. This implies that the coefficients are accurately estimated. However, the coefficients for ACORNOT are so low that they are, again, not economically significant. But, the coefficient for NEGDA is higher than the coefficients for ABSDA. This implies that the economic significance for the NEGDA model is higher than the economic significance of the ABSDA model. Opposite to the findings in the ABSDA model, the dependent variable in the
34 NEGDA model is affected by the control variables. I see an increase in the coefficient of the dependent variable between column one (I) and column two (II) / three (III). These findings are corroborated by the fact that, on average, the coefficient for the control variables increased. The POSDA model also shows in all three columns a significant relation with the dependent variable ACORNOT. This implies that the coefficients are accurately estimated. However, the coefficients for ACORNOT are so low that they are, just like in the ABSDA and NEGDA model, not economically significant. I found that, just like in the NEGDA model, the coefficient for the dependent variable is higher in the POSDA model than the coefficient in the ABSDA model. When I calculate the amount of times that the positive discretionary accruals need to increase for a company to see any effect of the increase in audit quality I see that this the number went down from 70 in the ABSDA model to approximately 35 in the POSDA model. However, like I stated before, despite the decrease, the findings are not economically significant. 4.2.1.3 Conclusion hypotheses 1 Overall, the coefficients for ACORNOT are for ABSDA and POSDA (NEGDA), and for each type of model, significantly positive (negative). This implies that companies without an audit committee will see an increase (decrease) in ABSDA and POSDA (NEGDA) and hence, see an increase in audit quality when they implement an audit committee. However, as stated before, all results are not economically significant. So, my first hypothesis, which states that audit committees increase audit quality is supported. But despite being supported, only a few firms, with very high discretionary accruals, would benefit from implementing an audit committee. These results imply that most of the companies implemented an audit committee before the introduction of the SOX. In addition, the findings show that, companies who did not implement an audit committee, would not see an economically significant increase in audit quality. Based on these findings I conclude that the SOX was implemented more for show than to increase audit quality. Also, as I stated earlier, the sample includes a lot of noise and the results can, because of the noise, be bias.
35
36
