CEO age, the pricing of an audit and
internal control quality.
A study into the relation between the age of the CEO, audit
fees and internal control quality using archival data.
Second version
Joost van der Ster (10266011) 10th of June 2016
First supervisor: G. Georgakopoulos Second supervisor:
Course code: 6314M0244
MSc Accountancy & Control, track Accountancy, 2015/2016 Amsterdam Business School Faculty of Economics and Business, University of Amsterdam
2
Abstract
This paper investigates the relation between CEO age and audit fees. The age of the CEO and the audit fees are obtained from the financial statements of U.S. based companies from 2010 to 2014. Prior research describes age-based differences in behavior between young and old individuals, and these findings could influence the pricing of an audit. This study focuses on the relation between CEO age and audit fees, and CEO age and internal control quality. I expect a negative relation between CEO age and audit fees, and a positive relation CEO age and internal control quality. The independent variable of interest, CEO age, is divided into three age groups; young CEOs, who are younger than 52, middle-aged CEOs, with ages ranging from 52 to 58, and old CEOs, who are older than 58. The results suggest that middle-aged CEOs experience significantly lower audit fees than both young and old CEOs. The proxy for internal control quality are financial restatements, and the independent variables are the CEO age groups. However, the research model used did not produce any viable results. This study contributes to the existing literature because it has not been investigated so far and it focuses on the effect CEO characteristic, age in this study, on audit fees. However, this study is subject to limitation due to the availability of data between various databases and due to the sample being restricted to one country.
3
Table of contents.
1. Introduction. ... 4
2. Literature review. ... 7
2.1 Determinants of audit fees. ... 7
2.2 Age-based differences in behavior. ... 10
2.3 Internal control and the Audit Risk Model. ... 13
3. Hypotheses ... 17
4. The research models. ... 19
4.1.1 The audit fee pricing model. ... 19
4.1.2 Variables of the audit fee pricing model ... 20
4.1.2 The financial restatement model. ... 22
4.2.2 Variables of the financial restatement model. ... 23
4.3.1 Sample of the audit fee pricing model. ... 24
4.3.2 Sample of the financial restatement model. ... 25
5. Results. ... 27
5.1 Results of the audit fee pricing model. ... 27
5.2 Results of the financial restatement model. ... 34
6. Sensitivity analysis. ... 40
6.1 Sensitivity analysis of the audit fee pricing model. ... 40
6.2 Sensitivity analysis of the financial restatement model. ... 42
7. Conclusion. ... 44
4
1. Introduction.
Dickins, et al (2008) have interviewed audit practitioners with a goal of increasing the understanding of the audit fee estimation process. The process has changed following the passage of the Sarbanes-Oxley Act of 2002. U.S. average public company audit fees have increased from $400.000 in 2001 to $900.000 in 2005. This suggests that audit fee pricing has been subject to change. They suggest that audit fees may be reduced by enhancing internal control effectiveness. According to prior research there are several factors that determine audit fees, such as: auditee size, auditor size, audit complexity. (Chan, et al., 1993; Cobbin, 2002; Hay, et al., 2006; Palmrose, 1986; Simunic, 1980)
Simunic (1980) was the first to examine the determinants of audit fees and developed a positive model of audit fee pricing. According to the positive model there are two quantities which affect the level of audit fees; the resources utilized by the auditee in the internal control system and the resources utilized by the auditor in the audit examination. Simunic argues that these quantities are complementing, which indicates that the CEO of the auditee can actively lower the audit fees by devoting more resources to the internal control system.
Hay et al. (2006) examined prior research on determinants of audit fees from the last 25 years. They find several determinants of audit fees which are frequently used in studies and countries. They summarize the main determinants of audit fees in their results section and divide them in three groups: client attributes, auditor attributes and engagement attributes. The client attributes are determinants like; size, complexity, risk, leverage and profitability. The auditor attributes are determinants like; size and quality. The engagement attributes are determinants like; non-audit services (Chan, et al., 1993, Cobbin, 2002, Hay, et al., 2006, Niemi, 2004, Palmrose, 1986, Simunic, 1980).
There has been extensive research into age-based differences in behavior of individuals (Dawson, 1997; Deshpande, 1997; Hambrick & Mason, 1984; Peterson, et al., 2001; Serfling, 2014; Terpstra, et al., 1993; Yim, 2013). Mudrack (1989) was on the first researchers to examine age-based differences in behavior. He finds that older individuals have a lower Machiavellian score. A high Machiavellian score indicates that the individual has a cynical disregard for morality and a focus on self-interest and personal gain. His results argue that older individuals are less likely to commit fraud for personal gain.
Further, prior research argues that as an individual ages they become more ethical (Dawson, 1997; Deshpande, 1997; Peterson, et al., 2001). Individuals become more ethical because they have experienced more personal and professional life events. Prior research
5
examined if there is a difference between the risk appetite of young and old individuals (Hambrick & Mason, 1984, Serfling, 2014, Yim, 2013). These researchers have found that older individuals are less likely to pursue risky strategies because they have more
responsibilities and a shorter career horizon. Their results suggest that as an individual ages, they become more risk-averse . From prior research I can conclude that as an individual ages he becomes more ethical and risk-averse.
More recently, Huang et al. (2012) examined the relation between CEO age and the firm’s financial reporting quality. The financial reporting qualities examined are the meeting or beating of analyst earnings forecasts and financial restatements. They hypothesize that older CEOs are associated with higher financial reporting quality. Their results suggest that CEO age is negatively associated with firms meeting or beating analyst earnings forecast and financial restatements.
Higher financial reporting quality indicates that there is a lower likelihood of a
financial restatement. This would in turn lead to lower audit risk for the auditor and therefore they might charge less audit fees. The purpose of this study is to examine if there is a relation between CEO age and the audit fees paid by the auditee. Therefore the research question is:
How is the pricing of an audit influenced by the age of the CEO? To examine this relationship
this study will look at U.S. firms in the years 2010, 2011, 2012, 2013 and 2014. I used the most recent five year sample to include sufficient observations and financial restatements in the sample.
The research found, by dividing CEO age into three age groups, that audit fees are significantly lower for middle-aged CEOs. Further, this study found that the previously explained determinants of audit fees by Hay et al. (2006) all have a significant relationship with audit fees. This study provides evidence on the already existing literature on the
determinants of audit fees and provides new evidence that there might be a relation between CEO age and audit fees.
This study will contribute to the existing literature because the relation between CEO age and audit fees has not been investigated so far. According to Simunic (1980) the CEO has the ability to reduce audit fees by devoting more resources to the internal control system. Therefore a risk-averse and ethical CEO can reduce the likelihood of financial restatements by devoting more resources into the internal control system. The risk of the audit examination is based upon the likelihood of financial restatements. An older CEO might have a lower audit risk which would result in lower audit fees. Additionally, I will examine the relation between CEO age and internal control quality.
6
The following part of the study consist of six more chapters. Chapter 2 reviews related prior research and literature. The hypotheses are developed in chapter 3. Chapter 4 explains the research models, the variables and the sample selection procedure. In chapter 5 the results of the descriptive statistics and regression results are explained for both research models. Further, in chapter 6 the sensitivity analyses are described. Finally, chapter 7 describes the conclusions, limitations and suggestions for future research of this study.
7
2. Literature review.
This section of the study provides information about the main theoretical concepts that will be used when conducting the research, based on prior research. This study will be based on auditing, finance and psychology literature. The first section will describe determinants of audit fees that have been used in prior research. The second section will describe age-based difference in behavior for individuals and for CEOs specifically. The third section will describe internal control quality and how it is assessed by the auditor.
2.1 Determinants of audit fees.
In 2000 the Securities and Exchange Commission (SEC) adopted a new rule that require companies to publicly disclose information about the audit fees paid to the auditor. The audit firm’s sole purpose is to provide audit- and non-audit services to companies. This means that audit- and non-audit fees are the main source of revenue for audit firms.
The determinants of audit fees have always been a great research topic in the history of accounting. In 1980 Simunic started research in the existence of competition among auditors and the monopolizing of the then so called ‘Big Eight’ auditors. During his research he derived a positive model of the process by which audit fees are determined. This positive model recognizes the essential interdependence of the auditees and auditor’s economic interests. The auditee demands a positive quantity of auditing because external auditors have advantages over internal controls. Both the auditee and the auditor want to minimize losses from possible ex-post litigation, due to financial statement errors. Therefore the auditor will provide that level of auditing required to minimize ex-post litigation, in order to prevent damage to his reputation. Simunic (1980) derived four determinants which influence the level of audit fees. Firstly, the quantity of resources utilized directly by the auditee, the firm which is being audited, in operating the internal accounting system. This represents the amount of effort and resources the auditee uses in order to operate their internal control system. Secondly, the quantity of resources utilized by the auditor in performing the audit
examination. This represents the amount of effort and resources used by the audit firm in order to perform the audit. The audit firm bases this on the amount of man-hours. Thirdly, the per-unit factor cost of internal accounting system resources to the auditee. This represents the fee per hour of the employees who operate the internal control system. Fourthly, the per-unit factor cost of external audit resources to the auditor, including all opportunity costs and a
8
provision for a normal profit (Simunic, 1980). This represents the fee per hour of the employees of the audit team.
Simunic (1980) argues in his positive model of the determinants of audit fees that the amount of resources utilized directly by the auditee in operating the internal control system is a complement to the quantity of resources utilized by the auditor in performing the audit examination. If these two factors are complementing then a higher amount in quantity of resources utilized directly by the auditee in operating the internal control system will lead to a lower amount of quantity of resources utilized by the auditor in performing the audit
examination, ceteris paribus. Therefore a stronger internal control system will lead to lower audit fees.
When Simunic (1980) conducted his research, there was no public data available about audit fees because there was no legislation to disclose this information. However during the 1990’s and 2000’s country after country began making laws for mandatory disclosure of audit fees of listed companies. Therefore researchers were able to begin conducting research into the determinants of audit fees using empirical data from listed companies and audit firms. In 1993 Chan et al. began using data from listed UK companies to determine the factors which influence audit fees. They empirically tested several determinants of which these are most significant.
Auditee size is by far the most significant explanatory variable in determining audit fees (Chan, et al., 1993). The majority of previous studies have used total assets as a measure of size (Cobbin, 2002; Jubb, et al., 1996; Palmrose, 1986; Simunic, 1980). However turnover is also a good variable depending on the nature of business researched (Chan, et al., 1993). These studies found that auditee size is positively related to audit fees (Chan, et al., 1993; Cobbin, 2002; Jubb, et al., 1996; Palmrose, 1986; Simunic, 1980). An audit firm charges a client based upon the time needed to complete the audit. The bigger the company, the more transactions need to be examined in order to successfully complete the audit. This leads to more hours spent by the audit firm on the job. Therefore the biggest part of the audit fees is explained by the size of the auditee.
The requisite audit effort may be expected to increase with increased complexity of the audit task, which in turn is likely to lead to higher fees. Auditee complexity costs are a
reflection of the nature of business, the location, the quality of internal control system and the proportion of unusual transactions (Chan, et al., 1993). Complexity measures used have either related to the nature of the business or to a particular balance sheet composition. Frequently used measures for the balance sheet are inventory to total assets ratio, and accounts receivable
9
to total assets ratio (Chan, et al., 1993; Hay, et al., 2006; Jubb, et al., 1996). These ratios are included because these assets bear greater costs of auditing. Inventory may consist of a great variety of heterogeneous items, and significant audit problems may arise when verifying the existence of ownership of inventory or determining whether realizable value is less than cost (Chan, et al., 1993). Accounts receivable can be hard to verify because of the existence of bad debtors, which can have a significant negative impact on the total amount of accounts
receivable.
Audit firms also base their audit approach on some form of an audit risk model in which the planned extent and scope of audit testing are determined by the perceived risk of audit failure (Chan, et al., 1993). One would expect higher audit risk to result in a higher audit fee either as a consequence of more hours needed for audit testing or as an audit premium. Turley and Cooper (1991) found support for this hypothesis by conducting interviews; all the interviewees agreed that audit risk was a significant factor in determining the extent of necessary audit work. Audit risk reflects the nature of the business and the control environment instituted by the company (Hay, et al., 2006). Previous studies used balance sheet measures such as liquidity ratio and current ratio. However, audit firms have also used subjective judgment to determine the audit risk, such as the integrity of senior management and assessments as to the strength of the internal control system (Chan, et al., 1993).
Palmrose (1986) examined the relation between audit fees and the size of the auditor. In particular, the difference in audit fees between Big-Eight firms and non-Big-Eight firms. Palmrose found that Big-Eight firms received higher audit fees, which is consistent with either monopoly pricing or higher audit quality. To distinguish between these two
explanations, Palmrose examined the relation between total audit hours and audit fees. He found that there were similar results as to non-Big-Eight firms. This similarity of results supports that Big-Eight firms provide higher audit quality to companies and therefore this
results in higher audit fees.
Most of the studies that researched the relation between auditor size and audit fees have focused on the difference between non-Big-Eight firms and Big-Eight firms (Chan, et al., 1993; Palmrose, 1986; Simunic, 1980). Niemi (2004) extends this literature by examining the same relationship, however using a sample of non-Big-Eight firms. He examines if there is a pricing difference between small audit firms. Using a sample of hourly billing rates of 100 small Finnish audit firms, he concludes that audit firms who have established a reputation higher than the average reputation charge higher hourly audit fees. This study suggests that audit firms with greater reputation are able to charge higher audit fees.
10
The opinion of audit partners as to whether there is a casual link between auditee profitability and audit fees is divided. There is an agreement that a client who faces financial pressure will likely need more audit work to assess the value of assets on a break-up basis in case of an adverse opinion on going concern. In order to assess whether additional valuation of assets is needed, audit partners have agreed that some form of audit profitability has an effect on audit fees. Measures which are most commonly used are return on shareholder’s equity and return on assets (Chan, et al., 1993; Hay, et al., 2006; Palmrose, 1986).
These determinants of audit fees have been frequently used in prior research as control variables. Therefore I will be using most of the determinants discussed in this section as control variables in both research models. The determinants of audit fees described in this section are the most explanatory variables according to prior research (Chan, et al., 1993; Cobbin, 2002; Gonthier-Besacier & Schatt, 2007; Hay, et al., 2006; Jubb, et al., 1996; Niemi, 2004; Palmrose, 1986; Simunic, 1980).
2.2 Age-based differences in behavior.
Prior research has given differences in behavior based on age a great deal of attention (Dawson, 1997; Deshpande, 1997; Hambrick & Mason, 1984; Mudrack, 1989; Peterson, et al., 2001; Serfling, 2014; Terpstra, et al., 1993; Yim, 2013). In 1989 Mudrack started doing research in Machiavellianism using an adult sample. One who is Machiavellian has little believe in the goodness of the world. Mudrack (1989) finds that as one gets older their Machiavellian score becomes lower. This would imply that older people are more able to believe in the goodness of the world. Mudrack (1989) explains the lower Machiavellian score by arguing that older individuals have greater experience in social situations than younger people do, simply because they have likely encountered a greater range of situations.
Unfortunately, a simple explanation for this finding remains somewhat elusive. Mudrack was one of the first of many to examine age-based differences in behavior.
Terpstra et al. (1993) investigated the influence of personality and demographic variables on an individual’s ethical decision related to insider trading. One of these variables was the age of the individual. They hypothesize that younger individuals would be more apt than older individuals to engage in insider trading. To examine this hypothesis they took a sample of men and women of an under-graduate business class and exposed them to several ethical dilemmas. They find that younger individuals are more likely to engage in insider
11
trading than older individuals. They believe that an individual’s value may change as they pass through certain personal or professional life experiences (Terpstra, et al., 1993).
Following this research Deshpande (1997) examined the impact of sex, age, and level of education on the perception of various business practices in a large non-profit organization. Many non-profit organizations have come under scrutiny in the popular press for unethical and questionable activities, however researchers have virtually ignored ethical practices in non-profit organizations. Deshpande (1997) hypothesizes that older subjects will be more ethical in the perception of various business practices. To examine this hypothesis, a questionnaire was delivered to 252 middle level managers of a non-profit charitable
organization. The results of this study support the hypothesis, that as age of the middle level managers increases, their ethical attitudes become more conservative.
Dawson (1997) addresses the question of whether men- and women in sales differ in the ethical attitudes and decision making with age and years of experience. A questionnaire was constructed containing 20 scenarios which described possible ethical misconduct by salespeople. This questionnaire was filled in by 203, of which were 109 male and 94 female. Dawson (1997) offers evidence that ethical standards do change with age for both men and women. The evidence suggests that as age progresses, ethical levels become higher for both genders.
Peterson et al (2001) examine how ethical beliefs and external factors affecting ethical beliefs are related to age and gender of business professionals. In this study, statistical tests were conducted to determine if ethical standards increase with age and whether the rate of increase depends on gender. A questionnaire was constructed containing 14 items of which only 9 were relevant to the study, these items required a response on a Likert-type scale. A total of 280 questionnaires were filled in and the median age of the respondents was 30. The results demonstrate that ethical levels were higher for business professionals in the group that were over 30 years of age.
Prior studies suggest that as an individual ages, the individual becomes more ethical (Dawson, 1997; Deshpande, 1997; Peterson, et al., 2001; Terpstra, et al., 1993). This might be due to that older individuals have experienced more personal and professional life events. It might also be that an older individual has gained more knowledge and expertise because they have experienced a greater range of events. Ethical conservatism can be linked to risk. As one becomes more ethical, the likelihood that he will knowingly harm to organization for his personal benefit will be lower.
12
Prior studies have found that risk-appetite shifts when individuals age (Hambrick & Mason, 1984; Serfling, 2014; Vroom & Pahl, 1971; Yim, 2013). The next subsections explains the relation between the age of an individual and his risk-taking behavior.
Vroom and Pahl (1971) examine the relationship between age and risk taking among individuals. Kogan and Wallach (1964) have found that group decisions tend to be riskier than those of individuals, which made Vroom and Pahl (1971) examine the relation between risk-appetite and the age of individual. A choice-dilemma questionnaire by Kogan and Wallach (1964) was presented to 1484 male managers employed in over 200 corporations in a large variety of positions. The results show a significant relationship between age and measures of both risk-taking and of the value placed on risk. They argue that an individual gains more responsibilities as he gets older. These responsibilities might make him more risk-averse and therefore he might take fewer risks than when he was younger.
Hambrick and Mason (1984) examine the upper echelon perspectives of top managers of organizations. One of the upper echelon characteristics examined is the age of top
managers. This study yield consistent results that managerial youth appears to be associated with corporate growth. They find that organizations with younger managers are inclined to pursue risky strategies and therefore will achieve greater growth and have higher variability in profitability than older managers.
Serfling (2014) more recently examined the relation between CEO age and risk-taking behavior. He uses the firm’s stock return volatility as a measure of overall risk, and he
controls for several industry-, year- and firm-fixed effects. He next examines the relation between CEO age and four corporate policies through which CEOs can influence their firms’ risk profile. Firms with low research & development expenditures, more diversified
operations and lower operating leverage are considered to be less risky. A sample of
ExecuComp firms from 1992 to 2010 is used. The results presented are consistent with prior studies that CEOs reduce firm risk as they age. Next, he examines whether there is a relation between CEO age and the risk-raking incentives provided to the CEO. He finds that older CEOs have a lower portion of stock option in their incentive plan, which indicates that older CEOs have less incentives to pursue risky strategies. Serfling (2014) also finds that firms with lower stock return volatility and firms that are more diversified tend to hire older CEOs. These results imply that firms tend to align the risk preference of the firm with risk preference of the CEO. Serfling (2014) identifies CEO age as a source of natural risk preference that affects their risk-taking behavior.
13
However, he uses a different proxy for risk-taking behavior, the acquisition behavior of the CEO. He demonstrates that incentives to pursue acquisitions vary with a CEOs career
horizon. A successful acquisition is accompanied by on average a $300,000 increase in annual compensation. Younger CEOs face a stronger incentive to pursue an acquisition, because they have a longer career horizon. These incentives predict a negative relation between CEO age and acquisition behavior. He finds that the age of the CEO is an important determinant of the firm’s acquisition behavior. A firm with a 20 year older CEO is 30% less likely to announce an acquisition. The incentive for younger CEOs to engage more frequently in acquisitions can lead to overinvestment and value-destruction. Therefore firms need to tailor compensation incentives to the person, not just to the firm (Yim, 2013).
In summary, prior research suggests that older individuals are more ethical than younger individuals because they have experienced more personal and professional events (Dawson, 1997; Deshpande, 1997; Peterson, et al., 2001; Terpstra, et al., 1993). Prior research also suggests that older individuals are more risk-averse than younger individuals because they have more responsibilities and a shorter career horizon (Hambrick & Mason, 1984; Serfling, 2014; Vroom & Pahl, 1971; Yim, 2013). Therefore older individuals are more likely to reduce the risk of the firm and are less likely to commit fraud than younger individuals.
2.3 Internal control and the Audit Risk Model.
Internal control for financial reporting has the task to provide reasonable assurance about the reliability of financial statements. Internal controls are policies and procedures related to: maintaining accounting records, authorizations and safeguarding assets. In order for these to be effective, they should prevent that material misstatements occur within a given area or that they will be detected and resolved by the management. If the internal controls cannot provide such assurance, then the financial reporting quality will decline (Hogan & Wilkins, 2008).
Two sections of the Sarbanes-Oxley Act relate specifically to internal control. SOX section 302 was finalized on August 29 of 2002 and requires management to periodically, quarterly and annually, report on the control processes and procedures. According to SOX section 404, the report should be conducted by external auditors and should reflect the effectiveness of the internal control processes and procedures. The Sarbanes-Oxley Act disclosures allow the stakeholders to identify the control risk of their investments (Hogan & Wilkins, 2008).
14
Hogan and Wilkins (2008) examined whether a firm with internal control deficiencies experience higher audit fees. They compare the audit fees of firms with internal control deficiencies, with firms that do not experience internal control deficiencies. Hogan and Wilkins (2008) show that a firm with internal control deficiencies experience significantly more audit fees. This research successfully establishes a link between internal control quality and audit fees.
Disclosures of internal control problems made under SOX 302 vary by type of problem and severity. One of the most important distinctions can be made between material weaknesses and significant deficiencies. Hogan and Wilkins (2008) report that market reactions associated with material weaknesses are significantly more negative than market reactions associated with significant deficiencies. A significant deficiency can be identified as an internal control problem which (a) only reports a control problem in one category and (b) the control problem is account-, subsidiary-specific or involves account reconciliations. A material weakness can be identified as an internal control problem which (a) has issues in a more problematic area or (b) has problems across multiple categories.
Hoitash et al. (2008) also examine the association between audit fees and internal control over financial reporting after the implementation of the Sarbanes-Oxley Act of 2002. SOX 404 disclosures made internal control weaknesses of public companies available to the public. These disclosures suggest that internal control weaknesses are valuable to investors because these represent a certain amount of risk. Hoitash et al. (2008) first establish a positive association between audit fees and disclosures of SOX 404. Second, they make a distinction between the disclosure of a material weakness and a significant deficiency. They find that material weaknesses do have a significant positive association with audit fees, however significant deficiencies have an insignificant positive association. Third, they make a
distinction between general and account-specific problems. They assume a greater association between general problems and audit fees than account-specific problems because auditors are more able to ‘audit around’ account specific problems. The results implicate that both
problems have a positive association with audit fees, but general problems have a stronger positive association with audit fees (Hoitash, et al., 2008).
Raghunandan and Rama (2006) also examine the association between audit fees and material weakness disclosures. They use a sample of manufacturing firms one year after the implementation of SOX section 404. They find that audit fees are significantly higher for the manufacturing firms in 2004 than in 2003, one year prior to SOX 404. Not only did audit fees increase due to SOX 404, but the audit fees were also higher for firms that report a material
15
weakness in their disclosure. Surprisingly, firms that voluntarily disclosed a material weakness did not suffer higher audit fees in both 2003 and 2004. Raghunandan and Rama (2006) make a distinction between systemic and non-systemic problems. This classification is similar as to a material weakness and a significant deficiency, where a systemic problem is a material weakness and a non-systemic problem is a significant deficiency. They find that audit fees are higher irrespective of whether internal control weaknesses can be categorized as systemic or non-systemic problems.
From these three studies I can conclude that internal control weaknesses have a positive association with audit fees. However there is mixed evidence that material
weaknesses have a stronger positive association with audit fees than significant deficiencies. To understand why internal control problems lead to higher audit fees I explain the Audit Risk Model. This model is used by auditors to assess how much evidence needs to be gathered to prevent the issuing of an inappropriate opinion of the financial statements (Blokdijk, 2004).
The Audit Risk Model is essentially based on the idea that an auditor’s detection risk is influenced by inherent risk and control risk of the auditee. The audit risk model uses the following formula: . ISA 400 defines detection risk (DR) as: the risk that an auditor’s substantive procedures will not detect a misstatement that exists in an account balance or class of transactions that could be material, individually or when aggregated with misstatements in other balances or classes. Inherent risk (IR) is defined as: the susceptibility of an account balance or class of transactions to misstatements that could be material,
individually or when aggregated with misstatements in other balances or classes. Control risk (CR) is defined as: the risk that a misstatement, that could occur in an account balance or class of transactions and that could be material, individually or when aggregated with misstatements in other classes, will not be prevented or detected and corrected on a timely basis by the accounting and internal control systems (Blokdijk, 2004) . Inherent- and control risks are influenced by the auditee whereas detection risk is depended upon the assessment of the auditor. These three forms of risk together are influencing audit risk (AR), which is defined as the risk that an auditor expresses an inappropriate opinion on the financial statements.
When confronted with financial statements to be audited, the auditor assesses the inherent risk and control risk of the auditee. The auditor should assess the levels of inherent risk and control risk to determine the nature, timing and extent of substantive procedures needed to reduce audit risk to an acceptable level. The purpose of this assessment is: the higher the assessed inherent risk and control risk, the more evidence the auditor has to gather
16
in performing the substantive procedures (Blokdijk, 2004). Inherent risk is largely determined by the activities and the external environment of the auditee. Control risk is determined by assessing the level of internal control implemented by the auditee based upon the decisions of management.
Management is able to actively lower control risk by providing more resources to the internal control system. Following the Audit Risk Model, if control risk is lower and inherent risk is given then detection risk can be higher to arrive at the same level of audit risk. This would imply that the auditor has to gather less evidence in performing the substantive procedures. Therefore management can impact the amount of resources the auditor has to utilize in the audit examination.
17
3. Hypotheses
Taken all findings of prior literature together, this chapter explains two hypotheses that will be tested in this study. This chapter gives a brief overview of the findings in the literature section, which will then be used to formulate the hypotheses.
According to prior literature there are several determinants of audit fees (Chan, et al., 1993; Cobbin, 2002; Hay, et al., 2006; Niemi, 2004; Palmrose, 1986; Simunic, 1980).Auditee size is by far the most explanatory variable, which is positively related to audit fees. Audit fees are likely to increase when the audit complexity increases, which is measured using inventory over total assets and accounts receivable over total assets. Auditor size is also positively related to audit fees, the size of the auditor is most commonly measured with a Big4 dummy variable.
The Audit Risk Model is commonly used by auditors to examine how much evidence needs to be gathered to conclude with an appropriate opinion about the financial statements (Blokdijk, 2004). Herein the auditor assesses the control risk and inherent risk of the company, which he uses to determine the amount of detection risk, in order to arrive at the planned audit risk. Blokdijk (2004) and Simunic (1980) argue that management is able to influence the amount of evidence the auditor has to gather by strengthening the internal control system.
Since the implementation of the Sarbanes-Oxley Act section 404 public companies have to disclose internal control weaknesses to the public. Various researchers have shown that a disclosure of a material weakness leads to higher audit fees for the company (Hogan & Wilkins, 2008; Hoitash, et al., 2008; Raghunandan & Rama, 2006). These results are in line with the Audit Risk Model. A material weakness increases the control risk and therefore the auditor has to decrease the detection risk. Lower detection risk leads to more work for the auditor, he has to gather more evidence in order to prevent the issuing of an inappropriate opinion on the financial statements (Blokdijk, 2004).
Prior studies have shown that older individuals are more risk-averse than younger individuals (Hambrick & Mason, 1984; Serfling, 2014; Vroom & Pahl, 1971; Yim, 2013). Older individuals are less likely to pursue risky strategies because they bare greater
responsibilities in their personal life’s and have a shorter career horizon. Not only are older individuals more risk-averse, they also are more ethical (Dawson, 1997; Deshpande, 1997; Peterson, et al., 2001). The findings from these studies suggest that older individuals are more ethical because they have gained more knowledge and expertise and have experienced a
18
greater range of events.
Huang et al. (2012) find that CEO age is a determinant of financial reporting quality. An older CEO leads to higher financial reporting quality of the firm. In their study financial reporting quality is measured using the likelihood of a financial restatement and the meeting or beating of analysts’ earnings forecast. A financial restatement is the revision of one or more of a company’s previous financial statements, which has a negative impact on financial reporting quality. The meeting or beating of analysts’ earnings forecast is associated with earnings management, which has a negative impact on financial reporting quality..
Higher financial reporting quality can therefore lead to less work for the audit firm. According to the audit risk model, lower control risk allows the auditor to increase detection risk, which results in less evidence gathering for the auditor. Control risk would be lower if there is a lower likelihood of a financial restatement. I expect that an older CEO, who is more risk-averse and ethical, would pursue a stronger internal control system, which therefore would lead to a lower amount of audit fees. This leads to my first hypothesis:
H1: Firms with older CEOs experience lower audit fees than firms with younger CEOs. In order to support this hypothesis I expect that older CEOs have higher internal control quality. Munsif, et al. (2011) examine the relation between audit fees after the
remediation of an internal control weakness. Their motivation comes from section 404 of the Sarbanes-Oxley Act. SOX 404 continues to be controversial and there are significant costs for firms with an adverse report on internal control. They have investigated firms that suffer from an internal control problem in the first 4 years after the implementation of SOX 404. They show that firms who remediate these internal control problems continue to pay higher audit firms in the years after the internal control problems. This provides strong evidence about the ‘stickiness’ of audit fees and suggests that even multiple clean SOX 404 reports are not able to convince an audit team to lower their fees (Munsif, et al., 2011). These results show the negative consequences of low internal control quality. A risk-averse and ethical CEO might want to avoid these consequences and therefore they would devote more resources into the internal control system. This would lead to a lower likelihood financial restatements. Therefore my second hypothesis is:
H2: Firms with older CEOs have higher internal control quality than firms with younger
19
4. The research models.
This chapter will explain the two research models. The first section will explain the research model for the relation between CEO age and audit fees. The second section will give a more detailed explanation of the variables used in the research model for the relation between CEO age and audit fees. The third section will explain the research model for the relation between CEO age and internal control quality. The fourth section will give a more detailed
explanantion of the variables used in the research model for the relation between CEO age and internal control quality. At last, the sample selection procedures are explained with data mutation tables.
4.1.1 The audit fee pricing model.
This study examines the association between CEO age and the pricing of an audit. To control for other variables which might affect audit fees, the following independent variables are included; total assets, non-audit services, auditor size, complexity of the audit, return on assets, debt over assets, and liquidity ratio. The independent variable which is particularly interesting is CEO age, which is included in terciles. The dependent variable will be the amount of audit fees paid by the auditee.
The pricing model I use was first developed by Simunic in 1980. His positive model of auditing shows the essential interdependence of the auditees and auditor’s economic interests. The model allows for auditor independence in the sense that the auditor implements a quantity of labor in the audit as a complement to the quantity of labor of the auditee in the internal control system. The auditee demands a positive quantity of labor of the auditor because external auditors have some advantages in certain aspects of control. Therefore the auditor is independent in the same sense that an supplier is independent in performing a task for a client.
Palmrose (1986) further developed Simunic’s positive model of auditing by examining the association between audit fees and auditor size. He was the first to show a significant association between auditor size and audit fees. As a proxy for auditor size he used a Big Eight/Non Big Eight dummy variable. The results suggest that a Big Eight audit firm has a positive association with audit fees. Jubb et al. (1996) have further developed this model and used it to examine the association between risk and audit fees. They argue that risk in the audit context is composed of two separate but related factor; audit risk and business risk. I will be using the model used by Jubb et al. (1996), however I include the independent variable
20
of interest (CEO age) and some control variables which are suggested by more recent research (Chan, et al., 1993; Hay, et al., 2006; Kinney & McDaniel, 1989; Palmrose, 1986).
EE 0 1O 2M E O S S E COM O 9 10 11 OSS
In which the variables are:
FEE The audit fees paid by auditee, measured as the log10 transformation of the total amount of audit fees for the fiscal year.
OLD The CEO age group for old CEOs, which equals 1 if the CEO is older than 58, else 0.
MIDDLE The CEO age group for middle-aged CEOs, which equals 1 if the CEO is older than 51 and younger than 59, else 0.
YOUNG The CEO age group for young CEOs, which equals 1 if the CEO is younger than 52, else 0.
AUD The auditor size dummy variable, which equals 1 if the auditor is a Big4 company, else 0.
NAS The non-audit fees paid by the auditee, measured as the log10 transformation of the total amount of non-audit fees for the fiscal year
SIZE The size of the auditee, measured as the log10 transformation of the total assets of the auditee.
COMP The complexity variable, measured as the inventory plus accounts receivable over total assets of the auditee.
ROA The return on assets ratio, measured as the net income over the total assets of the auditee.
DA The debt over assets ratio, measured as the total debt over total assets of the auditee.
LIQ The liquidity ratio, measured as current assets over current liabilities of the auditee. LOSS The net loss variable, which equals 1 if the auditee experienced a negative net
income (net loss) in the fiscal year, else 0.
Error term (assumed to have a normal distribution and constant variance).
4.1.2 Variables of the audit fee pricing model
The dependent variable, audit fees (FEE) charged to auditees, is measured by the log10 of the dollar value of audit fees paid by the firm to the auditor. Nowadays this dollar value must be disclosed by law. Fees are likely to vary in decreasing rate with increases in client size, therefore I use a log10 transformation (Jubb, et al., 1996).
21
investigate. It is measured as the age of the CEO in years. This data is widely available in the databases and audit reports. Following Yim (2013) I do not include age as an nominal value in the regression but I formulate age groups using dummy variables. Old CEOs are older than 58, middle-aged CEOs are 52 to 58 years of age, and young CEOs are younger than 52. The age group dummy variables are OLD, MIDDLE, and YOUNG respectively. I obtain the age of the CEOs from ExecuComp for a 5 year sample from fiscal year 2010 – 2014. In the research model I show the results for every age group as reference group.
The auditor size (AUD) is measured using a dummy variable, which is 1 if the audit is conducted by one of the Big4 auditors, or else 0. Palmrose (1986) found a significant positive association between auditor size and audit fees. The observed result by Palmrose is consistent with higher quality services provided by the big 4 firms in comparison to other audit firms.
The amount of non-audit services (NAS) is measured by the log10 of the dollar value of non-audit services purchased by the auditee from the auditor. Nowadays the dollar value of non-audit services have to be disclosed by law. Simunic (1980) and Simon (1985) found that the amount of non-audit services purchased have a spill-over effect to audit fees, potentially enlarging the effect on audit fee.
The client size (SIZE) is, consistent with other researchers, measured using the log10 of the book value of the auditees total assets. Although audit services may not increase
linearly with client size, clearly bigger clients will purchase more services than smaller clients (Palmrose, 1986).
The client complexity (COMP) is measured as the auditees inventory plus accounts receivable over total assets. Simunic (1980) shows that an audit team has to put in
significantly more effort into the testing and verifying of the posts inventory and accounts receivable. He argues that that these balance sheet items require specific audit procedures and even then may not be well audited because their valuation relies on future events. Further, if the auditee becomes more complex and its operations more dispersed, it may be harder for central management to maintain adequate control over inventory and debtors. Therefore, an increase in inventory and accounts receivable, as a portion of the total assets, increases the amount of effort by the audit team (Jubb, et al., 1996).
The variables return on assets (ROA), debt over assets (DA), liquidity ratio (LIQ) and
the net loss dummy (LOSS) are included to reflect the financial performance of the auditee. It
is argued that that these measures generally are associated with the potential, or actual level of financial distress. n auditor’s practice is vulnerable to damage through involvement with a particular auditee if the stakeholders in that auditee have suffered economic loss (Jubb, et al.,
22
1996). The stakeholders are more likely to suffer from economic loss if the auditee is in financial distress. Further, distressed firms are more likely to experience financial
restatements, which may lead to a greater likelihood of audit failure (Kinney & McDaniel, 1989).
4.1.2 The financial restatement model.
To test the second hypothesis I used the research model developed by Huang et al. (2012). This is a logistic regression model which investigates the association between CEO age and the internal control quality. The proxy used for internal control quality is a financial
restatement. According to the audit risk model, the auditor assesses the control risk of the firm by investigating the internal control system. A strong internal control system leads to low control risk and therefore a lower likelihood of financial restatements. The logistic regression model is slightly altered. I excluded variables related to the audit committee, board members and the firm’s industry. The audit committee and board member information are not available and the firm’s industry variable is not applicable to this study. I included variables related to the CEO age, audit fees and net loss. The CEO age group variables are included because these are the variables of interest. The audit fees variable is included to control for the effect of audit fees on financial restatements (Blankley, et al., 2012). The net loss variable is included to control for the effect of financial distress on financial restatements (Jubb, et al., 1996). As a proxy for auditee size, I used the log10 transformation of the total assets instead of the natural logarithm of the market value of shareholders equity following prior research (Chan, et al., 1993; Jubb, et al., 1996; Palmrose, 1986).
23
In which the variables are:
RESTA The financial restatement variable, which equals 1 if the firm experienced a financial restatement in the fiscal year, else 0.
OLD The CEO age group for old CEOs, which equals 1 if the CEO is older than 58, else 0.
MIDDLE The CEO age group for middle-aged CEOs, which equals 1 if the CEO is older than 51 and younger than 59, else 0.
YOUNG The CEO age group for young CEOs, which equals 1 if the CEO is younger than 52, else 0.
SIZE The size of the auditee, measured as the log10 transformation of the total assets of the auditee.
MB The market-to-book ratio, measured as the market value of shareholder's equity over the book value of shareholder's equity.
ROA The return on assets ratio, measured as the net income over the total assets of the auditee.
GROWTH The revenue growth ratio, measured as the percentage increase/decrease of the revenues from prior to current fiscal year.
AUD The auditor size dummy variable, which equals 1 if the auditor is a Big4 company, else 0.
FEE The audit fees paid by auditee, measured as the log10 transformation of the total amount of audit fees for the fiscal year.
LOSS The net loss variable, which equals 1 if the auditee experienced a negative net income (net loss) in the fiscal year, else 0.
Error term (assumed to have a normal distribution and constant variance).
4.2.2 Variables of the financial restatement model.
The dependent variable restatement (RESTA) is a dummy variable which is 1 if the firm experienced a financial restatement, or else 0. Financial restatements are material weaknesses which are found and corrected in the financial statements after publication. Most material weaknesses occur due to inadequate internal control systems (Huang, et al., 2012). Therefore I use financial restatements as a proxy for internal control quality. Restatement information is obtained from AuditAnalytics.
The age of the CEO is the independent variable which this research tends to
investigate. Just as in the audit fee pricing model, I made three CEO age groups. OLD is the old CEO age group, containing CEOs who are older than 58. MIDDLE is the middle CEO age group, containing CEOs in the age ranging from 52 to 58. YOUNG is the young CEO age group, containing CEOs who are younger than 52.
24
The control variables market-to-book ratio (MB) and growth ratio (GROWTH) are included to measure firm growth. The MB is calculated as the market value of shareholder’s equity over the book value of shareholder’s equity and OWTH is calculated as the
percentage of growth/decline in revenues from year-to-year. Huang et al. (2012) include these variables because a firm’s growth rate can negatively impact the ability of its internal control structure and accounting system to properly record and value the firm’s transactions.
The control variable audit fees charged to the auditee (FEE) is included to control for the effect of audit fees on restatements. Blankley et al. (2012) investigated the relation between audit fees and financial restatements. They found that abnormal audit fees are negatively associated with the likelihood of financial restatements.
The control variables SIZE, ROA, AUD and LOSS are previously defined in the audit fee pricing model.
4.3.1 Sample of the audit fee pricing model.
The sample used in this research will be based on archival data and needs to be selected in a way that unwanted factors do not interfere with the relationship between CEO age and audit fees. Therefore, the sample will contain North-American firms because these all use one similar accounting standard, the U.S. GAAP. Furthermore, the U.S. has a well developed legislation regarding the reporting of financial information. The sample will be based on observations from fiscal year 2010 to 2014 to examine relationship between CEO age and audit fees. I used the most recent five year sample to gain sufficient observations because the matching of databases lead to a decrease in observations, as shown in table 1. The first dataset included over 45.000 observations, however a large number of observations were deleted due to matching of the CompuStat dataset, the AuditAnalytics dataset and the ExecuComp
dataset. After matching, the observations with missing variables were deleted from the sample. The COMPUSTAT dataset included all financial information, the AuditAnalytics dataset included all audit related information, and the ExecuComp dataset included the information about the age of the CEO. At last, financial institutions were excluded from the sample using the SIC codes in range of 6000-6999. The balance sheets of financial institutions are significantly different than any other company included in the sample, which could have an adverse effect on the outcome of the research model. Further, some values were excluded
25
due to a high likelihood of error. The final sample contains 6.696 observations. Table 1 shows a detailed explanation of the data mutation. Chapter 6 shows the results of a sensitivity
analysis of the regression results using a winsorized sample. I winsorized the audit fee sample on the 2.5 and 97.5 percentiles, in total 335 observations of the dependent variable and
independent variables are winsorized. Table 1
Data mutation of CEO age on audit fees
Fiscal Year Observations
Matching Databases
Excluding Missing
Variables SIC Correction
2010 9205 1625 1620 1282 2011 9451 1667 1665 1325 2012 9509 1694 1688 1343 2013 9437 1744 1739 1381 2014 8973 1728 1726 1365 Total 46575 8458 8438 6696 Difference -38117 -20 -1742
4.3.2 Sample of the financial restatement model.
The sample of the restatement model is similar to the audit fee pricing model. Restatement information is gathered using the Audit Analytics database. Firm characteristics are gathered from the CompuStat database and CEO characteristics are gathered from the ExecuComp database. The sample contains U.S. based firms in a 5 year period from fiscal year 2010 to 2014. I used the most recent five year sample to gain sufficient financial statements with restatements, because only a few firms experience financial restatements each fiscal year. However to calculate growth control variables, financial information from 2009 is also gathered. The first dataset contains over 45,000 observations. The matching of databases excluded a large number of observations from the sample. Hereafter, the observations with missing financial information to compute control variables were excluded. At last, 535 observations of financial institutions (SIC 6000-6999) were excluded because the balance sheets of financial institutions differ from other industries. The final sample of the restatement model contains 5,645 observations. Chapter 6 shows the results of a sensitivity analysis of the regression results using a winsorized sample. I winsorized the financial restatement sample on
26
the 2.5 and 97.5 percentiles, in total 282 observations of the independent variables are winsorized.
Table 2
Data mutation of CEO age on restatements
Fiscal Year Observations
Matching Databases
Excluding Missing
Variables SIC correction
2010 8166 1632 1152 1017 2011 8758 1622 1210 1084 2012 9303 1614 1228 1132 2013 9664 1615 1279 1200 2014 9622 1615 1311 1212 Total 45513 8098 6180 5645 Difference -37415 -1918 -535
27
5. Results.
This chapter will explain the results of both research models. The first section will explain the descriptive statistics, Pearson correlation matrix, variation inflation analysis and regression results of the relation between CEO age and audit fees. The second section will explain the descriptive statistics, Pearson correlation matrix, variation inflation analysis and regression results of the relation between CEO age and financial restatements.
5.1 Results of the audit fee pricing model.
For the examination two research models are used. In this paragraph an overview is given of the descriptive statistics of the sample used for the audit fee pricing model. Table 3 shows an overview of all independent and the dependent variable used in this research model. This research model used group dummy variables for the age of the CEO, however included in the descriptive statistics is the actual age of the CEO to further clarify the groups made.
Table 3 N=6696
Mean Std. Deviation Minimum Maximum
FEE 6.2868 0.4523 4.3711 7.9533 AGE 55.1228 7.0707 27.0000 87.0000 OLD 0.2872 0.4525 0.0000 1.0000 MIDDLE 0.4101 0.4919 0.0000 1.0000 YOUNG 0.3027 0.4595 0.0000 1.0000 AUD 0.9050 0.2932 0.0000 1.0000 NAS 5.1037 1.5291 0.0000 7.8306 SIZE 9.3385 0.7346 6.1284 11.8758 COMP 0.2393 0.1651 0.0000 0.8784 ROA 0.0453 0.1314 -5.2016 1.0016 DA 0.5381 0.2636 0.0000 3.7928 LIQ 2.3820 2.2891 0.0000 69.2336 LOSS 0.1549 0.3618 0.0000 1.0000
Descriptive Statistics of CEO age on audit fees
The descriptive statistics in table 3 show the following results. The audit fees paid by the auditee (FEE) have a mean on 6.2868, which means that audit fees have a mean higher than a million dollars and this finding is similar to the results of Blankley et al. (2012). The
28
independent variable AGE represents the age of the CEO and the mean is 55.1228 years of age, this finding is similar to the results of Yim (2013). Following Yim (2013) the following age groups were formulated: OLD, MIDDLE and YOUNG. The dummy variable OLD has a mean of 0.2872, which means that 28.72% of the CEOs are included in the OLD age group. The dummy variable MIDDLE has a mean of 0.4101, which means that 41.01% of the CEOs are included in the MIDDLE age group. The dummy variable YOUNG has a mean of 0.3027, which means that 30.27% of the CEOs are included in the YOUNG group. In this research model, the age group dummy variables will all be used as reference groups, this allows for comparison between CEO age groups. The mean of the Big4 dummy variable (AUD) is 0.9050, which means that 90.50% of the observations was a client of a Big4 auditor. This sample contains more Big4 audit firms than non-Big4 audit firms, which is similar to the results of Chan et al. (1993) and Palmrose (1986). The amount of non-audit services provided by the auditor (NAS) has a mean of 5.1037, which indicates that most firms have also
purchased non-audit services when considering the standard deviation of 1.529. The mean of the auditee size (SIZE) of 9.3385 indicates that most companies have over a billion dollar in assets since the variable is transformed using the log10 value. The amount of inventory plus accounts receivable to total assets (COMP) has a mean of 0.2393, which indicates that the average amount of complexity in the audit was 23.93%. The findings from auditee size (SIZE) and complexity (COMP) are similar to the results of Blankley et al. (2012). The return on assets (ROA) has a mean of 0.0453, which indicates that most companies had a positive return on assets and this finding is similar to the results of Yim (2013). The debt over assets ratio (DA) indicates that on average the firms have about half of their assets in debt. The liquidity rate (LIQ) has a mean of 2.3820, which indicates that on average the firms have about two times the liquidity to pay their short-term liabilities. The leverage ratios DA and are similar to the findings of O’Keefe et al. (1994) and Raghunandan and Rama (2006). The loss dummy (LOSS), which indicates whether the firm made a net loss in the year, has a mean of 0.1549. On average about 15% of the audit reports in the sample contain firms with a net loss, which is lower than the results of Yim (2013). Overall, these data seem good enough to run the multiple regression.
Table 4 shows the Pearson correlation matrix for the variables of the audit fee pricing model. The correlation coefficients are shown including the significance values of the
coefficients. A coefficient is significant if their p-value is below 0.1. The following two-tailed significance levels are used; 0.1 for 10%, 0.05 for 5% and 0.01 for a 1% significance.
Table 4
N=6696FEE
OLD
MIDDLE YOUNG
AUD
NAS
SIZE
SIZE
ROA
DA
LIQ
LOSS
FEE
1.0000
OLD
-0.0062
1.0000
MIDDLE 0.0468*** -0.5292***1.0000
YOUNG -0.0440***-0.4182***-0.5494***1.0000
AUD
0.3869*** -0.0679***0.0309** 0.0338*** 1.0000
NAS
0.5022*** -0.0324** 0.0355*** -0.0060
0.2661*** 1.0000
SIZE
0.8282*** -0.0164
0.0688*** -0.0574***0.3699*** 0.4507*** 1.0000
COMP
-0.0529***0.0536*** 0.0178
-0.0717***-0.0868***-0.0568***-0.2011***1.0000
ROA
0.0617*** -0.0110
0.0001
0.0107
0.0580*** 0.0917*** 0.1358*** 0.0187
1.0000
DA
0.3081*** -0.0307** 0.0634*** -0.0377***0.1870*** 0.1660*** 0.3183*** -0.0186
-0.1457***1.0000
LIQ
-0.2570***0.0513*** -0.0432***-0.0043
-0.1544***-0.1555***-0.3220***0.0188
0.0259** -0.4184***1.0000
-0.1025***-0.0044
0.0140
0.0198
-0.1190***-0.1000***-0.2036***0.0044
-0.5346***0.1399*** 0.02395* 1.0000
*. Correlation is significant at the 0.1 level (2-tailed).
LOSS
***. Correlation is significant at the 0.01 level (2-tailed).
**. Correlation is significant at the 0.05 level (2-tailed).
Pearson Correlation Matrix of the audit fee pricing model
The coefficients in the Pearson correlation matrix show the underlying relation between the variables used. A correlation coefficient of 0.8 or higher might indicate a problem called multicollinearity. Multicollinearity is phenomenon in which two or more independent variables in a multiple regression model are highly correlated, meaning that one can be linearly predicted by the other with a high degree of accuracy (Gujarati, 2004).
A multiple regression model with correlated variables can indicate how well the entire bundle of variables predict the outcome variable, but it may not give valid results about any variable, or about which variables are redundant to another. Table 4 shows that audit fees and auditee size have a correlation coefficient of 0.8282, which is higher than the 0.8 threshold. This does not necessarily indicate multicollinearity but additional testing is needed. To examine if there exists a multicollinearity problem in the multiple regression a variation inflation factor (VIF) analysis will be performed on the regression formula.
The variable inflation factor (VIF) shows how the variance of an variable is inflated by Table 5 N=6696 VIF 1/VIF Intercept OLD 1.2205 0.8193 YOUNG 1.2285 0.8140 AUD 1.1945 0.8371 NAS 1.2783 0.7823 SIZE 1.6591 0.6027 COMP 1.0602 0.9432 ROA 1.4233 0.7026 DA 1.3592 0.7357 LIQ 1.2806 0.7809 LOSS 1.4740 0.6784 Mean 1.3178 0.7696
Variable inflation factor analysis on FEE Collinearity Statistics
31
the presence of multicollinearity. Table 5 shows the VIF and tolerance levels (1/VIF) of the variables in the regression model. As a rule of thumb, a VIF of 5 or higher or a tolerance level of 0.2 or lower indicates a multicollinearity problem. The correlation coefficient of audit fees and auditee size was 0.8282, however the coefficient has a VIF value of 1.6591 and a
tolerance level of 0.6027. These values indicate no issues regarding multicollinearity. The correlation coefficient of audit fees and auditee size is similar to the findings of prior research (Blankley, et al., 2012; Chan, et al., 1993; Jubb, et al., 1996). The other VIF and tolerance levels further indicate no issues regarding multicollinearity, with the mean VIF being 1.3178. The correlation matrix shows that there exists high correlation coefficient between non-audit services (NAS) and audit fees (FEE). This indicates that a firm who purchases audit services is likely to also purchase non-audit services and this finding is similar to the results of Jubb et al. (1996). The correlation coefficient of audit fees (FEE) and complexity (COMP) is -0.0529, this finding is lower than the results of Blankley et al. (2012) and Jubb et al. (1996). The correlation coefficient of auditee size (SIZE) and non-audit services (NAS) is relatively high, indicating that larger firms are more likely to purchase non-audit services. The correlation coefficient of auditee size (SIZE) and non-audit services (NAS) is similar to the results of Jubb et al. (1996) Further, return on assets (ROA) is highly negatively correlated with the loss dummy (LOSS), which is expected because both variables are derived using net income. Debt over assets (DA) has a relatively high correlation with the liquidity ratio (LIQ), which is expected because both variables are derived from debts and assets. The correlation coefficient of auditor size (AUD) and auditee size (SIZE) has a relatively high coefficient, this finding is similar to the results of Jubb et al. (1996) and Raghunandan and Rama (2006)
In this section the results of table 6 are explained. Table 6 shows the regression analysis of the independent variables on the dependent variable audit fees. The adjusted R-squared of the model is 72.84%. This indicates that the model explains a large part of the variability of the response data around the mean. This means that the model has great
explanatory power. Further, the Durbin-Watson statistic is 1.8945, indicating that there is no statistic evidence that the error terms are auto-correlated.
The coefficient needs to be significant to draw conclusions (with P-value<0.1). The regression analysis shows a significant positive association between the old CEO age group and audit fees with a coefficient of 0.0138 and a p-value lower than 0.05. This result indicates that in comparison to the reference group (middle-aged CEO group) old CEOs pay
significantly more audit fees than middle-aged CEOs. The regression analysis also shows a significant positive association between the young CEO age group and audit fees, with a
32
coefficient of 0.0150 and a p-value lower than 0.05. This results indicates that in comparison to the reference group (middle-aged CEO group) young CEOs pay significantly more audit fees that middle-aged CEOs. When using a reference group in a multiple regression model, the other group variables (young and old CEOs) can also be compared to each other. The coefficient of old CEOs is lower than the coefficient of the young CEOs (0.0138<0.0150). To compare the young CEO age group with the old CEO age group I also ran the regression with the young and the old CEO age groups as reference group. The old CEO age group has a coefficient of -0.0011 with the young CEO age group as reference group.
Table 6
Regression Results CEO age on audit fees
EE 0 1O 2 O 2 S S E COM O 9 OSS Adjusted R-squared = 0.7284 Durbin-Watson statistic = 1.8945 YOUNG as reference group MIDDLE as reference group OLD as reference group Intercept 1.3599*** 1.3450*** 1.3588*** OLD -0.0011 0.0138** -MIDDLE -0.0150** - -0.0138** YOUNG - 0.0150** 0.0011 AUD 0.0723*** 0.0723*** 0.0723*** NAS 0.0450*** 0.0450*** 0.0450*** SIZE 0.4829*** 0.4829*** 0.4829*** COMP 0.3235*** 0.3235*** 0.3235*** ROA -0.1046*** -0.1046*** -0.1046*** DA 0.0519*** 0.0519*** 0.0519*** LIQ 0.0071*** 0.0071*** 0.0071*** LOSS 0.0698*** 0.0698*** 0.0698***
1. Dependent Variable: FEE
***. Correlation is significant at the 0.01 level (2-tailed). **. Correlation is significant at the 0.05 level (2-tailed). *. Correlation is significant at the 0.1 level (2-tailed).
This indicates that in this sample old CEOs experience lower audit fees than young CEOs, however the coefficient is not significant. The middle CEO age group has a coefficient of -0,0150 and a p-value lower than 0.05. When taking the old CEO age group as reference group, the coefficient of the young CEO age group is 0.0011, however the coefficient is not
33
significant. The middle CEO age group has a coefficient of -0,0138 and is significant with a p-value lower than 0.05. These results indicate that there is no significant difference between the young CEO age group and the old CEO age group. However, these results also indicate that middle-aged CEO group experience lower audit fees than both young and old CEOs.
The control variable auditor size (AUD), which is a Big4 dummy, has a coefficient of 0.0723 and a p-value lower than 0.01. This result confirms the prior examination of Palmrose (1986), that auditor size is positively associated with audit firms due to higher quality audit services. The control variable non-audit services (NAS) has a coefficient of 0.0450 and a p-value lower than 0.01. This result indicates that firms who purchase non-audit services are likely to experience higher audit fees. Further, this results confirms prior studies of Simunic (1980) and Simon (1985) that non-audit services have a spill-over effect on audit fees. The control variable audit complexity (COMP) has a coefficient of 0.3235 and a p-value lower than 0.01. This result indicates that an increase in complexity will result in an increase in audit fees. The positive association between complexity and audit fees confirms a prior study by Chan et al (1993) that an increase in inventory and accounts receivable as a portion of total assets will increase audit fees.
The financial performance ratios used as control variables are all significant at the 1% level (p-value<0.01). Return on assets (ROA) has coefficient of -0.1046 indicating that an increase in profitability results in lower audit fees. The net loss dummy (LOSS) has a
coefficient of 0.0698, indicating that an negative net income results in higher audit fees. These variables both verify the results of Simunic (1980) that a firm in financial distress experiences larger audit fees. Debt over assets (DA) and the liquidity ratio (LIQ) are both positively associated to audit fees, with coefficients of 0.0519 and 0.0071 respectively, and significant at the 1% level.
