What is the influence of Enterprise Risk Management implementation on audit fees and how is this contingent to firm characteristics?

What is the influence of Enterprise Risk

Management implementation on audit

fees and how is this contingent to firm

characteristics?

By

ARJAN KLAASSEN

Drs. S. Sibum, supervisor

Thesis in Accountancy & Controlling, for the requirements of the degree of Master of Science

UNIVERSITY OF GRONINGEN Faculty of Economics & Business

What is the influence of Enterprise Risk

Management implementation on audit

fees and how is this contingent to firm

characteristics?

Name Arjan Klaassen

Date of birth 12/03/1991

Student number 2584697

E-mail Arjan.klaassen.3@student.rug.nl

E-mail (alternative) Arjanklaassen91@hotmail.com

Address Koeriersterweg 23a, Groningen

Postal code 9727 AC

Phone 06-22892944

Study program Accountancy & Controlling

Supervisor Drs. S. Sibum

Abstract

Recent developments in the field of enterprise risk management make it an interesting area to research, since a lot of firms are now forced either by regulatory instances or by the market to implement ERM in their firm and also disclose information about it. This results in a lot of new information available for the public, which results in a lot of new research opportunities. In this thesis the implementation of ERM is measured by using a proxy, being the announcement of a Chief Risk Officer position being created. Based on the Agency theory a trade-off among internal control and external assurance can be expected, therefore the first hypothesis is there is a negative relation between the implementation of ERM as measured by the appointment of a CRO and the external audit fees. This hypothesis was tested using a regression analysis, which found that there is a negative relation between CRO position creation and audit fees. But the results of this analysis were not significant, so the hypothesis could not be confirmed. Previous research on ERM has also stressed the importance of the contingency theory in studying the effects of ERM, therefore I also tested if the relationship between CRO position creation and the audit fee is contingent on the firm size, financial leverage and earnings volatility of the company. This was tested using a moderated regression analysis, some small effects were found, but results were not significant and thus didn’t allow me to confirm the hypothesis. The lack of statistical significance is most likely caused by the small sample in this research. Future research will have more years to include in the research and should try to redo the research with a larger and more diverse sample to see if this finds statistically significant results.

Table of Contents

Abstract ... 2

1.0 Introduction ... 4

2.0 Theoretical framework and Hypotheses development ... 5

2.1 Enterprise Risk Management ... 5

2.2 Audit fees ... 6

2.3 Chief Risk Officer ... 6

2.4 Agency theory and hypothesis ... 7

2.5 Contingency theory and hypothesis ... 7

2.5.1 Firm size ... 8 2.5.2 Earnings volatility ... 8 2.5.3 Financial leverage ... 9 3.0 Methodology ... 10 3.1 Sample ... 10 3.2 Variables ... 13 3.2.1 Dependent variable ... 13 3.2.2 Independent variable ... 13 3.2.3 Moderating variables ... 13 3.2.4 Control variables ... 14 3.3 Descriptive statistics ... 16

3.4 Moderated multiple regression analysis... 18

3.5 Testing the assumptions for MMR ... 19

4.0 Results ... 21

5.0 Conclusion ... 23

5.1 Discussion ... 23

5.2 limitations and future research ... 24

1.0 Introduction

Recent failures in governance have resulted in the passage of new rules and guidelines, some examples of corporate failures are Société Générale, Enron and Worldcomm. These corporate failures have made board members, executives and regulatory instances more aware of the consequences of ineffective risk management. Different countries reacted with different rules and guidelines to the failures. In the United States the Sarbanes Oxley Act (SOX) was enacted on July 30, 2002 and the Auditing Standard No. 5 was issued by the PCAOB in 2007 which placed increasing scrutiny on top-down risk assessment and included a requirement to perform a fraud risk assessment (PCAOB, 2007-005a). The Basel II Capital Accord and the revised combined code (2003) in the UK sought to minimize the risk of future major corporate failures via tighter regulation of internal control systems (Woods, 2008). The New York Stock Exchange(NYSE) changed its requirements for NYSE registrants audit committees, the requirements state that the audit committee must discuss guidelines and policies to govern the process by which risk assessment and management is undertaken (NYSE, 2003). Rating agencies are also giving increasing attention to Enterprise Risk Management (ERM) practices within firms and are incorporating these into their credit-rating analyses not only for financial firms, but also for nonfinancial firms as of 2008 (Standard & Poor’s, 2008). These items all show an increasing demand for effective control of the ever-changing portfolio of risks facing the enterprise.

The aforementioned developments make enterprise risk management an interesting area for research, since a lot of firms are now forced either by regulatory instances or by the market to implement ERM in their firm and also disclose information about it. This results in a lot of new information available for the public, which result in a lot of new research opportunities.

Corporate governance from an agency theory perspective is clear in indicating that adequate monitoring or control mechanism need to be present to protect shareholders from management’s conflict of interest (Fama and Jensen, 1983). The external auditor attests that the financial statements are conform to contractual commitments and that all shareholders are treated equally. Felix et al (2001) found that internal audit contribution is a significant determinant of the external audit fee. It is therefore

reasonable to expect that the implementation of ERM would influence the audit fees. It is argued that the effectiveness of an ERM system is contingent on firm-specific factors (COSO, 2004; Gorden et al., 2009; Woods, 2008). Following their reasoning I take into consideration that the relation between ERM implementation and audit fees can be contingent on firm-specific characteristics. The main research question will therefore be: What is the influence of ERM implementation on audit fees and how is this contingent to firm characteristics?

This study is structured as follows. First there is a brief summary of the literature regarding enterprise risk management and audit fees. Then I motivate the variables I used in this study and describe the data and method used to test my hypothesis. In the fourth chapter I report the results and in the last chapter I conclude with a discussion of the found results and the limitations of the research.

2.0 Theoretical framework and Hypotheses development

In the theoretical framework the theory around audit fees and enterprise risk management are reviewed. First I will look into previous research in the area of enterprise risk management, audit fees and the chief risk officer. The last two paragraphs will focus on the agency theory and contingency theory and the hypothesis I have made based on these theories.

2.1 Enterprise Risk Management

Enterprise risk management (ERM) is moving up in the agenda of practitioners, academics and the government. Gordon et al. (2009) state that managing risks is a fundamental concern in today’s dynamic global environment and that in recent years a paradigm shift has occurred regarding the way to view risk management. Instead of looking at risk management from a silo-based perspective, the trend is to take a holistic view of risk management commonly referred to as ERM. Gordon found that the relation between ERM and firm performance is contingent upon the appropriate match between ERM and the following five factors affecting a firm: environmental uncertainty, industry competition, firm size, firm complexity, and board of directors monitoring.

There has been some research to measure the effect of implementation of ERM on the capital market. Beasley et al. (2008) found no aggregate significant market reaction to the hiring of Chief Risk Officers (CRO) for either the financial service or nonfinancial service firms. But found that shareholders value ERM in cases in which ERM can enhance value by overcoming market distortions or agency cost. They also found certain characteristics that influence the perceived value of ERM, being volatile earnings, low amounts of leverage, and low amounts of cash on hand. Pagach & Warr (2007) found that firms that are more levered, have more volatile earnings and have poorer stock market performance are more likely to initiate an ERM program.

Beasley, Clune & Hermanson (2005) found that the stage of ERM implementation is positively related to the presence of a chief risk officer, board independence, Chief Executive Officer (CEO) and Chief

Financial Officer (CFO) apparent support for ERM, the presence of a Big Four auditor, entity size, and entities in the banking, education, and insurance industries.

Previous studies as stated above have shown that implementation of ERM can have significant effect on operational performance, firm value and firm performance. But there has not been a research yet that looks into the influence of Enterprise risk management on audit fees. This research is one of the first to test for this relationship. There is one paper by Desender & Lafuente (2011) that looks into the

relationship between ERM and external audit fees in the pharmaceutical industry. Their research is pretty limited in scope as it only covers one year and one industry and doesn’t take into consideration the contingency perspective. In this research these limitations will be resolved, this will become clearer in the methodology section.

2.2 Audit fees

Audit services are demanded as monitoring mechanisms because of the potential conflicts of interest between managers and owners and also among different classes of security holders (Watts, 1977; Watts and Zimmerman, 1981; Benston, 1985). In some cases the external audit of the financial statements is the least-cost contractual response to owner-manager and intra-owner conflicts of interest also known as agency costs. Agency costs usually vary across firms and time for a given client. For example it is know that firms going public often switch to one of the big auditors (Carpenter and Strawser (1971).

Differential agency costs across firms and over time for a given firm imply a heterogeneous demand for audit services, so it can be expected that differing levels of auditing are demanded (DeAngelo, 1981). The level of auditing demanded or the auditor effort can be measured using the height of the audit fees (Hope, Langli, & Thomas, 2012)

Prawitt et al. (2011) find in their research that internal audit function hours spent directly assisting external auditors reduce the audit fees. Abbott et al. (2012a) found the same results as Prawitt, and that external auditors fee reduction is magnified as the relative amount of influence over the internal audit function shifts to the audit committee (as opposed to management) and as internal audit function funding increases.

2.3 Chief Risk Officer

Although the concept of ERM is pretty straightforward, its implementation can be challenging. For example, determining what the appropriate leadership structure is to manage the identification, assessment, measurement, and response to the ever-changing risks facing the firm can be tricky. Nocco et al. (2006) state that for ERM to be successful all levels of the organization have to understand how ERM can create value and that ERM is a critical tool for executing the firm’s strategy. A ERM

implementation requires strong support from the senior management level to succeed, because the scope and the impact on the firm (Walker, Shenkir & Barton, 2002). Beasley, Clune and Hermanson (2005) found that the tone at the top is crucial for the implementation of ERM. COSO (2011) is in line with walker et al. (2002) and Beasley et al. (2005) in that ERM must be understood and embraced by its personnel, and driven by the board and senior management in clear and consistent communication and messaging. COSO is also clear in stating that there should be an effective ERM leader in place who has accepted “ownership” for the overall ERM implementation, this actor is responsible for overall ERM leadership, resources and support to accomplish the ERM effort. The senior executive leadership over ERM helps to communicate and integrate the firm’s risk strategy and philosophy for ERM consistently throughout the firm (Beasley et al. 2009).

Many organization are responsive to the statements above and are appointing a member of the senior executive team to assume “ownership” over the risk management process, this actor is often given the title of Chief Risk Officer, henceforth referred to as CRO in this paper (Economist intelligence unit, 2005). A main roadblock to the research of ERM is the difficulty in developing a valid and reliable measure for ERM (McShane et al., 2011). Gordon, Loeb and Tseng (2009) have done a good attempt at developing a ERM index, but their index fails to measure the implementation. Their index determines if ERM is implanted by measuring the effects of ERM, which is not a valid and reliable measure to determine if ERM is implemented. Previous research has found that the appointment of a CRO is associated to the implementation of ERM in an organization (Beasley, Clune and Hermanson, 2005; Desender & Lafuente

2011; Hoyt & Liebenberg, 2003). Currently a lot of research concerning ERM is using the appointment of a CRO as a proxy for the implementation of ERM in a firm (Beasley, Pagach, and Warr, 2008; Pagach & Warr, 2007). The use of CRO as proxy for adoption of ERM by a firm has a few drawbacks. The CRO appointment can be a replacement of an existing CRO or a CRO appointment may be little more than a title change to better reflect the manager’s responsibility. Both these drawbacks add noise to findings and will bias the research findings towards the null of finding no effect. Another drawback noted by Beasley et al. (2008) is that the appointment of a CRO doesn’t capture the extent of the ERM implementation. In this research the appointment of a CRO will be used as a proxy for ERM

implementation, although the proxy has a few drawbacks it is currently the best quantifiable measure available and commonly used in the research area. I do make a few improvements to the measure, which will be discussed in the methodology section

2.4 Agency theory and hypothesis

The agency theory proposes a series of mechanisms that seek to reconcile the interests of shareholders and managers, including the utilization of internal control mechanisms such as monitoring by large shareholders (Shleifer and Vishny, 1986), monitoring by non-executive directors (Fama and Jensen, 1983), the incentive effects of executive share ownership (Jensen and Meckling, 1976), and the

implementation of internal controls (Matsumura and Tucker, 1992). The statutory audit is an additional instrument of shareholders monitoring, whereby independent auditors report annually to shareholders on the appropriateness of the financial statements prepared by management (Watts and Zimmerman, 1983). The literature on agency theory suggests that some control mechanisms may be substituted by other control mechanisms, so that there could be a trade-off among various sources of control available to individual stakeholders, including external assurance (Jensen and Meckling, 1976). Therefore it can be expected that when ERM is implemented in an organization (implementation of internal control), that this will result in a trade-off with the statutory audit as source of control, thus leading to lower audit fees.

Hypothesis 1: There is a negative relation between the implementation of ERM as measured by the appointment of a CRO and the external audit fees.

2.5 Contingency theory and hypothesis

The Committee of Sponsoring Organizations (COSO, 2004) recognized in their ERM framework that the appropriate ERM system will likely vary from firm to firm and in essence suggest a contingency

perspective toward the appropriate ERM system for a particular organization. The relationship between the dependent and independent variable in the contingency theory is dependent on whether or not a third variable is present and in which quantity (Venkatraman, 1989). The contingency theory essentially states that efficient organization structures vary with organizational contextual factors such as

technology and environment. It further implies that the efficacy of certain managerial techniques for example Enterprise risk Management is contingent on the organization’s context and structure. So when measuring the effect of managerial techniques like ERM, it is good to take into account key factors in the relation.

There is no general theoretical framework or model available that can predict the key factors influencing the relation between a firm’s ERM and the audit fees. Although there are some firm-specific factors identified in similar ERM research, mostly concerning the relation between ERM and firm performance and factors concerned with the extent of implementation of ERM in a firm and the equity market reaction. But these factors are not readily usable in my research. Therefore based on prior research (Beasley et al., 2008; Mcshane et al., 2011; Gordon et al., 2009; Anderson et al. 1994; Hay et al, 2004) I identified the following contingency factors: firm size, financial leverage of the firm and earnings volatility. In the next section an explanation follows for the selection of these contingency factors.

2.5.1 Firm size

COSO (2004) states that ERM implementation may depend on the size and level of complexity of the institution. Smaller and mid-sized firms may apply less formal and less structured ERM than larger firms. Previous research by (Beasley, 2008) has found that larger firms benefit more from ERM than smaller firms, where the implementation of ERM was measured using the appointment of a CRO. Larger firms should be more capable of capturing economies of scale, which helps them implement ERM less costly. Based on previous research firm size seems to be a key factor when studying audit fees(simunic,1980; Craswell et al., 1995; Carcello et al., 2002). The relation between firm size and audit fees is very logical, since auditor’s fees are paid according to the amount of time spent on an audit. Larger companies are involved in a greater number of transactions which require more hours for an auditor to audit. The relation between company size and audit fees is confirmed in the meta-analysis by Hay et al. (2004), they confirm it as a well-established control variable for audit fees. So company size plays a role in determining the effect of ERM and the height of the audit fees. Following this it is not strange to expect company size to enhance the negative relation between ERM implementation and audit fees. Following this I anticipate that the ERM and audit fees relation will be dependent on the size of a firm. Therefore I hypothesize that:

Hypothesis 2: The negative relation between the implementation of ERM and the external audit fees is moderated by firm size, such that in larger firms the implementation of ERM has a stronger negative effect than in smaller firms.

2.5.2 Earnings volatility

One of the main goals of ERM is to reduce the likelihood that multiple negative events occur

simultaneously, in other words the avoidance of downside risk or lower-tail outcomes (Stulz 1996, 2003; Pagach & Warr, 2007). Previous research states that firms with a history of volatile earnings are more likely to benefit from ERM implementation. The reason for this is that firms with larger earnings volatility have a greater chance of seeing a lower-tail earnings outcome. Also firms with larger earnings volatility have a greater chance of missing analysts’ earnings forecasts and violating accounting based debt covenants (Bartov, 1993). These can lead to financial distress or bankruptcy of a firm. Simunic (1980) has found that auditors raise their audit fee in response to risks and financial distress, because the auditor becomes exposed to loss if the client is not financially viable and fails. In general, the more insecurity there is the more risk the auditor has, which leads to higher audit fees. It can be expected that when earnings are volatile in a company and it implements ERM, that this will have a moderating effect on the relation between ERM and audit fees. Therefore I expect that the relation between the

Hypothesis 3: The negative relation between the implementation of ERM and the external audit fees is moderated by earnings volatility, such that in firms with high earnings volatility the implementation of ERM has a stronger negative effect than in firms with low earnings volatility.

2.5.3 Financial leverage

As stated above one of the main goals of ERM is to reduce or eliminate the likelihood of costly lower-tail outcomes. High financial leverage increases the likelihood of financial distress, so in a firm with high leverage a downside risk or lower-tail outcomes have an increased chance to lead to financial distress or bankruptcy (Pagach & Warr, 2007). Leverage measures the risk of an auditor’s client failing, which exposes the auditor to the possibility of a loss (Simunic, 1980). Leverage also proxies for the agency costs between a company and its outside debtholders, an increase in leverage can therefore lead to an increase in audit fees to provide more security to the outside debtholders (Watts and Zimmerman, 1986). Therefore it can be expected that ERM implementation in firms with high leverage will have more influence on the audit fees than firms with low leverage. Following this I hypothesize that:

Hypothesis 4: The negative relation between the implementation of ERM and the external audit fees is moderated by financial leverage, such that in firms with high financial leverage the implementation of ERM has a stronger negative effect than in firms with low financial leverage.

The next chapter will cover the methodology used in this research to test the hypothesis that were formed in this chapter.

3.0 Methodology

This chapter will focus on the research design. This chapter starts with a description of the research sample and the research period. Followed by the discussion of the dependent, independent, control and moderating variables used in this research. The third paragraph discusses the descriptive statistics which contains information about the dataset used and the Pearson correlation. Following this the used statistical method moderated multiple regression is explained. The chapter closes with the testing of the assumptions for the moderated multiple regression.

3.1 Sample

The research sample consist out of American organizations who created the position of CRO in their organization. Following the approach of Hoyt & Liebenberg (2011), Pagach & Warr(2007) and Gordon et al.(2009) text based search of announcements and news articles will be used. The method used in this research was that of Pagach & Warr (2007) with a small adjustment made to their approach. Pagach & Warr (2007) identified firms appointing a CRO by searching for announcements containing the words “announced”, “named”, or “appointed”, in conjunction with position descriptions of “Chief Risk Officer” and “Risk management”. The problem with this is that using the terms “announced”, “named” and “appointed” leads to a lot of results which are not relevant for this study. An appointed or named CRO can be a replacement of an old CRO or it can be a title change that more accurately reflects the actor’s responsibilities. Both cannot be relied upon as proxies for first-time ERM adoption and will therefore add noise to the sample and will bias the tests towards the null of finding no effect. Only the creation of the CRO position for the first time can be reasonably relied upon as a signal of first-time ERM adoption. Therefore to try and mitigate the noise from previous mentioned method, the following search terms were used in this research to find the creation of a new position: “newly”, “created”, “create”, “build”, “constructed”, “designed”, “devised”, “established”, “formed”, “founded”, “generated”, “initiated”, “invented”, “organized”, “produced”, “set-upped”, “sets up”, “shaped”, “first” or “first-ever” in conjunction with “chief risk officer” or “CRO”. But even this way an observation can still be little more than a title change that more accurately reflects the actor’s responsibilities and this will add noise to the sample and reduce the power of the tests. The search engines used were Dow Jones Factiva,

Compliance Week and Lexis-Nexis which were set to search only for announcements and news articles originating from American sources.

The search was done for the years 2000-2014 and was focused on organizations in America. The reason for this time period was that the New York Stock Exchange (NYSE) changed its requirements for NYSE registrants audit committees in 2003. The new requirements state that the audit committee must discuss guidelines and policies to govern the process by which risk assessment and management is undertaken (NYSE, 2003), which should lead to organizations appointing a CRO. Also in the time-period of 2006-2010 organizations saw the added value of CRO’s in their organization and were scrambling to add chief risk officers to their top management ranks. Liberum research found that 25 individuals at public companies were hired as CRO in 2007, which was a 25% increase over the previous year and that in 2008 it was on track to outpace 2007 by 140%. Therefore the time period 2000-2014 should net enough observations for this research.

Each search “hit” was manually inspected to determine that it was a creation of the CRO position in the organization. If it turned out not to be a first time creation of a CRO position, then it was excluded from the sample. This resulted in a total sample of 126 organizations that had created the position of CRO in 2000-2013. There were also a four CRO position creations in 2014 but due to the lack of financial information for the year 2014 these were dropped. Upon further inspection of the sample several organization had to be removed from the sample for lacking the needed financial information. Some examples are: not being audited at all or being audited by state auditors; private companies that did not disclose any financial information publicly; daughters of a parent company in the sample; or companies that were not subject to American regulation. This resulted in a final sample of 91 companies, which translates to 182 observations. Table 3.1 shows the main industry of the company based on Standard Industry Classification (SIC) code and graph 3.1 shows in what year the position was created.

Table 3.1 Industry overview of selected companies

Industry Number of companies

01. Agriculture, Forestry & Fishing 0

10. Mining 1

15. Construction 0

20. Manufacturing 6

40. Transportation & Public Utilities 7

50. Wholesale Trade 1

52. Retail Trade 3

60. Finance, Insurance, Real Estate 69

70. Services 4

91. Public Administration 0

Total 182

As can be seen in table 3.1 most positions were created in the Finance, Insurance and Real Estate. The second in line is Transportation & Public Utilities with 7 creations. This is in line with predictions by Liebenberg & Hoyt (2003) that CRO appointments are most prevalent among less transparent firms such as those in the financial services and utilities industry. McShane et al. (2011) also notes that financial institutions have been leaders in implementing ERM. The composition of the sample in this research seems to confirm these claims. Companies in the finance, insurance and real estate industry usually have relatively large assets, but are generally easier to audit than companies with extensive receivables, inventory or knowledge-based assets (hay et al., 2004). Because a large part of the sample consist out of companies from this industry and for the above reason a control variable is introduced for this industry which will be further discussed in the paragraph 4.2.4 Control variables.

Graph 3.1 Creation of CRO position per year

Looking at graph 3.1 you can see that most creations were in the year 2004, this is most likely caused by the changing of the requirements by the NYSE for NYSE registrants audit committees in 2003. This made audit committees obliged to discuss guidelines and policies to govern the process by which risk

management and assessment is undertaken (NYSE, 2003). As expected this made companies create the position of CRO in their organization. The increase in creations after 2007 can be explained by the passing of Auditing Standard No. 5 by the PCAOB in 2007, which placed increasing scrutiny on top-down risk assessment and included a requirement to perform a fraud risk assessment (PCAOB, 2007-005a). Also in the time-period of 2006-2010 organizations began to see the added value of CRO’s in their organization and were scrambling to add chief risk officers to their top management ranks.

6 2 7 3 13 1 5 5 9 12 9 5 8 6 0 2 4 6 8 10 12 14 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Total

3.2 Variables

In this paragraph the dependent (DV), independent (IV), moderator (MV) and control variables (CV) are discussed.

3.2.1 Dependent variable

The dependent variable in this research are the audit fees (Fee). This is the fee paid by the organization to the independent auditor for audit services provided. Information about these audit fees can be obtained through the published definitive proxy statements (def 14a forms). These can be found in the US security and exchange commission EDGAR database and on the website of the organization. In the regression model the natural log of the audit fees is used, as prior audit fee research has found that the relation between audit fee measures and size measures are nonlinear (Gist 1992).

3.2.2 Independent variable

There is one independent variable in this research being Chief Risk Officer (CRO). Based on the creation of a CRO position the sample for this research was determined. So every organization in the sample has created a CRO position. For the regression analysis a reference or control group is needed to determine the effect of the CRO. In this research the year before the creation of the position is used as a reference group. The CRO position creation was measured using a dummy variable. In the year of the creation the organization got the value 1 and the year before the creation got the value 0.

3.2.3 Moderating variables

There are three moderator variables, which are expected to influence the direct relation between CRO appointment and audit fees. These are company size, earnings volatility and financial leverage

- Firm size (SIZE) There are multiple ways to measure firm size, you can use the number of

employees, the total sales of the firm, total assets of the firm and total market value of the equity. The two commonly used measures for firm size are total assets of the firm(McShane et al., 2011; Liebenberg & Hoyt, 2003; Gordon et al., 2009) and the total market value of the equity(Beasley et al., 2008). Shalit and Sankar (1977) state that there is no obviously ideal measure of firm size and that the choice of a measure depends on the purpose of the study. External auditors traditionally rely on the fact that verification of balance sheet components indirectly verifies reported income, so the general approach to the audit process is through the ending balance sheet (Simunic, 1980). Therefore the total assets of the firm seems like the most appropriate measure in my study. This is assumption is confirmed by the use of total assets as measure for firm size in previous audit fee research (Felix et al., 2001; Bedard & Johnstone, 2004; Hay et al., 2004). The information was extracted from the DataStream database and annual reports (10-K).

- Earning volatility I will apply the same measure for earnings volatility as Liebenberg & Hoyt

(2003), the measure is the coefficient of variation of EBIT for three years prior to CRO appointment. Liebenberg & Hoyt (2003) use the quarterly EBIT in their research, but this information was not readily available for all my companies so the yearly EBIT for years prior to the announcement was used. Yearly information to measure volatility will lack in strength compared to quarterly information, but is not uncommon to use in the research field (McShane et al., 2011). The coefficient of variation is used since it allows for comparisons across industries of different sizes, because the variance is standardized by the magnitude (Kren, 1992). The information was extracted from the DataStream database and annual reports (10-K). In the regression model the natural log of the total assets were used.

- Leverage Rajan and Zingales (1995) defined six measures of financial leverage and stated their pros and cons. In this research I use the measure total liabilities divided by total assets, this is the broadest definition of leverage and proxies for what is left for shareholders in case of liquidation which best suits my research. Also this measure is commonly used in ERM and audit fee research (Pagach & Warr, 2007; Felix et al., 2001; Desender & Lafuente, 2011, Hay et al., 2004). The con of this measure as stated by Rajan and Zingales (1995) is that it is not a good indication of whether the firm is at risk of default in the near future, because it doesn’t distinguish between short and long term items. The information was extracted from the DataStream database and annual reports (10-K).

3.2.4 Control variables

There are four control variables in this research being Auditor quality, Inherent risk, Complexity and industry. These control variables as measured as followed.

- Auditor quality (BIG4) Higher audit fees can be expected when an auditor is recognized as being

of superior quality (Hay et al., 2004). The most common used measure for this in previous audit fee research is a dummy variable for the audit firm being part of the big five or big four auditing firms according to the meta-analyses of audit fees done by Hay et al. (2004) . The big five became the big four by the demise of Arthur Anderson in 2002 following its involvement in the Enron scandal. The big four are Deloitte Touche Tohmatsu, PWC, Ernst & young and KPMG. Therefore one control variable for the audit fees is when the external audit is conducted by big 4/5 audit firm. As stated before this is done by using a dummy variable which takes the value 1 If the company is audited by a big 4/5 audit firm and 0 if audited by a non-big 4/5 audit firm.

- Inherent Risk As suggested by previous researchers some parts of the audit may have higher risk

of error and require specialized audit procedures (Simunic, 1980; Stice, 1991). The engagement can then be classified as having a higher inherent risk. The parts most often identified as being difficult to audit are inventory and receivables (Newton and Ashton, 1989; Simunic, 1980). Hay et al. 2004 found that inventories and receivables combined and divided by total assets is the most used proxy for inherent risks and also found the strongest significant positive results. As opposed to considering these accounts separately, therefore in this research the combination of these accounts is used as proxy to control for inherent risk.

- Complexity (SUBS) An audit can be expected to be harder and more time-consuming when a

client is complex. This expectation has often been mentioned by audit fee researchers

(Hackenbrack and Knechel, 1997; Simunic, 1980). Over the years the concept of complexity has been measured in a variety of ways, the most common measures are the number of SIC codes comprising the client ,the number of subsidiaries, the number of foreign subsidiaries, the number of business segments and the number of audit locations. The results for these

commonly-used complexity variables leave little doubt that the relationship between audit fees and complexity is positive and significant. The strongest results in the meta-analysis by hay et al. (2004) are for the number of subsidiaries measure, therefore we use this measure to control for complexity of the client. Companies have a legal obligation to report on the number of

subsidiaries in exhibit 21 of the 10-K form, so this information is readily available.

- Industry (SIC60) It is a common assertion among auditors and researchers that some industries

are less difficult to audit than others (Turpen, 1990; Simunic, 1980; Pearson and Trompeter 1994). As an example, financial institutions have relatively large assets, but are generally easier

to audit than companies with extensive inventory, receivables or knowledge based assets. Following previous research we control for companies that have finance, insurance and real estate (SIC-code 60) as main industry. The main industry SIC-code can be easily found through numerous sources, in this research the Compustat database was used.

Below is an overview of the used variables in this research with their expected effect on the dependent variable and the way it was measured.

Table 3.2 Overview of used variables

Variables Expected sign Measurement

Audit fees D.V. Natural logarithm of audit fees

CRO I.V. - Measured using a dummy variable 1 = CRO,

0 = No CRO

Firm Size M.V. + Natural logarithm of total asset at end of

the year

Earnings Volatility M.V. - Coefficient of variation of yearly EBIT over the previous 3 years

Financial leverage M.V. + Total liabilities divided by Total assets

Auditor quality(BIG4) C.V. + Measured using a dummy variable, which is coded one if a Big 4/5 company is the auditor and coded zero for non-big 4/5 Complexity (SUBS) C.V + Natural logarithm of the number of

subsidiaries

Inherent risk C.V. + Inventories and receivables divided by total assets

Industry (SIC60) C.V. - Dummy variables coded one when company

is in Finance, insurance and Real estate industry. Coded zero for other industries.

3.3 Descriptive statistics

Table 3.3 gives an overview of the descriptive statistics of the whole dataset. All variables except the dummy variables are winsorized to mitigate the impact of outliers in the dataset. Winsorizing means that values larger than three times the standard deviation above or below the average are set equal to three times the standard deviation above or below the average.

Table 3.3 Descriptive statistics

Variables N Average Std. Dev. Min Max

Dependent variable Ln Audit fees 182 8,142 1,464 4,710 11,555 Independent variables Dummy CRO 182 0,500 0,5014 0,000 1,000 Ln Firm Size 182 16,484 1,973 11,984 20,768 Volatility 182 0,325 1,856 -11,642 12,697 Financial leverage Control variables 182 0,794 0,195 0,180 1,204 Auditor quality 182 0,885 0,3204 0,000 1,000 Ln Subsidiaries 182 3,194 1,674 0,000 7,200 Ln Inherent risk 182 0,449 0,306 0,000 1,766

Dummy Finance Industry 182 55,341 12,661 10,000 70,000

The descriptive statistics show that the sample has a large variance in sizes of the companies as shown by the natural log of Firm size with a highest value of 20,768 and a lowest value of 11,984. This can also be seen in the audit fees. The variable financial leverage shows that there are both high leveraged companies (max 1,204) and low leveraged companies (min 0,180), which should help to see if this makes a large difference in the determination of the audit fees. It can also be observed by the mean of 0,885 that most companies in the sample are audited by a big four accounting company. This was to be expected since these four companies hold a large portion of the market share in the auditing market. The dummy Finance industry shows that, as stated before, that there are no companies from

Agriculture, forestry & Fishing (SIC 01) and Public administration (SIC 91) in the sample. Overall the descriptive statistics do not show anything unexpected.

The correlation table on the next page show the relationships between all the variables. This includes the dependent variable, the predictor variables, control variables and interaction variables. The Pearson correlation table is often used to determine if a sample suffers from multicollinearity. When two

variables have a correlation above 0,8 then multicollinearity exists(Blumberg, Cooper and Schindler, 2005). As you can see in the table the variables CRO and CRO x leverage (0,944), CRO and CRO x size (0,986), Volatility and CRO x volatility (0,816), CRO x firm size and CRO x leverage (0,951) all have values above the threshold and are statistically significant, indicating that the sample might suffer from multicollinearity. I deal with this problem in the chapter testing the assumptions of MMR, were I apply I method to reduce the multicollinearity in the sample.

Ta b le 3 .4 Pea rs o n c o rr ela ti o n ma tri x. 1 2 3 4 5 6 7 8 9 10 11 1. Au d it fe es 2 . Fi rm size 0 ,7 5 5** 3 . Fi n . L ev erag e 0 ,1 6 8* 0 ,4 0 ** 4 . Vo la tili ty 0 ,0 1 9 0 ,0 7 0 -0 ,8 9 5 . In h erent ri sk -0 ,7 2 6** -0 ,1 3 3 0 ,3 5 8** 0 ,0 1 7 6. Su b sid ia ri es 0,5 82 ** 0,3 59 ** -0,1 30 0,1 05 -0,1 67 * 7 . CRO -0 ,0 1 7 0 ,0 1 7 0 ,0 0 5 -0 ,0 5 7 -0 ,0 1 1 0 ,0 0 0 8 . Big 4 0 ,4 4 3** 0 ,4 2 8** -0 ,0 7 9 0 ,0 3 7 -0 ,1 8 1 * 0 ,3 1 7 ** -0 ,0 1 7 9 . In d u stry si c 60 -0 ,0 9 5 0 ,2 6 5** 0 ,6 0 3** -0 ,0 6 7 0 ,2 4 9 ** -0 ,2 5 7 ** 0 ,0 0 0 -0 ,0 4 3 1 0 . C RO x Fir m Siz e 0 ,0 7 3 0 ,1 3 2 0 ,0 4 9 -0 ,0 3 9 -0 ,0 2 6 0 ,0 4 1 0 ,9 8 6 ** 0 ,0 3 7 0 ,0 3 1 1 1 . C RO x le verag e 0 ,0 2 3 0 ,1 0 3 0 ,2 4 1** -0 ,0 6 8 0 ,0 7 0 -0 ,0 3 7 0 ,9 4 4 ** -0 ,0 3 5 0 ,1 4 2 0 ,9 5 1 ** 1 2 . C RO x v o latil it y 0 ,0 1 9 0 ,0 9 0 -0 ,0 3 6 0 ,8 1 6 ** 0 ,0 1 6 0 .07 0 0 ,0 7 2 0 ,0 1 6 -0 ,0 2 2 0 ,0 9 2 0 ,0 5 1

3.4 Moderated multiple regression analysis

The standard method of determining whether a moderating effect exists uses the addition of an interaction term in a multiple regression model. Therefore, this type of analysis is often referred to as a moderated multiple regression (MMR) (Aguinis, 2004). A moderator analysis is actually just a multiple regression equation with interaction terms. What makes it a moderator analysis is the theory and hypotheses that accompany this statistical test (Aguinis, 2004; Jaccard & Turrisi, 2003; Jose, 2013). In a moderation analysis the hypothesis is that the effect of an independent variable on a dependent variable depends on the value of a moderator variable. To illustrate this figure 4.1 shows the conceptual model for the hypotheses of this research.

In a moderated multiple regression you determine if a moderator effect exist by adding an interaction term to the multiple regression analysis. The interaction term is created by multiplying the independent variable with the moderator variable. By adding this term to the multiple regression analysis you create a moderated multiple regression. Figure 3.3 on the next page is an alternate presentation of the conceptual model shown above which includes the control variables and interaction terms.

An interaction term is symmetrical, so it can mean that either the first independent variable depends on the second independent variable or vice versa or both. Therefore as stated above the theory and hypothesis are key in making it a moderation analysis. The theory and hypothesis are clear in this research as to which independent variable is the moderator variable, thus making this a moderation analysis.

A moderated regression analysis uses two formulas, one to determine the additive effect and one to determine the additive and interaction effect. The first formula measures the additive effect of the independent and moderator variables on the dependent variable. The second formula measures both the additive and interaction by adding the interaction terms.

𝐴𝑢𝑑𝑖𝑡 𝑓𝑒𝑒𝑖 = 𝛽0+ 𝛽1𝐶𝑅𝑂𝑖+ 𝛽2𝑆𝑖𝑧𝑒𝑖+ 𝛽3𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖+ 𝛽4𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖+ 𝛽5𝑄𝑢𝑎𝑙𝑖𝑡𝑦𝑖+ 𝛽6𝑆𝑢𝑏𝑠𝑖𝑑𝑖𝑎𝑟𝑖𝑒𝑠𝑖+ 𝛽7𝐼𝑛ℎ𝑒𝑟𝑒𝑛𝑡 𝑅𝑖𝑠𝑘𝑖+ 𝛽7𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦𝑖+ 𝜀𝑖 (1) 𝐴𝑢𝑑𝑖𝑡 𝑓𝑒𝑒𝑖 = 𝛽0+ 𝛽1𝐶𝑅𝑂𝑖+ 𝛽2𝑆𝑖𝑧𝑒𝑖+ 𝛽3𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖+ 𝛽4𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖+ 𝛽5𝑄𝑢𝑎𝑙𝑖𝑡𝑦𝑖+ 𝛽6𝑆𝑢𝑏𝑠𝑖𝑑𝑖𝑎𝑟𝑖𝑒𝑠𝑖+ 𝛽7𝐼𝑛ℎ𝑒𝑟𝑒𝑛𝑡 𝑅𝑖𝑠𝑘𝑖+ 𝛽7𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦𝑖+ 𝛽8𝐶𝑅𝑂𝑖∗ 𝑆𝑖𝑧𝑒𝑖+ 𝛽9𝐶𝑅𝑂𝑖∗ 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒𝑖+ 𝛽10𝐶𝑅𝑂𝑖∗ 𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑖+ 𝜀𝑖 (2) Moderator variables: - Firm size - Earnings volatility - Financial leverage

+

- Chief Risk Officer

− Dependent variable: - Audit fees

Figure 3.3 Alternate presentation of the conceptual model

So the first formula measures the additive effect of the independent, moderator and control variables directly on audit fees. The second formula also measures this direct additive effect but adds the interaction terms to measure the interaction effect. By subtracting the values from formula one from formula two you are left with the interaction effect. This interaction effect is used to determine if a moderator effect exist, the interaction terms have to be tested for statistical significance. If they are statistically significant then a moderating effect exists.

3.5 Testing the assumptions for MMR

There are certain assumptions that must be met before conducting a moderated regression analysis. These are testing for linearity; testing for outliers, leverage points and influential cases; testing for homoscedasticity and testing for normality. In the part below I test if these assumptions are met and if not what was done to reduce or fix the problem

It is often recommended that the continuous independent variables are mean centered when performing a moderated regression analysis (West et al., 1991; Cohen et al., 2004). The decision to mean center is made based on the presence of multicollinearity. The testing of the presence of multicollinearity can be done in multiple ways, one way is the use of the Pearson correlation matrix. When two variables have a value greater than 0,8 you can say there is a presence of

multicollinearity(Blumberg, Cooper and Schindler, 2005). As can be seen in Table Pearson correlation the interaction variables show multicollinearity with their respective moderator variables. This was

Additive effect:

Interaction effect:

Dependent variable: - Audit fees

Independent variable: - Chief Risk Officer

Moderator variables: - Firm size - Earnings volatility - Financial leverage Control variables: - Auditor quality - Subsidiaries - Inherent Risk - Financial industry Interaction terms: - CRO x Firm Size

- CRO x Earnings volatility - CRO x Financial leverage

confirmed by running a moderated regression analysis and checking the variance inflation factor (VIF) and tolerance value. Evidence of multicollinearity is shown if the VIF is greater than 10 or a tolerance value lower than 0,1. The sample in this research showed a VIF higher than 10 for CRO (72,676), CROxsize(87,984) and CROxleverage (22,945). Therefore to reduce multicollinearity the independent variables and moderator variables were centered. The Interaction terms are not centered, since these are calculated using the new centered variables. After centering the independent and moderating variables and rerunning the test it seemed that the problem of multicollinearity was solved. Variance inflation factors and tolerance were all within acceptable limits, VIF’s are ranging from 1,005 to 3,592 and tolerance values were between 0,278 and 0,995. All correlations between variables in the Pearson correlation matrix also no longer showed a value greater than 0,8.

Linearity is another assumption that must be met before conducting a moderated regression analysis. A grouped scatterplot of the dependent variable against the moderator variable grouped by the

independent variable was used to check if a linear relationship existed. Linearity was confirmed by visual inspection of the grouped scatterplot.

To test for outliers I used the method provided by Cohen et al. (2003), which involves examining the studentized deleted residuals (SDR). When the SDR are greater than ± 3 standard deviations you can classify them as potential outliers. There was one observations which was greater than three times the standard deviation, overseas shipholding group had a SDR of -4,007. This turned out to be a data entry error for the variable audit fee which was entered as 327.000, but was actually 1.785.000. After fixing this error all values for SDR were between ± 3 standard deviations. I also tested for influential cases, which are observations that can alter the regression line. This was done using cook’s distance, a cook’s distance value greater than 1 could mean that there is an influential case (Cook & Weisberg, 1982). The highest value was 0,20318 which is below the threshold for influential cases, so there seems to be no influential cases in the sample.

Another important assumption of MMR is that the data needs to have constant variance better known as homoscedasticity. Homoscedasticity means that the dependent variables have the same variance in their errors. This assumption can be checked by plotting the studentized residuals against the predicted values (Kutner et al., 2005; Weisberg, 2014). There is homoscedasticity as assessed by visual inspection of the studentized residuals plotted against the predicted values for both groups (CRO and NO-CRO). The last assumption for regression is the assumption of normality (Weisberg, 2014). A variety of tests can be used to test this assumption, in this research the Shapiro-Wilk test for normality is used. The assumption of normality is met when the significance level is higher than p>0,05 in the Shapiro-Wilk test. Running the test in SPSS resulted in a significance level of 0,115 meaning that the assumption for normality is met.

Now that it is verified that all data meets the requirements of regression analysis, the moderated regression analysis can be run in SPSS. The result from this are shown in the next chapter.

4.0 Results

The results of the Moderated multiple regression are shown in table 4.2. The first model only includes the four control variables, in the second model the independent variable and moderator variables are added. The third model includes the control, independent and moderator variables, plus the interaction variables created to test the moderating effect. Model one has an R square of 0,458 and model two has an r square of 0,759 which shows that the models explain respectively 45,8% and 75,9% of the variance in the audit fees. Thus the models have relatively high predictive value for the dependent variable. Table 4.1 Results of moderated multiple regression with dependent variable Audit fees

Model 1 B SE Model 2 B SE Model 3 B SE Constant 6,910** (0,283) 8,553** (0,228) 8,560** (0,230) Control variables Auditor quality 1,180** (0,268) 0,331+ (0,198) 0,323 (0,200) Ln Subsidiaries 0,733** (0,088) 0,423** (0,064) 0,424** (0,065) Inherent risk -0,253** (0,085) -0,197** (0,062) -0,197** (0,062) Financial Industry 0,261 (0,199) -0,868** (0,165) -0,869** (0,166) Independent variable Dummy CRO -0,073 (0,109) -0,074 (0,109) Moderator variables Ln Firm Size 0,888** (0,078) 0,851** (0,098) Financial leverage 0,231** (0,080) 0,220* (0,104) Volatility -0,076 (0,055) -0,071 (0,098) Interaction variables

CRO x Firm Size 0,078 (0,120)

CRO x Leverage 0,023 (0,121) CRO x Volatility -0,012 (0,119) R Square 0,458 0,759 0,760 Adjusted R Square 0,446 0,747 0,744 R Square change 0,458** 0,300** 0,001 F change 37,438** 53,817** 0,225 N=182 + _p<0,1 * p<0,05 ** P<0,01

One interesting finding is that the control variable for inherent risk seems to have a negative effect on audit fees, while a positive effect was expected based on prior literature. This opposite effect could be caused by the large amount of financial institutions in the sample. This will be further discussed in the next chapter discussion, first the results for the tested hypothesis. The rest of the control variables have the expected effect on the audit fees and are mostly significant.

Hypothesis 1 tests whether the creation of a CRO position has a negative relation with the external audit fees. Model two is used to test this hypothesis. The model shows a coefficient of -0,073 so there seems to be a negative relationship, but it is not significant (p<0,501). The non-significance means that the found effect can be caused due to a sampling error, meaning that the sample is not representative for the population. Therefore hypothesis one cannot be confirmed.

Hypothesis 2, 3 and 4 test whether the relation between the creation of a CRO position and the height of audit fees is moderated by respectively firm size, financial leverage and earnings volatility. To test this a moderated multiple regression was run to assess the increase in variation explained by the addition of the interaction terms between CRO position creation and audit fee height. The results of this are found in model 3. The R Square change of 0,001 in model 3 means that the interaction terms model adds 0,1% increase in total variation explained, which is very low and not statistically significant (p<0,879).

There seems to be a small interaction effect between firm size and CRO position creation, as evidenced by B=0,078 but this effect is not statistically significant (P<0,517). So hypothesis 2 is not confirmed, there is no statistically significant interaction between firm size and CRO position creation. The same goes for hypothesis 3 and 4 there seems to be a small interaction effect between financial leverage and CRO position creation B=0,023 and earnings volatility and CRO position creation B= -0,012, but both are not statistically significant with values of p<0,851 and p<0,920 respectively. Therefore none of the three hypothesis testing for the moderating effect can be confirmed, due to missing statistical significance

5.0 Conclusion

In this chapter I will first discuss the results from the research and compare these results with similar research done by other auteurs. The chapter will close with the limitations of my research and recommendations I give that future research can consider.

5.1 Discussion

In this research I wanted to see if the implementation of ERM has an influence on the audit fees and if this relation is influenced by firm characteristics. This research was possible due to the increase in popularity of ERM. It seems that companies are seeing the added value of ERM in their organization and are appointing a CRO to oversee the ERM process. This allowed me to find enough CRO position

creations to test my hypotheses.

Before discussing the results for the hypotheses tested, I first want to discuss one interesting finding for one of the control variables. The model shows a significant negative effect of inherent risk on audit fees, while based on prior literature a positive effect was expected. A meta-analysis of audit fees research by Hay et al. (2004) showed that 28 studies found a positive relationship between inherent risk and audit fees and four studies did not find any significant results. There were no studies in the meta-analysis that found a significant negative effect as in this study. Even further literature research by myself didn’t turn up any previous research that found the same effect. Research in audit fees routinely omit financial institutions due to the special nature of this industry(Fan and Wong, 2005). It could be that the opposite effect is caused by the large amount of financial institutions in this sample. Future research will have to prove that this is the cause.

The first hypothesis tested if there was a relationship between the implementation of ERM as measured by the creation of a CRO position and the audit fees billed by the auditor of the company. The

expectation was that the implementation of ERM would lower the audit fees billed by the auditor. This expectation is based on the literature on agency theory, which suggests that some control mechanisms may be substituted by other control mechanisms, so that there would be a trade-off among internal control and external assurance (Jensen and Meckling, 1976). Model 2 of the moderated regression analysis shows some evidence of a relationship in this direction, but the results are not statistically significant so no conclusions can be drawn. So hypothesis one is not confirmed. This is a common problem when using the CRO appointment as a proxy for ERM implementation, for example Desender & Lafuente (2010) and Gordon et al. (2009) were also not able to find a significant relationship using CRO appointment while they were able to find a significant relationship using another proxy for ERM implementation.

The second hypothesis tested if the relationship between the implementation of ERM and the audit fees was moderated by the size of the firm as measured by the total assets. The expectation for a moderating effect was based on the direct effect of firm size on both the effect of ERM and the height of audit fees, which makes it reasonable to expect an enhancing effect when both are present. To test this hypothesis model 3 was used to see if the adding of an interaction term to the model would increase the

explanatory power of the model and if a statistically significant effect of the interaction term was found. The coefficient was 0,078, this implies that the interaction effect of firm size and CRO increase the audit fees by 7,8%. But the adding of the interaction term CROxSize did not add any explanatory power to the

model and the interaction term was also not statistically significant. Therefore hypothesis 2 cannot be confirmed.

Hypothesis three and four are largely the same as hypothesis two with the only difference being the variable moderating the effect between audit fees and ERM implementation. Both these hypothesis were also tested using model three were interaction terms were added to test for a moderating effect. Hypothesis four tested if earnings volatility has an moderating effect on the negative relationship between audit fees and ERM implementation. The interaction term VolatilityxCRO had a coefficient of -0,012 implying that an increase of 1% in the total value of the interaction term lowers the audit fees by 1,2%. But this interaction term was also not significant, so hypothesis four could not be confirmed. Hypothesis four tested if the financial leverage moderates the negative relation between audit fees and ERM implementation. The interaction term LeveragexCRO had a coefficient of 0,023 meaning a 2,3% increase for every 1% increase in the total value of the interaction term, but the interaction term was not statistically significant. Therefore hypothesis three cannot be confirmed, just like hypothesis two and three.

So it seems no conclusion can be drawn as none of the hypothesis test results have any significance. In the next section some possible reasons and solutions for this problem are discussed.

5.2 limitations and future research

There are some limitations to this study that are worth noting. First the study uses a proxy for ERM implementation being the creation of a CRO. The use of CRO position creation as proxy for adoption of ERM by a firm has a few drawbacks. The CRO position creation can just be a title change to better reflect the manager’s responsibility and it doesn’t capture the extent of the ERM implementation, it only captures that a senior executives is overseeing the risk management practices. There could be large differences in the extents of implementation of ERM in the firms that appoint a CRO. Although the proxy has drawbacks it is currently the best quantifiable measure available and the most used measure in the research area. Future research should look into better determinants or proxies for the implementation and the extent of ERM implementations in companies.

Another limitation is that the research is conducted for CRO creations during the years 2000 till 2013 there are several drawbacks from studying such a long period. For one there are a lot of things that can happen that have an effect on any of the variables in the research and bias the results. Some examples can be new laws and regulation that are passed, new technology or work methods that are developed or a market downturn like the financial crisis can influence the variables and results for certain years. Future research can try to incorporate these effect into their research analysis or can study one year if there are enough observations for one year.

In this research I measured the effect of CRO position creation on the audit fees in the same year, it could be that it can take a few years before positive effects of CRO position creation is seen in the audit fees. Future research could take into account multiple years after the CRO position creation to see if this changes the results.

A large part of the Sample in this research are financial institutions (sic-code 60). This is in line with predictions by Liebenberg & Hoyt (2003) and McShane et al. (2014) that financial institutions have been leaders in implementing ERM. This makes the results less generalizable since one industry is dominating

the sample. This was partially solved by adding a dummy control variable for the finance industry, but it would be preferable if future research would look for a more diverse sample. This could be done by looking in other countries or future years.

Some companies are more likely to create the CRO position than others. For example, Liebenberg & Hoyt (2003) have found that financial leveraged and more firms with volatile earnings are more likely to appoint a CRO. This make the sample biased towards companies with certain characteristics. Future research can look for a research setting were CRO appointment is mandatory, since this will lead to a more objective sample.

In this research the control group or reference group was the same company only the year before the appointment. The same company a year before can also have some extent of ERM, but did not have a CRO yet. Future research could repeat this research and use a similar company in the same industry as a control or reference group as done by Liebenberg and Hoyt (2003).

The reason for not finding any statistically significant results for our hypothesis could be due to the sample being too small. A larger sample could fix the problem of not finding any statistical significance, unfortunately at this time this was not possible for me. But future research should be able to redo this research with a larger sample. It would be interesting to see if the effects will stay the same and if statistical significance is found.

References

Abbott, L. J., S. Parker, and G. F. Peters. 2012a. Internal audit assistance and external audit timeliness. Auditing: A Journal of Practice & Theory. 34 (4): 3–20.

Aguinis, H. (2004). Regression analysis for categorical moderators. New York, NY: Guilford Press. Anderson, D., Francis, J. R., & Stokes, D. J. (1994). Auditing, directorships and the demand for monitoring. Journal of Accounting and Public Policy, 12(4), 353-375.

Arnold, H. J. (1982). Moderator variables: A clarification of conceptual, analytic, and psychometric issues. Organizational Behavior and Human Performance, 29(2), 143-174.

Bedard, J. C., & Johnstone, K. M. (2004). Earnings manipulation risk, corporate governance risk, and auditors' planning and pricing decisions. The Accounting Review, 79(2), 277-304.

Bank for International Settlements (BIS) Joint Forum, August 2003. Trends in Risk Integration and Aggregation. Accessed on www.bis.org on 20-03-2015

Barton, T. L., Shenkir, W. G., & Walker, P. L. (2002). Enterprise Risk Management: Pulling It All Together. The Institute of Internal Auditors Research Foundation. Altamonte Springs, Florida.

Bartov E. The Timing of Asset Sales and Earnings Manipulation. Accounting Review [serial online]. October 1993;68(4):840-855.

Beasley, M. S., Clune, R., & Hermanson, D. R. (2005). Enterprise risk management: An empirical analysis of factors associated with the extent of implementation. Journal of Accounting and Public Policy, 24(6), 521-531.

Beasley, M., Pagach, D., & Warr, R. (2008). Information conveyed in hiring announcements of senior executives overseeing enterprise-wide risk management processes. Journal of Accounting, Auditing & Finance, 23(3), 311-332.

Benston, G. J. (1985). The market for public accounting services: Demand, supply and regulation. Journal of Accounting and Public Policy, 4(1), 33-79.

Blumberg, B., Cooper, D. R., & Schindler, R. S. (2005). Business Research Methods. London: The McGrawHiIl Companies.

Carcello, J. V., Hermanson, D. R., Neal, T. L., & Riley, R. A. (2002). Board Characteristics and Audit Fees*. Contemporary Accounting Research, 19(3), 365-384.

Carpente. CG, & Strawser, R. H. (1971). Displacement of auditors when clients go public. Journal of Accountancy, 131(6), 55-58.

Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied multiple regression/correlation analysis for the behavioral sciences (3rd ed.). Mahwah, NJ: Lawrence Erlbaum Associates.

Committee of Sponsoring Organisations of the Treadway Commission (COSO), 2004. Enterprise Risk Management—Integrated Framework. AICPA, New York, NY.

Committee of Sponsoring Organisations of the Treadway Commission (COSO), 2011. Frigo, M. L., & Anderson, R. J. Embracing Enterprise Risk Management – Practical approaches for getting started. http://www.coso.org/documents/EmbracingERM-gettingStartedforWebPostingDec110_000.pdf

Cook, R. D., & Weisberg, S. (1982). Residuals and influence in regression. New York, NY: Chapman & Hall. Craswell, A. T., Francis, J. R., & Taylor, S. L. (1995). Auditor brand name reputations and industry

specializations. Journal of accounting and economics, 20(3), 297-322.

DeAngelo, L. E. (1981). Auditor size and audit quality. Journal of accounting and economics, 3(3), 183-199.

Desender, K. A., & Lafuente, E. (2011). The Relationship Between Enterprise Risk Management and External Audit Fees: Are They Complements or Substitutes?. risk management and corporate governance, Jalilvand & Malliaris, eds., Routledge.

Desender, K., Lafuente, E. (2011), "The influence of Board composition, audit fees and ownership concentration on enterprise risk management", Electronic copy available at:

http://ssrn.com/abstract=1495856

Economist Intelligence Unit. (2005). The Evolving Role of the CRO. London, New York, Hong Kong: The Economist Intelligence Unit.

Fama, E. F., & Jensen, M. C. (1983). Separation of ownership and control. Journal of law and economics, 301-325.

Felix, W. L., Gramling, G. A. A., & Maletta, M. J. (2001). The contribution of internal audit as a determinant of external audit fees and factors influencing this contribution. Journal of Accounting Research, 39 (3), 513-34.

Frigo, M. L., & Anderson, R. J. (2011). Embracing enterprise risk management: Practical approaches for getting started. Retrieved from

http://www.coso.org/documents/embracingerm-gettingstartedforwebpostingdec110_000.pdf.

Gist, W. E. (1992). Explaining variability in external audit fees. Accounting and Business Research, 23(89), 79-84.

Gordon, L. A., Loeb, M. P., & Tseng, C. Y. (2009). Enterprise risk management and firm performance: A contingency perspective. Journal of Accounting and Public Policy, 28(4), 301-327.

Hackenbrack, K., & Knechel, W. R. (1997). Resource Allocation Decisions in Audit Engagements*. Contemporary Accounting Research, 14(3), 481-499.

Hay, D., Knechel, W. R., & Wong, N. (2004). Audit Fees: A Meta-Analysis of the Effect of Supply and Demand Attributes. SSRN Working Paper Series.

Hope, O. K., Langli, J. C., & Thomas, W. B. (2012). Agency conflicts and auditing in private firms. Accounting, Organizations and Society, 37(7), 500-517.

Hoyt, R. E., & Liebenberg, A. P. (2011). The value of enterprise risk management. Journal of Risk and Insurance, 78(4), 795-822.

Jaccard, J., & Turrisi, R. (2003). Interaction effects in multiple regression (2nd ed.). Thousand Oaks, CA: Sage Publications.

Jensen, M. C. and W. Meckling. (1976). The Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure. Journal of Financial Economics 3: 305–360.

Jose, P. E. (2013). Doing statistical mediation & moderation. New York, NY: Guilford Press. Kren L. Budgetary Participation and Managerial Performance: The Impact of Information and Environmental Volatility. Accounting Review [serial online]. July 1992;67(3):511-526.

Kutner, M. H., Nachtsheim, C. J., Neter, J., & Li, W. (2005). Applied linear statistical models (5th ed.). New York, NY: McGraw-Hill.

Liebenberg, A. P., & Hoyt, R. E. (2003). The determinants of enterprise risk management: Evidence from the appointment of chief risk officers. Risk Management and Insurance Review, 6(1), 37-52.

Matsumura, E., and Tucker, R. (1992). Fraud detection: A theoretical foundation. Accounting Review 67 (Fall): 753–782.

McShane, M. K., Nair, A., & Rustambekov, E. (2011). Does enterprise risk management increase firm value?. Journal of Accounting, Auditing & Finance, 26(4), 641-658.

Nocco, B. W., & Stulz, R. M. (2006). Enterprise risk management: Theory and practice. Journal of Applied Corporate Finance, 18(4), 8-20.

New York Stock Exchange (NYSE), 2003. Accessed on www.nysemanual.nyse.com on 19-03-2015 Newton, J. D., & Ashton, R. H. (1989). The association between audit technology and audit delay. Auditing-a Journal of Practice & Theory, 8, 22-37.

Otley, D. T. (1992). The contingency theory of management accounting: achievement and prognosis. In Readings in Accounting for Management Control (pp. 83-106). Springer US.

Pagach, D., & Warr, R. (2007). An empirical investigation of the characteristics of firms adopting enterprise risk management. North Carolina State University working paper.

Pearson, T., & Trompeter, G. (1994). Competition in the market for audit services: The effect of supplier concentration on audit fees*. Contemporary Accounting Research, 11(1), 115-135.

Prawitt, D. F., N. Y. Sharp, and D. A. Wood. 2011. Reconciling archival and experimental research: Does internal audit contribution affect the external audit fee? Behavioral Research in Accounting 23 (2): 187– 206.

Prescott, J. E. (1986). Environments as moderators of the relationship between strategy and performance. Academy of Management Journal, 29(2), 329-346.

Public Company Accounting Oversight Board (PCAOB) Release No. 2007-005A, June 12, 2007. An audit of internal control over financial reporting that is integrated with an audit of financial statements and related independence rule and conforming amendments. Auditing Standard No. 5 Washington, D.C.