Fair value level 3 accounting : implications on auditing and governance

(1)

Fair value level 3 accounting: implications on

auditing and governance

Master Thesis

Student name: René van Lier

Student number: 10440372

Date: 14-7-2014

Study: Msc Accountancy & Control

Faculty of Economics and Business, University of Amsterdam

Supervisor: dr. J.J.F. van Raak

(2)

2

Abstract

This paper investigates the impact of fair value accounting on the cost of debt, the quality of audit committees and the costs of external audits. In the literature review I develop three hypotheses and test them with three OLS regression models.

The first hypothesis states that firms with a higher proportion of fair value level 3 assets have a higher cost of debt. The results of the cost of debt model indicate that investors require lower interest rates when fair values of all the three types are used because these values seem to improve decision usefulness.

The second hypothesis states that firms with a higher proportion of fair value level 3 assets have higher quality audit committees. The results of the audit committee quality model do not indicate any significant reaction of firms to the accounting nature of their assets. No evidence suggest that firms show reactive behavior to the potential agency problems and risks that are generally associated with fair value and in particular fair value level 3 assets.

The third hypothesis states that firms with a higher proportion of fair value level 3 assets have higher audit fees. The results of the audit fees model indicate that firms with a bigger proportion of fair value level 3 assets have higher audit fees, while they tend to be lower when fair value level 2 assets are used. This dynamic does not continue for fair value level 1 assets, which seem to be more expensive to audit. While the third hypothesis in not rejected, the underlying explanation for it is not comprehensive and other dynamics in auditing could have different impacts on audit fees between and within the different categories of fair value assets.

(3)

3

1. Introduction

1.1 Background and academic contribution

Fair Value Accounting is generally associated with an increase in the relevance of financial statements. However, the use of fair value accounting also has a background of controversy because of the decrease in reliability. This issue is particularly present in the case of fair value level 3 assets (mark to model). The values of assets in this category are usually calculated by the present value of future cash flows (earnings) or option pricing models, which implies that multiple variables, such as interest rates, growth rates and other forecasts are used. This means that fair value level 3 assets might be less reliable or

transparent in more complex and volatile environments.

Because a absent of liquid markets, some degree of subjectivity exists in the preparation and auditing of financial reports that contain fair value level 3 assets. This degree of subjectivity can lead to less accurate and less reliable financial reporting and causes additional difficulties for investors and auditors to process and evaluate financial information. Furthermore, an increased degree of subjectivity can also lead to an increased level of discretion, earnings management, risks and fraud.

One way to mitigate these kinds of agency problems is the employment of an independent board of directors and an audit committee. The quality of these governance bodies is determined by the financial expertise, devotion and independency of its members. Prior literature identifies that firms with weaker corporate governance mechanisms have greater agency problems (core et all. 1999). However, Dey (2008) finds that firms with more agency problems have stronger corporate governance mechanisms, which implies that firms with a larger potency of agency problems respond to these risks by extending and improving their governance mechanisms.

This paper expands existing accounting and corporate governance literature by determining whether the different asset types in the fair value hierarchy, with their corresponding risks and agency problems, are related to the strength of audit committee mechanisms. Furthermore, this paper provides additional insights in the relation between the three types of fair value assets and the associated costs for firms in terms of cost of debt and cost of external auditing.

(5)

5

1.2 Societal relevance

From a societal point of view, high financial reporting and audit quality ensure reliable financial information for shareholders, debt holders and all other stakeholders. Corporate governance mechanisms such as audit committees are important to preserve financial transparency and to keep financial reporting quality at a high level.

Except reliability, decision usefulness is another important determinant of the quality of financial information. The usage of fair values increase the timeliness and therefore the relevance of financial information at some costs in terms of complexity and reliability. The combination of these aspects determine the quality of financial information and high quality information improves the effectiveness and efficiency of the decision making processes with regard to allocation of financial resources.

This study contributes to the discussion about the advantages and disadvantages of fair value accounting by providing insights about the structure and quality of audit

committee structure in response to different asset structures in which fair value accounting is adopted. Additionally, this research provides evidence of cost implications with regard to cost of debt and cost of audits for different amounts and types of fair value assets, which may help companies in assessing the impact of their assets and activities on their cost of debt and the costs of their external audits.

1.3 Structure of the paper

In the next chapter, I first provide a literature review in which the hypotheses of this research are developed. The literature review covers the notion of fair value accounting and its history and development along with its advantages and disadvantages, in particular with regard to information asymmetry and auditing implications. Subsequently, I present an overview of the governance mechanisms that are developed to mitigate these problems, with a particular focus on the audit committee. In the third chapter, I describe the research methodology and the sample used for the analysis in this paper. In the fourth chapter, I provide and discuss the results of my research and in the last chapter, I provide a conclusion and directions for future research.

(6)

6

2. Literature review and hypothesis development

2.1 Accounting history and transition to fair value accounting

Fair value accounting has been a controversial issue since its emergence in the last century. Its alternative, historic cost accounting (HCA), only enjoyed an episodic legitimacy in the 1940s–70s and outside this period, mixed measurement incorporating market values were routinized and taken for granted. (Georgiou & Jack, 2011). In 1961 the Accounting Research Study (ARS) issued a study which provided a set of fundamental principles that served as a foundation for subsequent accounting principles. In the following of this, the ARS released a study which questioned the primacy of historical cost for asset valuation; the study recommended that any changes in the value of assets that could be “objectively determined” should be recognized (Accounting Principles Board, 1962). Philips (1963) Provided 5 methods in which income can be determined: (1) psychic income: Subjective income, what the individual perceives it to be, (2) economic present value added

(discounted present value of future cash flows), (3) cash income (Income determined by cash inflows and outflows), (4) accrual income (recognizes revenues when earned and expenses when incurred in support of revenues) and (5) accretion income, defined by the increase in economic power as measured by changes in the market value of assets. This last method is now known as fair value accounting. The conjunction of these studies eventually lead to the issuance of a report by a committee of American Institute of Certified Public Accountants in 1973, which emphasized the primacy of providing “information that is useful to users in making economic decisions” and also stated that this objective “could not be served by the use of a single valuation basis such as historical cost.”

The newly created Financial Accounting Standards Board (FASB) in 1973 later developed a Statement of Financial Accounting Concept (SFAC), which was issued in 1984 and it contained five alternative methods for valuation: (1) historical cost, (2) current cost, (3) current market value, (4) net realizable value, and (5) present value. In later years, the FASB did revise the use of different valuation methods in multiple Statements of Financial Accounting Standards (SFAS), most importantly the numbers 87, 105, 107, 115, 119, 121, 123, 123R, 133, 157, and 159. Emerson et al. (2010) argue that these particular statements of financial standards have incrementally and systematically advanced the use of valuation

(7)

7

methods that provide a more relevant measure of value than that provided by historical cost. In 2007, the FASB extended the list of financial statement items that may be valued at fair value to include: (1) loans receivable and payable, (2) investments in equity securities, (3) rights and obligations under insurance contracts, (4) rights and obligations related to warranty agreements, (5) host financial instruments that are separated from embedded derivative instruments, (6) firm commitments involving financial instruments and (7) written loan commitments. The types of fair value assets investigated in this paper are all composed of these 7 groups.

Due to the larges differences in nature between and within these groups of assets, further guidance is needed to attain consistency and clarity about valuation methods with regard to financial reporting. In the next section, I provide an explanation of the most contemporary definitions of fair value accounting and its current uses that are relevant for this paper.

2.2 Fair value hierarchy and definition of level 3 assets

The FASB issued FAS 157 in September 2006 to provide guidance about how entities should determine fair value estimations for financial reporting purposes. This standard was subsumed in Accounting Standards Codification (ASC) Topic 820, which defines fair value as "The price that would be received to sell an asset or paid to transfer a liability in an orderly transaction between market participants at the measurement date". The topic emphasizes the use market inputs to determine the fair value of an asset or liability. In some cases however, no active market with quoted prices is available and estimations have to be made. For these purposes, the topic 820 contains a framework which defines 3 types of fair value assets/liabilities.

The first category contains assets and liabilities that are visible as “quoted prices in active markets for identical assets or liabilities”. Financial information in this category is based on direct observations of transactions involving the identical assets or liabilities being valued, without any assumptions. These characteristics mean that the entity must have access to an active market for the item being valued in this category and values in this category are the most reliable.

The second category contains assets and liabilities that are valued based on market observables and values can fall in this category in three different situations. The first

(8)

8

situation is the presence of ‘’less-active markets for identical assets and liabilities’’. The second situation is the presence of owned assets and liabilities are similar to, but not the same as, those traded in a market. These situations demand the company to make

assumptions about what the fair value of the reported items might be in a market. The third situation arises when no active or less-active markets exist for similar assets and liabilities, but other market inputs can be used to make a reliable estimation. This requires the entity to derive the market value from quoted prices of other items and market data. Examples of these methods are option pricing models, such as the Black-Scholes model.

The third category of fair values is based on market prices which the FASB describes as “unobservable”. The FASB states that this residual category exists “for situations in which there is little, if any, market activity for the asset or liability at the measurement date.” In contrast to “observable inputs” mentioned in the two other categories, these “unobservable inputs” are not based in external and independent sources, but on “the reporting entity’s own assumptions about the assumptions market participants would use.” As these characteristics indicate, the values in this category contain the most subjectivity and assumptions of internal uses, which makes this category the least reliable and the most controversial.

Despite the controversy, Barth and Landsman (1995) state that fair values are

conceptually relevant to financial statement users in assessing firm value. They also define a financial statement item to be value-relevant when the underlying information it reflects helps financial statement users to assess firm value. Other more recent research, such as Herrmann et al. (2008) shows that fair value measures for property, plant, and equipment are superior to historical cost based on the characteristics of predictive value, feedback value, timeliness, neutrality, representational faithfulness, comparability, and consistency.

While there are legitimate concerns about marking to market (level 3 fair values), much of the controversy results from confusion about what is new and different about FVA (Laux & Leuz, 2009). In the next section of this literature review, some important problems and issues about fair value that motivate this research are highlighted.

(9)

9

2.3 Fair value issues: transparency, risk and agency problems

A clear example of difficulties with fair value level 3 assets is illustrated by Benston (2008), who shows how exit values such as work-in-process, inventories, special-purpose machines and also assets and liabilities restated at exit prices yield balance sheets and income statements that are of little, if any, value to investors in ongoing firms, at much higher cost than historical cost accounting.

Webinger et al. (2013), however, states that historical cost prices are also not a suitable alternative in these situations, and they show that extra disclosures are helpful in these cases of illiquid or inactive markets. Lee and Park (2013) also show that the problem of increased subjectivity in the measurement of these fair values can be partly reduced by bigger accounting firms (big 4 auditors), which seem to be able increase the value-relevancy of these measurements better than smaller auditing firms.

In case of financial instruments, however, Khurana & Kim (2003) were unable to detect a discernible difference in the informativeness of fair value measures collectively relative to historical cost measures. They also find that that historical cost measures of loans and deposits are more informative than fair values. Gaynor et al. (2011) supports this last point by illustrating that when liabilities are accounted for at fair value, a deterioration of a company’s credit risk results in the reporting of an income statement gain; an improvement in a company’s credit risk results in a loss. Their research suggests that these

counterintuitive income statement effects cause confusion by financial statement-users who are likely to misinterpret fair value gains as positive signals and fair value losses as negative signals. This illustrates the importance of high quality disclosure of all value relevant items in financial reports.

One of the most famous cases of deception of investors and failure of fair value accounting is the Enron case, where level 3 measurements played a substantial role in the fraud (Benston, 2006). Gwilliam & Jackson (2008) state that the unreliability of valuation estimates put the management of Enron in the position to commit the fraud.

The risks of fair value accounting still exists in the contemporary world however. Glaser et al. (2013) present evidence that the use of fair value assets with a higher degree of accounting discretion is a good predictor for default risk in banks. Fang et al. (2013) also illustrate that the incremental explanatory power of fair-value accounting information on financial instruments was affected during the 2008 financial crisis. Other risks that fair value

(10)

10

accounting may cause lie in increased earnings management possibilities and agency problems.

Xiaolu (2013) examines the relation between earnings volatility and fair value measurements. He finds that banks who recognize more level 3 fair value assets and liabilities report smoother earnings over the time. Prior to that study, Andrews (2012) also found that earnings management significantly increased after fair value accounting was introduced because of changes in regulations. The unreliability and controversy which are associated with fair value accounting do, because of possible asymmetrical information and discretion, imply potential agency problems and risks, which may have an impact on

investors behavior.

In a literature review about the usefulness of fair value accounting information to investors, Landsman (2007) suggests that disclosed and recognized fair values are

informative to investors, but that the level of informativeness is affected by the amount of measurement error and source of the estimates - management or external appraisers. These aggregation of prior literature suggest that relevance and reliability characteristics play a major role in investors behavior. To investigate how fair value and its associated risks influence investors assessments on capital provision, I derive the following hypothesis:

H1: Firms with a higher proportion of fair value level 3 assets have a higher cost of debt

In the next section of this literature review, the way auditors and audit committees are deployed to mitigate these risks is covered and possible relationships between these two phenomena and fair value accounting are elaborated. As a result, the last two hypotheses of this paper are developed in this section.

2.4 Corporate governance, auditing and audit committees

To protect the interests of shareholders, corporate governance mechanisms are developed and the audit committee is an important body to investigate when fair value accounting issues are the main topic. Vera-Munoz (2005) states that audit committees have a broad range of responsibilities with regard to approving the corporate strategy and

monitoring the operating and internal control systems. She defines three broad areas of oversight. The first one is the audit process, which includes the authority to hire and

(11)

11

terminate an external auditor and the approval of all audit and non audit services. The second part includes the financial reporting duties of the firm, which includes critical choices in accounting methods with regard to the recording of income, expenses, assets and

liabilities and a review of audit processes of internal and external auditors. And thirdly, the audit committee has duties with regard to whistleblowers protection and treatment of complaints and concerns by other employees. In this next part of this section, I explain the different quality characteristics of audit committees and the effect of this quality.

Regarding earnings management, Nelson & Devi (2013) find that the presence of both non-accounting experts and accounting experts reduces the magnitude of earnings management. Krishnan et al. (2011) reports that that the presence (and proportion) of directors with legal backgrounds on the audit committee is associated with higher financial reporting quality. They also find that the presence of directors with both accounting and legal expertise enhances financial reporting quality beyond the contribution of the individual forms of expertise, suggesting that these forms of expertise supplement each other. A recent study of Cohen et. Al. (2014) extends this literature by suggesting that industry expertise does also have a positive effect on financial reporting quality. Abernethy et al. (2012) also find a significant association between accounting and financial expertise on the audit committee and analyst earnings forecasts that are more accurate and less dispersed. All of the results mentioned above highlight the importance of various kinds of expertise within the audit committee to function properly.

Some effects of expertise in audit committees are illustrated by Barua et al. (2010), who find that the investment in internal auditing (internal audit budget) is negatively related to the presence of auditing experts on the committee and the average tenure of audit committee members, but positively related to the number of audit committee meetings (a proxy for audit committee diligence). These observations suggest that complementary and substitutionary relationships between the audit committee and internal auditing exist. Rainsbury et al. (2009), however, find that the quality of audit committees has little impact on the level of fees paid to external auditors. The results suggest that the benefits of ‘best practice’ audit committees may be less than anticipated by regulators and policymakers.

Another aspect that audit committees are supposed to improve is the prevention and mitigation of earnings management and accounting fraud. Beasley (1996) and Gerety & Lehn (1997) ﬁnd no relation between the existence of an audit committee and accounting fraud,

(12)

12

but Dechow et al. (1996) ﬁnd that the incidence of Securities and Exchange Commission (SEC) accounting enforcement actions is lower for ﬁrms with a formal audit committee. These result suggests that the presence of audit committees have limited impact in revealing fraud, but they seem to mitigate the number of fraud attempts by managers. A more recent study of Zhang et all (2007) finds that firms are more likely to be identified with an internal control weakness if their audit committees have less financial expertise or, more specifically, have less accounting expertise. They also find that firms are also more likely to be identified with an internal control weakness, if their auditors are more independent.

The quality of audit committees seems to be important to investors, because

financial markets react to appointments of audit committees. Davidson (2004) finds positive stock price reactions when new members of audit committees have financial expertise, which suggests that the market rewards firms that appoint financial experts to their audit committees. Singhvi et al. (2013), however, finds that in the post-SOX period, the market reaction to the appointment of different types of expert directors is not significantly different from zero. These results suggest that almost all companies have financial experts on audit committees, which removed the focus of markets on appointments of financial experts in the post-SOX period.

The use of experts in audit committees indeed seems to be ordinary practice in the post-SOX period and prior research (Dey, 2008) also suggest other proxies for audit

committee quality, such as size, frequency of meetings and independency, to assure a decent level of reliability of financial statements. This paper goal is to examine the relations of these quality characteristics with the presence of (mark to model) fair value accounting practices, which seem to imply potential agency problems and risks, as indicated in the prior section.

Core et al. (1999) find that firms with weaker corporate governance mechanisms have greater agency problems. However, Dey (2008) finds that firms with more agency problems have stronger corporate governance mechanisms, which suggests that firms that are more imposed to potential risks react to these settings by increasing the quality of their

governance mechanisms. In the context of fair value, Song et al. (2010) find that the value relevance of fair values (especially Level 3 fair values) is greater for firms with strong corporate governance. They argue that their overall results support the relevance of fair

(13)

13

value measurements under FAS No. 157, but weaker corporate governance mechanisms may reduce the relevance of these measures.

To investigate to which extend firms with a high proportion of fair value level 3 assets are aware of and react to these governance implications, I derive the following hypothesis:

H2: Firms with a higher proportion of fair value level 3 assets have higher quality audit committees

The increasing usage of fair value accounting increased the complexity of financial reporting and the effort that is needed to perform the audits. However, Goncharov et al. (2014) document lower audit fees for firms reporting property assets at fair value relative to those employing depreciated cost, This difference appears in part to be driven by

impairment tests that occur only under depreciated cost. They further find that audit fees are decreasing in firms’ exposure to fair value and increasing both in the complexity of the fair value estimation and for recognition (versus only disclosure) of fair values. This indicates that any reductions in audit fees will vary with different characteristics of the reported fair values. This includes the difficulty to measure or estimate the values and the treatment within the financial statements. To measure the impact of fair value accounting on audit fees, I derive the following hypothesis.

(14)

14

3. Research methodology and sample

3.1 Sample and data selection

Using publicly available information from the Compustat, Risk Metrics and Audit Analytics database, I constructed three OLS regression models that cover multiple research periods between 2010 and 2013. The three samples consist of available financial, audit and governance related information of the publicly listed firms in the USA. The next three sections provide an overview of the sample selection process.

Cost of debt model - sample selection

The first regression model has the cost of debt as outcome variable to measure the impact of three predictor variables and five control variables. The data used for this regression model is collected from the Compustat database. The research period covers 4 years; from 2010 to 2013.

The initial sample consisted out of 2729 firm years of 820 firms in the USA. After removing all firms without debt and all cases with missing values, the sample consisted out of 2040 firm years of 626 firms. I provide descriptive statistics of this dataset before the removal of extreme values in appendix A, including range, minimum, maximum, skewness and kurtosis. After the removal of extreme values with regard to all variables, the sample used in the model consists out of 1950 firm years of 599 firms.

Data Firms Firmyears

820 2729

626 2040

599 1950

Dataset after removal of missing values Final dataset after removal of extremes

Merger dataset before cleaning

(15)

15 Audit committee quality model – sample selection

The second regression model has an audit committee quality score as outcome variable to measure the impact of three predictor variables and four control variables. The data used for this regression model is collected from the Compustat and Risk Metrics database. The research period covers 3 years; from 2010 to 2012. There were no date available for 2013 in the Risk Metrics database.

The initial sample consisted out of 811 firm years of 335 firms in the USA. After removing all cases with missing values, the sample consisted out of 807 firm years of 335 firms. Also for this dataset before the removal of extreme values, I provide descriptive statistics in appendix B, including range, minimum, maximum, skewness and kurtosis. After the removal of extreme values with regard to all variables, the sample used in the model consists out of 786 firm years of 327 firms.

Audit fees model – sample selection

The third regression model has audit fees as outcome variable to measure the impact of three predictor and thirteen control variables. The data used for this regression model is collected from the Compustat and Audit Analytics Database. The research period covers 4 years; from 2010 to 2013.

335 811

335 807

327 786

Audit Comittee Model - Sample Selection

(16)

16

The initial sample consisted out of 2420 firm years of 938 firms in the USA. After removing all cases with missing values, the sample consisted out of 2138 firm years of 863 firms. Descriptive statistics of this dataset before the removal of extreme values are

provided in appendix C, including range, minimum, maximum, skewness and kurtosis. After the removal of extreme values with regard to all variables, the sample used in the model consists of 1996 firm years of 835 firms.

3.2 Research methodology

For this study, I use three OLS regression models to test the three hypothesis

developed in chapter two of this paper. These three models are constructed and explained in the following three sections.

The first model is used to test hypothesis 1 and it measures the impact of fair value assets on the cost of debt. The second model corresponds to hypothesis 2 and uses an audit committee quality scores to measure their relations with fair value assets. The third model is used for hypothesis 3 and it measures the impact of fair assets on audit fees.

3.2.1 Explanation of the cost of debt model

To examine the first hypothesis, I use a regression model with the cost of debt as outcome variable. I calculate the cost of debt in the same way as Gray et al. (2008) and Francis et al. (2005), who use interest expense divided by the average total debt as proxy for the cost of debt. I do this by adding average short-term debt and average long-term debt variables in the Compustat dataset. The test variables consist of the proportion of fair value

938 2420

863 2138

835 1996

(17)

17

level 1, level 2 and level 3 assets compared to the proportion of total assets of each

company. I expect all these variables to be positively correlated with the outcome variable.

CoD = β0 + β1FVL3 + β2FVL2 + β3FVL3 + β4Fsize + β5Lev + β6ROA + β7Dcov + β8Evol + ε

To account for other variables which might have an impact on cost of debt, I use several control variables. Pinches and Mingo (1973) identify five independent variables that give an empirically proved indication for bond quality and therefore may have an impact on the cost of debt. These variables are leverage (Lev), firm size (Fsize), return on assets (ROA), debt coverage (Dcov), and earnings volatility (Evol). I use debt-to-equity ratio’s to measure leverage, the natural logarithm of total assets for firm size, debt coverage by debt coverage ratios (operating income divided by interest expenses + short & long term debt due in 1 year), and earnings stability by using the standard deviation of net income before

extraordinary items (IB), scaled by the total assets of the company. I expect leverage and earnings volatility to be positively correlated with the cost of debt. I expect firm size, return on assets and debt coverage ratio to be negatively related with the cost of debt.

This model might not capture all determinants of cost of debt. Himme &Fischer (2014) developed a model with additional determinants such as customer satisfaction, brand value, and corporate reputation, which is outside the scope of this paper. Furthermore, I did not use credit ratings because these might have incorporated effects of fair value level 3

CoD cost of debt

FVL3 proportion of FVL3 assets

Fsize natural logaritm of total assets

Lev ratio of total debt to total assets

ROA return on assets

Dcov debt coverage ratio

Evol earnings volatility

Outcome variable: Test predictor variables:

(18)

18

assets and this reduces the model’s ability to isolate this influence. For the same reason, I omit a control variable for accrual quality and earnings quality. Because of practical reasons, I also did not incorporate control variables with regard to governance quality, which might have an additional impact on cost of debt.

3.2.2 Explanation of the audit committee model

To find evidence for the second hypothesis, I adapt a similar approach as Dey (2008) for building a regression model which incorporates companies reactions to agency problems in terms of governance quality. Dey (2008) used audit committee size, number of meetings of the audit committee, audit committee (financial & accounting) expertise and audit

committee independency (proportion of outside directors in the audit committee) As proxies for audit committee quality. My outcome variable is an audit committee quality score that incorporates the same quality determinants as Dey (2008). I omit the number of audit committee meetings in my model, because no data was available for this proxy in the Risk Metrics database. Furthermore, I include an age variable, because a higher average age of directors indicates a more experienced committee.

The table below illustrates the way in which the audit committee quality score is calculated for each firm year in the dataset. The the positive components of the audit committee quality score are the audit committee size (total members x 10), the financial expertise of the audit committee (total number of members with financial expertise x 10) and the average age of these members. To get a more balanced score with regard to all quality determinants, I deduct 45 from the average age of the audit committee members in each firm year: 45 This was the lowest average age found in the total dataset. The negative components of the audit committee quality score exist of two penalty scores for absent members (total absent members x 10) and dependent members (total interlinked or inside members x 10)

(19)

19

The test variables consist of the proportion of fair value level 1, level 2 and level 3 assets compared to the proportion of total assets of each company. I expect the proportion of fair value level 3 assets to be positively related with the audit committee quality score. I have no particular expectations about the directions of the other two predictor variable coefficients.

ACQ = β0 + β1FVL3 + β2FVL2 + β3FVL3 + β4Fsize + β5Empl + β6lev + β7Loss + ε

I use several control variables which are used by Dey (2008) to make proxies for agency problems. Firm size (Fsize) is measured by the logarithm of total assets. I use the total number of employees as a proxy for firm complexity (Empl), because a higher number of employees generally indicate more subsidiaries, divisions and departments, which may cause higher potential for information asymmetry (Swanson, 2008). I also use leverage as a proxy for agency problems (Lev), because I expect managers in firms with higher levels of debt to be more inclined to manipulate earnings (Dey, 2008). Finally, I use a dummy variable for accounting loss for the same reasons (Loss). I expect all these control variables to be positively correlated with the outcome variable.

+

(total members x 10)

+

(average age members - 45)

+

(total financial expertise members X 10)

-

(total absent members X 10)

-

(total interlinked or inside members X 10)

=

Total audit committee quality score

Quality score calculation

(20)

20 3.2.3 Explanation of the audit fees model

To test the third hypothesis, I use a regression model with the natural logarithm of audit fees as outcome variable (AUFEE). The test variable consist of the proportion of fair value level 3 assets compared to the proportion of total assets of each company. I also use the proportion of fair value level 2 en level 1 assets as predictor variables to account for the differences within the three different categories of fair value assets.

AUFEE = β0 + β1FVL3 + β2FVL2 + β3FVL3 + β4Fsize

+ β5Inv + β6Rec + β7Lev + β8Dcov + β9Loss + β10Qopi + β11Empl + β12For + β13Ncrtl + β14Big4 + β15Res + β16MA + ε

To account for other variables which might have an impact on audit fees, I use several control variables. Simunic (1980), Chan et al. (1983) and Kim et al. (2012) find a positive relation between audit fees and firms size, measured as the natural logarithm of total assets. Similar to these papers, I use the natural logarithm of total assets to measure client size (Fsize). Client size can also be measured in terms of total sales, however, research of Hay et al. (2006) illustrates that this measure is less common used compared to total assets. For this reason, I omit this variable from the model. Other variables that I include are inventory and receivables, which increase audit work for audit companies (Kim et al., 2012).

ACQ

Audit committee quality score

FVL3

proportion of FVL3 assets

FVL2

FVL1

Fsize

natural logaritm of total assets

Lev

ratio of total debt to total assets

Empl

number of employees

Loss

Loss (dummy)

(21)

21

Like Kim et al. (2012), I measure inventory (Inv) and receivables (Rec) as the natural logarithms of the reported book values.

To account for complexities which arise when firms are in financial distress, I use leverage (Lev) and debt coverage (Dcov) to measure company risk due to debt. Chan et al. (1993) find in their interviews with several partners at audit firms that facing financial pressure could result in more audit work (e.g. audit effort) due to increased awareness on possible company’s balance sheet. For this reason, I also include dummy variables for firms that have made losses (Loss) or received a qualified or going concern opinion (Qopi) in the last recorded year.

Another aspect that influences audit fees is the complexity of the audited firm. Chan et al. (1993) find that the number of subsidiaries, auditee diversification and ownership structure of the auditee has an impact on audit fees. I use the total number of employees (Empl) as a proxy for firm complexity, because a higher number of employees indicate more subsidiaries, divisions and departments, which cause higher potential information

asymmetry (Swanson, 2008). Furthermore, I use a foreign income dummy (For) and the proportion of non controlled assets (Nctl) as additional proxies for firm complexity.

The size of the auditor is another factor that has an impact on audit fees. Chan et al. (1993), Ferguson et al. (2003) and Kim et al. (2012) find that big 4 audit firms demand higher audit fees than non big 4 audit firms. Therefore, I use a dummy variable to control for the effects of big 4 auditors (Big4).

Additionally, I use dummy variables for companies that had a merger or a significant financial restructuring in the last recorded year, because these firms tend to have additional accounting complexities with corresponding increases in audit fees (Kim et al., 2012). I expect debt coverage ratios to be negatively correlated with the outcome variable. I expect all other control variables to be positively correlated with the outcome variable.

For practical reasons, I do not incorporate any corporate governance measures in the model for audit fees like Carcello et al. (2002) and Abbott (2003) did. For similar reasons, I do not incorporate some other control variables regarding non audit fees and other

complexities identified in Hay et al. (2006) and Kim et al. (2012). The complete overview of outcome, predictor and control variables are illustrated in the table below.

(22)

22

AUFEE

audit fees

FVL3

proportie FV3 van het totaal (AUL3/AT)

FVL2

FVL1

Fsize

natural logaritm of total assets

Inv

natural logaritm of total inventories

Rec

natural logaritm of total receivables

Lev

ratio of total debt to average total assets

Dcov

debt coverage ratio

Loss

losses (dummy)

Qopi

qualified opinion (dummy)

Empl

number of employees

For

Foreign income (dummy)

Ncrtl

proportion non controlled equity

Big4

big four auditor (dummy)

Res

Restructure (dummy)

MA

merger/acquisition (dummy)

(23)

23

4. Results

4.1 Cost of debt model

4.1.1 Descriptive statistics

In this section, the descriptive statistics of the cost of debt model sample are provided.

The average fair value proportions are relatively low. This is especially the case with fair value level 1 and level 2 assets. These values all have a high standard deviation, which means that a high dispersion between values exists. This also results in a relatively high skewness and high kurtosis. Debt coverage, return on assets and earnings volatility also have high standard deviations and kurtosis.

The outcome variable does not have a particularly high standard deviation, which means that there is not a very high dispersion. However, the kurtosis is quite high, which means that there are relatively many outliers. For this reason, some extreme values were removed from the original sample. In appendix A, I provide the complete descriptive statistics including skewness and kurtosis for the original sample (before removal of extremes) and the adjusted sample used in this paper.

N Range Minimum Maximum Mean

Std. Deviation CoD 1950 ,6809 ,0015 ,6824 ,056249 ,0387863 FVL3 1950 ,9739 ,0000 ,9739 ,080144 ,2115883 FVL2 1950 ,9282 0,0000 ,9282 ,126855 ,2086337 FVL1 1950 ,9392 0,0000 ,9392 ,051201 ,0927040 Fsize 1950 11,5557 3,4446 15,0003 9,151471 1,9381624 Lev 1950 ,9891 ,0041 ,9933 ,316131 ,2072241 ROA 1950 1,6162 -,5805 1,0356 ,029856 ,0631378 Dcov 1950 57,8278 -8,5078 49,3200 4,252855 6,6408695 Evol 1950 ,3533 ,0010 ,3543 ,021042 ,0299317

(24)

24 4.1.2 Correlation matrixes

In this section, the correlation matrix of cost of debt model sample is presented. The correlations are based on the Pearson test. The Pearson correlation measures the

correlation between different variables. The strength of the correlation between two

variables is illustrated by the coefficients. When high correlations exist, multicollinearity can occur, which may impair the explanatory power of the regression model.

The correlations of the variables in the cost of debt model sample are displayed in the table above. The correlations are relatively low on average. Negative correlations exist between firm size and fair value level 3 assets (- 0,371) and between firm size and earnings volatility (- 0,242). This implies that these bigger firms tend to have more stable earnings. Another negative correlation exist between debt coverage and leverage (- 0,355). This result seems logical because firms with low leverage should have less problems with paying off debt. Debt coverage and return on assets are positively correlated (0.362). This result also seems logical because firms with high returns should have less problems with paying off debt.

CoD FVL3 FVL2 FVL1 Fsize Lev ROA Dcov Evol

CoD 1,000 -,015 -,042 -,047 -,201 -,157 -,064 ,031 ,217 FVL3 -,015 1,000 -,073 -,058 -,371 ,143 ,057 -,055 ,091 FVL2 -,042 -,073 1,000 ,212 ,133 -,308 -,069 ,089 -,048 FVL1 -,047 -,058 ,212 1,000 ,106 -,250 -,025 ,039 -,005 Fsize -,201 -,371 ,133 ,106 1,000 -,060 ,036 -,119 -,242 Lev -,157 ,143 -,308 -,250 -,060 1,000 -,162 -,355 ,041 ROA -,064 ,057 -,069 -,025 ,036 -,162 1,000 ,362 -,048 Dcov ,031 -,055 ,089 ,039 -,119 -,355 ,362 1,000 ,095 Evol ,217 ,091 -,048 -,005 -,242 ,041 -,048 ,095 1,000 CoD ,249 ,031 ,018 ,000 ,000 ,002 ,086 ,000 FVL3 ,249 ,001 ,005 ,000 ,000 ,006 ,008 ,000 FVL2 ,031 ,001 ,000 ,000 ,000 ,001 ,000 ,016 FVL1 ,018 ,005 ,000 ,000 ,000 ,138 ,042 ,420 Fsize ,000 ,000 ,000 ,000 ,004 ,057 ,000 ,000 Lev ,000 ,000 ,000 ,000 ,004 ,000 ,000 ,035 ROA ,002 ,006 ,001 ,138 ,057 ,000 ,000 ,017 Dcov ,086 ,008 ,000 ,042 ,000 ,000 ,000 ,000 Evol ,000 ,000 ,016 ,420 ,000 ,035 ,017 ,000 Sig. (1-tailed)

Correlations - Cost of Debt Model

Pearson Correlation

(25)

25

Other negative correlations exist between fair value level 2 assets and leverage (- 0,308) and between fair value level 1 assets and leverage (- 0,250). These results can be partly explained because of characteristics of certain industries, such as pension funds, insurance companies and private equity funds, who tend to hold high amounts of fair value 1 and level 2 assets. These companies use relatively little debt.

4.1.3 Regression analysis

In this section, the results of the regression analysis of the cost of debt model are described and discussed. Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate Change Statistics

R Square Change F Change df1 df2 Sig. F Change

1 ,355 ,126 ,122 ,0363338 ,126 35,000 8 1941 ,000

The R-square and the adjusted R-square for this model are significant. This means that some of the independent variables have a significant impact on the outcome variable. This is illustrated by the ANOVA table, in which the regression sum of squares is compared with the residual sum of squares and the total sum of squares. The significant F-statistic proves that this model has explanatory power. This model has no major issues with

multicollinearity and the results for the different types of fair value assets investigated in this paper are not impaired by this issue.

ANOVA

Model Sum of Squares df Mean Square F Sig.

1

Regression ,370 8 ,046 35,000 ,000 Residual 2,562 1941 ,001

Total 2,932 1949

The proportion of fair value level 1, level 2 and level 3 assets of firms are negatively associated with Cost of debt. This means that cost of debt tend to decrease when firms hold a higher proportion of assets that are valued at fair value. Although the model does only partly explain the variances in cost of debt of firms, it seems that investors demand lower interest rates when asset values are measured at fair value. This implies, as illustrated by

(26)

26

Barth and Landsman (1995), that investors may consider information that fair value measurements hold about current and future cash flows useful for investment decisions.

These results also seem to hold for the less reliable fair value level 3 measurements. Investors seem to value the additional gain in accuracy more than the potential

disadvantages that these measurement methods hold in terms of reliability and

corresponding agency costs. Despite the controversy illustrated Khurana & Kim (2003), Glaser et al. (2013) and Xiaolu (2013), the regression result seem to be in favor of fair vale accounting.

This paper does not investigate specific industries, which may have different issues with fair value accounting. Khurana & Kim (2003) for example find that that historical cost measures of loans and deposits are more informative than fair values. In a similar context, Glaser et al. (2013) present evidence that the use of fair value assets with a higher degree of accounting discretion is a good predictor for default risk in banks. These results are

supported by Xiaolu (2013), who examines the relation between earnings volatility and fair value measurements. He finds that banks recognizing more level 3 fair value assets and liabilities report smoother earnings over the time.

The benefit of fair value accounting in other industry types may outweigh the

disadvantages more clearly as illustrated by Herrmann et al. (2008) who show that fair value measures for property, plant, and equipment are superior to historical cost based on the

Standardized Coefficients B

Std.

Error Beta Tolerance VIF

(Constant) ,107 ,005 20,800 ,000 FVL3 -,014 ,004 -,077 -3,287 ,001 ,829 1,207 FVL2 -,013 ,004 -,071 -3,082 ,002 ,859 1,164 FVL1 -,031 ,009 -,075 -3,376 ,001 ,908 1,102 Fsize -,004 ,000 -,187 -7,742 ,000 ,776 1,289 Lev -,045 ,005 -,240 -9,764 ,000 ,747 1,339 ROA -,041 ,014 -,066 -2,836 ,005 ,825 1,212 Dcov ,000 ,000 -,065 -2,643 ,008 ,739 1,354 Evol ,244 ,029 ,188 8,541 ,000 ,928 1,077 1

Coefficients - Cost of Debt Model

Model

Unstandardized Coefficients

t Sig.

(27)

27

characteristics of predictive value, feedback value, timeliness, neutrality, representational faithfulness, comparability, and consistency. This may be partly explained improved

disclosure requirements and guidance by the FASB. Other research, such as Webinger et al. (2013) shows that extra disclosures are helpful in these cases of illiquid or inactive markets. The overall results of this regression model suggest that fair value accounting might be a good alternative to historical cost accounting in most cases and the first hypothesis of this paper is therefore rejected.

The remaining control variables in the model show that bigger firms with high levels of assets, firms with a high return on assets and firms with high debt coverage ratios have lower cost of debt. Firms with high leverage also appear to have lower cost of debt. This can be partly explained by the fact that firms with better access to debt capital (against low interest rates) may be more inclined to finance themselves to a bigger extent with debt. The earnings volatility coefficient shows that higher earnings volatility leads to higher cost of debt. This result is expected, logically and consistent with prior literature, because investors require a risk premium for investments with higher risk.

4.2 Audit committee model

In this section, the descriptive statistics of the audit committee model sample are provided.

N Range Minimum Maximum Mean

Std. Deviation ACQ 786 164,0000 37,0000 201,0000 82,1460 21,2019 FVL3 786 ,3052 ,0001 ,3054 ,0136 ,0239 FVL2 786 ,8983 ,0000 ,8983 ,1842 ,2245 FVL1 786 ,6401 ,0000 ,6401 ,0491 ,0837 Empl 786 304,9870 ,0130 305,0000 18,5043 41,5821 Fsize 786 10,1459 4,4871 14,6330 9,3845 1,8339 Lev 786 ,8077 ,0000 ,8077 ,1998 ,1678 Loss 786 1,0000 ,0000 1,0000 ,1043 ,3059

(28)

28

The average fair value proportions are relatively low. This is especially the case with fair value level 3 assets. These values all have a high standard deviation, which means that a high dispersion between the values exists. This also results in a high skewness and kurtosis for the fair value level 1 and level 3 assets. The number of employees variable also has a very high standard deviation, skewness and kurtosis.

The outcome variable has a relatively low standard deviation, which means that there is not a very high dispersion between the values. This also did not result in a high skewness and kurtosis, although some extreme values were removed from the original sample. In appendix B, I provide the complete descriptive statistics including skewness and kurtosis for the original sample (before removal of extremes) and the adjusted sample used in this paper.

4.2.2 Correlation matrixes

In this section, the correlation matrix of the cost of debt model sample is presented. The correlations are based on the Pearson test. The Pearson correlation measures the correlation between different variables.

ACQ FVL3 FVL2 FVL1 Fsize Empl Lev Loss

ACQ 1,000 -,027 ,033 -,055 ,403 ,246 ,168 -,072 FVL3 -,027 1,000 ,108 ,157 -,045 ,072 -,078 -,035 FVL2 ,033 ,108 1,000 ,190 ,314 ,114 -,433 -,066 FVL1 -,055 ,157 ,190 1,000 ,029 ,069 -,199 -,011 Fsize ,403 -,045 ,314 ,029 1,000 ,543 ,090 -,157 Empl ,246 ,072 ,114 ,069 ,543 1,000 ,137 -,070 Lev ,168 -,078 -,433 -,199 ,090 ,137 1,000 ,177 Loss -,072 -,035 -,066 -,011 -,157 -,070 ,177 1,000 ACQ ,227 ,181 ,061 ,000 ,000 ,000 ,022 FVL3 ,227 ,001 ,000 ,102 ,022 ,014 ,166 FVL2 ,181 ,001 ,000 ,000 ,001 ,000 ,032 FVL1 ,061 ,000 ,000 ,212 ,027 ,000 ,378 Fsize ,000 ,102 ,000 ,212 ,000 ,006 ,000 Empl ,000 ,022 ,001 ,027 ,000 ,000 ,025 Lev ,000 ,014 ,000 ,000 ,006 ,000 ,000 Loss ,022 ,166 ,032 ,378 ,000 ,025 ,000

Correlations - Audit Committee Model

Pearson Correlation

Sig. (1-tailed)

(29)

29

The correlations of the variables in the cost of debt model sample are displayed in the table above. The correlations between the variables in this sample are quite low on average. Positive correlations exist between firm size (measured in terms of assets) and the number of employees (0,543), which is logical because firms with larger amounts of assets tend to have more employees. Firm size is also positively correlated with audit committee quality (0,403), which implies that bigger firms tend to have more advanced governance mechanisms and audit committees. Another positive correlation exist between leverage and accounting losses (0.177). This result also seems logical because firms with high leverage should have more fixed costs and therefore more risk to make accounting losses.

Other negative correlations exists between fair value level 2 assets and leverage (- 0,433) and between fair value level 1 assets and leverage (- 0,199). These results can be partly explained because of characteristics of certain industries, such as pension funds, insurance companies and private equity funds, who tend to hold more fair value level 1 and level 2 assets. These companies use relatively little debt.

4.2.3 Regression analysis

In this section, the results of the regression analysis of the audit committee model are described and discussed.

Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate Change Statistics R Square Change F Change df1 df2 Sig. F Change 1 ,430 ,184 ,177 19,2324858 ,184 25,143 7 778 ,000

The R-square and the adjusted R-square for this model are significant. This means that some of the independent variables have a significant impact on the outcome variable. This is illustrated by the ANOVA table, in which the regression sum of squares is compared with the residual sum of squares and the total sum of squares. The significant F-statistic proves that this model has explanatory power.

(30)

30

ANOVA

1

Regression 65099,577 7 9299,940 25,143 ,000 Residual 287773,262 778 369,889

Total 352872,839 785

This model has no major issues with multicollinearity. Although firm size is somewhat interrelated with the other independent variables. This result was expected, since firm size plays a major role in the dynamics of most other variables in this model. This fact does not impair the results for the different types of fair value assets investigated in this paper.

The coefficients in the table indicate no specific results to support reactions of companies governance mechanisms to any of the three types of fair value assets. It seems that company size has a significant positive influence on the quality of audit committees when measured in the form of total assets. This result does not hold when company size is measured by the number of employees, which also serves as a proxy for the level of complexity of firms. There are also no specific results that indicate that losses do have a significant impact on the quality of audit committees. Leverage does seem to have a positive influence on the quality of audit committees.

The importance of governance mechanisms are illustrated by Core et al. (1999), who find that firms with weaker corporate governance mechanisms have greater agency

Standardized Coefficients

B Std. Error Beta Tolerance VIF

(Constant) 38,230 4,334 8,822 ,000 FVL3 6,446 29,485 ,007 ,219 ,827 ,948 1,055 FVL2 -3,780 3,747 -,040 -1,009 ,313 ,666 1,502 FVL1 -10,097 8,537 -,040 -1,183 ,237 ,923 1,083 Fsize 4,500 ,488 ,389 9,230 ,000 ,589 1,697 Empl ,012 ,020 ,024 ,624 ,533 ,682 1,466 Lev 13,938 4,879 ,110 2,857 ,004 ,703 1,422 Loss -2,167 2,329 -,031 -,931 ,352 ,929 1,077 1

Coefficients - Audit Committee Model

Model Unstandardized Coefficients t Sig. Collinearity Statistics

(31)

31

problems. The potential for discretion and earnings management are illustrated by Xiaolu (2013) who finds that banks recognizing more level 3 fair value assets and liabilities report smoother earnings over the time. Other results of Glaser et al. (2013) present evidence that the use of fair value assets with a higher degree of accounting discretion is a good predictor for default risk in banks.

Evidence suggest that firms with more agency problems have stronger corporate governance mechanisms (Dey, 2008). It seems that firms that are more imposed to potential risks react to these settings by increasing the quality of their governance mechanisms. In the context of fair values, good corporate governance quality also seems to have an impact. Song et al. (2010) find that the value relevance of fair values is greater for firms with strong corporate governance. These results are the strongest in case of level 3 fair values.

Despite of the illustrated dynamics in the context of firms reactions to potential agency problems and risks, the regression results seem to implicate that the nature of assets do not significantly influence the audit committee quality characteristics, while the size of the firm (measured in total assets), as well the way it is financed seem to have a significant impact on the quality of audit committees. This last aspect can be explained by the

possibility that firms which high quality governance, in which audit committees form a key component, might be more able to attract larger amounts of debt. Another explanation might lie in characteristics of certain highly leveraged financial institutions, such as banks, which are exposed to stricter governance standards or expectations and therefore, require higher quality governance mechanisms.

The overall results of this regression model suggest that audit committee quality is not related with fair value assets and other characteristics of large firms, who have more resources available and bigger interests, in particular seem to be more influential.

(32)

32

4.3 Audit fees model

In this section, the descriptive statistics of the audit fees model sample are provided.

The average fair value proportions are relatively low. This is especially the case with fair value level 1 and level 3 assets. These values all have a high standard deviation, which means that a high dispersion between the values in the sample exists. This also results in a relatively high skewness and high kurtosis for the fair value level 1 and level 3 asset

proportions. Debt coverage and the number of employees also have high standard deviations, skewness and kurtosis.

The outcome variable has a relatively low standard deviation, which means that there is not a very high dispersion. This also did not result in a high skewness and kurtosis,

although some extreme values were removed from the original sample. In appendix C, I provide the complete descriptive statistics including skewness and kurtosis for the original sample (before removal of extremes) and the adjusted sample used in this paper.

N Range Minimum Maximum Mean Std. Deviation

AUFEE 1996 8,36800 10,42079 18,78879 14,42430 1,45415 FVL3 1996 0,97391 0,00002 0,97393 0,05319 0,16038 FVL2 1996 0,99411 0,00000 0,99411 0,17078 0,21807 FVL1 1996 0,87314 0,00000 0,87314 0,04359 0,08638 Fsize 1996 13,52898 1,31882 14,84780 8,81497 2,04528 Inv 1434 17,87844 -4,96185 12,91660 5,00225 2,39352 Rec 1974 16,68426 -2,95651 13,72775 6,81053 2,41992 Lev 1996 0,95283 0,00000 0,95283 0,24601 0,19983 Dcov 1996 73,58388 -24,52273 49,06116 3,42873 6,76235 Empl 1996 400,60000 0,00000 400,60000 16,04809 42,03429 Big4 1996 1 0 1 ,80 ,404 For 1996 1 0 1 ,24 ,425 Nctrl 1996 ,995438 0 ,995438 ,005058 ,046392 Loss 1996 1 0 1 ,17 ,378 Opi 1996 1 0 1 ,21 ,405 Res 1996 1 0 1 ,21 ,410 MA 1996 1 0 1 ,23 ,423

(33)

33 4.3.2 Correlation matrixes

In this section, the correlation matrix of the audit fees model sample is presented. The correlations are based on the Pearson test. The Pearson correlation measures the correlation between different variables. The correlations of the variables in the audit fees model sample are displayed in the table below.

The correlations are relatively low on average, but some variables are highly

interrelated. Positive correlations exist between firm size (measured in terms of total assets) and the number of employees (0,514), which is logical because firms with larger amounts of assets tend to have more employees. Firm size is also positively correlated with audit fees (0,818) and big four audit firms (0,483), which implies that the auditing of bigger firms tends to more expensive due to complexity and time effort, which is usually done by big 4

auditors. In addition to this, the number of employees is also correlated with audit fees (0,559). The correlation between complexity of auditing and audit fees is also illustrated by the positive correlation between audit fees with foreign income (0,379) and financial restructuring (0,355). Restructuring also has a positive correlation with foreign income (0,343), which implies that international firms have access to more capital markets and tend to do more restructurings.

Other significant correlations are inventories and receivables (0,471). These two variables are also related to firm size (0,707 for inventories and 0,825 for receivables). This relationship also hold for employees with inventories (0,473) and with receivables (0,428). Furthermore, a negative correlation exists between fair value level 2 and leverage (- 0,274) These results can be partly explained because of characteristics of certain industries, such as pension funds, insurance companies and private equity funds, who tend to hold more fair value 1 and level 2 assets. These companies use relatively little debt.

(34)

FVL3 -,074 1,000 ,032 ,062 -,208 -,032 -,228 ,106 -,077 -,015 -,057 ,005 ,045 ,117 ,058 -,026 -,050 FVL2 -,100 ,032 1,000 ,091 ,096 -,063 ,311 -,274 -,134 ,072 -,210 -,115 ,034 -,025 -,057 -,096 -,033 FVL1 ,227 ,062 ,091 1,000 ,135 ,163 ,002 -,016 ,066 ,124 ,126 ,183 ,242 ,044 ,021 ,164 ,039 Fsize ,818 -,208 ,096 ,135 1,000 ,707 ,825 ,184 -,067 ,514 ,483 ,089 ,039 -,184 ,089 ,155 ,099 Inv ,760 -,032 -,063 ,163 ,707 1,000 ,471 ,277 ,025 ,473 ,503 ,293 ,021 -,063 ,093 ,264 ,093 Rec ,529 -,228 ,311 ,002 ,825 ,471 1,000 -,115 -,131 ,428 ,192 -,011 ,008 -,191 ,004 ,041 ,083 Lev ,306 ,106 -,274 -,016 ,184 ,277 -,115 1,000 -,164 ,103 ,324 ,068 ,015 ,130 ,170 ,101 -,068 Dcov ,075 -,077 -,134 ,066 -,067 ,025 -,131 -,164 1,000 -,017 ,124 ,281 -,012 -,182 ,029 ,137 ,122 Empl ,559 -,015 ,072 ,124 ,514 ,473 ,428 ,103 -,017 1,000 ,192 ,195 -,008 -,086 ,058 ,203 ,059 Big4 ,622 -,057 -,210 ,126 ,483 ,503 ,192 ,324 ,124 ,192 1,000 ,242 ,015 -,060 ,111 ,230 ,051 For ,379 ,005 -,115 ,183 ,089 ,293 -,011 ,068 ,281 ,195 ,242 1,000 ,071 -,011 ,101 ,416 ,113 Nctrl ,082 ,045 ,034 ,242 ,039 ,021 ,008 ,015 -,012 -,008 ,015 ,071 1,000 ,064 ,029 ,081 ,004 Loss -,082 ,117 -,025 ,044 -,184 -,063 -,191 ,130 -,182 -,086 -,060 -,011 ,064 1,000 ,063 ,069 -,057 Opi ,142 ,058 -,057 ,021 ,089 ,093 ,004 ,170 ,029 ,058 ,111 ,101 ,029 ,063 1,000 ,058 ,028 Res ,355 -,026 -,096 ,164 ,155 ,264 ,041 ,101 ,137 ,203 ,230 ,416 ,081 ,069 ,058 1,000 ,126 MA ,160 -,050 -,033 ,039 ,099 ,093 ,083 -,068 ,122 ,059 ,051 ,113 ,004 -,057 ,028 ,126 1,000 AUFEE ,002 ,000 ,000 0,000 ,000 ,000 ,000 ,002 ,000 ,000 ,000 ,001 ,001 ,000 ,000 ,000 FVL3 ,002 ,113 ,010 ,000 ,110 ,000 ,000 ,002 ,282 ,016 ,432 ,045 ,000 ,015 ,159 ,029 FVL2 ,000 ,113 ,000 ,000 ,009 ,000 ,000 ,000 ,003 ,000 ,000 ,101 ,177 ,015 ,000 ,103 FVL1 ,000 ,010 ,000 ,000 ,000 ,464 ,278 ,006 ,000 ,000 ,000 ,000 ,050 ,211 ,000 ,073 Fsize 0,000 ,000 ,000 ,000 ,000 ,000 ,000 ,006 ,000 ,000 ,000 ,071 ,000 ,000 ,000 ,000 Inv ,000 ,110 ,009 ,000 ,000 ,000 ,000 ,177 ,000 ,000 ,000 ,217 ,008 ,000 ,000 ,000 Rec ,000 ,000 ,000 ,464 ,000 ,000 ,000 ,000 ,000 ,000 ,343 ,376 ,000 ,440 ,063 ,001 Lev ,000 ,000 ,000 ,278 ,000 ,000 ,000 ,000 ,000 ,000 ,005 ,281 ,000 ,000 ,000 ,005 Dcov ,002 ,002 ,000 ,006 ,006 ,177 ,000 ,000 ,256 ,000 ,000 ,328 ,000 ,140 ,000 ,000 Empl ,000 ,282 ,003 ,000 ,000 ,000 ,000 ,000 ,256 ,000 ,000 ,382 ,001 ,014 ,000 ,013 Big4 ,000 ,016 ,000 ,000 ,000 ,000 ,000 ,000 ,000 ,000 ,000 ,288 ,012 ,000 ,000 ,027 For ,000 ,432 ,000 ,000 ,000 ,000 ,343 ,005 ,000 ,000 ,000 ,004 ,338 ,000 ,000 ,000 Nctrl ,001 ,045 ,101 ,000 ,071 ,217 ,376 ,281 ,328 ,382 ,288 ,004 ,007 ,136 ,001 ,441 Loss ,001 ,000 ,177 ,050 ,000 ,008 ,000 ,000 ,000 ,001 ,012 ,338 ,007 ,009 ,004 ,016 Opi ,000 ,015 ,015 ,211 ,000 ,000 ,440 ,000 ,140 ,014 ,000 ,000 ,136 ,009 ,015 ,142 Res ,000 ,159 ,000 ,000 ,000 ,000 ,063 ,000 ,000 ,000 ,000 ,000 ,001 ,004 ,015 ,000 MA ,000 ,029 ,103 ,073 ,000 ,000 ,001 ,005 ,000 ,013 ,027 ,000 ,441 ,016 ,142 ,000 Correlation Sig. (1-tailed)

(35)

and discussed. Model Summary Model R R Square Adjusted R Square Std. Error of the Estimate Change Statistics R Square Change F Change df1 df2 Sig. F Change 1 ,929 ,864 ,862 ,5581651 ,864 559,295 16 1412 ,000

The R-square and the adjusted R-square for this model are highly significant. This means that some of the independent variables have a significant impact on the outcome variable. This is illustrated by the ANOVA table, in which the regression sum of squares is compared with the residual sum of squares and the total sum of squares. The significant F-statistic proves that this model has explanatory power.

ANOVA

1

Regression 2787,957 16 174,247 559,295 ,000b Residual 439,906 1412 ,312

Total 3227,864 1428

This model has some issues with multicollinearity because firm size, inventories and receivables are highly interrelated. This result is logical, but it might not be reasonable to draw valid conclusions about specific relations with regard to the interplay between these variables and their impact on the outcome variable. This will, however, not impair the results for the different types of fair value assets investigated in this paper.

(36)

36

The coefficients in the table indicate that firms with a bigger proportion of fair value level 3 assets have higher audit fees. This positive relation is also found with high

proportions of fair value level 1 assets. These results do not hold for fair value level 2 assets, who have a negative relation with audit fees.

The higher cost of accounting for exit values are illustrated by Benston (2008), who shows how that exit values, such as work-in-process, inventories, special-purpose machines and also assets and liabilities, have significantly higher costs than historical cost accounting. These extra costs can be explained by an increase in subjectivity as stated in the third

hypothesis. The regression results support this explanation, while audit fees tend to go down when less subjectivity is needed in the auditing of fair values. However, this explanation is not comprehensive for all the three fair value categories, because audit fees tend to go up when relatively more fair value level 1 assets are audited.

Standardized Coefficients

B

Std.

Error Beta Tolerance VIF

(Constant) 9,106 ,092 99,509 0,000 FVL3 1,514 ,319 ,050 4,747 ,000 ,878 1,140 FVL2 -,556 ,115 -,055 -4,832 ,000 ,750 1,333 FVL1 ,538 ,231 ,025 2,329 ,020 ,818 1,222 Fsize ,561 ,021 ,739 26,473 ,000 ,124 8,065 Inv ,066 ,010 ,105 6,672 ,000 ,388 2,579 Rec -,119 ,015 -,176 -7,660 ,000 ,182 5,495 Lev ,216 ,104 ,026 2,069 ,039 ,630 1,586 Dcov ,007 ,003 ,028 2,537 ,011 ,788 1,270 Empl ,004 ,000 ,132 10,966 ,000 ,670 1,492 Big4 ,514 ,048 ,140 10,798 ,000 ,570 1,754 For ,577 ,041 ,165 13,982 ,000 ,694 1,441 Nctrl 1,714 ,685 ,026 2,501 ,012 ,928 1,078 Loss ,155 ,043 ,038 3,639 ,000 ,879 1,137 Opi ,028 ,041 ,007 0,683 ,494 ,949 1,054 Res ,237 ,040 ,067 5,968 ,000 ,771 1,297 MA ,175 ,035 ,050 4,996 ,000 ,953 1,049 1

Coefficients - Audit Fees Model

Model

Unstandardized Coefficients

t Sig.

Fair value level 3 accounting : implications on auditing and governance