Mandatory audit firm rotation and audit quality

(1)

Amsterdam Business School

MSc Accountancy & Control, variant Accountancy

Faculty of Economics and Business, University of Amsterdam

Mandatory Audit Firm Rotation and Audit Quality

Jennifer Ngai (10128174) Supervisor: dr. A. Sikalidis

Date of final version: June 23, 2014 Amsterdam 2014

(2)

ABSTRACT

For more than 35 years, limited audit firm tenure has been considered by the US, indicating that “mandatory change of accountants after a given period of time is one alternative to exercise independent action to protect investors and the public”. This topic has been discussed for a relative long time and the debates are still going on. My motivation for this study comes from the limited empirical evidence regarding the relation between audit quality and audit firm tenure and the interests of legislators and regulators in MAFR to increase audit quality. Another reason why I’m interested in MAFR, is because MAFR is recently promulgated in The Netherlands and will be effective as of the beginning of 2016. This study examines the long auditor-client relationship and the effect on audit quality to see whether MAFR is supported. For this study, I used two proxies for the measurement of audit quality: (1) propensity to issue a going-concern opinion, and (2) discretionary accruals. For the second measure, the modified Jones model including the ROA as suggested by Kothari et al. (2005) is used. Using a sample of 4,077 companies for the period of 2000-2012, I can conclude that audit quality is decreased when measured with the audit opinion measure, but increased with the discretionary accruals measure. However, the finding of the second measure is not significant, and therefore should be interpreted carefully. Also when I divide my sample in pre- and post-SOX period, both measures show a decrease in audit quality. Based on my findings, I can cautiously state that audit quality is decreased with long audit firm tenure and therefore MAFR is supported. Another notable finding is that when I compare the results of different auditor tenure years (TENURE5, TENURE7, and TENURE10), I find a nonlinear relation between auditor tenure and audit quality when I measure audit quality with the audit opinion measure. One limitation of this study is that it uses data from voluntary audit firm switches to draw conclusions about a regime of mandatory switches. Therefore it is not certain whether these findings will hold in a MAFR regime. However, due to availability issues, this study is restricted to a voluntary setting.

Keywords: Auditor independence, auditor’s propensity to issue a going-concern opinion,

(3)

I. INTRODUCTION

Since 2000, many accounting scandals occurred, affecting the public’s perception of auditor independence. In response, the Securities Exchange Commissions (SEC) included several rules in the Sarbanes-Oxley Act (SOX) of 2002 to restore investors’ confidence. Also in Europe, new rules and regulations were being discussed to strengthen the audit profession regime as a response to the economic crisis in 2007 (European Commission, 2010). As a result of these events, legislators, regulators, politicians, and the public raised their concern about the reliability of the financial statements and thereby questioning the credibility of the auditors. Even though the primary responsibility for preparing the financial statements belongs to the managers, auditors play a crucial role in providing an independent opinion about the financial statements.

Auditor independence is one of the most important characteristics that an auditor should possess. Auditor independence requires the characteristic of professional skepticism during the audit performance, but also requires an auditor to ensure the integrity of the audit process to protect shareholders and other stakeholders from incorrect claims made by the audited companies. An important critic regarding the independence of auditors is on the tenure of the audit engagement. Long auditor-client relationship can be seen as an impairment of auditor independence and audit quality, due to the familiarity threat. There are many conflicting views on the relation between long tenure and the auditor’s independence and some of them are proposing the mandatory audit firm rotation (MAFR) as a solution. This study examines the long auditor-client relationship and the effect on audit quality to see whether MAFR is supported. Mandatory audit firm rotation (MAFR) attracted my attention because of the questions from legislators and regulators in many countries concerning the effect of MAFR on auditor independence and audit quality1. Another reason is that MAFR is recently promulgated in The Netherlands and will be effective as of the beginning of 2016. And finally my motivation for this study comes from the limited empirical evidence regarding the relation between audit quality and audit firm tenure.

The main supporting argument for MAFR is that long auditor-client relationship results in a high level of familiarity between the auditor and the client, and as a result decreases audit quality. Mautz & Sharaf (1961) suggest that the longer the auditor tenure, the

1_{During the promulgation of the Sarbanas-Oxley Act in 2002, the US Congress asked the General Accounting}

Office (GAO) to examine the effects of MAFR. As of today, there still has not been a formal proposal or active project. In Europe, this question is also asked by the European Commission in their Green paper in 2010.

(5)

less objective an auditor is towards the client. It is also argued that the economic interest of incumbent auditors and the desire to retain clients increase management’s abilities to influence the auditor’s opinion, affecting the quality of the audit. Further, longer auditor tenure can increases earnings manipulation, as managers are more aware of how the auditors work and therefore gain more confidence. Audit firm rotation by limiting the number of sequent years that a particular audit firm can audit a company is seen as a way of preventing this, which improves the audit quality and decreases the probability of audit failures. Finally, having a new auditor will also bring a fresh look to the company’s financial statements because of the different approach.

On the contrary, opponents of MAFR believe that short auditor-client relationship is more susceptible to audit failures, as the new auditor would be unfamiliar with the client and lack some client-specific knowledge. This leads to a decrease in the effective working and the probability of detecting material misstatements. It has been argued that when the auditor tenure is long, client-specific knowledge increases, and therefore auditors rely less on managerial estimates and become more independent. Fraudulent financial reporting and litigation risks are also greater at the beginning of the audit firm’s tenure and substantially lower for longer auditor tenure. Further, it is suggested that MAFR will increase the start-up costs to understand the client’s organizational structure and business model, leading audit fees also to be higher. Questions raise about whether the potential MAFR benefits outweigh the associated costs. Finally, opponents of MAFR argue that the market provide strong economic and institutional incentives for auditor independence, making MAFR unnecessary.

For this study, I used two measures of audit quality: (1) propensity to issue a going-concern opinion, and (2) discretionary accruals. Many studies use either the propensity to issue a going-concern opinion or the level of abnormal accruals as a proxy for audit quality. This study use both proxies to measure audit quality. For the second measure I use the modified Jones model, including the ROA as suggested by Kothari et al. (2005). The control variables are derived from prior literature, mainly from Carey & Simnett (2006), Jackson et al. (2008), and Ruiz-Barbadillo et al. (2009). I used a sample of all US companies for the period of 2000-2012. To my knowledge, this is the first research to combine prior literature in the way it is explained.

The main analysis show that long tenure (7 or more years) decreases the audit quality as measured by the propensity to issue a going-concern opinion. This finding is significant and in line with my expectation. This means that the MAFR regime is supported. When audit quality is measured with the discretionary accruals measure, I find that audit quality increases

(6)

with longer tenure and therefore not support the hypothesis and the introduction of MAFR. However, this finding is not significant, and therefore should be interpreted carefully. Even though the two measures are not both supported, the result of the first measure is stronger as that one is significant. Also when I divide my sample in pre- and post-SOX period, both measures show a decrease in audit quality. Thus, I can cautiously state that audit quality is decreased with longer tenure and therefore MAFR is supported. Another notable finding is that when I compare the results of different auditor tenure years, I find a nonlinear relation between auditor tenure and audit quality when I measure audit quality with the audit opinion measure.

One limitation of this study is that it uses data from voluntary audit firm switches to draw conclusions about a regime of mandatory switches. Therefore it is not certain whether these findings will hold in a MAFR regime. However, due to availability issues, this study is restricted to a voluntary setting.

This paper is structured as follows. Section II provides the literature review and hypothesis. Section III outlines the research design. Section IV presents the results and discussion. Finally, section V concludes.

(7)

II. LITERATURE REVIEW AND HYPOTHESIS

1. Auditor independence

1.1 Importance of independence

Auditor independence refers to the independence of the external auditor. An auditor is considered independent if he or she operates objectively and with integrity. Auditor independence is one of the most important characteristics that an auditor should possess, which is also emphasized by many auditing standards boards like the American Institute of Certified Public Accountants, the Public Company Accounting Oversight Board, and the Institute of Chartered Accountants in England and Wales.

As the focus of this research is on US companies, explanation of the US Framework for auditor independence is necessary. The world’s largest member association representing the accounting profession, the American Institute of Certified Public Accountants (AICPA), defines specific rules about independence in their ‘Code of Professional Guidance (Code)’, which give guidance and rules for all AICPA members. Rule 101 from the Code defines that: “A member in public practice shall be independent in the performance of professional services as required by standards promulgated by bodies designated by Council” (AICPA, 1988).

Rule 101 also describes ways in which independence shall be considered to be impaired. To give a brief overview, three important causes for impairment are described. First of all, independence will impair if, after the audit engagement period, a partner or professional employee is employed by the associated client in a key position. Auditor independence will not occur if the conditions regarding this rule are met, e.g. operations of the accounting firm could not be influenced by the former partner or professional employee if the retirement benefits of the former partner or professional employee are not material to the firm. Another way independence is considered to be impaired is when the audit firm perform non-audit services, unless the non-audit services were provided prior to the audit engagement and the financial statements were audited or reviewed by another firm. And last but not least, independence can be impaired if there is a financial relationship between the client and the auditor. Independence will impair if, during the professional engagement period, a covered member: (1) acquired any direct or indirect material financial interest in the client, (2) was a trustee, executor or administrator of any trust or estate if it acquired any direct or indirect material financial interest in the client, (3) had a material joint closely held investment, or (4)

(8)

had a loan from or to the client, or individual owning of 10 percent or more outstanding equity securities or other ownership interests from the client (AICPA, 2006).

The association also differentiate independence as independence in mind and independence in appearance. The importance of the two elements of auditor independence is widely accepted in theory and in practice (Kaplan & Mauldin, 2006). Independence in mind is a state of mind that perform an attestation in an objective manner, without their professional judgment being influenced. This allows an auditor to act objectively, professional, and with integrity. Independence in appearance is the perceived independence, expressing how others perceive auditors as independent (AICPA, 2006). It is not only important that an auditor acts independently, but also appears so. If an auditor in independent in fact, but does not appears so, the public could potentially infer that the auditor does not represent a true and fair view of the report. Ghosh & Moon (2005) states that the perception of capital market on independence and audit quality is consistent with the conceptual framework of the Financial Accounting Standards Board (FASB) for financial reporting and principles for auditor independence. They also state that the business of audit firms rely heavily on their reputations, and that is affected by appearance. Accordingly, both aspects are crucial to attain the goal of independence.

Since 2000, many accounting scandals occurred, affecting the public’s perception of auditor independence. In response, the Securities Exchange Commissions (SEC) included several rules in the Sarbanes-Oxley Act (SOX) of 2002 to restore investors’ confidence. First of all, it prohibits auditors to provide any non-audit services. Section 201 lists nine non-audit services, including bookkeeping or other services related to the financial statements of the audit client and recommending or advising audit clients to hire a specific candidate for a specific job. This rule was created to eliminate the conflict of interest that occur when an audit firm has a high percentage of total revenues in one client. Another rule is partner rotation, described in section 203 of SOX. This rule states that audit partners should rotate every five years and are subject to a five year cooling off period. Other rules, like the communication with the Audit Committee and compensations, are implemented to strengthen auditor independence (SEC, 2003). Future developments regarding the auditor independence are still being discussed, and one of the prominent topic is the mandatory audit firm rotation (Nagy, 2005).

(9)

1.2 Independence and the role of the auditor

Auditors are aimed to enhance the credibility of financial statements of a company by providing reasonable assurance that the numbers present a true and fair view in accordance with the applicable standard. Therefore auditor independence is crucial to obtain and maintain credibility of the audit opinion. Auditor independence requires the characteristic of professional skepticism during the audit performance, but also requires an auditor to ensure the integrity of the audit process to protect shareholders and other stakeholders from incorrect claims made by the audited companies. Ghosh & Moon (2005) even considered independent auditors as the “gatekeepers” of the public securities markets. Auditors are not independent if they are influenced by other parties (e.g. managers of the audited company) or by conflicting interests (e.g. auditor owns shares of the audited company).

Watts & Zimmerman (1983) state that an audit is successful is it reduces the opportunistic behavior costs, also referred to as agency costs, borne by managers. The agency costs derives from the agency theory, where principle-agent problems occurs when the agent (in this case: managers) is motivated to act in this own best interests rather than those of the principal (in this case: shareholders). Auditors can reduce the agency costs by reporting the discovered breach in the contract. Watts & Zimmerman (1983) define the probability to report this breach as the definition of auditor’s independence. Puro (1984) describes that the agency theory also implies to auditors and clients, whereby auditors (in this case: agent) are lobbying for rules which benefit their clients (in this case: principal), but in the process benefit the audit firm.

Due to prior audit scandals and especially due to the many fraud scandals at the beginning of the last century2, regulators and stakeholders question the reliability and credibility of the audit profession (Arel et al., 2006). These financial reporting failures have caused an increased focus on the work of external auditors, which led to calls for more measures and rules to strengthen the audit profession regime.

2. Audit quality

2.1 Definition of audit quality

In the Conceptual Framework of the FASB, audit quality is defined as the

“usefulness of financial statements to investors”, whereby the usefulness is associated with relevant and reliable information (FASB, 1978 no.1, p.15 and no.2, p. 2).

(10)

Financial statements communicates the financial information of a company to the outside world. Since there is a conflict of interest (agency theory) and cases where

managers have more or better information than the outside world (information asymmetry), auditors are needed to enhance the quality of financial statements that are reported by managers (Johnson et al., 2002). The IAASB recently issued a framework for audit quality that described the in- and output factors that contribute to audit quality. Inputs based on values, ethics, and attitudes involves an auditor’s objectivity, integrity, independence, competence, and professional skepticism, while inputs based on knowledge, experience, and time emphasize an auditor’s understanding and judgments. Outputs relate to an auditors’ report to users of the financial statements, auditors’ report to management, and the reports’ transparency (IAASB, 2013, p. 20-23). DeAngelo (1981b) defines audit quality as: (1) the probability that an auditor will discover a breach in the client’s accounting system, and (2) the likelihood that the auditor will report the discovered breach. An auditor with economic interest in their client is unlikely to report an observed breach and this results in a decrease in audit quality. The economic interest is seen as a quasi-rent that represents the present value of future revenues over the auditor-client relationship period. The incentive of

auditors to earn long-term quasi-rents can thus impair auditor independence. Therefore it is important that auditors maintain their professional skepticism in accordance with the auditing standards (Arel et al., 2006).

2.2 Proxies for audit quality

2.2.1 Audit opinion

As a proxy of audit quality, the propensity of issuing a going-concern opinion is widely used in many studies (Firth et al., 2010). The going-concern reporting standard, SAS no. 59, requires the auditors to evaluate the ability of a company to continue as a going-concern in a period of one year. If the auditor is uncertain about the ability of continuance, the audit opinion should be to reflect the doubt (Geiger & Rama, 2006). A going-concern opinion is a significant reason for clients to change auditors (Chow and Rice, 1982), but an auditor must be able to withstand client pressure concerning the issuance of a clean opinion and objectively evaluate firm performance. The auditor communicate the findings through the audit report and warn the public about problems by issuing a going-concern opinion. Ceteris paribus, the auditor’s propensity to issue a going-concern opinion is positively correlated with the auditor’s level of independence (DeFond et al., 2002).

(11)

for audit quality to examine the effect of MAFR on audit quality in Australia, where currently no legislative requirement is for audit firm rotation, and found that audit quality increases with audit firm tenure. Firth et al. (2010) test the effect of auditor rotation on audit quality in China, which provides the authors to be able to compare auditor rotation in different settings (voluntary versus mandatory), and found that firm rotation in a voluntary setting are associated with a significantly higher likelihood of an propensity to issue a going-concern opinion and no effect on the propensity to issue a going-concern opinion in a MAFR. Ruiz-Barbadillo et al. (2009) study the relationship between MAFR and auditor independence in a mandatory regime (Spain), by testing two competing hypotheses regarding the impact of mandatory rotation on the propensity of issuing a going-concern opinion, and found no support to suggest that MAFR is associated with a higher likelihood of issuing going-concern opinions.

2.2.2 Discretionary accruals

Another commonly used proxy for audit quality is the level of discretionary

accruals. Prior literature argue that accounting accruals are an acceptable measure of audit quality. The Business Dictionary describe accruals as the amount recorded in the balance sheet as assets or liabilities, that are not received or paid yet. Since these amounts have to be estimated, there are many different ways to do so. Dechow et al. (1995) compares five models in their paper and concluded that the modified Jones model (1991) has the greatest ability to detect earnings management.

The original Jones model is based on DeAngelo (1985) and Healy (1985) and distinguish total accruals to discretionary and non-discretionary accruals. The equation is as follows:

TA = α + β1(∆REV) + β2PPE + ε (1)

Here TA = total accruals in year t for firm i; ∆REV = revenues in year t less revenues in year t-1 for firm i; PPE = gross property, plant, and equipment in year t for firm i; ε = regression residual, which is the discretionary accruals in year t for firm i.

Later, the change of receivables is included, which will be subtracted from the change in revenues, to remove the bias that arise in the Jones model due to the fact that the Jones model assumes that receivables depend on manager’s decisions. The equation of the modified Jones model is as follows:

(12)

TA = α + β1(∆REV-∆REC) + β2PPE + ε (2)

The only difference with equation (1) is the ∆REC = receivables in year t less receivables in year t-1 for firm i.

As the amount of accruals differ from the actual amount, managers can

opportunistically manipulate these numbers in their favor. Managers can choose to present a higher or lower amount on the balance sheet, in the hope to e.g. raise more capital or get a better bonus next year. If the auditor cannot detect these misstatements or ignore these on purpose, the audit quality decreases. Thus, abnormal accruals can be used to determine the level of audit quality (Becker et al., 1998; DeAngelo, 1986; Jones, 1991); Myers et al., 2003).

Higher discretionary accruals are positively associated with auditor litigation (Heninger, 2001), auditor changes (DeFond & Subramanyam, 1998), and audit failures (Geiger & Raghunandan, 2002), while lower discretionary accruals are associated with greater auditor conservatism (Becker et al., 1998). Based on these findings, it can be assumed that high audit quality mitigate extreme management reporting decisions, and suggest that these reporting decisions can be identified using accruals (Myers et al., 2003).

Johnson et al. (2002) argue that, using the absolute value of unexpected accruals and the persistence of accruals as proxies for audit quality, short audit-firm tenures of two to three years are associated with lower-quality financial reports and find no support of reduced audit quality for longer audit-firm tenures. Chen et al. (2009) used discretionary accruals to investigate the relation between audit firm tenure and audit quality and found that discretionary accruals decrease significantly with audit firm tenure. Myers et al. (2003) used both the absolute Jones-model abnormal accruals and the absolute current accruals as proxies for audit quality and found that auditors are placing greater constraints on extreme management decisions in the reporting of financial performance when there is a longer auditor tenure.

3. Mandatory audit firm rotation

For more than 35 years, limited audit firm tenure has been considered by the US, indicating that “mandatory change of accountants after a given period of time is one alternative to exercise independent action to protect investors and the public” (Metcalf Report, 1977). The Cohen Commission responded one year later with a different conclusion

(13)

and was against mandatory audit firm rotation, stating that: “cost of mandatory rotation would be high and the benefits that financial statement users might gain would be offset by the loss of benefits that resulted from a continuing relationship” (AICPA, 1978). Since then, MAFR was not considered until 2001, with the Enron and WorldCom scandals. The General Accounting Office (GAO), asked by the US Congress to examine the potential effects of MAFR, reported in 2003 that it needed some time before determining whether MAFR is necessary to enhance audit quality and auditor independence due to the implementation of SOX in 2002 (Jenkins & Vermeer, 2013). Thus, the GAO decided to postpone the

consideration of MAFR, but noted that the PCAOB, in the future, should evaluate whether MAFR may be needed to further protect the stakeholders’ interest and to restore the

investor’s confidence. GAO indicated in their report that even though the current audit partner rotation increases the auditor independence in fact, mandatory audit firm rotation might be needed to strengthen the auditor independence in appearance (GAO, 2003). In Europe, the MAFR subject is also a much discussed topic. In 2010, this question raised again in the Green Paper of the European Commission, after several discussions about measurements for enhancing auditor’s independence as a response to the economic crisis in 2007 (European Commission, 2010). Since many parties, like regulators, audit practitioners, and academics, are interested, the European Commission proposed a discussion on MAFR, requesting for more research on the pros and cons of the implementation of MAFR. Recently, The Netherlands promulgated to be the next European country with the MAFR regime, and this regime will be effective as of the beginning of 2016.

In 2011, the Public Company Accounting Oversight Board (PCAOB) issued a

concept release soliciting public recommendations to improve auditor independence and audit quality, where the focus is on MAFR (PCAOB, 2011). As a response, many comments were received with complaints about the increased costs and the decreased audit quality due to less client-specific knowledge. Three years later, the PCAOB still haven’t had a formal proposal or active project, and mention that it will continue think about what impacts auditor

independence. Thus, as to date, MAFR has not been enacted in the United States. Other countries such as Brazil, Italy, and India have already implemented the MAFR.

3.1 Arguments supporting mandatory audit firm rotation

MAFR is proposed in order to improve the auditor independence and audit quality. In general, proponents of MAFR believe that long auditor-client relationship results in a high level of familiarity between the auditor and client, and thereby decreases audit quality. Mautz

(14)

& Sharaf (1961) suggest that the longer the auditor tenure, the less objective an auditor is towards the client. The economic interest of incumbent auditors, also called the economic dependence, and the desire to retain clients increase management’s abilities to influence the auditor’s opinion (Ruiz-Barbadillo et al., 2009). Audit firm rotation by limiting the number of sequent years that a particular audit firm can audit a company is seen as a way of preventing this, which improves the audit quality and decreases the probability of audit failures

(Casterella & Johnston, 2013; Dopuch et al., 2001). Davis et al. (2009) argue that longer auditor tenure increases earnings manipulation, as managers are more aware of how the auditors work and therefore gain more confidence. This is in line with the findings of Myers et al. (2003). Arel et al. (2006) did an experimental study and find that experienced auditors believed their firm would be more likely to issue a going-concern opinion under MAFR than when there is no rotation. Although DeAngelo (1981a) state that low-balling3 does not impair auditor independence as it is a competitive response to the expected future quasi-rents,

Dopuch et al. (2001) find that low-balling does impair auditor independence and decreases audit quality and therefore are in favor of MAFR. Further, having a new auditor will bring a fresh look to the company’s financial statements because of the different approach (Lu & Sivaramakrishnan, 2009).

3.2 Arguments against mandatory audit firm rotation

On the contrary, opponents of MAFR believe that short auditor-client relationship is more susceptible to audit failures, as the new auditor would be unfamiliar with the client and lack some client-specific knowledge (Casterella & Johnston, 2013). Client-specific knowledge is necessary for auditors in order to detect material misstatements. Prior research suggests that audit quality is lower in the initial years of the engagement, due to a loss of client-specific knowledge and expertise. DeAngelo (1981b) refers this as the ‘learning curve’ for new auditors, which results in significant start-up costs. Less client-specific knowledge in the early years means a lower probability of detecting material misstatements in the financial statements, but over the years it will give the incumbent auditors a comparative advantage in detecting the errors. Johnson et al. (2002) find that the lack of sufficient client-specific knowledge during the first years of the engagement decreases the effective working and the probability of detecting material misstatements. Solomon et al. (1999) argue that when the auditor tenure is long, client-specific knowledge increases, and therefore auditors rely less on

3_{DeAngelo (1981a) defines low-balling as setting audit fees below total current costs on initial audit engagements,}

(15)

managerial estimates and become more independent. Most audit firm rotation studies look at the relation between auditor tenure and audit quality and find that audit quality is higher when there is a longer auditor-client relationship (Stefaniak et al., 2009). Further, Carcello & Nagy (2004), Geiger & Raghunandan, (2002), and Myers et al. (2003) find that fraudulent financial reporting and litigation risks are greater at the beginning of the audit firm tenure and substantially lower for longer auditor tenure. Myers et al. (2003) also suggest that MAFR will increase the start-up costs to understand the client’s organizational structure and business model, which is consistent with Arruñanda & Paz-Ares (1997), who finds that under MAFR, the audit costs tend to increase, leading audit fees also to be higher. There are also doubts, using cost-benefit analysis, about whether the potential MAFR benefits outweigh the associated costs (AICPA 1978, 1992). Finally, opponents of MAFR argue that the market provide strong economic and institutional incentives for auditor independence, making MAFR unnecessary (Ruiz-Barbadillo et al., 2009). For example, the loss of reputation, due to the lack of independence, will increase the risk of losing future quasi-rents (AICPA, 1992). Thus, the incentive to maintain a good reputation is strong enough to prevent the risk of auditor’s failure.

3.4 Limitations on current research

Casterella & Johnston (2013) find that conclusions made about MAFR depends on the data that is used – voluntary vs. mandatory auditor switches. They examined a total of 24 empirical archival and experimental research studies on audit firm rotation from the years 2001-2013, that are published in the top 25 accounting journals, and found that two of the 13 studies that are based on voluntary auditor changes support MAFR. On the other hand, eight of the 10 studies using mandatory auditor changes support MAFR. Based on these findings, it can be concluded that the findings depends on the used data and studies that use data from mandatory settings are more in favor of MAFR. However, due to availability issues, this study is restricted to a voluntary setting.

4. Hypothesis development

Prior studies show that a company is more likely to retain its auditor when it is less likely to give a going-concern opinion (Antle & Nalebuff, 1991). Managers who are avoiding the going-concern opinion, and have some influence over the engagement of the incumbent auditor, may want to switch auditors to avoid receiving this report (Jackson et

(16)

al., 2008). Teoh (1992) find that the manager will choose to switch if the new auditor is less likely to give a going-concern opinion, or can obtain an unqualified report by threatening the incumbent auditor to switch. Thus, an auditor is more likely to be dismissed if the auditor gives a going-concern opinion. This makes the auditor dependent, as the auditor wants to maintain quasi-rents from existing clients. Prior studies find that auditor

independence is less likely to be threatened when the audit firm rotation is mandatory, as the managers cannot influence the engagement of the incumbent auditor (Copley and

Doucet, 1993). Dopuch et al. (2001) find that auditors that are in a voluntary regime are less likely to issue a going-concern opinion. Based on the above, it can be argued that in a mandatory rotation regime an auditor will more likely issue a going-concern opinion. Dechow & Dichev (2002) argue that one role of accruals is to shift the cash flow so that the adjusted numbers (earnings) better measure firm performance. By manipulating the reported earnings, managers are able to decrease restrictiveness of the debt covenants and increase their compensation through higher bonuses (Jones, 1991). The skill to restrain management earnings vary with the quality of the auditor. Becker et al. (1998) argue that high quality auditors, in comparison to low quality auditors, are more likely to detect doubtful accounting practices and to object and report a qualified audit report. High quality auditing reduce earnings management, because auditors want to prevent reputation damage if

misreporting is detected. Therefore, the earnings quality is higher in firms with higher quality auditors. Prior studies on earnings quality are important to the measurement of audit quality, because they usually find that the quality of audit declines with extreme accruals (Dechow & Dichev, 2002; Sloan, 1996; Xie, 2001). Thus, the relation between the extent of accruals behavior and auditor tenure is informative to the current debate of mandatory rotation.

Based on the above, I will use the propensity to issue a going-concern opinion and the level of discretionary accruals as proxies to measure audit quality. I formulate the following hypothesis regarding the relation between audit firm tenure and audit quality:

(17)

III. RESEARCH DESIGN

1. Methodology

First, audit quality is measured as the propensity of an auditor to issue a going-concern opinion. There is a correlation between the issuance of a going-going-concern opinion and auditor independence, as the auditor must withstand client pressure to issue a clean report and objectively evaluate company performance when issuing a going-concern report (DeFond et al., 2002). Based on prior studies, I suggest that increased tenure leads to a lower propensity to issue going-concern opinions, and consequently independence becomes impaired. This inverse relation induce a decrease in audit quality and provides convincing evidence for mandatory rotation (Carey & Simnett, 2006; Geiger & Raghunandan, 2002; Jackson et al., 2008).

Consistent with DeFond et al. (2002), Carey & Simnett (2006), and Ruiz-Barbadillo et al. (2009), I use the following logistic regression model to estimate the likelihood the auditor issues a going-concern opinion:

GCO = α + β1 TENURE7 + β2 PROBBANK + β3 SIZE + β4 LEV + β5 CLEV + β6 LLOSS + β7 BIG4 + β8 CFFO + β9 FEERATIO + β10 TENURE2 + ε (3)

where:

GCO = A dummy variable equal to 1 if a going-concern opinion is issued,

and 0 otherwise;

TENURE7 = A dummy variable equal to 1 if the audit firm is engaged with the

client for seven years or more, and 0 otherwise;

PROBBANK = The probability of bankruptcy using the financial condition score of

Zmijewski (1984)4;

SIZE = The natural logarithm of total assets at fiscal year-end;

LEV = The ratio of total debt to total assets;

CLEV = The change in leverage during the year;

LLOSS = A dummy variable equal to 1 if the company reported a loss for the

4_{Based on the modified calculation of Zmijewski’s score by Carcello et al. (1995): -4.803 – 3.6 (net income/total}

(18)

previous year, and 0 otherwise;

BIG4 = A variable equal to 1 if the auditor is a Big 4 firm, and 0 otherwise5;

CFFO = The operating cash flows divided by total assets at fiscal year-end;

FEERATIO = The ratio of non-audit fees to total fees paid to the incumbent auditor;

TENURE2 = A dummy variable equal to 1 if the audit firm is engaged with the

client for two years or less, and 0 otherwise;

ε = The regression residual.

The measurement of earnings management discretion changed from the ‘change in total accruals’ (e.g., DeAngelo, 1986) in early studies, to the modified ‘Jones Model’ (Jones, 1991). Nowadays, many studies use the abnormal discretionary accruals as the proxy for extreme managerial discretion (Myers et al., 2003). Following DeFond and Jiambalvo (1994), Subramanyam (1996), Becker et al. (1998), and Johnson et al. (2002), discretionary accruals are estimated from the residuals of the modified Jones 1991 model. Jones included the

change in revenues and the level of property, plant and equipment to measure the normal total accruals. Since revenues and PPE control for the change in total accruals that is due to

changing economic conditions, as opposed to manipulation of accruals, both variables are important (DeFond & Jiambalvo, 1994). As Kothari et al. (2005) describe that performance-matched discretionary accrual measures enhance the reliability of the inferences from earnings management research, the ROA is included as an additional control variable. Following DeFond & Subramanyam (1998) and Kothari et al. (2005), the level of discretionary accruals will derive from the following equation:

TA = α + β1 (1/ASSETS) + β2 (∆REV-∆REC/ASSETS) + β3 PPE/ASSETS +

β4 ROA/ASSETS + ε (4)

where:

TA = The total accruals6;

ASSETS = The total assets;

∆REV = The change in revenues;

5_{Big 4 firms: Deloitte, Ernst & Young, KPMG, and PricewaterhouseCoopers.}

6_{As in Kothari et al. (2005), the total accruals is measured as the change in non-cash current assets less the change}

(19)

∆AR = The change in accounts receivables;

PPE = Gross property, plant and equipment

ROA = The rate of return on total assets;

ε = The discretionary accruals estimate.

The OLS regression for the second proxy of audit quality, discretionary accruals, uses the same independent and control variables as those described in equation (3). All the

variables in equation (4) are scaled by the total assets in year t-1 to reduce potential heteroskedasticity. With the second measure I look at the change in the discretionary component. Since discretionary accruals is a component of total accruals, high (low) total accruals would indicate that either the discretionary (non-discretionary) component has increased (decreased). Following prior literature, I used the absolute numbers for the change in revenues and the change in receivables.

The choice of control variables for both measures (3) and (4) are derived from prior literature, in particular from Carey & Simnett (2006), Jackonson et al. (2008), and Ruiz-Barbadillo et al. (2009). TENURE7 is the variable of interest for both proxies. Following the Integrity in Auditing Acts of 2002, the Truth and Accountability in Accounting Act of 2002 (Myers et al., 2003), and Carey & Simnett (2006), I have used 7 or more years as the length of the auditor tenure. A positive significant coefficient of TENURE7 from the ‘propensity to issue a going-concern opinion’ regression would mean no support for the hypothesis, as this means that auditor will issue a going-concern opinion with a tenure of 7 or more years. On the contrary, a positive significant coefficient of TENURE7 from the ‘discretionary accruals’ regression indicates a support for the hypothesis, as discretionary accruals increase if the audit engagement is 7 or more years and therefore MAFR might be an option.

The variable PROBBANK with a higher value indicates a higher probability of bankruptcy. This control variable is likely to be positively related to both proxies, as a firm with a high probability of bankruptcy is expected to receive a going-concern opinion and those firm managers are more likely to manipulate the numbers to present a more fruitful situation. SIZE is included because large firms are less prone to bankruptcy, as they have a stronger negotiation power. This slope coefficient is expected to be negative related to both proxies. This is because, looking at both measures, large companies auditors have more resources and great incentives to protect their reputation, and therefore are less likely to get a going-concern opinion and less likely to manage earnings (Ruiz-Barbadillo et al., 2009; DeFond et al., 2002). LEV is included because it presents the risk that is associated with debt,

(20)

and CLEV indicates the change. A positive relation is expected for both measures, as a higher ratio of total liabilities to total assets is more likely to get a going-concern opinion and to engage in earnings manipulations. LLOSS is included because companies with a negative result is more likely to fail and receive a qualified opinion and have discretionary accruals. BIG4 is included to capture the difference between audit by a big 4 and a non-big 4. It is expected that a Big 4 is less likely to issue a going-concern opinion and also less likely to approve a managers’ earnings manipulation due to reputation protection (Ruiz-Barbadillo et al., 2009), and therefore a negative relation is predicted with both measures. CFFO is also included and is usually associated with a lower probability of financial distress as the company has more cash and therefore negatively associated with the audit opinion and discretionary accruals proxies (DeFond et al., 2002; Myers et al., 2003; Sloan, 1996). FEERATIO is included because it is expected to decrease audit quality and therefore is negatively related to the audit opinion measure and positively related to the discretionary accruals measure (Frankel et al., 2002). ROA is included in the discretionary accruals proxy. Similarly to Kothari et al. (2005), I believe that ROA increase the probability in detecting abnormal earnings. ROA is expected to be positively related with discretionary accruals, as the larger the accruals, the better the firm performance.

Finally, TENURE2 is included to control for instances when the auditor tenure is short. A positive relation with the first measure would mean that with a tenure of 2 or less years, the audit quality is high as the auditor is likely to issue a going-concern opinion. A negation relation would mean that the audit quality is low, probably due to the lack of client-specific knowledge. Further, a positive relation with the discretionary accruals measure would mean that earnings manipulation is high in the first two years and subsequently audit quality low. Negative relation would indicate that managers are less likely to manipulate the numbers as it might be difficult to due to less familiarity and the different and fresh approach of the new auditors.

2. Sample selection

My research is conducted using a sample of all US companies extracted from the Wharton Research Data Services (WRDS) database during the period of 2000-2012. Financial statement information data for the variables used for the ‘propensity to issue a going-concern opinion’ measure and the ‘discretionary accruals’ measure are obtained from Compustat North America Fundamental Annual and information about audit opinion and

(21)

auditor changes are obtained from Audit Analytics. The tracking of auditor changes and other auditor reports in Audit Analytics are disclosed since the year 2000 and therefore my sample initiates from that year.

A total of 82.775 firm-years observations (7.314 US companies) were obtained in the initial sample. Since Compustat North America Fundamental Annual also include Canadian companies, I removed these from my sample. I also deleted all companies with the SIC codes from 6,021 to 6,999, as those firms are financial and insurance companies with different financial characteristics, which could lead to misleading results. As Deloitte took over a large part of Arthur Andersen’s former business and many partners and employees went to Deloitte after the demise of Arthur Andersen in 2002, the switch from Arthur Andersen to Deloitte is not defined as a switch to another auditor. Also, all variables are winsorized at the 1st and 99th percentiles to remove the effect of significant outliers. Finally, in order to run both regression equations (3) and (4), it is important to have sufficient data for all variables in the models. Thus, observations with missing data for any variable are excluded from the analysis. This resulted in a sample 40.557 firm-year observations (4.077 US companies).

3. Descriptive results

3.1 Descriptive statistics

The descriptive statistics of the audit opinion and discretionary accruals measures are presented in TABLE 2. As can be seen, for 9,30% of the 4.077 companies, a going-concern opinion has been issued. The discretionary accruals is negative, which means that companies on average use earnings management to decrease their earnings. 69,19% of the companies already have the same auditor for 7 or more years, whereas 12,03% have the same auditor for

(22)

2 or less years. This means that for 18,78% of the companies, the auditor tenure is between 3 and 6 years. Further, 66,23% of these companies are audited by one of the big 4. 37,71% reported a loss last year, which is in line with the financial condition score of Zmijewski (1984) of 3.23.

(23)

3.2 Correlation matrices

TABLE 3 presents both the Pearson and Spearman rank correlation matrices between the variables for the audit opinion measure and the discretionary accruals measure. DA and ROA give the strongest correlation with a value of 0.93767. This explains that when the rate of return is high, the level of discretionary accruals are also high, which is also in line with the strong negative correlation between DA and PROBBANK. What is interesting to note, is the fact that the Pearson correlation matrix shows a very strong positive correlation between these two variables, while the Spearman rank correlation matrix shows only a weak positive correlation (0.0878) between DA and ROA. The PROBBANK and ROA are also very strong negatively correlated in both the Pearson and Spearman rank correlation matrices. This is expected, as when the return on assets is high, the probability of bankruptcy will be in the adverse direction, which is also in line with the strong positive correlation between

PROBBANK and LLOSS in the Spearman rank correlation matrix. Further, you can see that CFFO and PROBBANK are also very strong correlated in the Pearson correlation matrix. This strong negative correlation can be explained by the fact that the CFFO is associated with a low probability of financial distress and therefore the PROBBANK is lower. Thus, if CFFO is high, PROBBANK will be low. This correlation is lower in the Spearman rank matrix, but still strong. Another strong correlation in both correlation matrices, is the positive correlation between SIZE and BIG4. This is expected, as big 4 firms have more resources and therefore also serves clients with more capital. Another strong negative correlation that was expected, is the correlation between TENURE7 and TENURE2. The reason for this, is because the auditor tenure can be either a tenure of 7 or more years or 2 or less years, it cannot be both. A stronger correlation might be expected, but this can be explained due to the fact that there are also years in between (3-6 years).

I check for multicollinearity with the variance inflation factor (VIF). In TABLE 4 you can see that there are no multicollinearity problems in the audit opinion measure, as according to the rules of thumbs, none of these VIF are between 5 and 10. However, the VIF of the PROBBANK in the discretionary accruals measure has a size of 6.61, which means that multicollinearity is high. As some prior literature argue that the cutoff value is 10, which is also explained in the book of Kutner (2004), I assume that there are no multicollinearity problems.

7_{Correlation of variables is between +1 and -1, whereby a correlation of +1 means that there is a perfect positive}

(24)

(25)

(26)

IV. RESULTS

1. Univariate regression analysis

Before performing the main analysis, a preliminary univariate regression analysis is performed in order to see whether there is a link between the audit firm tenure and the audit quality. TABLE 5 presents the univariate regression analysis using the propensity to issue a going-concern opinion measure and the discretionary accruals measure. The overall model is highly significant with a p-value of 0.0000 for both measures, which means that the

association between tenure and both measures for audit quality are very strong. The R-squared in the first measure is 0.0804, which indicates that 8,04% of the audit quality

measure is explained by TENURE7. In the discretionary accruals measure this percentage is 2,16%. Looking at the output of the audit opinion measure, TENURE7 has a negative coefficient, which is significant (β1 -1.5741; p-value 0.0000). This is in line with the expectation of this study and supports the hypothesis that claim that long audit firm tenure will decrease the propensity of issuing a going-concern opinion and therefore deteriorates audit quality. The result of the second measure is also in line with the expectation, as the discretionary accruals measure has a positive significant coefficient (β1 0.2132; p-value 0.0000).

The results are in accordance with my expectation, but the analysis is only a

preliminary univariate analysis where I use the variable of interest as my control variable. In the next section, a multivariate analysis will be performed to see whether other factors will also influence the audit quality.

2. Multivariate regression analyses

Seeing in the previous section that the univariate regression analysis, with the variable of interest TENURE7, is in line with my expectation, a multivariate analysis is necessary to conclude about the audit-client relationship with a higher degree of certainty.

First, I will do a logistic regression for the audit opinion measure. As a binary response is predicted, a logistic regression is very suitable for this measure. After that, an ordinary least squared regression will be performed for the discretionary accruals measure.

(27)

2.1. Logistic regression

TABLE 6 shows the logistic regression analysis, which test my hypothesis based on the audit opinion measure. The model shows that the overall model is highly significant, with a p-value of 0.000. 36,26% of the variables in this model explain the audit quality as

measured by the audit opinion. This percentage is not very high, which means that the results for this measure is not very strong.

Again, the variable of interest TENURE7 is in line with my expectation for the hypothesis, as it shows a negative significant coefficient (β1 -0.7361; p-value 0.000). This means that audit quality impairs with a tenure of 7 or more years. This is in line with Carey & Simnett (2006), Arel et al. (2006), and Jennings et al. (2006) who find a lower probability to issue a going-concern opinion and a decrease of audit quality with longer tenure. The control variable TENURE2 also presents a negative coefficient (β10 -0.0266; p-value 0.640), which can be explained by the fact that audit quality is low at the beginning of the audit engagement due to unfamiliarity and lower client-specific knowledge. Although this finding is consistent with the findings of Casterella & Johnston (2013), Johnson et al. (2002), Solomon et al. (1999), and DeAngelo (1981b), this estimate is not significant and therefore should be interpreted with caution.

(28)

Further, most of the other control variables are consistent with prior literature. As expected, PROBBANK has a positive coefficient (β2 0.0051; p-value 0.011). This can be explained due to the fact that companies with a higher probability of bankruptcy are more likely to get a going-concern opinion. However, the significance with 0.0110 is not very

(29)

strong. SIZE has a negative significant coefficient (β3 -0.3448; p-value 0.000). This is

expected, as large companies usually have more incentives to protect their reputation and also have more resources and therefore are less likely to get a going-concern opinion. LEV has a positive significant coefficient (β4 0.7525; p-value 0.000), confirming the finding of Carey & Simnett (2006) that highly leveraged companies are more likely to get a going-concern opinion. This is related to the variable CLEV, as the change will affect the probability to get a going-concern opinion. In this case, a decrease in leverage will decrease the propensity for auditors to issue a going-concern opinion (β5 -0.7539; p-value 0.000). LLOSS has a positive significant coefficient (β6 0.6445; p-value 0.000), which is as expected. BIG4 has a negative significant coefficient (β7 -0.9011; p-value 0.000), suggesting that big audit firms are less likely to issue a going-concern opinion due to e.g. client pressure (Myers et al., 2003). As the cash of a company increases, the probability to get a going-concern opinion decreases. This prediction is in line with the CFFO variable (β8 -0.3744; p-value 0.000). FEERATIO is also as expected and has a negative significant coefficient (β9 -0.8727; p-value 0.000).

2.2. OLS regression

TABLE 7 presents the ordinary least squared (OLS) regression for the discretionary accruals measure. The overall model shows a p-value of 0.000, which is highly significant. R-squared is 0.8644, which means that 86,44% of the variation of audit quality is explained by the use of this second measure.

Inconsistent with the univariate regression analysis and the prediction, TENURE 7 shows a negative coefficient (β1 -0.0114; p-value 0.023). This is not in line with the hypothesis, as long audit firm tenure is expected to have a higher level of discretionary accruals which leads to a decrease in audit quality. It is important to note that this estimate is not significant and therefore the result should be interpreted carefully. The same goes for the interpretation of TENURE2 (pvalue 0.130). TENURE2 shows a negative coefficient (β11 -0.0101), which indicates that managers are less likely to manipulate the numbers in the first two years as if might be difficult due to less familiarity and the different and fresh approach of the new auditors. This can be seen as a support for MAFR, but since the estimate is not significant, this result should be interpreted with caution. PROBBANK has a positive significant coefficient (β2 0.0041; p-value 0.000), which is as expected, as managers are more likely to manipulate the numbers when the company is not doing so well, to present a more healthy situation in order to attract investors. SIZE has a positive significant coefficient (β3 0.0496; p-value 0.000), which is also not in line with my expectation and the findings of

(30)

(31)

Ruiz-Barbadillo et al. (2009) and DeFond et al. (2002), as we expect that large companies have great incentives to protect their reputation, and therefore are less likely to manipulate earnings. This result can be explained by the fact that large companies can afford to be risky, as they have more resources available and also due to their strong negotiation power, which can influence the auditor (Nelson et al., 2002). LEV has a negative significant coefficient (β4 -0.0827; p-value 0.000), which means that with a high risk that is associated with debt, a lower level of discretionary accruals is predicted. This is related to the variable CLEV, as the change will affect the level of discretionary accruals. In this case, an increase in leverage will increase the level of discretionary accruals (β5 0.0297; p-value 0.000). LLOSS has, as

expected, a positive significant coefficient (β6 0.0073; p-value 0.000). BIG4 shows a negative significant coefficient(β7 -0.1049; p-value 0.000), which is as predicted and in line with the findings of Ruiz-Barbadillo et al. (2009) due to reputation protection. CFFO is expected to show a negative coefficient, but instead it shows a coefficient of (β8 0.0438) at a significance level of 0.000, which means that whenever cash increases, the level of

discretionary accruals also increases. A possible explanation might be that in times of economic grow, the company wants to show an even better performance to attract more investors, or another reason can be that earnings management is involved and this increases the cash flows. FEERATIO shows an negative coefficient (β9 -0.0134; p-value 0.333), although it was expected to show a positive coefficient, as a high FEERATIO decreases audit quality and therefore discretionary accruals is more likely to exist (Frankel et al., 2002). This estimate is insignificant and should therefore interpreted carefully. ROA shows a positive significant coefficient (β10 0.0623; p-value 0.000), which means that the ROA is positively related with discretionary accruals. This is expected, as the larger the accruals, the better the firm performance.

3. Sensitivity analyses

As the discretionary accruals measure shows many inconsistencies with my

expectations and the findings of prior literature, and in order to increase the validity of the results obtained, I will conduct several sensitivity analyses. First, I want to see whether there is a problem with the measurement of the discretionary accruals. In the correlation matrices in TABLE 3 you can see that the highest correlation is between DA and ROA. Furthermore, there are also other variables that are strongly correlated with ROA. Therefore, I run the OLS regression for the discretionary accruals measure again as explained by the modified Jones

(32)

model, but without the ROA which was later included by Kothari et al. (2005). After running the OLS regression without the ROA, TENURE7 still has a negative coefficient and it is also not significant (β1 -0.0206; p-value 0.064) as you can see in TABLE 8. The SIZE, CFFO, and FEERATIO are also still positive, positive, and negative respectively, which means that it is still not in line with the expectations. You can also see that the overall R-squared is only 0.4363 now, while it was 0.8644 with ROA.

It is interesting to note that when I don’t cluster the firms for the panel data (use command: regress instead of xtreg in STATA), I do get the expected positive coefficient for the variables TENURE7 (β1 0.0067; p-value 0.001) and FEERATIO (β9 0.0391; p-value 0.000). For brevity I did not include this table.

Further, to estimate the discretionary accruals, I used the absolute numbers for the change in revenues and the change in receivables in equation (4) (results in TABLE 7). As the use of absolute numbers, in contrast to relative numbers, can make a difference for the coefficients in the OLS regression, I will rerun this regression with the relative numbers for the change in revenues and the change in receivables. Nevertheless, TENURE7, SIZE, CFFO, and FEERATIO are still not as expected (TENURE7 (β1 -0.0156; p-value 0.001), SIZE (β3 0418; pvalue 0.000), CFFO (β8 0. 0522; pvalue 0.000), and FEERATIO (β9 -0.0057; p-value 0.666). The only difference here is that TENURE7 is significant now, whereas TENURE7 in TABLE 7 en TABLE 8 were insignificant. For brevity I did not include this table.

I also made a distinction between pre-SOX (2000-2001) and post-SOX period (2002-2012), to test the impact of SOX on the audit profession. I rerun both the logistic and OLS regression for the two measures of audit quality with the variable of interest TENURE7. The results are shown in TABLE 9. The results of this analysis is in line with my hypothesis, as it is predicted that with longer tenure, the audit quality will decrease. Here you can see that TENURE7 shows a negative coefficient in the audit opinion measure and a positive

coefficient in the discretionary accruals measure. This means that with a tenure of 7 or more years, the propensity to issue a going-concern opinion is lower and the level of discretionary accruals higher, both leading to a decrease in audit quality. The results apply to both the pre- and post-SOX period, which means that the SOX (2002) rules did not increase the audit quality and therefore MAFR might be necessary.

And last but not least, I want to see the results of different auditor tenure years. In the main regression results (TABLE 6, TABLE 7), I run the regression analyses whereby the tenure is 7 or more years, based on prior literature. As there is no standard yet about the

(33)

auditor tenure years, I want to see what the impact would be if I take different auditor tenure years. Jenkins & Vermeer (2013) and Davis et al. (2009) also mentioned that auditor tenure and audit quality can have a nonlinear relation, as opposed to the usual linear relation

(34)

described in prior literature. With this test, I also want to test if there is a nonlinear relation like the authors claim. In TABLE 10 you can see the logistic regression analysis for

TENURE5, TENURE7, and TENURE10, as measured by the audit opinion. All three TENURES show a negative significant coefficient of around -0,7, which is in line with the hypothesis. TENURE10 shows the lowest coefficient, which means that audit quality is the lowest when the audit firm is engaged with the client for 10 or more years. However, the table also shows that when the tenure is 7 years or more, it is more likely to give an going-concern opinion than when the tenure is 5 years or more. The results are as follows:

TENURE5: -0.7574; TENURE7: -0.7361; TENURE10: -0.7605. This concludes that there is indeed a nonlinear relation between auditor tenure and audit quality. The results in TABLE 11 shows the OLS regression analysis for the discretionary accruals measure and I find that all three TENURES show a negative coefficient, which means that discretionary accruals decrease with a tenure of 5, 7, or 10 years and therefore does not support the hypothesis. These results, in contrast to the results of the audit opinion measure, are showing a linear

(35)

relation (downwards) between auditor tenure and audit quality. However, the coefficients of TENURE 5 and TENURE 7 are not significant, and should therefore be carefully interpreted.

(36)

(37)

V. CONCLUSION

Mandatory audit firm rotation (MAFR) attracted my attention because of the

questions from legislators and regulators in many countries concerning the effect of MAFR on auditor independence and audit quality. Another reason is that MAFR is currently a hot topic in Europe, as it is promulgated in The Netherlands and will be effective as of the beginning of 2016. And finally my motivation for this study comes from the limited empirical evidence regarding the relation between audit quality and audit firm tenure. This study examines this long auditor-client relationship and the effect on audit quality and to see whether MAFR is supported. For this study, two measures of audit quality is used: (1) propensity to issue a going-concern opinion, and (2) discretionary accruals. For the second measure, the modified Jones model including the ROA as suggested by Kothari et al. (2005) is used.

Using a sample of 4,077 companies for the period of 2000-2012, I can conclude that audit quality is decreased when measured with the audit opinion measure (β1 -0.7361; p-value 0.000), but increased with the discretionary accruals measure (β1 -0.0114; p-p-value 0.023). However, the finding of the second measure is not significant, and therefore should be interpreted carefully. My hypothesis with the audit opinion measure is supported, but not with the discretionary accruals measure. As the hypothesis with the second measure is not supported, I did several sensitivity analysis to test whether I will get a positive result for the discretionary accruals measure if I had run the test differently and to test the robustness of my results. I find that when I rerun the OLS regression based on the modified Jones model

without the ROA as explained by Kothari et al. (2005), the results is still negative and not significant (β1 -0.0206; p-value 0.064). However, when I distinct my sample in a pre-SOX and a post-SOX period and rerun all my main regressions, I find that both measures show a decrease in audit quality, which support my hypothesis. Another notable finding is that when I compare the results of different auditor tenure years (TENURE5, TENURE7, and

TENURE10), I find a nonlinear relation between auditor tenure and audit quality when I measure audit quality with the audit opinion measure.

Based on my findings, I can cautiously state that audit quality is decreased with long audit firm tenure and therefore MAFR is supported. However, there are two concerns for the validity of these findings. First of all, the samples for the post-SOX period is larger than the pre-SOX period, thus the positive results can be explained by the sample difference. And

(38)

secondly, this study uses data from voluntary switches to draw conclusions about a regime of mandatory audit firm switches. Therefore it is not certain whether these findings will hold in a MAFR regime and the results should be interpreted carefully. Based on my last finding, I would also recommend further research on nonlinear relations between auditor tenure and audit quality, to see how and why there is a rise or fall of audit quality during the audit engagement years and when exactly the audit quality is the highest and lowest.