1
Amsterdam Business School
The impact of auditor’s independence on earnings
management
Evidence from the Netherlands
Name: Kees Hoogendoorn Student number: 5980755 Date: 12 August 2014 Master Thesis
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam First supervisor: Ir. drs. A.C.M de Bakker
2
Acknowledgement
In September 2008, I started with the premaster Accountancy & Control, as part of my job. After almost six years I finished this study with this master thesis. Three years ago I faced serious illness, with heavy complications. This illness had an enormous impact on my life. This illness also delayed finishing my master. But now being close to the finish, I’m very proud on myself having almost finished my master.
During the thesis writing process, I received valuable feedback from Toon de Bakker, for what I’m very grateful. I also want to thank my employers, which I served during my study, for giving me the time and opportunity to finish this thesis.
Finally I hope that everyone enjoys reading this thesis.
Schagen, August 2014
3
Abstract
In this research we examine whether auditor independence, proxied by auditors’ fees specifications, impairs audit quality, proxied by the degree of earnings management. This research is motivated by the fears of the AFM that the provision of non-audit services impairs auditors independence, which affects audit quality negatively. In this research, we do not find support for this fear. Auditors are more likely to protect their reputation. Our findings are consistent with many other papers regarding this subject. Non-audit fees seem to be mainly a perception problem.
4 Table of contents
Acknowledgement ... 2
Abstract ... 3
1. Introduction ... 6
2. Background literature & hypothesis ... 8
2.1 Frankel et al. (2002) ... 8
2.2 Antle et al. (2006) ... 9
2.3 Other papers regarding auditors’ fees and earnings management ... 9
2.4 Auditors’ fees and audit opinions ... 11
2.5 Auditors’ fees and value relevance ... 11
2.6 Audit quality ... 11
2.7 Summary of the background literature on auditors’ fees ... 13
2.8 Hypothesis development ... 13
3. Research design ... 14
3.1 Model specification Frankel et al. (2002) ... 14
3.1.1 Auditor’s fees specification ... 14
3.1.2 Earnings management specification ... 14
3.1.3 Control variables ... 15
3.1.4 Model ... 16
3.2 Model specification Antle et al. (2006) ... 16
3.2.1 Auditor’s fee specification ... 17
3.2.2 Earnings management specification ... 17
3.2.3 Control variables ... 17
3.2.4 Model ... 18
4. Data ... 20
4.1 Sample ... 20
4.2 Performing Modified Jones ... 21
4.2.1 Frankel et al. (2002)-model ... 21
4.2.2 Antle et al. (2006)-model ... 22
4.3 Stage regressions Antle et al. (2006)-model ... 22
4.4 Descriptive statistics ... 23
4.4.1 Frankel et al. (2002)-model ... 23
4.4.2 Antle et al. (2006)-model ... 24
4.5 Correlation ... 25
4.5.1 Frankel et al. (2002)-model ... 26
4.5.2 Antle et al. (2006)-model ... 27
5
5.1 Frankel et al. model ... 28
5.1.1 Results FEERATIO... 28
5.1.2 Results LOGAUD ... 28
5.1.3 Results LOGNON ... 29
5.1.4 Results LOGTOT ... 29
5.1.5 Summary findings Frankel et al. ... 30
5.2 Antle et al. model ... 30
5.2.1 Results LOGAF ... 30
5.2.2 Results LOGNAF ... 31
5.2.3 Results AA ... 32
5.2.4 Summary findings Antle et al. model ... 32
6. Conclusion ... 33
Literature list ... 35
Appendix 1: ... 37
6
1.
Introduction
In 2010 the AFM (Dutch Authority of Financial Markets, comparable with the SEC) presented a report in which they criticized the audit quality of the Big Four. They stated that a ‘fundamental change in behavior’ of auditors is necessary to improve audit quality. However, the AFM also stated that none of the issued opinions were wrong. One of the findings of the AFM was insufficient professional skepticism (AFM, 2010).
One of the criterions for professional skepticism is the independence of the auditor. In the wake of the Enron bankruptcy, concerns about auditor independence have prompted the American Congress to enact legislation that bans most auditor-provided non-audit services. Before this ban, the SEC required disclosure of audit fees and non-audit fees in the belief that such disclosures would attract the scrutiny of shareholders. In 2008, the Dutch law required such disclosures for Dutch companies.
Also in 2010, the European Commission presented a green paper about the role of the audit as well as the scope of the audit. Their reason for issuing this green paper is “the fact that numerous banks revealed huge losses from 2007 to 2009 on the positions they had held both on and off balance sheet raises not only the question of how auditors could give clean audit reports to their clients for those periods but also about the suitability and adequacy of the current legislative framework”. The Commission stated that “the independence of auditors should be the bedrock of the audit environment” (EC, 2010). In this paper, the Commission launches ideas to improve auditors’ independence, for instance auditors’ appointment by a third party (not being the client), mandatory rotation of the audit firm, and banning all auditor-provided non-audit services.
Getting ahead of the definitive proposals of the European Commission, the Dutch government agreed to new rules to improve auditors’ independence, from 2013 all auditor-provided non-audit services to listed companies are banned, and from 2016 all listed companies are required to rotate their audit firm each eight years.
Based on the issued report and green paper, some worries exist about audit quality and the independence of the auditor. Therefore the goal of this research is to examine the impact of auditors’ independence on audit quality. Because the report of the AFM only focuses on the Dutch audit market and such a research has never been performed in The Netherlands, the research will be focused on the Netherlands.
One of the fears of the European Commission and the SEC is that auditors’ provision of non-audit services has a negative impact on auditors’ independence. Because Dutch companies are required to disclose auditors’ fees since 2008, it is possible to use the non-audit fees related to
7
the total fees as a measure of auditors’ independence. Therefore the research question is the following:
Does the independence of the auditor impact audit quality?
The remainder of this thesis is organized as follows. In the second chapter we discuss background literature and we develop our hypotheses. In the third chapter we mention the research design. In the fourth chapter we present the data, for example descriptive statistics. In the fifth chapter we show the results of the regressions. We discuss these results and conclude in the sixth and last chapter.
8
2.
Background literature & hypothesis
The relationship between (non-)audit fees and earnings management has already been examined in previous research in other countries. As stated in the introduction, the SEC required disclosure of auditors’ fees in the USA in 2001. In countries like the UK, Australia and New Zealand such disclosure is required since the mid 90’s. In this chapter two researches are extensively mentioned, namely Frankel et al. (2002) and Antle et al. (2006), because in these researches a significant relationship has been found.
2.1 Frankel et al. (2002)
Frankel et al. (2002) examined the relationship between provision of audit services as well as auditors’ provision of non-audit services, and earnings management in the USA, short after the requirement of auditors’ fees disclosure. They examined both relationships, because Hansen and Watts (1997) and Reynolds and Francis (2000) argued that audit and non-audit fees create similar incentive effects. These effects are two-sided. If (non-)audit fees are high, auditors have an economic bond with the client so they admit earlier to client pressure to accept earnings management. On the other hand, if (non-)audit fees are high, auditors’ knowledge about the client and its industry will be high. So Frankel et al. (2002) tested their hypotheses stated in null form. They used four specifications of auditors’ fees: the ratio of non-audit fees to total fees (fee ratio), the rank in percentiles of audit fees, the rank in percentiles of non-audit fees, and the rank in percentiles of total fees. They also used three specifications of earnings management: small earnings surprises, small increases in earnings, and the absolute value of discretionary accruals. The results of their regressions are shown in table 1. In summary, they found a significant negative relationship between audit fees and earnings management. The relationship between non-audit fees and earnings management is positive and significant. There is no relationship between total fees and earnings management.
Table 1. Summary of the findings of Frankel et al. (2002)
Small earnings surprise Small increase in earnings Discretionary accruals
Fee ratio + 0 +
Ranked non-audit fees + 0 +
Ranked audit fees -/- -/- -/-
9 2.2 Antle et al. (2006)
Antle et al. (2006) criticized the models used in previous research, e.g. Frankel et al. (2002) and Ashbaugh et al. (2003). Antle et al. (2006) argued that prior research has estimated piece-meal the determinants of audit fees, non-audit fees and abnormal accruals. Antle et al. (2006) view audit fees, non-audit fees, and abnormal accruals as being jointly determined.
Antle et al. (2006) tested these models in the UK for the period 1994-2000 and in the USA for fiscal year 2000. The test of US firms was meant as a comparative analysis. The results of their UK regressions are shown the table 2. These results are contradictory with the results of Frankel et al. (2002), who found a negative impact of audit fees and a positive impact of non-audit fees on earnings management. The results of USA regressions are comparable with the UK regressions, although less significant.
Table 2. Summary of the findings of Antle et al. (2006) Audit fees Non-audit fees Abnormal accruals
Audit fees X -/- 0
Non-audit fees + X +
Abnormal accruals + -/- X
In both researches, a different measurement of abnormal/discretionary accruals has been used. Frankel et al. (2002) used discretionary accruals estimated with a cross-sectional modified Jones (1991) model. Antle et al. (2006) used a variation of the modified-Jones model that is advocated by Dechow et al. (1995). An interesting question will be whether the results of both researches will be different if they change their abnormal/discretionary accruals measurement.
2.3 Other papers regarding auditors’ fees and earnings management
Kinney and Libby (2002) discussed the research design and result interpretation of the paper of Frankel et al. (2002). They suggested that the conceptual relations between management and auditor incentives and earnings management are more complex than suggested in Frankel et al. (2002). Management incentives affect both the magnitude of earnings management attempts and auditor incentives to resist those attempts (Nelson et al. 2002). Kinney and Libby (2002) criticized the conclusion of Frankel et al. (2002) that the economic bond between auditor and client impaired auditors’ independence. This is contradictory to the negative relationship between audit fees and earnings management, and the lack of relationship between total fees and earnings management.
Ashbaugh et al. (2003) examined the relationship between auditors’ fees and meeting analyst forecasts in the USA. They found no relation between positive discretionary accruals and
10
any of the auditor fee metrics when discretionary accruals are adjusted for firm performance and sample firms are partitioned by income-increasing versus income-decreasing accruals. Second, in the earnings benchmark tests, they found no relation between fee ratio and the likelihood that firms beat analysts' forecasts. They further stated that the findings of Frankel et al. (2002) are sensitive to research design choices.
Kinney et al. (2004) examined the relationship between auditor fees and restatements of financial statements in the period 1995-2000. They found no relationship.
Larcker and Richardson (2004) examined the relationship between auditor fees, the choice of accruals, and corporate governance. They found a significant positive relationship between the ratio of non-audit fees to total fees and unexpected accruals. They found a negative relationship between other ratios of auditor fees and unexpected accruals, especially for firms with weak governance. Their results suggest that auditors are likely to protect their reputation.
Ruddock et al. (2006) examined the relationship between non-audit fees and the timeliness of recognizing bad news in Australia. Less timelier recognition of bad news also indicates earnings management. They consistently found that higher than expected levels of non-audit fees are not associated with reduced conservatism.
Gul et al. (2007) examined with the same research design choices as Ashbaugh et al. (2003) did, the relationship between non-audit fees, discretionary accruals and auditor tenure. They found that when auditor tenure is short, and non-audit fees are high, discretionary accruals are also high.
Cahan et al. (2008) examined the relationship between non-audit fee characteristics (like growth of non-audit fees and duration of non-audit services provision) and discretionary accruals, their measure for earnings management, in New Zealand. They found no significant relationship. Lin and Hwang (2010) examined which factors have a significant relationship with earnings management. They found out that auditor independence has a negative relationship with earnings management.
Gleason and Mills (2011) examined the relationship between auditor provided tax services and the adequacy and accuracy of the estimate of the tax reserves. They found a significant positive relationship between the adequacy and the accuracy of the tax reserves and auditor provided tax services, concluding that auditor provided tax services leads to a knowledge spillover, useful for the auditor.
11 2.4 Auditors’ fees and audit opinions
DeFond et al. (2002) examined the relationship between non-audit fees and going concern opinions, their indicator of auditors’ independence. They found no significant relationship. They interpreted from this finding that auditors won’t issue a wrong opinion, because of possible loss of reputation and litigation costs.
Craswell et al. (2002) examined the relationship between fee dependence and qualified opinions in Australia. Fee dependence is measured by the client fee related to the total fee of the audit firm. They found no significant relationship, consistent with the findings of DeFond et al. (2002).
Hay et al. (2006) examined the relationship between non-audit fees and modified or qualified opinions in New Zealand. Consistent with the findings of DeFond et al. (2002) and Craswell et al. (2002), they found no significant relationship.
2.5 Auditors’ fees and value relevance
Eilifsen and Knivsfla (2008) examined the relationship between non-audit fees and the responsiveness of annual stock market returns to reported earnings in Norway in the period 2003-2006. They found that annual stock market returns are less responsive to reported earnings when auditors’ provision of non-audit services to the reporting firms is relatively high. This was especially the case in 2003, a year with escalating scandals and severe criticism of the audit profession.
Gul et al. (2006) also examined the relationship between non-audit services and the value relevance of earnings in Australia. Value relevance of earnings is measured by the earnings response coefficient. Their results are consistent with the findings of Eilifsen and Knivsfla (2008).
Holland and Lane (2012) examined the value relevance of the disclosed non-audit fees in the UK. They found that shareholders perceive a threat of auditor’s independence at higher non-audit fees.
2.6 Audit quality
It’s difficult to assess audit quality ex ante because the only observable outcome of the audit is the audit report which is a generic template and the overwhelming majority of reports are standard clean opinions. While it is possible to assess audit quality ex post in the case of outright audit failures, these are relatively infrequent occurrences (Francis, 2004). However, in previous
12
research the quality of Big 4 firms related to smaller firms has been examined by focusing on audit outcomes. There are two observable audit outcomes examined in previous research, namely audit reports and audited financial statements. In previous research of Francis and Krishan (1999) the quality of audit reports is measured by using lower thresholds for issuing modified audit reports, which indicates greater reporting conservatism for a given set of client characteristics. Also in previous research, the level of abnormal accruals has been used as a measure of the quality of financial statements (Becker et al., 1998; Francis et al., 1999). Lower abnormal accruals imply less aggressive earnings management behaviour and therefore higher earnings quality. In previous research of Teoh and Wong (1993) used the degree of valuing earnings surprises by the stock market as a measure of the quality of financial statements as interpreted by the stock market.
In this research, the level of abnormal accruals will be used as a measure for audit quality. Because The Netherlands have a small stock market and the period of required disclosure of auditors’ fees is short, there will be only a few modified audit reports. Furthermore the AFM reported that none of the examined issued opinions were wrong. So this is not a good measure for this research. The degree of valuing earnings surprises by the stock market isn’t a good measure either, because this measures the interpretation of audit quality by the stock markets, not audit quality itself. The level of abnormal accruals is a good measure of audit quality, because auditing accruals requires professional judgement.
For examining whether there is a relationship between independence of the auditor and earnings management, it is important to know how to measure earnings management. Dechow et al. (1995) examined several models used to detect earnings management. All of these models focus on discretionary accruals. They found that a modified version of the model developed by Jones (1991) provided the most powerful tests of earnings management. This model provides the fewest Type II-errors. They also found that all researched models are potentially misspecified if the partitioning variable is correlated with firm performance. Beneish (1997) concluded that existing accrual models are not useful in studying incentives among firms with extreme financial performance. He presented an alternative model for detecting earnings management for firms with extreme financial performance. He added lagged total accruals and a measure of past price performance. His findings have implications for researchers investigating manager’s accrual decisions where extreme financial performance is likely, like security offerings or financial distress.
13
In this research, extreme financial performance is supposed to be a small part of the population, and isn´t likely for the whole population. Antle et al. (2006) found no relationship between auditors’ fees and earnings management, while they used the Modified Jones model.
2.7 Summary of the background literature on auditors’ fees
Prior research on the relationship between auditors’ fees and earnings management provided mixed results. Prior research on the relationship between auditors’ fees and audit opinions provided no evidence that there is a significant relationship. However, stock markets reactions are different for financial statements audited by an auditor who also provide non-audit services in comparison with financial statements which are only audited by the auditor (no non-audit services provided). This seems somewhat conflicting.
Francis (2006) also collected evidence about this matter, summarized the evidence and came to the conclusion that auditor provided non-audit services is mainly a perception problem. Namely that there is only a supposed relationship between (non-)audit fees and impaired audit quality. Only Frankel et al. (2002) found a positive relationship, this finding is possibly caused by research design choices. This subject has been investigated in much other countries, no one found “smoking gun” evidence.
2.8 Hypothesis development
Because of the ban on auditor-provided non-audit services, we test whether the provision of non-audit services is related to earnings management:
H1: Auditor provided non-audit services aren’t associated with earnings management The theory behind this supposed relationship is the economic bond between the auditor and his client. If this is the case, then the economic bond also has effect on the relationship between audit fees and earnings management:
H2: The provision of audit services isn’t associated with earnings management
Gleason (2011) found that non-audit fees leads to a knowledge spill over. Despite theories about economic bond between the auditor and client, there’s evidence that auditors are more likely to protect their reputation than admitting to client pressure (e.g. DeFond et al. (2002), Craswell et al. (2002)).
Because of the mixed results from previous researches, these hypotheses are tested in null form.
14
3.
Research design
In this chapter the research design is mentioned. Both hypotheses are tested with the models of Frankel et al. (2002) and Antle et al. (2006), because both researches reported a significant relationship, however the sign differed. This approach increases the robustness of the findings. First, the model specification of Frankel et al. (2002) is mentioned. Second, the model specification of Antle et al. (2006) is mentioned. The biggest difference between both models is that the Antle et al. (2006)-model is a simultaneous regression, the Frankel et al. (2002)-model is a linear regression. Another difference is the estimation of earnings management.
3.1 Model specification Frankel et al. (2002)
3.1.1 Auditor’s fees specification
Frankel et al. (2002) used four specifications of the auditor’s fees, namely FEERATIO, RANKAUD, RANKNON and RANKTOT. FEERATIO is the ratio of non-audit fees related to total fees. In Dutch annual statements, a distinction has been made between audit fees, other audit services, tax services and other services. In the calculation of the FEERATIO in this research other audit services, tax services and other services are classified as non-audit fees. The rank variables, RANKAUD, RANKNON and RANKTOT are percentile ranks of respectively audit fees, non-audit fees and total fees, per auditor. Because our sample is too small to make percentile ranks per auditor, some auditors have only one firm year, other specifications has been chosen. In several other researches, the logarithm of audit fees and non-audit fees has been chosen as a variable, like in researches of Antle et al. (2006) and DeFond et al. (2002). In our researches we use also these logarithms, calling them LOGAUD, LOGNON and LOGTOT.
3.1.2 Earnings management specification
Frankel et al. (2002) used three specifications for earnings management: absolute discretionary accruals, earnings surprise and small earnings increase. Earnings surprise is based on the real earnings versus market expectations. Because market expectations aren’t available for Dutch firms, this specification is excluded from this research. Because it is questionable whether there will remain enough firm years in the sample if small earnings increase is used as an earnings management specification. So small earnings increase is excluded from this research.
Frankel et al. (2002) used a discretionary accruals estimated with a cross-sectional modified Jones (1991) model. DACC (discretionary accruals) are calculated as follows:
15 DACC = TA - (α + β1(∆REV - ∆REC) + β2PPE),
where TA stands for total accruals, defined as net income less cash from operations. REV stands for revenue, REC stands for accounts receivable, PPE stands for property, plant & equipment. All variables are scaled by total assets at the beginning of the year. The estimates for α, β1 and β2
are obtained from the following regression at industry level: TA = α + β1∆REV + β2PPE + ε.
The industry level is identified by the first two digits of the SIC code. Due to limitations which are normal for a master thesis (e.g. reduction of time), the estimates for α, β1 and β2 are
obtained from the regression above for the total sample. Because there are also too many Dutch companies for some industries in a given year, it´s practically impossible to obtain these estimates by running this regression for each industry in a given year. The estimates are obtained by running the above mentioned regression for all firm years for which all for this regression necessary data are available in the period 2008-2012, this population consists of 489 firm years. Financial institutions (recognized by the SIC codes 6000-6999) are excluded, because of the unique way of estimating discretionary accruals for financial institutions. The results of this regression are shown in subparagraph 4.2.1.
3.1.3 Control variables
Frankel et al. (2002) used the following control variables: BIGFIVE is a dummy variable which is equal to 1 for companies that has been audited by a bigfive auditor and 0 otherwise. Because of the bankruptcy of Arthur Andersen these days there is only a bigfour, so we call it BIGFOUR. This variable has been added because prior research suggested that bigfour auditors are less likely to allow earnings management. AUDTEN stands for auditor tenure, this variable has been added because prior research suggested that auditor independence decrease in the length of auditor tenure. CFO stands for cash flow from operations deflated by total assets and is a measure for firm performance. ABSCFO stands for the absolute value of cash flow from operations, deflated by total assets and is also a measure for firm performance. Frankel et al. (2002) added two more measures for firm performance, because Dechow et al. (1995) noted that the Modified Jones model doesn’t completely extract non-discretionary accruals, so Frankel et al. (2002) added ACC (accruals deflated by total assets) and ABSACC (absolute accruals deflated by total assets) as a control variable. They also added LEVERAGE (ratio of total liabilities to total assets) as control variable, because prior research suggested that leverage is associated with
16
discretional accruals. LITRISK is a dummy variable which is equal to 1 for companies that operate in a high-risk industry, as identified by Francis et al. (1994), and 0 otherwise. M/B is the market-to-book-value, added as a proxy for growth. %INST stands for the percentage shares held by institutions, as reported by Spectrum. These data aren’t available for Dutch firms, so the variable has been excluded from the model. This variable is meant to control for institutional ownership, so we don’t control for this. LOSS is a dummy variable which is equal to 1 for companies who reported a loss and 0 otherwise. FIN/ACQ is a dummy variable which is equal to 1 for companies who made an acquisition of issued shares and 0 otherwise, because these circumstances often lead to higher (non-)audit fees. ANNRET is a variable for the percentage compounded monthly return in 2000 adjusted for the CRSP value weighted market index, added as a measure for firm performance. Because this variable is not available for Dutch companies in Datastream, we use ROA, return on assets (net income divided by average total assets) as an alternative measure for firm performance. LOGMVE is added as an indicator for firm size and stands for the logarithm of the market value of equity.
3.1.4 Model
Based on the previous subparagraphs, we estimate the following model:
ABSDACC = α + β1FEEVAR + β2BIGFOUR + β3AUDTEN + β4CFO +
β5ABSCFO + β6ACC + β7ABSACC + β8LEVERAGE + β9LITRISK +
β10M/B + β11LOSS + β12FIN/ACQ + β13ROA+ β14LOGMVE + ε
ABSDACC stands for the absolute value of discretionary accruals. FEEVAR stands for each of the auditor’s fee specifications, FEERATIO, LOGAUD, LOGNON and LOGTOT. So we run this regression four times. Refer to paragraph 5.1 for empirical results.
3.2 Model specification Antle et al. (2006)
Antle et al. (2006) view audit fees, non-audit fees and abnormal accruals as being jointly determined. As a result of this view, Antle et al. (2006) constructed three models and simultaneously regressed them. Because it’s impossible to run a simultaneous regression in SPSS, we run this simultaneous regression in two stages of linear regressions, following an approach described by the University of California at Berkeley (1998). We describe this process more detailed in Appendix 1.
17
3.2.1 Auditor’s fee specification
Antle et al. (2006) used only two auditor’s fee specifications, namely LOGAF and LOGNAF. LOGAF stands for the log on audit fees, LOGNAF stands for the log on non-audit fees. The classification of other audit services, tax services and other services as non-audit fees, is the same as in the Frankel et al. (2002)-model.
3.2.2 Earnings management specification
As stated in the introduction of this chapter, Frankel et al. (2002) and Antle et al. (2006) used another method to estimate the degree of earnings management. Antle et al. (2006) focused on working capital accruals, and estimated the following model:
WCA/TA = α(1/TA) + β((∆REV - ∆REC)/TA) + ε,
where WCA stands for current working capital accruals, defined as accounts receivable less accounts payable plus inventories. TA stands for total assets at the beginning of the year, REV stands for revenue, REC stands for accounts receivable.
This regression has to be run for each industry, due to the limited population we run this regression for all available firm years, only for the firm years which have the necessary data available, comparable to the estimation of earnings management in the Frankel et al. (2002)-model. This sample consists of 572 firm years. Financial institutions (recognized by the SIC codes 6000-6999) are excluded, because of the unique way of estimating abnormal accruals for financial institutions. By running this regression we obtain the estimates for α and β. With these estimates we can calculate the expected value of WCA/TA for a given firm year, then we calculate the real WCA/TA, and the difference are abnormal accruals. The results of the regression are shown in subparagraph 4.2.2.
3.2.3 Control variables
Antle et al. (2006) used the following control variables: FISDEC is a dummy variable which is equal to 1 for fiscal years ending in December and 0 otherwise, because firms with a fiscal year end in December are supposed to have higher audit fees, because most of the firms have a fiscal year end in December. LEV stands for leverage and is meant to capture agency costs and also a proxy for financial distress. QUICK stands for the quick ratio, defined as current assets less current liabilities. BMB stands for the market-to-book-value. LOSS is a dummy variable which is equal to 1 for firm years who reported a loss and 0 otherwise. LITI is a dummy variable which is equal to 1 for companies that operate in a high-risk industry, as identified by
18
Francis et al. (1994), and 0 otherwise. QUAL is a dummy variable which is equal to 1 for companies who received a qualified opinion and 0 otherwise. This variable is excluded from the models, because all firms in the sample received an unqualified opinion. LOGAR is the natural logarithm of accounts receivable. LOGINV is the natural logarithm of inventories. These logs are a proxy for complex audits, what often lead to higher audit fees. BIG is a dummy variable which is equal to 1 for companies who’s auditor is a big-four-member, and 0 otherwise. In the Antle et al. (2006) research there was a big six. RETAIN is a dummy variable which is equal to 1 for companies who have the new auditor for that firm year and 0 otherwise. BLOGTA is the natural logarithm of total assets at beginning of the year and is meant to control for firm size. LAA stands for lagged abnormal accruals. Because it is questionable whether there will remain enough firm years in the sample if LAA is added as a control variable, LAA is excluded from the models of Antle et al. (2006). LROA stands for lagged return on assets and is replaced by ROA, the actual return on total assets. LROA and LAA are meant to control for past firm performance. For the non-audit fees equation the following control variables are added, LOGTAX stands for the natural logarithm of income taxes, because this is a proxy for the demand of tax advisory services. SOCF stands for cash flow from operations, scaled by total assets at beginning of the year and is meant to control for operational performance.
3.2.4 Model
Based on the previous subparagraphs, we estimate the following system of simultaneous equations:
LOGAF = αAF0 + αAF1LOGNAF + αAF2AA + αAF3FISDEC + αAF4LEV +
αAF5QUICK + αAF6BMB + αAF7LOSS + αAF8LITI + αAF9LOGAR +
αAF10LOGINV + αAF11BIG + αAF12RETAIN + αAF13BLOGTA +
αAF14ROA + εAF
LOGNAF = αNAF0 + αNAF1LOGAF + αNAF2AA + αNAF3LOGTAX + αNAF4LEV +
αNAF5QUICK + αNAF6BMB + αNAF7LOSS + αNAF8LITI + αNAF9ROA +
αNAF10LOGINV + αNAF11BIG + αNAF12RETAIN + αNAF13BLOGTA +
εNAF
AA = αAA0 + αAA1LOGAF + αAA2LOGNAF + αAA3SOCF + αAA4BLOGTA +
αAA5QUICK + αAA6BMB + αAA7LOSS + αAA8LITI + αAA9ROA +
19
AA stands for abnormal accruals. As stated in the introduction of this paragraph, the system of equations of Antle et al. (2006) are simultaneously equations, which cannot be run in SPSS. To run this system of equations in SPSS, estimates of LOGAF, LOGNAF and AA has to be obtained. The way of obtaining these estimates is mentioned in appendix 1.
20
4.
Data
In this chapter we mention the data gathering process. We start with a description of the sample. Then we perform the Modified Jones-models. Then we perform the estimations of the Antle et al. (2006)-model. After this, we present descriptive statistics. In the end, we test whether there is correlation between the variables used in the regression models.
4.1 Sample
We collected the data for the control variables from Datastream. We collected data for Dutch firms in the period 2008-2012, because the disclosure on auditor’s fees became required in 2008. Financial institutions are excluded from the research, because of their unique way to
estimate accruals. The determination of whether firms are financial institutions is based on the general industry classification in Datastream. In this period 2.161 firm years are available, only 290 firm years are non-financial institutions for which all required control variables are available for the models of Antle et al. (2006) and Frankel et al. (2002). For these 290 firm years we collected the fee data manually from annual reports, for 36 firm years the data aren´t disclosed in their annual report, for 48 firm years there is no annual report found, due to several reasons. In the end, for 206 firm years complete data sets are available. For distribution per year and per industry, refer to tables 3 and 4.
Table 3. Distribution per year
Year N % 2008 47 23% 2009 20 10% 2010 21 10% 2011 56 27% 2012 62 30% Total 206 100%
21
Table 4. Distribution per industry
Industry N % SIC codes
Agriculture 1 0% 0100-0999
Mining and constrution 23 11% 1000-1999, excluding 1300-1399
Food 15 7% 2000-2111
Textiles and printing/publishing 20 10% 2200-2799
Chemicals 11 5% 2800-2899
Extractive 5 2% 1300-1399
Durable manufacturers 50 24% 3000-3999, excluding 3670-3679
Transportation 15 7% 4000-4899
Retail 17 8% 5000-5999
Services 22 11% 7000-8999, excluding 7370-7379
Computers 27 13% 3670-3679, 7370-7379
Total 206 100%
The industry classification is determined by the SIC codes. We compared the general industry classification in Datastream with the SIC codes, the SIC codes 6000-6999 belong to financial institutions. In the final sample, 6 firm years have a SIC code between 6000 and 6999. We assessed these firm years by reading the annual reports. We noted that the core activities of these six companies have nothing to do with banking and insurance. We also noted that these six companies have another industry classification in other years.
4.2 Performing Modified Jones
4.2.1 Frankel et al. (2002)-model
Following subparagraph 3.1.2 we determine the estimates for α, β1 and β2, the results are
shown in the table below, table 5.
Table 5. Coefficients Modified Jones Frankel et al. (2002)
B t-value p-value. (Constant) -5,380 -8,122 ,000** ∆REV 1,469 4,163 ,000** PPE 10,145 12,419 ,000** Adjusted R Square ,284 F 97,669 ,000** N 489
** significant at the 0,01 level
* significant at the 0,05 level
22
4.2.2 Antle et al. (2006)-model
Following subparagraph 3.2.2 we determine the estimates for α and β, the results are shown in the table on the next page, table 6. The regression resulted in values for α and β of -7,814 and -0,0003.
Table 6. Coefficients Modified Jones Antle et al. (2006)
B t-value p-value (Constant) 0,196 21,806 ,000** (1/TA) -7,814 -1,771 ,077 ((∆REV - ∆REC)/TA) -0,0003 -0,540 ,590 Adjusted R Square ,002 F 1,573 0,208 N 572
** significant at the 0,01 level
* significant at the 0,05 level 4.3 Stage regressions Antle et al. (2006)-model
In this paragraph, we perform the first stage regressions of the Antle et al. (2006)-model, we also test the data for multicollinearity. We test whether a group of variables correlate with other variables. We test this with the variance influance factor (VIF). In general a VIF higher than 5 decreases the reliability of a model.
To obtain the estimates mentioned in appendix 1, we run the following linear regression models.
LOGAF = αAF0 + αAF3FISDEC + αAF4LEV + αAF5QUICK + αAF6BMB +
αAF7LOSS + αAF8LITI + αAF9LOGAR + αAF10LOGINV + αAF11BIG +
αAF12RETAIN + αAF13BLOGTA + αAF14ROA + αAF15LOGTAX +
αAF16SOCF + εAF
LOGNAF = αNAF0 + αNAF3FISDEC + αNAF4LEV + αNAF5QUICK + αNAF6BMB +
αNAF7LOSS + αNAF8LITI + αNAF9LOGAR + αNAF10LOGINV +
αNAF11BIG + αNAF12RETAIN + αNAF13BLOGTA + αNAF14ROA +
αNAF15LOGTAX + αNAF16SOCF + εNAF
AA = αAA0 + αAA3FISDEC + αAA4LEV + αAA5QUICK + αAA6BMB +
αAA7LOSS + αAA8LITI + αAA9LOGAR + αAA10LOGINV + αAA11BIG +
αAA12RETAIN + αAA13BLOGTA + αAA14ROA + αAA15LOGTAX +
23
Untabulated results show that LOGAR has a VIF of 8,54 and BLOGTA has a VIF of 9,7. So we run new regressions without BLOGTA. Refer to table 7 for the results. Now all VIF´s are below the threshold of 5. With these results we calculate the estimations LOGAF1, LOGNAF1 and AA1. With these estimations we run the final regressions, refer to paragraph 5.2.
4.4 Descriptive statistics
4.4.1 Frankel et al. (2002)-model
Refer to table 8 for the descriptive statistics of the variables used in the Frankel et al. (2002)-model. Frankel et al. (2002) reported a mean (standard deviation) of the FEERATIO of 0,49 (0,24). Frankel et al. (2002) reported a mean of ABSDACC of 0,15. The difference with our mean of ABSDACC (4,327) can possibly be explained by the fact that with we performed the Modified Jones-model for the whole sample instead of for each industry. We noted that high values of ABSDACC are not driven by one specific industry. 188 firm years have a big four auditor, 41 firm years operate in a high litigation risk industry, 43 firm years reported a loss, 156 firm years did an acquisition or issued shares. The mean audit fee of our sample is 2.122.000 euros. The mean audit fee of the Frankel et al. (2002)-sample is 511.000 US dollars. The mean
Table 7. Coefficients first stage estimations
LOGAF1 LOGNAF1 AA1
B t-value p-value B t-value p-value B t-value p-value VIF (Constant) -0,240 -0,885 0,377 -0,476 -0,605 0,546 0,251 1,373 0,171 FISDEC -0,043 -0,711 0,478 0,025 0,141 0,888 0,000 -0,005 0,996 1,170 LEV 0,347 2,545 0,012* -0,418 -1,057 0,292 0,079 0,857 0,392 1,984 QUICK 0,010 0,338 0,736 -0,156 -1,873 0,063 0,071 3,653 0,000** 1,513 BMB 0,004 7,055 0,000** 0,002 1,438 0,152 0,000 -0,777 0,438 1,989 LOSS 0,105 1,929 0,055 0,200 1,270 0,205 -0,070 -1,897 0,059 1,304 LITI -0,045 -0,884 0,378 -0,165 -1,113 0,267 -0,006 -0,167 0,868 1,107 LOGAR 0,412 9,948 0,000** 0,689 5,732 0,000** -0,034 -1,206 0,229 4,820 LOGINV 0,011 0,788 0,431 -0,004 -0,104 0,917 0,015 1,563 0,120 1,538 BIG -0,104 -1,168 0,244 -0,462 -1,786 0,076 0,127 2,107 0,036* 1,695 RETAIN -0,005 -0,024 0,981 -0,519 -0,890 0,375 -0,012 -0,090 0,929 1,035 ROA 0,003 0,121 0,903 -0,009 -0,127 0,899 0,003 0,203 0,839 1,078 LOGTAX 0,233 7,143 0,000** 0,127 1,338 0,183 -0,038 -1,709 0,089 3,817 SOCF -0,560 -1,856 0,065 0,461 0,526 0,599 -0,199 -0,976 0,330 2,274 Adjusted R Square ,847 ,442 ,118 F 88,351 0,000** 13,481 0,000** 3,115 0,000** N 206 206 206
** significant at the 0,01 level * significant at the 0,05 level
24
non-audit fee of our sample is 835.000 euro’s. The mean non-audit fee of the Frankel et al. (2002)-sample is 1.258.000 US dollars. The mean total fee of our sample is 3.045.000 euro’s. The mean total fee of the Frankel et al. (2002)-sample is 1.769.000 US dollars.
Table 8. Descriptive statistics Frankel et al. (2002)-model
N Minimum Maximum Mean Std. Deviation
ABSDACC 206 ,01 25,76 4,3270 4,38226 FEERATIO 206 0,00 ,81 ,2472 ,20091 LOGAUD 206 1,40 4,34 2,7783 ,71041 LOGNON 206 0,00 4,08 2,0575 1,07891 LOGTOT 206 1,48 4,45 2,9216 ,71940 BIGFOUR 206 0,00 1,00 0,9126 0,28308 AUDTEN 206 1,00 43,00 40,7379 5,04122 CFO 206 -,48 ,36 ,0853 ,08540 ABSCFO 206 ,00 ,48 ,0992 ,06855 ACC 206 -2,11 10,93 -,0103 ,82137 ABSACC 206 ,00 10,93 ,1827 ,80075 LEVERAGE 206 ,00 103,24 1,7444 9,87661 LITRISK 206 0,00 1,00 ,1990 ,40024 MB 206 -24,53 673,45 4,6589 46,93590 LOSS 206 0,00 1,00 ,2087 ,40740 FINACQ 206 0,00 1,00 ,7573 ,42977 ROA 206 -2,09 11,63 ,0779 ,86000 LOGMVE 206 3,36 7,87 5,6089 ,95247
4.4.2 Antle et al. (2006)-model
Refer to table 9 for the descriptive statistics of the variables used in the Antle et al. (2006)-model. Antle et al. (2006) reported a mean (standard deviation) of audit fees of 452.000 pounds (1.263.000). Antle et al. (2006) reported a mean (standard deviation) of non-audit fees of 500.000 pounds (2.098.000). Antle et al. (2006) reported a mean (standard deviation) of AA of 0,01 (0,062). 177 firm years have a fiscal year end in December. Only 2 firm years changed their auditor in the given year.
25
Table 9. Descriptive statistics Antle et al. (2006)-model
N Minimum Maximum Mean
Std. Deviation LOGAF 206 1,40 4,34 2,7783 0,71041 LOGNAF 206 0,00 4,08 2,0575 1,07891 AA 206 -0,26 1,97 0,2173 ,19968 FISDEC 206 0,00 1,00 0,8592 0,34864 LEV 206 -0,05 1,31 0,5752 ,20067 QUICK 206 0,23 7,29 1,3895 ,83375 BMB 206 -24,53 673,45 4,6589 46,93590 LOSS 206 ,00 1,00 ,2087 ,40740 LITI 206 ,00 1,00 ,1990 ,40024 LOGAR 206 0,00 6,82 4,9021 1,02830 LOGINV 206 ,00 7,35 4,2966 1,73237 BIG 206 ,00 1,00 0,9126 0,28308 RETAIN 206 0,00 1,00 ,9903 ,09829 BLOGTA 206 1,40 7,90 5,8063 0,98498 ROA 206 -2,09 11,63 ,0779 ,86000 LOGTAX 206 0,00 6,27 3,8212 1,16085 SOCF 206 -0,72 0,38 ,0877 ,09701 LOGAF1 206 1,16 4,14 2,7783 ,65758 LOGNAF1 206 -0,3783 10,2353 2,101 ,95199 AA1 206 -518,58 26,0757 -1,0783 36,60333 4.5 Correlation
In this paragraph we test whether the variables used correlate with each other. We test this with the Pearson test. In general a Pearson correlation higher (lower) than 0,7 (-0,7) decreases the reliability of a model. We only mention the highest Pearson correlations, the Pearson correlations which are outside the mentioned threshold.
26
4.5.1 Frankel et al. (2002)-model
Refer to table 10 for the Pearson values for the variables used in the Frankel et al. (2002)-model. We only show the most important Pearson values. All other Pearson values are between 0,7 and -0,7. We found Pearson correlations between LOGAUD on the one hand and LOGNON (0,703) and LOGTOT (0,98) on the other hand. And we found a Pearson correlation between LOGNON and LOGTOT (0,801). These correlations are significant at the 0,01 level (2-tailed). These correlations aren’t a problem because these variables aren’t used in the same regression. We also found Pearson correlations between LOGMVE, our indicator for firm size, on the one hand and LOGAUD (0,863) and LOGTOT (0,849) on the other hand. Both correlations are significant at the 0,01 level (2-tailed). We found Pearson correlations between ROA on the one hand and ACC (0,994) and ABSACC (0,836) on the other hand. Both correlations are significant at the 0,01 level (2-tailed). We also found a Pearson correlation between ACC and ABSACC (0,825), significant at the 0,01 level (2-tailed). This correlation can be explained by the fact that ABSACC is an absolute value of ACC.
Table 10. Pearson values Frankel et al. (2002)-model
ABSDACC FEERATIO LOGAUD LOGNON LOGTOT CFO ABSCFO ACC ABSACC ROA LOGMVE ABSDACC Pearson Correlation 1 ,071 -,360** -,200** -,339** ,070 ,128 -,037 -,008 -,029 -,327**
Sig. (2-tailed) ,310 ,000 ,004 ,000 ,315 ,067 ,599 ,907 ,679 ,000 FEERATIO Pearson Correlation ,071 1 ,013 ,615** ,210** ,055 -,097 -,035 -,009 -,029 ,023
Sig. (2-tailed) ,310 ,848 ,000 ,002 ,436 ,167 ,618 ,903 ,676 ,742 LOGAUD Pearson Correlation -,360** ,013 1 ,703** ,980** -,097 -,179* -,065 -,206** -,077 ,863**
Sig. (2-tailed) ,000 ,848 ,000 ,000 ,166 ,010 ,352 ,003 ,270 ,000 LOGNON Pearson Correlation -,200** ,615** ,703** 1 ,801** ,030 -,215** -,047 -,127 -,044 ,645**
Sig. (2-tailed) ,004 ,000 ,000 ,000 ,671 ,002 ,506 ,069 ,529 ,000 LOGTOT Pearson Correlation -,339** ,210** ,980** ,801** 1 -,085 -,191** -,073 -,206** -,084 ,849**
Sig. (2-tailed) ,000 ,002 ,000 ,000 ,224 ,006 ,299 ,003 ,232 ,000 CFO Pearson Correlation ,070 ,055 -,097 ,030 -,085 1 ,478** -,031 -,009 ,072 ,182**
Sig. (2-tailed) ,315 ,436 ,166 ,671 ,224 ,000 ,657 ,898 ,305 ,009 ABSCFO Pearson Correlation ,128 -,097 -,179* -,215** -,191** ,478** 1 -,056 ,012 -,002 -,051
Sig. (2-tailed) ,067 ,167 ,010 ,002 ,006 ,000 ,422 ,866 ,979 ,465 ACC Pearson Correlation -,037 -,035 -,065 -,047 -,073 -,031 -,056 1 ,825** ,994** -,075
Sig. (2-tailed) ,599 ,618 ,352 ,506 ,299 ,657 ,422 ,000 ,000 ,285 ABSACC Pearson Correlation -,008 -,009 -,206** -,127 -,206** -,009 ,012 ,825** 1 ,836** -,221**
Sig. (2-tailed) ,907 ,903 ,003 ,069 ,003 ,898 ,866 ,000 ,000 ,001 ROA Pearson Correlation -,029 -,029 -,077 -,044 -,084 ,072 -,002 ,994** ,836** 1 -,060
Sig. (2-tailed) ,679 ,676 ,270 ,529 ,232 ,305 ,979 ,000 ,000 ,388 LOGMVE Pearson Correlation -,327** ,023 ,863** ,645** ,849** ,182** -,051 -,075 -,221** -,060 1
Sig. (2-tailed) ,000 ,742 ,000 ,000 ,000 ,009 ,465 ,285 ,001 ,388 **. Correlation is significant at the 0.01 level (2-tailed).
27 4.5.2 Antle et al. (2006)-model
Refer to table 11 for the Pearson values for the variables used in the Antle et al. (2006)-models. We only show the most important Pearson values. All other Pearson values are between 0,7 and -0,7. We found a Pearson correlation between LOGAF and LOGNAF (0,703), significant at the 0,01 level (2-tailed). This correlation isn’t a problem, because these variables aren’t used in the same regression. We also found a Pearson correlation between LOGAF and LOGAR (0,822). We found Pearson correlations between BLOGTA on the one hand and
LOGAF (0,861) and LOGNAF (0,704) on the other hand. Both correlations are significant at the 0,01 level (2-tailed). We found Pearson correlations between LOGAF on the one hand and LOGTAX (0,791) and LOGAF1 (0,926) on the other hand. Both correlations are significant at the 0,01 level (2-tailed). Both correlations aren’t a problem, because in the equation with LOGAF as dependent variable, the correlating variables aren’t used. LOGAF1 correlates with LOGAR (0,888), BLOGTA (0,86) and LOGTAX (0,855). All these correlations are significant at the 0,01 level (2-tailed). The correlations with LOGAR and BLOGTA aren’t a problem, because these variables aren’t used in the estimation of LOGAF1.
Table 11. Pearson values Antle et al. (2006)-model
LOGAF LOGNAF LOGAR BLOGTA LOGTAX LOGAF1 LOGNAF1 LOGAF Pearson Correlation
1 ,703** ,822** ,861** ,791** ,926** ,646**
Sig. (2-tailed) ,000 ,000 ,000 ,000 ,000 ,000 LOGNAF Pearson Correlation
,703** 1 ,657** ,704** ,574** ,643** ,546**
Sig. (2-tailed) ,000 ,000 ,000 ,000 ,000 ,000 QUICK Pearson Correlation
-,203** -,186** -,163* -,222** -,138* -,219** -,236**
Sig. (2-tailed) ,003 ,007 ,019 ,001 ,047 ,002 ,001 LOGAR Pearson Correlation
,822** ,657** 1 ,926** ,802** ,888** ,720**
Sig. (2-tailed) ,000 ,000 ,000 ,000 ,000 ,000 BLOGTA Pearson Correlation
,861** ,704** ,926** 1 ,832** ,860** ,689**
Sig. (2-tailed) ,000 ,000 ,000 ,000 ,000 ,000 LOGTAX Pearson Correlation
,791** ,574** ,802** ,832** 1 ,855** ,643**
Sig. (2-tailed) ,000 ,000 ,000 ,000 ,000 ,000 LOGAF1 Pearson Correlation
,926** ,643** ,888** ,860** ,855** 1 ,699**
Sig. (2-tailed) ,000 ,000 ,000 ,000 ,000 ,000 LOGNAF1 Pearson Correlation
,646** ,546** ,720** ,689** ,643** ,699** 1
Sig. (2-tailed) ,000 ,000 ,000 ,000 ,000 ,000 **. Correlation is significant at the 0.01 level (2-tailed).
28
5.
Empirical results
5.1 Frankel et al. model
5.1.1 Results FEERATIO
Refer to table 12 for the results on the FEERATIO-equation. The most important finding of this regression is the fact that FEERATIO isn’t a significant predictor of ABSDACC, contradictory to the results of Frankel et al. (2002), who found a significant positive relationship. Only ABSACC, LITRISK and LOGMVE are significant negative predictors. The negative relationship of ABSACC with ABSDACC seems somewhat strange, because this says that higher accruals reduce earnings management.
Table 12. Results ABSDACC (FEERATIO) Table 13. Results ABSDACC (LOGAUD)
B t-value p-value B t-value p-value
(Constant) 13,103 4,165 ,000** (Constant) 13,051 4,265 ,000** FEERATIO 1,134 ,785 ,433 LOGAUD -2,284 -2,345 ,020* BIGFOUR -1,701 -1,248 ,213 BIGFOUR -1,861 -1,382 ,169 AUDTEN ,023 ,406 ,685 AUDTEN -,008 -,137 ,891 CFO 11,760 ,966 ,335 CFO 9,464 ,786 ,433 ABSCFO -,638 -,079 ,937 ABSCFO -4,927 -,616 ,538 ACC -1,415 -,155 ,877 ACC -3,040 -,335 ,738 ABSACC -1,973 -2,240 ,026* ABSACC -2,236 -2,547 ,012* LEVERAGE ,026 ,873 ,384 LEVERAGE ,034 1,154 ,250 LITRISK -2,252 -2,936 ,004** LITRISK -2,448 -3,240 ,001** M/B ,009 ,756 ,450 M/B ,015 1,328 ,186 LOSS 1,489 1,818 ,071 LOSS 1,970 2,368 ,019 FIN/ACQ -1,389 -1,919 ,057 FIN/ACQ -1,225 -1,705 ,090 ROA 2,635 ,290 ,772 ROA 4,406 ,489 ,626 LOGMVE -1,449 -3,718 ,000** LOGMVE ,041 ,054 ,957
Adjusted R Square ,169 Adjusted R Square ,189
F 3,973 0,000** F 4,422 0,000**
N 206 N 206
** significant at the 0,01 level ** significant at the 0,01 level * significant at the 0,05 level * significant at the 0,05 level
5.1.2 Results LOGAUD
Refer to table 13 for the results on the LOGAUD-equation. The most important finding of this regression is that LOGAUD has a significant negative relationship with ABSDACC, this means the higher the log of audit fees, the lower the earnings management, which is not really surprising. This finding is consistent with the results of Frankel et al. (2002). ABSACC (negative), LOSS (positive) and LITRISK (negative) are also significant.
29
5.1.3 Results LOGNON
Refer to table 14 for the results on the LOGNON-equation. LOGNON has no significant relationship with ABSDACC, contradictory to the results of Frankel et al. (2002), who found a significant positive relationship. Only ABSACC, LITRISK and LOGMVE have a significant negative relationship.
Table 14. Results ABSDACC (LOGNON) Table 15. Results ABSDACC (LOGTOT)
B t-value p-value B t-value p-value
(Constant) 13,520 4,353 ,000** (Constant) 13,627 4,437 ,000** LOGNON -0,062 -,170 ,865 LOGTOT -1,577 -1,742 ,083 BIGFOUR -1,733 -1,262 ,208 BIGFOUR -1,758 -1,298 ,196 AUDTEN ,019 ,344 ,731 AUDTEN ,000 -,009 ,993 CFO 12,785 1,054 ,293 CFO 11,420 ,947 ,345 ABSCFO -1,854 -,228 ,820 ABSCFO -4,643 -,573 ,567 ACC -1,334 -,145 ,885 ACC -2,249 -,247 ,805 ABSACC -1,965 -2,221 ,028* ABSACC -2,115 -2,403 ,017* LEVERAGE ,031 1,050 ,295 LEVERAGE ,038 1,268 ,206 LITRISK -2,325 -3,033 ,003** LITRISK -2,473 -3,239 ,001** M/B ,009 ,837 ,404 MB ,014 1,216 ,225 LOSS 1,528 1,849 ,066 LOSS 1,833 2,198 ,029* FIN/ACQ -1,401 -1,927 ,055 FINACQ -1,264 -1,749 ,082 ROA 2,540 ,278 ,781 ROA 3,535 ,390 ,697 LOGMVE -1,414 -2,953 ,004** LOGMVE -,456 -,656 ,512
Adjusted R Square ,166 Adjusted R Square ,179
F 3,919 0,000** F 4,195 0,000**
N 206 N 206
** significant at the 0,01 level ** significant at the 0,01 level * significant at the 0,05 level * significant at the 0,05 level
5.1.4 Results LOGTOT
Refer to table 15 for the results on the LOGTOT-equation. LOGTOT has no significant relationship with ABSDACC, consistent with the results of Frankel et al. (2002). Only ABSACC (negative), LITRISK (negative) and LOSS (positive) has a significant relationship.
30
5.1.5 Summary findings Frankel et al.
The results of the findings from the Frankel et al. model are summarized in table 16. Only evidence has been found on H2, namely a significant negative relationship between (the log of) audit fees and discretionary accruals. All other auditor fee specifications lacked a significant relationship and therefore they do not support one of hypotheses.
Table 16. Summary of the findings of the Frankel et al. model Discretionary accruals
Fee ratio 0
Log on non-audit fees (Hypothesis 1) 0
Log on audit fees (Hypothesis 2) -/-
Log on total fees 0
5.2 Antle et al. model
In this section the second stage regression results of the Antle et al. Model are presented. See also Appendix 1.
5.2.1 Results LOGAF
Refer to table 17 for the results on the LOGAF-equation. We found significant relationships for AA (negative), LEV (positive), QUICK (positive), BMB (positive), LOGINV (negative), BLOGTA (positive). Some of these relationships seems somewhat strange, others are more logical. For example, the higher the level of abnormal accruals, the lower the audit fees. Also a higher level of inventories and the presence of a big four-auditor reduces audit fees. More logical is the coefficient on BLOGTA, indicating that firms with higher assets have higher audit fees. Antle et al. (2006) found a significant relationship between LOGAF and LOGNAF (positive).
31
5.2.2 Results LOGNAF
Refer to table 18 for the results on the LOGNAF-equation. These are inconsistent with the findings of Antle et al. (2006), who found a significant positive relationship between LOGNAF on the one hand and LOGAF and AA on the other hand. Only BLOGTA (positive) and BIG (negative) has a relationship with LOGNAF.
Table 17. Results LOGAF Table 18. Results LOGNAF
B t-value p-value B t-value p-value
(Constant) -0,901 -3,729 ,000** (Constant) -1,539 -2,085 ,038* LOGNAF1 -,011 -,415 ,679 LOGAF1 ,244 ,743 ,458 AA1 -0,056 -3,533 ,001** AA1 -,002 -,673 ,501 FISDEC -,069 -1,345 ,180 LOGTAX -0,097 -0,841 ,401 LEV ,437 3,761 ,000** LEV -,524 -1,354 ,177 QUICK ,254 4,131 ,000** QUICK -,070 -0,860 ,391 BMB -,039 -3,102 ,002** LOSS ,060 0,404 ,687 LOSS -,017 -0,352 ,725 LITI -,123 -0,878 ,381 LITI -,051 -1,161 ,247 ROA ,008 ,122 ,903 LOGAR ,103 1,945 ,053 LOGINV -,051 -1,310 ,192 LOGINV ,071 2,417 ,017* BIG -,481 -2,025 ,044* BIG -,022 -0,268 ,789 RETAIN -,803 -1,469 ,143 RETAIN -,204 -1,183 ,238 BLOGTA ,888 5,395 ,000** BLOGTA 0,465 8,939 ,000** Adjusted R Square ,511 ROA ,020 ,993 ,322 F 18,874 0,000** Adjusted R Square ,889 N 206
F 118,099 0,000** ** significant at the 0,01 level
N 206 * significant at the 0,05 level
** significant at the 0,01 level * significant at the 0,05 level
32
5.2.3 Results AA
Refer to table 19 for the results on the AA-equation. LOGAF have a significant positive relationship with AA. The significant positive relationship between LOGAF and AA is consistent with the findings of Antle et al. (2006). The lack of relationship between LOGNAF and AA is inconsistent with the findings of Antle et al. (2006), who found a significant negative relationship between LOGNAF and AA. AA also has a significant relationship with BLOGTA (negative), QUICK (negative), BMB (negative), LOSS (negative) and BIG (positive).
Table 19. Results AA B t-value p-value (Constant) 0,829 5,416 ,000** LOGAF1 ,324 6,094 ,000** LOGNAF1 -0,017 -0,959 ,339 SOCF -,260 -1,690 ,093 BLOGTA -,302 -9,026 ,000** QUICK ,048 2,871 ,005** BMB -,002 -5,836 ,000** LOSS -,087 -2,817 ,005** LITI -,012 -,410 ,682 ROA ,001 0,048 ,962 LEV -,007 -0,089 ,929 BIG ,192 3,861 ,000** RETAIN ,097 ,849 ,397 Adjusted R Square ,376 F 11,276 0,000** N 206
** significant at the 0,01 level * significant at the 0,05 level
5.2.4 Summary findings Antle et al. model
Our results from the Antle et al. model are summarized in table 20. Audit fees have a significant positive influence on abnormal accruals, supporting H2, that the provision of audit services is associated with earnings management.
Table 20. Summary of the findings of the Antle et al. (2006)-model Audit fees Non-audit fees Abnormal accruals
Audit fees X 0 + (Hypothesis 2)
Non-audit fees 0 X 0 (Hypothesis 1)
33
6.
Conclusion
In this research, we examined whether there’s a relationship between auditor independence and audit quality, represented by the degree of earnings management. To examine this, we used two models which presented mixed results in other countries and other period. We hoped that using two models produced robust results, but they produced mixed results. According to the models of Frankel et al. (2002), there’s a significant negative relationship between audit fees and earnings management, indicating that when an auditor performs more work, there’ll be less earnings management. The opposite is indicated by the results of the models of Antle et al. (2006). According to the simultaneous equations of Antle et al. (2006), there’s a significant positive relationship between audit fees and earnings management. This finding can also support the claim that there’s an economic bond between the auditor and his client. This claim can be refuted by the lack of a significant relationship between non-audit fees or total fees and earnings management, in both the models of Frankel et al. (2002) and Antle et al. (2006). The positive relationship between audit fees and earnings management can also be explained as follows: the auditor performed a lot of work to check whether the earnings management is within the boundaries of generally accepted accounting principles. That’s one of the limitations of this kind of research, we focused on earnings management, while earnings management isn’t forbidden. Earnings manipulation is forbidden. Because there are no common used models to detect earnings manipulation available, it’s hard to detect earnings manipulation without auditing the financial statements. Only Beneish (1999) presented a model to detect earnings manipulation, this model is very complicated. This model isn’t common used, it’s cited about 300 times, compared with about 4.000 cites of the Dechow et al. (1995) paper.
Another limitation of this research in particular, by applying Modified Jones we didn’t obtain the estimations for α and the β’s by running regressions for each industry, due to the limited firm years available per year per industry, we obtained this estimations by running these regressions for the entire sample. This resulted in a big difference between the mean ABSDACC reported by Frankel et al. (2002) of 0,15 and our mean ABSDACC of 4,32. There’s no particular industry what explains this difference. Another limitation of this research in particular, is that we didn’t use the same control variables as Frankel et al. (2002) and Antle et al. (2006), due to non-availability. In the case of Frankel et al. (2002), we didn’t proxy for institutional ownership. In the case of Antle et al. (2006), we didn’t proxy for past firm performance. Both proxies were significant. Despite these limitations, our results are consistent with the findings of (for instance) Ashbaugh et al. (2003), De Fond et al. (2002) and Ruddock et al. (2006), namely that there’s no significant relationship between auditor’s independence and earnings management. Auditors are
34
more likely to protect their reputation. Our suggestion for future research is that case studies have to examine, in cases of audit failure, what´s really going wrong. Maybe the role of provided non-audit services can be investigated.
35
Literature list
Antle, R., E. Gordon, G. Narayanamoorthy, and L. Zhou (2006). The joint determination of audit fees, non-audit fees, and abnormal accruals. The Review of Quantitative Finance and
Accounting, 27, 235–266.
Ashbaugh, H., R. LaFond, and B.W. Mayhew (2003). Do Nonaudit Services Compromise Auditor Independence? Further Evidence. The Accounting Review, 78, 3, 611-639.
Autoriteit Financiële Markten. Rapport algemene bevindingen kwaliteit accountantscontrole en kwaliteitsbewaking. Amsterdam, The Netherlands. AFM, 1 September 2010.
Becker, C.L., M.L. DeFond, J. Jiambalvo, and K.R. Subramanyam (1998). The effect of audit quality on earnings management. Contemporary Accounting Research, 15, 1, 1–24.
Beneish, M.D. (1997). Detecting GAAP Violation: Implications for Assessing Earnings Management among Firms with Extreme Financial Performance. Journal of Accounting and
Public Policy, 16, 271-309.
Beneish, M.D. (1999). The Detection of Earnings Manipulation. Financial Analysts Journal, 55, 5, 24-36
Cahan, S., D. Emanuel, D. Hay, and N. Wong (2008). Non-audit fees, long-term auditor–client relationships and earnings management. Accounting and Finance, 48, 181-207.
Craswell, A., D.J. Stokes, and J. Laughton (2002). Auditor independence and fee dependence.
Journal of Accounting & Economics, 33, 253-275.
Dechow, P.M., R.G. Sloan, and A.P. Sweeney (1995). Detecting Earnings Management. The
Accounting Review, 70, 2, 193-225.
DeFond, M.L., K. Raghunandan, and K.R. Subramanyam (2002). Do Non-Audit Service Fees Impair Auditor Independence? Evidence from Going Concern Audit Opinions. Journal of
Accounting Research, 40, 1247-1274.
Economics Laboratory. Simultaneous Equations. Regression Analysis Tutorial, University of California at Berkeley, 23-27 March 1998.
Eilifsen, A., and K. Knivsfla (2008). Non-Audit Services and Audit Quality: Investors’ Concerns Post-Enron.
European Commission. Audit Policies: Lessons from the Crisis. Brussels, Belgium. EC, 13 October 2010.
Francis, J.R. (2004). What do we know about audit quality? The British Accounting Review, 36, 345-368.
Francis, J.R. (2006). Are Auditors Compromised by Nonaudit Services? Contemporary Accounting
