The effects of Firm Ownership Concentration on the level of Audit Risk; and the mediating effect of CEO Narcissism

The effects of Firm Ownership Concentration

on the level of Audit Risk; and the mediating

effect of CEO Narcissism

An archival study of 405 listed firms from the

United Kingdom over the years 2013 to 2016.

Master thesis, MSc Accountancy and Controlling University of Groningen, Faculty of Economics and Business

June 25th, 2018 Word count: 10.884 LEONIE SMIT Student number: S2359448 Address: Metaallaan 203 9743 BT, Groningen Phone number.: +31(0)6 13 16 80 42 E-mail: l.smit.7@student.rug.nl

Supervisor: dr. Y. (Yasemin) Karaibrahimoglu Assessor: prof. dr. J.A. (Jim) Emanuels Co-assessor: G.C. (Gerard) Helminck RA MSc EMA

-2-

The Effects of Firm Ownership Concentration on the level of Audit Risk; and the

mediating effect of CEO Narcissism

ABSTRACT

This research investigates the direct influence of firm ownership concentration and the mediating role of Chief Executive (CEO) narcissism on the level of audit risk from both agency and upper echelons perspective. For the initial sample 536 U.K. and Dutch listed firms were selected over the years 2013 to 2016. This archival study uses both hand-collected data from audit and annual reports, and data from several databases. Due to data unavailability, the final sample resulted in 857 firm-year observations from the U.K.. Consistent with my expectations, the results show that firm ownership concentration has a statistically significant negative effect on the level of audit risk. This implies that firms with ownership concentration face lower levels of audit risk, compared to firms with ownership dispersion. However, no statistical evidence is found for the mediating role of CEO narcissism on this relation. This implies that CEO narcissism does not mediate the relationship between firm ownership concentration and the level of audit risk. The findings of this research could be useful for several parties, in particular the board of directors, shareholders and the external auditor.

Keywords: Audit risk ª Firm ownership concentration ª Narcissism ª Agency theory ª Corporate Governance ª Upper Echelons Theory

-3-

TABLE OF CONTENTS

3.2 Upper echelons theory 9

4 LITERATURE REVIEW AND HYPOTHESIS GENERATION 10

4.1 Firm Ownership Concentration and Audit Risk 10

4.2 The mediating effect of CEO Narcissism 12

5 RESEARCH METHODOLOGY 14

5.1 Sample selection 14

5.2 Measurement of Audit Risk 15

5.3 Measurement of CEO Narcissism 16

5.4 Measurement of Ownership Concentration 17

5.5 Control variables 17

5.6 Statistic model and testing specification 18

6 RESULTS 20

6.1 Descriptive statistics 20

6.2 Correlation analysis 21

6.3 Primary regression analysis 23

6.4 Additional analysis 25

6.4.1 Test for alternative measure of CEO Narcissism 26

6.4.2 Test for audit company and audit partner fixed effects 26

6.4.3 Test for Dutch sample 27

6.4.4 Test for effect of using other control variables 27

-4-

7 DISCUSSION AND CONCLUSION 28

7.1 Findings 28

7.2 Theoretical implications 29

7.3 Practical implications 29

7.4 Research limitations and future research 30

8 REFERENCES 32

APPENDIX 40

Appendix A 40

Appendix B 43

-5-

1

INTRODUCTION

The last few years the accounting profession has attracted a lot of media attention due to several accounting scandals of Big 4 audit firms, caused by failing their duty of care. For example, KPMG gave an unqualified report while this was not appropriate (Murphy, 2017), and PwC was negligent in a mortgage fraud case (Nesmith and Pearson, 2016). By law, audit firms need to ensure that their auditors comply with professional standards (AFM, 2016). The inspection of the AFM showed that several audits performed by Big 4 audit firms contained shortcomings that the auditor did not obtain sufficient and appropriate audit evidence in order to support the auditor’s opinion. In other words, the audit firms acted contrary to the law and failed their duty of care. In return they get high administrative fines for that (AFM, 2016). The accounting scandals may also result in a decreased public confidence in the financial reporting process and related audit functions, and it even may lead to reputation losses (Rezaee, 2004).

The loss of trust and credibility in the auditor made clear that something has to be done to restore the confidence in the accounting profession, which has led to more regulations and supervision. In 2014, the Dutch Institute of Chartered Accountants (NBA, 2014a) published a report of 53 measures to improve the quality and independence of the auditor. Furthermore, the AFM (Authority Financial Markets) aims to be a demonstrably ground-breaking supervisory authority by 2022 and therefore focuses on strengthening and renewing their supervision on auditors (AFM, 2017).

According to professional accounting standards it is necessary to take into account all the risks associated with providing audit services to potential clients, when making the client-acceptance decision (Johnstone, 2000). Auditors need to evaluate the relevant risks (client-business risk, audit risk and auditor’s business risks), and use that evaluation to determine the audit firm’s risk of loss on the engagement (Gramling et al., 1998; Johnstone, 2000). The legal liability for auditors, which is increased by more regulations and supervision, could cause a higher levels of audit risk and may therefore lead to earlier rejection of the client (Gramling et al., 1998; Venkataraman et al., 2008).

Because of the increasing attention paid to the accounting profession and the influence on the assessment of the level of audit risk by the auditor, this research focusses on the level of audit risk. Previous studies investigated the impact of corporate governance mechanisms or ownership characteristics on financial reporting, audit quality and the level of audit fees (e.g. Cohen et al., 2002; Mitra et al., 2007). This research is focussing on firm ownership concentration, as part of corporate governance. According to several studies, firm ownership structure consists of both ownership concentration and ownership type (e.g. Kumar and Zattoni, 2011; Madhani, 2016). Khan et al. (2011) argue that the financial reporting process is influenced by firm ownership concentration, which in return affects the risk assessment of

-6-

auditors. Hence, it would be interesting to consider the impact of firm ownership concentration in determining audit risk.

According to the upper echelons theory, organisational choices reflect the individual characteristics of the top management (Hambrick and Mason, 1984; Hambrick, 2007). In other words, decisions are affected by experiences, values and personal traits of the management. The Chief Executive Officer (CEO) has a strong influence on the ‘’tone at the top’’ or ethical climate of their firms (Judd et al., 2017). Recently, research showed a shift towards greater individualism around the world, promoting the view of the self as self-directed, autonomous and separate from others (Santos et al., 2017). Therefore, people are more likely to endorse individualistic traits such as self-esteem and narcissism (Twenge and Foster, 2010; Twenge et al., 2012). Even in the business world, narcissists are more likely to be chosen as the leading figure of the company than non-narcissists (Chamorro-Premuzic, 2016; O’Reilly et al., 2014). Narcissism can be seen as an essential condition for effective leadership, but there is also a dark side to narcissism because it may lead to an excessive desire for recognition and power which could harm the company (Maccoby, 2000; Rijsenbilt et al., 2011). Shareholders play a central role in appointing CEOs (Lee and O'Neill, 2003). As they want to select the best CEO, they need to consider the skills, characteristics and abilities of the CEO (Kaplan et al., 2008). Besides, the risk assessment of the auditor is influenced by the personality of the CEO, as it reflects a client’s ‘’tone at the top’’ (Judd et al., 2017). According to Judd et al. (2017) narcissistic CEOs want to improve perceptions of themselves. For example, by assuming greater business risk and financial reporting risk, the auditor’s risk assessment and therefore the level of audit risk is influenced.

In sum, firm ownership concentration may influence the level of audit risk, and this relation may possibly be influenced by CEO narcissism. Since the main purpose of this research is to evaluate the link between firm ownership concentration, CEO narcissism and the level of audit risk, the following research question is discussed in this research:

‘’To what extend does firm ownership concentration influence the level of audit risk, and does CEO Narcissism mediate this relationship?’’

The remainder of this paper is organised as follows. The next section discusses the academic contribution. The third section describes two relevant theories concerning the above research question. The theoretical framework of this research focusses on both agency and upper echelons theory. Thereafter, the hypotheses generation is addressed in section four, which makes predictions about the relationship between firm ownership concentration and the level of audit risk and the mediating role of CEO narcissism. The research methodology and associated data collection are described in section five and the results of

-7-

the analysis are provided in section six. In the last section, a conclusion is drawn, and limitations and suggestions of future research are offered.

2

ACADEMIC CONTRIBUTION

This research aims to contribute to the literature on the agency theory, which is further discussed in the theoretical background. Regarding corporate governance, the role of firm ownership concentration may be relevant. Previous studies already investigated the influence of firm ownership structure. For example, the effects of firm ownership concentration on firm performance (Kumar and Zattoni, 2015), earnings management (Ding et al., 2007) and managerial outlook and mentality (Peng et al., 2004). Also some research focused on the effects of firm ownership concentration on audit fees (Mitra et al., 2007) and the risk-taking behaviour (Srairi, 2013).

However, the literature does not provide insights about the direct relationship between firm ownership concentration and the level of audit risk. This is remarkable, as research indicates that the level of ownership concentration is potentially related to different incentives of shareholders whether (or not) to monitor or replace managers (e.g. Mitra et al., 2011; Madhani, 2016), and therefore may influence the level of audit risk. That is why this research tries to explore this gap in the literature. Thereby, the firm ownership concentration is of major importance because it affects the incentives of managers to take more (or less) risks, or to have more rigorous audits performed (Khan et al., 2011; Srairi, 2013). Although some studies investigated the role of firm ownership concentration, it has not been examined in combination with CEO narcissism and the level of audit risk. Therefore, this research contributes to the agency theory literature by extending the literature on the effect that corporate governance may have on the audit risk assessment.

Also, the literature about CEO narcissism in the accounting and audit area is quite new and open to explore. One stream of the accounting literature related to CEO narcissism explained the relationship between CEO narcissism and several actions. For example, Chatterjee and Hambrick (2007) examined the relationship between CEO narcissism and the firm’s strategy and performance. They argued that narcissism is seen as an ingredient that stimulates extreme managerial actions. Research of Judd et al. (2017) has been done on the influence of the clients’ risk behaviour and the auditors risk judgement. They concluded that more narcissistic CEOs tend to take riskier decisions, which causes higher audit fees, indicating that auditors work more and/or charge a higher risk premium. Ham et al. (2018) found that narcissistic CEOs pursue ‘high-exposure’ investments, such as R&D expenditures and M&A expenditures, which could embellish the reputation of the CEO, while they do not invest in routine capital expenditures necessary to maintain the condition of the firm’s assets. Research of Rijsenbilt and Commandeur (2013) indicates that narcissistic

-8-

CEOs are more likely to commit fraud, because they are strongly motivated to strive for self-admiration. Research of Johnson et al. (2013) related the fraud risk assessment to the audit risk. They found a significant relationship between CEO narcissism and the level of audit risk by performing an experiment.

As stated above, previous studies have provided insights about the influence of CEO narcissism on several actions, such as decision-making (Chatterjee and Hambrick, 2007), the risk judgement of auditors (Judd et al., 2017), ‘high-exposure’ investments (Ham et al., 2018) and committing fraud (Rijsenbild and Commandeur, 2013). However, little research has been done on the direct effects of CEO narcissism on the level of audit risk yet (e.g. Johnson et al., 2013). Therefore, this research complement the experimental research of Johnson et al. (2013) and examines the mediating relationship of CEO narcissism on the relationship between firm ownership concentration and the level of audit risk.

This research also enriches the current literature by adding value to the measurement of audit risk by hand-collecting the data from audit reports. Instead of using audit fees as a proxy for the audit risk (e.g. Brockman et al., 2013; Hogan and Wilkins, 2008), in this research the level of audit risk is determined by using the number of risks of material misstatement disclosed by the auditor, as stated in the audit report of the companies.

3

THEORETICAL FRAMEWORK

In this section, two theories are explained to have a closer look at the relationship between firm ownership concentration, CEO narcissism and audit risk. Starting with the agency theory to describe the monitoring function of corporate governance, in this case ownership concentration. Afterwards, the upper echelons theory is used to explain the influence of CEO narcissism.

3.1

Agency theory

The agency theory is as a dominant theoretical framework used to explain corporate governance problems (Kultys, 2016; Dalton et al., 2007). According to Jensen and Meckling (1976) the agency relationship is a contract between two or more persons, whereby the agent is hired by the principal(s) to perform some services on their behalf. Specifically, some decision-making authority is delegated to the agent. A common example is the relationship between the management (i.e. agent) and the shareholders (i.e. principal).

The key idea of the agency theory is that ‘‘principal-agent relationships should reflect efficient organisation of information and risk-bearing costs’’ (Eisenhardt, 1989, p. 59). But in practice there are often problems with the ‘’separation of ownership and control’’, which the agency theory tries to resolve. Agency problems arise when the agent does not act in the best interest of the principal or when the

-9-

principal does not know exactly what the agent is doing. Also the problem of risk sharing can occur in agency relationships, which arises when both parties have different risk-attitudes (Eisenhardt, 1989).

The involvement of an external auditor could contribute to governance efforts in addressing those agency problems (Fan and Wong, 2005). Through an appropriate application of accounting policies, the external auditor can impact the risk-taking behaviour of the CEO. In other words, by imposing fines on managers who intentionally manipulate accounting numbers and financial statements, this behaviour is discouraged (Kashyap and Tomar, 2016). Other research showed that a lack of ownership concentration leads to lower performance caused by agency problems (Gaur et al., 2015). Therefore, firm ownership concentration as a corporate governance mechanism may mitigate the traditional agency problem, by aligning managers and shareholders’ interests (Mitra et al., 2007; Madhani, 2016). The owners of the firm are an important group that governs the actions of the managers (Connelly et al., 2010). Monitoring managers by large and sophisticated shareholders reduces the risk-taking behaviour of managers (Srairi, 2013).

Agency problems may occur more often in organisations with a narcissistic CEO. This can be explained by the fact that CEO narcissism is linked to unethical behaviour, which can be harmful to others (Harrison et al., 2016). For example, narcissistic CEOs are more likely to engage in risky business practices or even commit fraud (Harrison et al., 2016; Judd et al., 2017). Therefore, the self-serving behaviour of narcissistic CEOs causes misalignment between the interests of the CEO and the shareholders. The shareholders cannot always verify that the CEO behaves appropriately. Hence, it is important that organisations have governance mechanisms, such as monitoring, that limit the agent’s self-serving behaviour in order to ensure that the agent behaves in the interest of the principal (Eisenhardt, 1989).

3.2

Upper echelons theory

A common theory used to observe the effects of individual characteristics, for example narcissism, on strategic decision-making is the upper echelons theory (e.g. Judd et al., 2017). Hambrick and Mason (1984) were the first who issued the impact of individual characteristics on the behaviour of top managers. According to their research, organisational outcomes are reflections of the strategic choices of top managers. Moreover, the strategic choices are predetermined by a manager-specific interpretation of a situation (Oesterle et al., 2016). Therefore, managers act differently in the same situation.

The limited perception of the manager is caused by the manager’s constrained cognitive capacity and the situation’s complexity (Oesterle et al., 2016). On the one hand, the cognitive base and values to a decision, the psychological characteristics, create a screen between the actual situation and the manager’s perception of it. On the other hand, the manager’s perception is influenced by the observable characteristics, such as age, tenure, education and functional background (Hambrick and Mason, 1984).

-10-

Combining the above, organisational outcomes are a reflection of the strategic choices of managers, which in return are the result of both psychological and observable upper echelon characteristics. Previous studies mainly focused on the influence of observable characteristics, such as age, gender and experience (e.g. Heavey et al., 2015), whereas research about the psychological characteristics remains a little bit underexposed. Nevertheless, research may include the personality of the CEO, such as narcissism, as one of his or her psychological characteristics.

According to Judd et al. (2017) those characteristics have a strong influence on organisational outcomes, including the ‘’tone at the top’’ of a company. In addition, the firm’s tone at the top, as a reflection of the CEO’s personality, influences the auditor’s inherent and control risk assessment.

4

LITERATURE REVIEW AND HYPOTHESIS GENERATION

This section provides predictions about the direct relationship between firm ownership concentration and the level of audit risk, and the mediation effects of CEO narcissism. Furthermore, it provides the definitions of the concepts of audit risk, firm ownership concentration and CEO narcissism.

4.1

Firm Ownership Concentration and Audit Risk

To get a better understanding of the concept of audit risk, a description of audit risk is given. Audit risk is defined as ‘’the risk that the auditor expresses an inappropriate opinion when the financial statements are materially misstated’’ (IFAC, 2017, p. 73). The American Institute of Certified Public Accountants (AICPA) developed the audit risk model, which helps auditors to assess the audit risk by identifying the determinants of audit risk (Hogan and Wilkins, 2008). The audit risk model is expressed as follows:

AR = IR x CR x DR

According to the audit risk model, the audit risk consists of three components: inherent risk (IR), control risk (CR) and detection risk (DR). Inherent risk is the probability of material misstatement before consideration of any related controls; whereas control risk is the probability that a material misstatement will not be prevented, or detected and corrected, by the entity’s internal control; and detection risk is the probability that a material misstatement is not detected by the procedures performed by the auditor (IFAC, 2017). Several studies describe how those risks are interrelated. The inherent and control risk are based on the auditor’s evaluation of the client. Subsequently, the auditor establishes the level of detection risk by increasing or decreasing substantive testing, in order to achieve an acceptable level of audit risk (Demartini and Trucco, 2016; Hogan and Wilkins, 2008).

-11-

According to Cohen et al. (2002), the auditors’ risk assessments and clients’ acceptance decisions are influenced by corporate governance. In order to help auditors assess various client risks, it is important to get an understanding of the type of corporate governance in place. In absence of other control devices, such as an audit committee or board of directors, firm ownership concentration can serve as an internal corporate governance mechanism (Bozec and Bozec, 2007). Firm ownership concentration relates to the power and control between management and shareholders (Kumar and Zattoni, 2015), whereby shareholders can exert influence on managers to protect their interests (Gaur et al., 2015).

In case of concentrated ownership, shareholders have stronger incentives to monitor managers and have more power within the firm, leading to better monitoring, while when ownership is dispersed, shareholders are weaker because of poor monitoring due to the ‘’free-rider’’ problem (Madhani, 2016). Also, research of Mitra et al. (2007) showed that firm ownership concentration is related to the effectiveness of monitoring. Large, sophisticated shareholders actively monitor managers and reduce the probability of managers to manipulate financial statement information.

On the other hand, dispersion of ownership leads to greater management opportunism (Gaur et al., 2015), which is linked to higher levels of restatements and earnings management (Archambeault et al., 2008; Ghazali et al., 2015). Thus, dispersed ownership is related to misconduct in financial reporting. This is supported by research from Kumar and Zattoni (2015), which indicates that controlling shareholders address opportunistic behavior by managers. Furthermore, firm ownership concentration is negatively associated with the risk-taking behavior of managers (Srairi, 2013). The managerial initiative to pursue risky actions, such as new investment opportunities, decreases when the monitoring effort by large shareholders increases.

As firm ownership concentration might impact the financial reporting process, auditors need to take this into account in their risk assessments (Khan et al., 2011). When assessing the client’s situation, the auditor involves both aggressive management and inadequate corporate governance (Bedard and Johstone, 2004). Because firm ownership concentration ensures better monitoring, the inherent risk of material misstatements in financial reporting decreases (Mitra et al., 2011). Conversely, dispersed ownership leads to an increase of inherent risk, caused by a greater likelihood of misconduct in financial reporting (Gaur et al., 2015). Also due to the diminishing effect of risk-taking behaviour in case of firm ownership concentration, the client’s business risk and thus the inherent risk decreases (Srairi, 2013; Judd et al., 2017). Therefore, if I combine these thoughts, firm ownership concentration is expected to cause lower levels of audit risk. Based on the arguments mentioned above, the following hypothesis is provided:

-12-

4.2

The mediating effect of CEO Narcissism

A possible linkage between firm ownership concentration and the level of audit risk is CEO narcissism. The term ‘narcissism’ originated from Greek methodology, where the young Narcissus fell in love with his own image. The construct of narcissism is most of the time linked with issues of love, esteem or self-regard (Campbell and Campbell, 2009). In the literature, CEO narcissism is defined in different contexts, either as a clinical disorder or a personality dimension (Chatterjee and Hambrick, 2007). In the Diagnostic and Statistical Manual of Mental Disorders (DSM-V) narcissism is described as ‘‘a pervasive pattern of grandiosity (in fantasy or behavior), need for admiration, and lack of empathy, beginning by early adulthood’’ (American Psychiatric Association, 2012). This mental disorder is caused by a combination of genetic factors and early parental relationships, but may change over time as a result of adult life experiences (Chatterjee and Hambrick, 2007). Based on DSM-II, Raskin and Hall (1979) developed the Narcissistic Personality Inventory (NPI), in order to measure narcissism as a personality dimension and assign scores to individuals along that dimension. In this research narcissism can be seen as a personality dimension and the following definition is used: ‘’a positive, inflated, and agentic view of the self; and a self-regulatory strategy to maintain and enhance this positive self-view’’ (Campbell et al., 2004, p. 298). Narcissism is mostly characterised by arrogance entitlement and hostility (Rosenthal and Pittinsky, 2006).

CEOs have an important and influential role within organisations because they have the ultimate authority in making decisions (Bertrand, 2009; Glick, 2011). Since CEOs affect firm performance and firm policies (Hambrick and Quigley, 2014; Bertrand and Schoar, 2003), the selection of a new CEO is an important process. The board of directors, representing the shareholders, are responsible for overseeing and monitoring top management and the CEO selection process, and serves as a ‘’watchdog’’ for the shareholders (Adams et al., 2010). In addition, shareholders can actively monitor managers themselves, and play a central role in appointing CEOs (Lee and O'Neill, 2003). For the shareholders, it is of great importance that the new CEOs fits the needs of the firm and acts in the best interests of the firm (Hoffman et al., 2004). Therefore, a prediction of the future needs should be established. Furthermore, during the CEO selection process, personal skills, characteristics (e.g. background, expertise, age and personality) and abilities of the candidates needs to be assessed (Kaplan et al., 2008; Schmitz, 2012). It can be inferred that shareholders attach value to skills and characteristics that create shareholder value when appointing a new CEO. For example, leadership capability, integrity, having concern for others, being open to criticism, and teamwork can be considered as key factors. As narcissistic CEOs lack integrity and have little concern for others (Peterson et al., 2012; Yukl, 2006), and may even harm the company due to an excessive desire for recognition (Maccoby, 2000), they may be seen as unsuitable candidates.

According to De Jong et al. (2017), large shareholders have interest in the long-term performance of an organisation, and have the voting power to influence firm decisions. Opposed to small shareholders, large

-13-

shareholders are more likely to be actively involved in decisions regarding organisational strategies and governance, such as the appointment or replacement of the CEO. Therefore, it is reasonable to expect that firms with higher levels of firm ownership concentration appoint a less narcissistic CEO, compared to firms with lower levels of firm ownership concentration.

As to the linkage between CEO narcissism and the level of audit risk, previous research stated that narcissism affects the auditors response (Judd et al., 2017). Although some researchers pointed out that narcissism has a positive side for business leaders, CEO narcissism could also be related to detrimental organisational outcomes (e.g. Rosenthal and Pittinsky, 2006; Judge et al., 2009). According to Chatterjee and Hambrick (2007) narcissists tend to generate more extreme and irregular performance. Their belief in themselves may lead to disastrous decisions (Barnard, 2008). Narcissistic CEOs are more likely to pursue bold strategic actions and to engage in risky business practices, leading to either big gains or big losses (Judd et al., 2017).

Furthermore, CEO narcissism is linked to aggressive financial reporting, whereby accounting numbers are distorted to portray a more favorable situation (Hales et al., 2012; Patelli and Pedrini, 2015). By falsifying financial accounting numbers, the CEO is guilty of fraudulent financial reporting (Rijsenbilt and Commandeur, 2013). CEO narcissism may also result in a greater likelihood of unethical behavior of CEOs in order to get whatever they think they are entitled to (Chen, 2010; Harrison et al., 2016). For example, since narcissistic CEOs are motivated to strive for self-admiration, they can behave unethically in order to meet the unrealistic high goals they have set out for themselves (Tracy and Robins, 2003; Rijsenbilt and Commandeur, 2013). According to Rijsenbilt and Commandeur (2013) those deliberate actions that result in financial misstatement to mislead stakeholders, can be seen as managerial fraud. Subsequently, it can be assumed that narcissistic CEOs lack integrity, since integrity refers to a trustworthy, ethical and honest person (Yukl, 2006).

In their audit risk assessment, the client’s fraud risks and business risks as well as personality traits, like narcissism, matter to the auditor (Brockman et al., 2013; Judd et al., 2017). Because narcissistic CEOs engage in more risky business practices and even commit fraud, the client’s business risk and thus the inherent risk increases (Judd et al., 2017). Furthermore, the integrity of the client negatively influences the risk assessment of the auditor. This means, a lower level of client’s integrity causes both higher inherent risk and control risk (Kizirian et al., 2005). According to the audit risk model, the level of audit risk increases due to heightened inherent risk and control risk caused by higher levels of CEO narcissism. Based on the arguments mentioned above, the following hypothesis is provided:

Hypothesis 2: CEO narcissism mediates the relationship between firm ownership concentration and the level of audit risk.

-14-

The figure below presents both hypotheses in a conceptual model.

FIGURE 1 – Conceptual model

5

RESEARCH METHODOLOGY

This section describes how the research is executed. The sample selection criteria, a description of the measurement of the variables and the statistical model used to test the hypotheses are elaborated in this section.

5.1

Sample selection

To conduct research on the relationship between firm ownership concentration, CEO narcissism and the level of audit risk, this study uses a panel data sample of (premium) listed non-financial firms from the United Kingdom and the Netherlands over the years between 2010 and 2016. This in an interesting timeframe because during this period more information became available in the independent auditor’s report due to the fact that the extended audit report became mandatory (Bédard et al., 2016). The U.K. and the Netherlands have been chosen because they can be considered as front liners when it comes to the adoption of the new auditor’s report. Specifically, extended auditor reporting was implemented successfully in the U.K. and the Netherlands in respectively 2013 and 2014, whereas globally it was implemented in 2016 (NBA, 2014; ICAEW, 2017). The relevant data on the variables is obtained from several databases and hand-collected from annual reports and audit reports.

For the initial sample, 405 U.K. and 131 Dutch listed firms were selected, resulting in total 3,621 firm-year observations. However, due to limited data availability for the Dutch firms only U.K. firms are used in the primary analysis. Therefore, 786 year-firm observations were excluded. As the audit report was implemented in 2013 in the U.K., there is no audit risk data available for the years 2010 to 2012. Consequently, several firms did not disclose their audit risk in the years 2013 to 2016. Resulting in total 1,530 year-firm observation with no audit risk data. Likewise, 28 firm-year observations did not have

-15-

complete ownership data. These observations were excluded because no causal relationship between the independent and dependent variable can be established for those firms. Next, 609 firms-year observations have not obtained a CEO narcissism score due to the lack of one of the four narcissism items. Since one third of those missing values (228 firm-year observations) is caused by the unavailability of the CEO’s compensation data, a 2-item narcissism measure is used in the primary test. As non-scoring firms are not relevant for the scope of this research, 381 firm-year observations were excluded. Also, 40 firm-year observations did not report about one or more financial indicators (i.e. market-to-book value and return on assets) and were therefore excluded. A detailed presentation of the sample selection is presented in Table 1. The final sample results in 857 year-firm observations.

TABLE 1 – Sample selection

Initial sample 3,621 year-firm observations

Less: Dutch firms

Firms that did not disclose their audit risks Firms with incomplete ownership data Firms that obtained no CEO narcissism score Firms that did not report on financial indicators1

(786) (1,530) (28) (381) (40)

Final sample 857 year-firm observations

5.2

Measurement of Audit Risk

In previous literature, some researchers used audit fees as a proxy for audit risk (e.g. Brockman et al., 2013; Hogan and Wilkins, 2008). They concluded that auditors need to perform substantive testing to maintain an acceptable level of audit risk, which is consistent with the level of audit fees. The last few years audit regulators focused on improving the auditor’s report, including the International Auditing and Assurance Standards Board (IAASB) across countries, the Financial Reporting Council (FRC) in the U.K., and the Public Company Accounting Oversight Board (PCAOB) in the U.S. As a reaction of the changing audit reporting standards, a more recent stream of research focused on the changing auditor’s report (e.g. Bédard et al, 2016; Gutierrez et al, 2016). The new audit report has to describe ‘’significant risks of material misstatement, the application of materiality, and the audit’s scope’’ (Gutierrez et al., 2016). The additional information, including key audit matters, became available in the independent auditor’s report

1 _{In order to increase the sample size, some missing values of the control variables are replaced by the mean of that specific} control variable, sorted by company.

-16-

after auditors were required to make additional disclosures for listed firms (Bédard et al., 2016). The FRC (2016, p.4) defined key audit matter as ‘’those matters that, in the auditor’s professional judgment, were of most significance in the audit of the financial statements of the current period and include the most significant assessed risks of material misstatement (whether or not due to fraud) identified by the auditor, including those which had the greatest effect on: the overall audit strategy; the allocation of resources in the audit; and directing the efforts of the engagement team’’.

The final standard is designed to obtain more information about the audit directly from the auditor (PCAOB, 2017). Therefore, I assume that it is interesting to use the key audit matters as the measurement of audit risk. Consequently, the level of audit risk is directly measured by the number of abnormal audit risks (NRISKS). That is the total number of risks of material misstatement disclosed by the auditor in the independent auditor’s report reduced by the fixed audit risks which were the same for each firm. The audit risk data is hand-collected from audit reports. Since the disclosure of key audit matters became mandatory for listed firms with a fiscal year starting on the 1st of October 2012 in the U.K., a timeframe from 2013 until 2016 is used for the audit risk data.

5.3

Measurement of CEO Narcissism

In previous research, the Narcissistic Personality Inventory (NPI) is often used as an instrument to measure narcissism (Raskin and Terry, 1988). In that case, the DSM-3 behavioural criteria are used for the measurement of narcissism. However, more recent research suggests to use unobtrusive indicators to measure narcissism (Chatterjee and Hambrick, 2007; Judd et al., 2017). According to Chatterjee and Hambrick (2007) this method is more feasible because questions about personality traits, such as narcissism, result in low response rates, and the given answers would be influenced by social desirability bias.

Therefore, in this research CEO narcissism is measured by a CEO narcissism index. This index is constructed using the four unobtrusive indicators of the study of Chatterjee and Hambrick (2007), which are the following: (1) the prominence of the CEO’s photograph in the company’s annual report; (2) the CEO’s use of first-person singular pronouns in interviews; and (3) the CEO’s relative cash compensation; and (4) the CEO’s relative non-cash compensation. Hence, because the third and fourth indicator would considerably limit the sample size due to insufficient data availability, they are not included in the primary analysis. So, for the primary data analysis, a 2-item narcissism measure (NARCISSISM) is used, excluding the CEO’s compensation data. However, for the additional analysis, all indicators are included (NARCISSISM_ALTERNATIVE). Data for the narcissism index is collected from Execucomp database and

-17-

hand-collected from annual reports for the years 2010-20122_{. Note that a higher score indicates a higher} level of CEO narcissism

5.4

Measurement of Ownership Concentration

This research measures firm ownership concentration as the percentage of closely held shares (OWNERSHIP), which represents the percentage of shares held by insiders. This includes shares held by officers and their immediate families, shares held by any other corporation (besides shares held in a fiduciary banks or other financial institutions), shares held by pension/benefit plans, and shares held by individuals who hold more than 5% or more of the outstanding shares. Note that a higher percentage means more governance power within the firm.

5.5

Control variables

Previous studies showed that several other variables may influence the level of audit risk. To control for confounding effects, control variables on different levels are included in this research. First, firm-level control variables are included to capture the effect of firm size, firm complexity, firm growth and client business risk in the analysis (Judd et al., 2017; Mitra et al., 2007). Firm size (LnASSETS) controls for the effect of various economic phenomena such as risk, profitability and information environment. The inherent risk (IHRISK), foreign sales (FOREIGN) and the number of days to the issuance of the audit opinion (AUDLAG) are used as proxies for the effect of firm complexity, whereas the ratio of the market value of equity to book value of equity (MTBV) is used to control for firm growth. Furthermore, ROA and LEVERAGE are used to capture the respective effects of performance and liquidity on the level of audit risk. LOSS is an indicator variable for firms reporting negative net income. I expect positive coefficients on all variables, with the exception of ROA.

Following, this research controls for audit-related variables, shown to have impact the level of audit risk (Judd et al., 2017). Specifically, I control for the type of auditor that a firm has (BIG4). Furthermore, the amount of audit fees (LnFEES) and the level of materiality during the audit (LnMAT) might influence the decisions of the auditor and therefore the level of audit risk. A negative coefficient on BIG4 and LnMAT and a positive coefficient on LnFEES is expected. Furthermore, CEO-related variables3 _{might impact the}

2 _{In accordance with Judd et al. (2017) this research considers that CEO narcissism is a time-invariant personality trait. This means} there is no variation in the level of CEO narcissism within the same firm, unless there is a CEO change. Therefore, for the years 2013 to 2016, the CEO narcissism score is composed of the average of the two or four indicators of the CEO narcissism measure over the years 2010 and 2011. In order not to lose too much data, some adjustments are made for this. If there is a CEO change between the years 2010 and 2011, the years 2011 and 2012 are used to determine the CEO narcissism score, whereas if there is a CEO change between the years 2011 and 2012, only 2012 is used.

3 _{As the CEO-related control variables would decrease the sample size by around 150 firm-year observations, those variables are} not included in the primary analysis, but only in the additional analysis (section 6.4.4).

-18-

decisions of the CEO and therefore influence the level of audit risk. Consequently, AGECEO, GENCEO and TENCEO are included to control for this effects.

Next, on corporate governance level4_{, board and audit committee characteristics might impact the level}

of audit risk (Cohen et al., 2002; Mitra et al., 2007). Therefore, several board and audit committee related items are included for the confounding effects on the level of audit risk to examine the relationship between firm ownership concentration and the level of audit risk. Based on prior literature (Carcello et al., 2002; Abbott et al., 2003), the following audit committee variables are included: audit committee independence (ACIND), audit committee expertise (ACEXP), and audit committee size (ACSIZE). Likewise, for board variables the number of annual meetings (BDMEET), the percentage of outside directors (BDIND), and the number of board members (BDSIZE) are included. I expect positive coefficients on all variables.

Finally, indicator variables are included to control for year, country and industry fixed effects. As the audit risk data is collected for the years 2013 to 2016, three year indicator variables (YEAR_dum*) are included to control for year-fixed effects. Although the Dutch firms are excluded, country indicator variables (COUNTRY_dum*) are included to control for the differences in the country of incorporation. To control for industry effects industry indicator variables (INDUSTRY_dum*) are included, based on the twelve industries classified by Fama and French.

The data for the above mentioned control variables is obtained from several databases (e.g. Worldscope and BoardEX). The variables are defined in Appendix A.

5.6

Statistic model and testing specification

This paragraph discusses in which way the hypotheses have been tested. This research uses ordinary least squares (OLS) methodology to test the hypotheses stated in section 4. This is done by using the following regression models:

NRISKS = α + β1 OWNERSHIPi,t + β2 CONTROLSi,t + εi [1]

NARCISSISM = α + β1 OWNERSHIPi,t + β2 CONTROLSi,t + εi [2]

NRISKS = α + β1 OWNERSHIPi,t + β2 NARCISSISM + β3 CONTROLSi,t + εi [3]

4 _{As the corporate governance level control variables would decrease the sample size by around 200 firm-year observations, those} variables are not included in the primary analysis, but only in the additional analysis (section 6.4.4).

-19-

where α is the constant coefficient, βi are coefficients and εi is the error. The other variables are further defined in Appendix A.

The first hypothesis, the relationship between firm ownership concentration and the level of audit risk, is tested by using equation 1. To test the second hypothesis, the causal steps procedure, as described in Baron and Kenny (1986), is followed. This procedure will be further explained with reference to Figure 2 below. The first step in the process consists of testing the total effect (path c) of firm ownership concentration on the level of audit risk. Actually this step is completed by testing the first hypothesis. Step 2 and 3 together test the indirect effect (path a x b), the influence of the mediating variable CEO narcissism on the relationship between firm ownership concentration and the level of audit risk. The last step of the process tests whether firm ownership concentration still has a significant effect on the level of audit risk, after controlling for CEO narcissism (path c’). Therefore, equation 2 and 3 are used to test hypothesis 2.

According to the causal steps procedure of Baron and Kenny (1986), mediation can be established if the following four criteria are met: (1) OWNERSHIP should relate to NRISKS, such that β1 in equation 1 is significant; (2) OWNERSHIP should relate to NARCISSISM, such that β1 in equation 2 is significant; (3) NARCISSISM should relate to NRISKS, such that β2 in equation 3 is significant, and (4) the relationship between OWNERSHIP and NRISKS (β1 in equation 3) should be not significant or significantly smaller than the relationship between OWNERSHIP and NRISKS in equation 1.

-20-

6

RESULTS

This section discusses the results of this research. First, the descriptive statistics and the correlation matrix are presented. Next, the multiple regression analysis show the results of the tested hypotheses, and thereafter the regression results are tested for robustness.

6.1

Descriptive statistics

Descriptive statistics for all variables used in the primary test are provided in Table 2, presenting the number of observations, mean, median, standard deviation, minimum and maximum of the variables used in the dataset. In order to reduce the effect of outliers, without losing data, all continuous variables are winsorised at the 1 and 99 percent levels. Due to unavailability of CEO’s compensation data, the 2-item narcissism measure (NARCISSISM) is used in the primary analysis. Furthermore, NARCISSISM is normalised for the ease of interpretation of this variable.

On average, the firms in the sample report 2.78 (3.00) audit risks. The mean (median) OWNERSHIP is 17.44 percent (7.22 percent), while standard deviation is 20.65 percent, indicating significant variation in firm ownership concentration within our sample. The mean (median) NARCISSISM is 0.33 (0.32). Furthermore, approximately 95 percent of the sample firms are audited by a Big 4 audit company (BIG4), and on average the audit opinion was issued within 66 days (AUDLAG). Mean (median) of unlogged total assets (ASSETS) is 9,071 million USD (1,454 million USD), while only 15.40 percent of the observations in the sample reported a loss (LOSS). This suggests that the sample firms are relatively large and profitable. I note that the descriptive statistics of IHRISK, FOREIGN, LEVERAGE, ROA and MTBV are generally in line with previous research (e.g. Judd et al., 2017; Mitra et al., 2007). This appears reasonable because the samples of those previous studies also consists of relatively large firms.

To save space, the statistics for the year-, country- and industry-indicator variables are not shown in Table 25_{. The observations of the sample are roughly distributed equally over the years 2013 to 2016.} Furthermore, approximately 86 percent of the firms is incorporated in the United Kingdom, while the remaining 14 percent is distributed among eleven other countries. The three most common industries which the firms operate in are wholesale, retail and some services (17 percent), manufacturing (13 percent) and consumer nondurables (10 percent). Another 32 percent is attributed to the category ‘other’ and the remaining percentage is distributed among eight other industries. The complete frequencies of

-21-

Variable n Mean Median Std. Dev. Min. Max.

Dependent variable NRISKS 857 2.7829 3.0000 1.3218 0.0000 6.0000 Independent variables OWNERSHIP 857 0.1744 0.07220 0.2065 0.0010 0.7701 NARCISSISM 857 0.3345 0.3257 0.1920 0.0000 1.0000 Control variables

ASSETS (in $ mil) 857 9,071 1,454 28,562 57,55 214,224

IHRISK 857 0.2672 0.2375 0.1847 0.0108 0.9187 FOREIGN 857 0.5038 0.5669 0.3824 0.0000 1.0141 AUDLAG 857 66.2753 63.000 19.6463 28.0000 154.0000 LEVERAGE 857 0.5563 0.5614 0.2105 0.0715 1.1433 ROA 857 0.0616 0.0606 0.08922 -0.3552 0.3012 MTBV 857 3.5706 2.4000 4.4423 -7.4200 29.6000 MAT (in $000) 857 37,622 6,725 109,342 237 800,000 FEES (in $000) 857 3,721 1,174 7,766 92 49,968 LOSS 857 0.1540 0.0000 0.3611 0.0000 1.0000 BIG4 857 0.9486 1.0000 0.2208 0.0000 1.0000

This table presents descriptive statistics for the variables employed in this study. All variables are winsorised at the 1 and 99 percent levels. All variables are defined in Appendix A.

6.2

Correlation analysis

Table 3 presents the correlation matrix among the variables employed in the primary tests. OWNERSHIP is significantly negatively correlated with NRISKS (p < 0.01), providing initial evidence to support the first hypothesis. Besides this, OWNERSHIP is significantly negatively correlated with NARCISSISM (p < 0.05), and NARCISSISM is significantly positively correlated with NRISKS (p < 0.01). Furthermore, 8 out of 11 control variables are significantly correlated with NRISKS at p < 0.10 or lower, suggesting the model is appropriate.

The three largest correlation coefficients with NRISKS are LnFEES (0.4239), LnASSETS (0.3808) and LnMAT (0.3239), adressing the importance of those variables as determinants of the level of audit risk. The correlation matrix helps to control for multicollinearity. Multicollinearity problems arise when a correlation coefficient between two variables is greater than |0.70|, which means those variables are strongly correlated. Table 3 shows that LnMAT and LnFEES are both strongly correlated with LnASSETS. An explanation for this is that often materiality and audit fees are used as a percentage of total assets. Since these variables might cause multicollinearity problems, I excluded LnMAT and LnFEES from the primary analysis. Furthermore, I acknowledge that LOSS strongly correlates with ROA (0.6412). However, the coefficient is still not greater than |0,70|, and is similar to the research of Judd et al (2017). Also the other correlations have not created any major problem of multicollinearity because the correlation

-22-

coefficients in this research remain below the limit, except for LnMAT and LnFEE. In addition to the correlation matrix, for each of the regression models the highest variation inflation factor (VIF) is provided. Table 4 shows that, after excluding LnMAT and LnFEE, the VIF’s of all the models stay below the threshold of 10.

TABLE 3 Correlations Panel A: Correlation Variables NRISKS to AUDLAG

Variables 1 2 3 4 5 6 7 1 NRISKS 1.000 2 OWNERSHIP -0.1083 1.000 3 NARCISSISM 0.1507 -0.0727 1.000 4 LnASSETS 0.3808 -0.2108 0.2303 1.000 5 IHRISK -0.1433 -0.0371 -0.1209 -0.2109 1.000 6 FOREIGN 0.1896 -0.0696 0.0192 0.2470 -0.1938 1.000 7 AUDLAG -0.0499 0.2073 -0.0820 -0.3175 -0.0316 -0.0960 1.000 8 LEVERAGE 0.1851 -0.0786 0.0946 0.2160 0.0646 -0.1550 -0.1388 9 ROA -0.1265 -0.0879 -0.0897 0.0171 0.1665 -0.0788 -0.1856 10 MTBV 0.0117 -0.0642 -0.0294 -0.0042 0.0711 -0.0478 -0.0844 11 LnMAT 0.3239 -0.1772 0.1906 0.9284 -0.1447 0.2368 -0.3018 12 LnFEES 0.4239 -0.2292 0.2188 0.8149 -0.1729 0.4128 -0.2640 13 LOSS 0.1166 0.1152 0.0853 -0.0279 -0.1369 0.0941 0.1466 14 BIG4 -0.0462 -0.2285 0.1317 0.1427 -0.0116 0.0535 -0.1947

Panel B: Correlation Variables LEVERAGE to BIG4

8 9 10 11 12 13 14 8 LEVERAGE 1.000 9 ROA -0.0306 1.000 10 MTBV 0.2262 0.2674 1.000 11 LnMAT 0.1413 0.1510 0.1061 1.000 12 LnFEE 0.2659 -0.0362 0.0591 0.7624 1.000 13 LOSS 0.0318 -0.6412 -0.1076 -0.0747 0.0106 1.000 14 BIG4 0.1251 -0.0130 0.0653 0.1045 0.1885 0.0114 1.000 This table presents Pearson correlations among the variables employed in the primary tests. Correlations in bold are significant at 10 percent level or lower. All variables are winsorised at the 1 and 99 percent levels and are defined in Appendix A.

-23-

6.3

Primary regression analysis

Table 5 presents the regression results for the tests of the direct effect of firm ownership concentration on the level of audit risk (hypothesis 1) and the mediating effect of CEO narcissism on the relationship between firm ownership concentration and the level of audit risk (hypothesis 2).

To illustrate the effects on the level of audit risk (NRISKS) this research uses different regression models. Model 1 regresses NRISKS on several firm-specific control variables. The direction of most control variables is in line with my expectations, based on previous research, such as Judd et al. (2017). I acknowledge that only 5 out of 9 variables show significant associations with NRISKS. Model 2 adds the independent variable, OWNERSHIP, to the regression. As predicted, a statistically significant negative association is documented between firm ownership concentration and the level of audit risk (β = -0.519, p < 0.05). Therefore, hypothesis 1 is supported. The negative effect suggests that firms with ownership concentration report less audit risks than firms with ownership dispersion.

As discussed in the method section, the causal steps approach is used to test the second hypothesis. For further explanation of this method I refer to section 5.6. Since the relationship between firm ownership concentration and the level of audit risk (total effect) is established, the next step includes testing the mediation effect of CEO narcissism (hypothesis 2). Model 3 in Table 5 shows that there is a not significant, negative relationship between firm ownership concentration and CEO narcissism (β=-0.0339, p=0.282), whereas model 4 shows a significant positive association between CEO narcissism and the level of audit risk (β=0.399, p < 0.10). The indirect effect is determined by multiplying both coefficients, resulting in a no significant negative effect of β=-0.0135 (p=0.35). Furthermore, model 4 shows that firm ownership concentration, after controlling for CEO narcissism, is still a significant predictor of the level audit risk, (β=-0.506, p < 0.05). Table 4 below summarises the results of testing hypothesis 2.

TABLE 4 - Summary results hypothesis 2

Step 1 Total effect (path c) - 0.519 Significant

Step 2/3 Indirect effect (path ab) - 0.0135 Not significant Step 4 Direct effect (path c’) - 0.506 Significant

This table presents the total, indirect and direct effects of firm ownership concentration on the level of audit risk.

When linking the results to the criteria for mediation, as discussed in section 5.6, not all conditions to establish mediation are satisfied. Because the results show no significant association between OWNERSHIP and NARCISSISM, causing no significant indirect effect, the second criteria is not fulfilled. In addition, the relationship between OWNERSHIP and NRISKS remains significant after controlling for

-24-

NARCISSISM, and is just slightly smaller than the total effect of OWNERSHIP on NRISKS. As a result, also criteria four criteria has not been met. Therefore, hypothesis 2 is not supported.

TABLE 5 - Regression results primary analysis

Hypothesis 1 Hypothesis 2

NRISKS NRISKS NARCISSISM NRISKS

Variable (1) (2) (3) (4) Constant 0.489 0.777 0.0120 0.772 (0.954) (0.958) (0.145) (0.957) NARCISSISM - - - 0.399* - - - (0.230) OWNERSHIP - -0.519** -0.0339 -0.506** - (0.208) (0.0315) (0.208) LnASSETS 0.237*** 0.226*** 0.0253*** 0.216*** (0.0280) (0.0283) (0.00428) (0.0288) IHRISK -0.0742 -0.114 -0.0876** -0.0786 (0.237) (0.236) (0.0358) (0.237) FOREIGN 0.541*** 0.542*** -0.0489** 0.561*** (0.127) (0.127) (0.0192) (0.127) AUDLAG 0.00261 0.00311 -0.00004 0.00313 (0.00231) (0.00231) (0.000350) (0.00231) LEVERAGE 0.947*** 0.933*** 0.0183 0.926*** (0.215) (0.214) (0.0324) (0.214) ROA -1.273** -1.263** -0.0867 -1.229* (0.631) (0.629) (0.0952) (0.628) MTBV 0.00777 0.00723 -0.00154 0.00785 (0.00976) (0.00973) (0.00147) (0.00973) LOSS 0.183 0.205 0.0426* 0.188 (0.146) (0.146) (0.0221) (0.146) BIG4 -0.612*** -0.694*** 0.0908*** -0.730*** (0.198) (0.200) (0.0303) (0.201)

Year Fixed Effects Yes Yes Yes Yes

Industry Fixed Effects Yes Yes Yes Yes

Country Fixed Effects Yes Yes Yes Yes

Observations 857 857 857 857

R-Squared 0.2685 0.2740 0.2113 0.2766

Adjusted R-squared 0.2382 0.2430 0.1777 0.2449

F-value 8.87 8.85 6.29 8.71

Highest VIF 2.04 2.04 2.04 2.04

This table presents results of the primary regression analysis. Standard errors in parentheses.

*, **, *** Denote statistical significance (two-tailed) at the 10 percent, 5 percent and 1 percent levels respectively. All variables are winsorised at the 1 and 99 percent levels. All variables are defined in Appendix A.

-25-

6.4

Additional analysis

To check the robustness of the regression results, five additional tests are conducted. Table 6 presents a summary of the regression results of the additional analysis, which are further explained in section 6.4.1 to 6.4.5. The complete results can be found in Table C1 to C8 of Appendix C.

TABLE 6 – Summary regression results additional analysis

Hypothesis 1 Hypothesis 2

NRISKS NARCISSISM NRISKS

Variable (2) (3) (4)

PANEL A: Test for alternative measure of CEO Narcissism (Table C1 Appendix C)

NARCISSISM_ALTERNATIVE - - 0.200

OWNERSHIP -0.475* -0.0204 -0.471*

PANEL B: Test for audit company fixed effects (Table C2 Appendix C)

NARCISSISM - - 0.278

OWNERSHIP -0.462** -0.0321 -0.453**

PANEL C: Test for audit partner fixed effects (Table C3 Appendix C)

NARCISSISM - - 0.731**

OWNERSHIP -1.029*** -0.0292 -1.010***

PANEL D: Test for Dutch sample (Table C4 Appendix C)

NARCISSISM - - 0.581

OWNERSHIP -1.045** -0.0833 -0.996**

PANEL E: Test for corporate governance control variables (Table C5 Appendix C)

NARCISSISM - - 0.148

OWNERSHIP -0.755*** 0.0292 -0.759***

PANEL F: Test for CEO-related control variables (Table C6 Appendix C)

NARCISSISM - - 0.176

OWNERSHIP -0.502** -0.0636* -0.491**

PANEL G: Test for LnFEE and LnMAT (Table C7 Appendix C)

NARCISSISM - - 0.433*

OWNERSHIP -0.442** -0.0380 -0.426**

PANEL H: Test for size effects (Table C8 Appendix C) Less than median total assets

NARCISSISM - - 0.814*

OWNERSHIP -0.430 -0.0972** -0.351

Greater than median total assets

NARCISSISM - - 0.308

OWNERSHIP -0.871*** 0.0418 -0.883***

-26-

6.4.1

Test for alternative measure of CEO Narcissism

As mentioned before, in the primary analysis a 2-item CEO narcissism index (NARCISSISM) is used to measure the level of CEO narcissism. This is due to the fact that the unavailability of the compensation data would reduce the number of observations in the sample. This additional test adds the CEO’s compensation data to CEO narcissism index. Consequently, CEO narcissism is measured by forming a 4-item CEO narcissism index (NARCISSISM_ALTERNATIVE), as done before in the research of Chatterjee and Hambrick (2007). The measurement of all other variables remains unchanged.

Panel A of Table 6 presents the results of this additional analysis. As expected, the sample size decreased from 857 to 654 firm-year observations. Consistent with the primary analysis, model 2 shows a significant negative relation between firm ownership concentration and the level of audit risk (β=-0.475, p < 0.10), supporting hypothesis 1. The coefficient is slightly lower than in the primary analysis, and due to the use of an alternative measure the significance level changed from the 5 percent to the 10 percent level. The results of model 3 show a negative, but not significant, association between OWNERSHIP and NARCISSISM_ALTERNATIVE (β=-0.0204, p=0.548). Furthermore, by using the more extensive narcissism measure, the relationship between CEO narcissism and the level of audit risk is no longer significant, but still positive (β=0.200, p=0.506). Together this results in even a lower indirect effect than in the primary analysis 0.00408). The direct effect of OWNERSHIP on NRISKS is again significant and negative (β=-0.471, p < 0.10). The results of this additional analysis point out that besides criteria 2 and 4, also criteria 3 is not fulfilled, indicating there is no mediation. In sum, from this addition analysis I am able to accept hypothesis 1, but it cannot be concluded that CEO narcissism mediates the relationship between firm ownership concentration and the level of audit risk. This is in accordance with the primary analysis.

6.4.2

Test for audit company and audit partner fixed effects

Year, country and industry fixed effects are used in the regression models of the primary analysis, to isolate the effects of firm ownership concentration and CEO narcissism on the level of audit risk. However, the results may also be driven by the effects of firms audited by a particular audit company, or even by a particular audit partner. Therefore, this additional test includes audit company fixed effects and audit partner fixed effects to control for these effects. To check whether the primary results are driven by audit company effects, five indicator variables are included, consisting of the big four audit companies (EY, PwC, KPMG and Deloitte), and a residual group (other). Because of collinearity, BIG4 is not included as a control variable. To control for audit partner fixed effects, indicator variables for the different audit partners are included.

-27-

Based on this additional analysis, for both audit company and audit partner fixed effects, there is a significant negative relation between OWNERSHIP and NRISKS (Panel B and C of Table 6), which is consistent with the primary analysis. Furthermore, Panel B of Table 6 exhibits that the relationship between NARCISSISM and NRISKS is no longer significant (β=0.278, p=0.194) when controlling for audit company fixed effects. However, this has no influence on whether or not to accept hypothesis 2, since other criteria for mediation still have not been met. Therefore, the primary results remain insensitive to the inclusion of audit company or audit partner effects in the analysis.

6.4.3

Test for Dutch sample

This additional analysis tests whether the results of the primary analysis differ if only Dutch firms are included in the analysis. Noteworthy is that the sample size decreased from 857 to 172 observations. Since the sample size drastically differs from the primary analysis, no conclusions regarding the hypotheses is based on this additional analysis. As can be read from Panel D of Table 6, the results obtained from this additional analysis are similar to the ones reported for the primary analysis.

6.4.4

Test for effect of using other control variables

In the primary analysis, only firm-level control variables are included in the regression models. However, corporate governance level and CEO-related control variables may influence the findings. Therefore the hypotheses are re-examined, and CEO, audit committee and board related control variables are included in the regression models. The regression results of including corporate governance level control variables (Panel E of Table 6), exhibits that the relationship between OWNERSHIP and NARCISSISM is still not significant, and the relationship between NARCISSISM and NRISKS is no longer significant (β=0.148, p=0.562). For the relationship between OWNERSHIP and NRISKS the significance level changed from the 5 percent to the 1 percent level, and the coefficient increased. When including CEO-related control variables, Panel F of Table 6 shows a significant negative association between OWNERSHIP and NARCISSISM (β=-0.0636, p<0.10), and a positive, but no longer significant, association between NARCISSISM and NRISKS (β=0.176, p=0.514). The relationship between OWNERSHIP and NRISKS remains the same. In sum, the results of this additional analysis are similar to the results of the primary analysis. Therefore, based on this additional analysis, hypotheses 1 can be accepted, while no conclusion can be drawn that a mediating effect exists, not supporting hypothesis 2.

Next, because of multicollinearity problems, LnMAT and LnFEE were excluded in the primary analysis. Since the correlation analysis showed that those variables are important determinants of the level of audit risk, an additional analysis is conducted in which LnMAT and LnFEE are included and LnASSETS is excluded in all regression models. After dropping LnASSETS, all VIF values remained well below 10. As seen in Table 6 Panel G, the results remain the same as in the primary analysis (Table 5).