Amsterdam Business School
The relation of client importance at local-office level on
audit quality in a financial crisis setting
Master Thesis
Name: Hidde Wyngaard Student number: 6074340 Date: 15-06-2016
Word Count: 20,371
Supervisor: dr. G. Georgakopoulos
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam
Statement of Originality
This document is written by student Hidde Wyngaard who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
This paper examines the influence of audit client importance at local-office level on audit quality. Client importance is measured as the ratio of the total fees of a specific client, divided by the total local office revenues. Additionally, total fees are split up between audit fees and non-audit fees. Audit quality is measured in two ways: (1) the absolute value of the discretionary accruals, and (2) the auditor’s propensity of issuing a going-concern opinion. The research is conducted in the United States in two timeframes. The first timeframe is the period from 2004 until 2007, which is considered as the pre-financial crisis period. The second timeframe is the period from 2009 until 2015, which is considered as the financial crisis period. No significant associations between client importance at local office level and audit quality are found for both models in both timeframes. This contradicts to the concerns of society about the lack of auditor independence and the possible influence that large clients have in the audit decision process.
Keywords:
Table of Content: 1. Introduction ... 6 1.1 Background ... 6 1.2 Research Question:... 7 1.3 Contribution ... 7 1.4 Structure... 8
2. Literature review & Hypotheses development ... 10
2.1 Audit Quality ... 10
2.2 Auditor Independence ... 10
2.3 Client importance ... 12
2.4 Hypotheses Development ... 13
2.4.1 Client importance and the treatment of earnings management ... 14
2.4.2 Client importance and going-concern opinions ... 18
2.5 Oversight ... 20
3. Research methodology ... 23
3.1 Research model and sample selection ... 23
3.1.1 Sample discretionary accruals model ... 24
3.1.2 Sample Going-Concern Model ... 27
3.2 Discretionary accruals model ... 30
3.3 Going-concern opinion model... 34
4. Empirical Results... 37
4.1 Discretionary accruals model ... 37
4.1.1 Descriptive statistics discretionary accruals model ... 37
4.1.2 Correlations discretionary accruals model ... 40
4.1.3 Regression results discretionary accruals model ... 44
4.2 Going-concern model ... 47
4.2.1 Descriptive statistics going-concern model ... 47
4.2.2 Correlations going-concern model ... 50
4.2.3 Regression results going-concern model ... 56
4.3 Oversight hypotheses and findings ... 62
5. Conclusion ... 63
References ... 66
5
List of Tables:
Table 1: Summary of prior research on audit quality ... 21
Table 2: Data reduction process model 1. ... 27
Table 3: Data reduction process model 2. ... 29
Table 4: Total Accruals Variables ... 31
Table 5: NDA regression variables ... 31
Table 6: SIC Code Table ... 32
Table 7: Discretionary Accruals regression model variables ... 33
Table 8: Going-concern opinion model regression variables ... 35
Table 9: Descriptive statistics discretionary accruals model 2004-2007 ... 38
Table 10: Descriptive statistics discretionary accruals model 2009-2015 ... 39
Table 11: Pearson correlation matrix discretionary accrual model, sample 2004-2007 ... 42
Table 12: Pearson correlation matrix discretionary accrual model, sample 2009-2015 ... 43
Table 13: OLS Regression values discretionary accruals model sample 2004-2007 ... 46
Table 14: OLS Regression values discretionary accruals model sample 2009-2015 ... 47
Table 15: Descriptive Statistics Going-concern Opinion Model 2004-2007 ... 48
Table 16: Descriptive Statistics Going-concern Opinion Model 2009-2015 ... 48
Table 17: Pearson correlation matrix going-concern opinion model, sample 2004-2007 ... 53
Table 18: Pearson correlation matrix going-concern opinion model, sample 2009-2015 ... 55
Table 19: Logisitic regression values going-concern opinion model, sample 2004-2007 ... 61
Table 20: Logisitic regression values going-concern opinion model, sample 2009-2015 ... 61
Table 21: Oversight hypotheses vs. findings ... 62
Table 22: VIF Values table discretionary accruals model, TFR regression ... 70
Table 23: VIF Values table discretionary accruals model, AFR regression ... 70
Table 24: VIF Values table discretionary accruals model, NAFR regression ... 71
Table 25: Variance Inflation Factor (VIF) Values TFR regression ... 71
Table 26: Variance Inflation Factor (VIF) Values AFR regression ... 72
Table 27: Variance Inflation Factor (VIF) Values NAFR regression... 72
List of Equations: Equation 1: Total Accruals ... 30
Equation 2: Total Accruals (2) ... 31
Equation 3: NDA regression ... 31
Equation 4: Discretionary Accruals regression model ... 32
Equation 5: Going-concern opinion model logistic regression: ... 35
List of Figures Figure 1: Financially distressed firms per year ... 30
Figure 2: Overview dependent and independent variables discretionary accruals model... 39
1. Introduction
1.1 Background
Scandals like the fraud at WorldCom and Enron made audit quality a well-discussed issue under politicians, economics and scientists. According to DeFond et al. (2002) the role in society for the auditor is to assess whether a firm’s financial statements provide a fair view of the reality. This role in society is of high importance, because there are lots of different users of financial statements, like investors, financial institutions and other stakeholders (DeFond et al., 2002). The confidence of society in the financial statements is therefore an essential feature of efficient working capital markets. This confidence is based on the level audit quality. Audit quality consists of two critical factors: the capabilities and independence of the auditor (Tepalagul & Lin, 2015).
Audit quality is defined as the joint probability that an existing material error is detected and reported by an auditor (DeAngelo, 1981). This implies that audit quality is dependent on two factors: detection and reporting. Detection is about the ability of the auditor to find material errors. Reporting is about the willingness of the auditor to report material errors, which is dependent on the independence of the auditor (DeAngelo, 1981). After the different scandals, the independence of auditors is well discussed in society. These discussions lead to the implementation of new audit standards, the Sarbanes Oxley Act (hereafter SOx), in 2002. Stricter rules on disclosures and internal controls should increase audit quality. However, the audit scandals did not stop after the implementation of SOx. Big scandals like Lehman Brothers and Bernard L. Madoff Investment Securities proved that despite the SOx implementation, there was still a lack of audit quality.
Prior research focused on possible threats of auditor independence and therefore threats of audit quality (DeFond et al., 2002; Li, 2009; Tepalagul & Lin, 2015). The main threats are the provision of non-audit services, auditor tenure, client’s affiliation with CPA firms and client importance (Tepalagul & Lin, 2015). In this paper the threat of client importance will be investigated. The auditor’s economic bond with the client may possibly influence the independence of the
7 effect of client importance on audit quality. Even while society is concerned about the lack of auditor independence, many researches cannot find negative associations between client importance and audit quality (e.g. DeFond et al., 2002; Ashbaugh et al., 2003; Callaghan et al., 2009; Chung & Kallapur, 2003; Zhou & Zhu, 2012). Some researches even find a positive association (e.g. Li, 2009; Reynolds & Francis, 2001).
During a financial crisis, governments are triggered to implement stricter regulations and supervision policies. Besides that, the auditing profession is under increased scrutiny during a financial crisis (Xu et al., 2013). Because of stricter regulations, there is a possibility that litigation risks are higher, so this increases the audit risk. In addition, because of the increased scrutiny there is a higher risk of reputation loss (Xu et al., 2013).
This research will examine the influence of client importance on audit quality at a local-office level. This is because according to prior literature, the effect of client importance is better determinable at local-office level (e.g. Reynolds & Francis, 2001; Li, 2009; Sharma et al., 2011). The impact of losing a client at local-office level is bigger than for national-offices (Reynolds & Francis, 2001). This is because on local-office level, one client will probably represent a bigger percentage of the total revenues, than a client at national-office level.
1.2 Research Question:
The research question of this paper is:
‘Does audit client importance at local-office level affect audit quality in a financial crisis setting?’
1.3 Contribution
This research has different contributions to existing knowledge. There are multiple analytical contributions. Firstly, to the best of my knowledge there is no research yet which is focused on the effect of client importance on audit quality in the recent financial crisis setting in the US. This may be explained by the fact that the financial crisis in the US is a very recent issue, and so on there is no publicized research available yet. Secondly, prior research in the research field of client importance related to audit quality is not conclusive (e.g. Frankel et al.,
2002; Sharma et al., 2011; Reynolds et al., 2004; Ashbaugh et al., 2003), so this research will contribute in the discussion about this issue and contributes to the call for further research. One of the most recent researches in the field of audit quality and client importance is the research of Tepalagul and Lin (2015). They conclude that the mixed evidence in this research area, together with the regulatory changes and the recent financial crisis provides opportunities for future research on audit client importance and audit quality. So, this research contributes in the ongoing debate, and adds new evidence about the recent financial crisis.
The practical contribution of this research is for different groups of users. First of all, this research can be interesting for auditors. This research will examine whether auditors’ independence may be affected by client importance. They could keep this in mind, while assessing their audits. This research is also interesting for regulators. There are concerns about auditor independence, and so on about audit quality, in society (Li, 2009). Regulators are responsible for regulation and supervision policies. The findings of this research are helpful for regulators to assess whether new regulations, or changes in current regulations are necessary in a financial crisis setting. Another group for who this research could be interesting are audited companies. These companies could use this research to assess whether the firm is an important client for their auditor. According to the client importance level they can evaluate whether their audit quality could be affected (Chung & Kallapur, 2003). At last this research is interesting for society in general. Since the different scandals and the financial crisis, auditors have been under greater scrutiny because of concerns about auditor independence in society (Xu et al., 2013). The findings of this research could show them if their concerns are correct or not.
1.4 Structure
The remainder of this paper will be structured as follows. In the next chapter prior literature about audit quality, auditor independence and client importance will be discussed. Based on this prior literature, the hypotheses will be developed. In the third chapter the research method will be discussed. This will
9 by a detailed description of both models used in this research: the discretionary accruals model and the going-concern opinion model. In the fourth chapter, the empirical results will be discussed. This chapter will be divided in two parts. The first part will exist of the descriptive statistics, correlations and the regression results of the discretionary accruals model. The second part will exist of the descriptive statistics, correlations and the regression results of the going-concern opinion model. In the last chapter, the conclusions will be made.
2. Literature review & Hypotheses development
In this chapter different important definitions are explained. In the first section audit quality is defined, this is followed by a detailed explanation of auditor independence. In the third section the concept of client importance is explained. In the fourth section, and the subsequent subsections the hypotheses are developed. The last section presents a table, which provides as an oversight of the main research papers in the field of audit quality and client importance.
2.1 Audit Quality
DeAngelo (1981) defines audit quality as follows: ‘the joint probability that an
existing material error is detected and reported by an auditor’. So, high audit
quality means that there is a high probability that an existing material error is detected and reported by the auditor, and this means that the financial statements provide a fair view of reality. As DeAngelo (1981) defines, audit quality is dependent on two factors: detection and reporting. Detection is about the ability of the auditor to find material errors, which is dependent on audit effort. Reporting is about the willingness of the auditor to report material errors, which is dependent on the independence of the auditor (DeAngelo, 1981). When the auditor finds an error, the auditor can choose to report or not to report the error. When the auditor chooses to report the material error, there is a risk of being fired by the client, and so on lose revenues. On the other hand, when the auditor does not report the error, and in the end the error is discovered, this could lead to litigation costs (Reynolds & Francis, 2001, p.378). Furthermore it could also lead to reputation loss, which will make it hard to retain current clients and attract new clients, and so on will decrease the revenues (DeAngelo, 1981).
2.2 Auditor Independence
As described in section 2.1, audit quality is also dependent on the willingness of the auditor to report material errors (DeAngelo, 1981). This willingness is dependent on the independency of the auditor. In the existing literature there are multiple definitions for auditor independence. The most accepted definition
11
statements’ (Watts & Zimmerman, 1966, p.615). When the independence of
auditors is doubted, financial statements will become less reliable, and this increases the financial risks (Quick & Warming-Rasmussen, 2009). So, it is important that society has confidence in the independency of accountants.
To earn this confidence from society, auditors should be independent in mind and in appearance (IFAC, 2006, Sec. 290.8). Independence in mind is described by Quick & Warming-Rasmussen (2009, pp.142) as ‘the state of mind
that permits the provision of an opinion without being affected by influences that compromise professional judgment, allowing an individual to act with integrity, and exercise objectivity and professional skepticism. Independence in appearance
is defined as: ‘the avoidance of facts and circumstances that are so significant that
a reasonable and informed third party would reasonably conclude that a firm’s integrity, objectivity or professional skepticism had been compromised (2009,
pp.142). So, in short, independence in mind is about the factual independence, and independence in appearance is about the representation of independence to society.
There are different incentives for auditors to stay independent. According to Watts and Zimmerman (1983, p.633) auditors have market-based and institutional based incentives to stay independent. Large clients generate a lot of revenues, but also a lot of attention. When audit failure is exposed at a large client, this could lead to substantial litigation costs and reputation loss (Reynolds & Francis, 2001, p.377). Reputation loss could be costly because losing clients is associated with losing revenues. Besides losing clients, a bad reputation will not be favorable in attracting new clients. So, reputation loss risks and litigation costs risks are incentives for auditors to act independent.
There are also factors that could threat auditor independence. As mentioned before, the main threats for auditor independence are the provision of non-audit services, auditor tenure, client’s affiliation with CPA firms and client importance (Tepalagul & Lin, 2015, p.103). This research will focus on the possible threat of client importance. Client importance will be discussed in the next section.
2.3 Client importance
As mentioned before, there are different factors that threat auditor independence. One of them is the important of a client. Auditors are hired and paid by their clients, so there always is a certain form of economic dependency. As a result of this dependency, a conflict could arise between maintaining a client and maintaining independence (DeAngelo, 1981). As described in section 2.1, the auditor has to choose whether he or she reports an error, or not. There are multiple factors that influence the auditor’s choice. Firstly there are incentives to report the error, because of possible litigation costs and reputation loss when the error is not reported, but later discovered by the users of the financial statements (Tepalagul & Lin,2015). But there are also factors that will probably withhold the auditor for reporting the error. An example is the fact that the client wants to fire the auditor, which leads to a decrease of the auditor’s revenues. When the auditor is not reporting the error, the auditor will not be fired, and so it could retain the future revenues from that client (Chung & Kallapur, 2003). So, the economic dependency could compromise the auditor’s independence, which could lead to a lower audit quality (Simunic, 1984).
Prior literature about the effect of audit client importance on auditor independence is not conclusive (e.g. Frankel et al., 2002; Ashbaugh et al., 2003; Chung & Kallapur, 2003; Reynolds & Francis, 2001). Beattie et al. (1999) research which factors could lead to an affection of auditor independence, by questioning audit partners, financial directors of listed firms and financial journalists. The factors were about the dependency of the audit partner’s income related to the client, the firm income related to the client and partner’s desire of maintaining the client because the client generates at least ten percent of the total office revenues. These factors are all related to the economic dependency of the client, and so Beattie et al. (1999) conclude that economic dependence on the client could affect auditor independence. Another research that implies that auditor independence is compromised because of the economic dependency of the client is the research of Frankel et al. (2002). They examined whether auditor’s provision of non-audit services leads to lower earnings quality, and so to lower audit quality. Frankel et al. (2002) found an association between the
13 statements of the clients. This is why they conclude that when there is a high economic dependency on the client, there is a higher probability that the auditor’s independence is compromised.
But, there are also a lot of researches that cannot find an association between audit client importance and auditor independence (e.g. Ashbaugh et al., 2003; Chung & Kallapur, 2003; Reynolds & Francis, 2001). This has led to an interesting discussion in prior literature. Ashbaugh et al. (2003) challenged the paper of Frankel et al. (2002). They used total fees, instead of a fee ratio and did not find an association between the provision of non-audit services and earnings quality. They conclude auditor independence is not compromised because of the amount of fees received from the client. Chung and Kallapur (2003) used fee ratios to measure client importance. They examined this at firm and local-office level and found no significant associations between client importance and auditor independence. This is why Chung and Kallapur (2003) also conclude that auditor independence is not compromised by client importance. Reynolds and Francis (2001) examined the influence of client importance on auditor independence on local-office level. According to Reynolds and Francis (2001), the impact of losing a client is higher for a local office, than for the nationally operating firm. This is why local offices are more likely to compromise their independence to retain a client. They conclude that client importance on local-office level, does not affect auditor independence. Li (2009) also examines the impact of client importance on local-office level. This research is a comparison of two different timeframes: the pre-SOx period and the post-SOx period. Li (2009) finds a positive association between client importance and the propensity of the auditor to issue a going-concern opinion, which implies that the more important clients, are, the more likely it is to receive a going-concern audit opinion.
So, even while there is a lot of research available in the field of auditor independence and audit quality, there is no conclusive information. The discussion is still going on.
2.4 Hypotheses Development
In this section the hypotheses will be developed. In this paper, there are two types of measures chosen to measure audit quality. The first measure is an
earnings management measure, namely the absolute value of discretionary accruals. The second measure is the propensity of the auditor to issue a going-concern audit opinion. In the first subsection the hypotheses in the field of earnings management are developed. In the second subsection the hypotheses related to the propensity of issuing a going-concern opinion are developed.
2.4.1 Client importance and the treatment of earnings management
The potential incentives of auditors might be affected by the audit treatment of accruals in the client’s financial statements. Jones (1991) stated that companies may use accruals to manage reported earnings towards desirable outcomes. Prior research provides different insights in the field of auditor independence measured by earnings management measures (Chung and Kallapur, 2003; Frankel et al., 2002; Sharma et al., 2011; Reynolds et al., 2004; Ashbaugh et al., 2003; Reynolds & Francis, 2001). A high level of earnings management is related to a lower earnings quality, and so on a lower audit quality. Frankel et al. (2002) examined the effect of the delivery of non-audit services on audit quality, by researching if the delivery of non-audit services leads to biased financial reporting. Their sample consists of US firms in a certain period in 2001. They measure biased financial reporting in 2 manners: (1) discretionary accruals and (2) the likelihood of firms to meet earnings benchmarks. Frankel et al. (2002) find a positive association between non-audit fees and the magnitude of discretionary accruals, which implies that a higher amount of non-audit fees paid to the auditor, is associated with higher discretionary accruals. This is why they conclude that when an audit firm receives high non-audit fees, it is more likely that auditors compromise their independence.
Sharma et al. (2011) examined the moderating effect of the audit committee on the association between client importance and earnings management. They conducted their research in the US on local office level. The proxy used for earnings management is the value of the discretionary accruals, and client importance is measured in two ways: (1) the ratio of non-audit fees divided by the total office revenues and (2) the ratio of audit fees divided by the total office revenues. Sharma et al. (2011) find a positive association between
15 Frankel et al. (2002). They also find that auditors only compromise their independence, when the audit committee provides weak oversight.
There are other researches that contradict the findings of Frankel et al. (2002) and Sharma et al. (2011)(Tepalagul & Lin, 2015). This leads to an interesting discussion in prior literature. Reynolds et al. (2004) suggest that the findings of Frankel et al. (2002) can be explained by their specific sample. In the research of Reynolds et al. (2004) they also use data from the US in 2001, to examine whether the objectivity of an auditor is influenced by non-audit services or economic dependence on a specific client. To measure the auditor’s objectivity, they use discretionary accruals as measure. Following their findings, they conclude that an auditor’s objectivity will not be influenced by providing audit and non-audit services. Ashbaugh et al. (2003) also conduct a research with a sample of US firms in 2001. They examine whether non-audit services compromise auditor independence. They measure auditor independence by testing discretionary accruals. Ashbaugh et al. (2003) are not able to find an association between non-audit fees and discretionary accruals, and so they conclude that providing non-audit services does not affect auditor independence. When an auditor is more dependent on the income of a certain client, the possibility exists that the auditor will treat accruals of this customer with more flexibility or will allow a greater discretion in these accruals. This means that the client is more flexible in manipulating, adjusting and managing the earnings of the firm. According to Reynolds and Francis (2001) the impact of losing a client is bigger for firms at a local office level than for firms at national level. This is because of the audit firm size. A firm at local level is smaller than a firm at national level, and will therefore be more dependent on certain clients than firms at national level. When their dependence is higher, the possibility may exist that auditors will compromise their independence.
In the pre-financial crisis setting, there is a lower risk to become financially distressed for as well audit firms as their clients than during the financial crisis. Besides that, auditors were more focused on their independence since SOx was introduced in 2002, which was introduced to increase auditor independency (Li, 2009). The low financial distress situation and the recent
introduction of SOx leads to hypothesis 1a:
H1a: There is no association between client importance at local-office level and earnings management in a pre-financial crisis setting.
During a financial crisis, not only clients are exposed to a higher risk of getting in a financially distressed situation, but audit firms themselves as well. As mentioned before, the impact of losing a client is bigger for firms at a local office level than for firms at national level (Reynolds and Francis, 2001). This impact is greater because of the economical dependency on the client. When their economic dependence is higher, the possibility may exist that auditors will compromise their independence. This brings us to the following hypothesis:
H1b: There is a positive association between client importance at local-office level and earnings management in a financial crisis setting.
On the other hand, prior literature shows that during a financial crisis regulators are more willing to change their policies to increase financial reporting quality (Vichitsarawong et al., 2010). Some researches show that high litigation risk and possible reputation loss would decrease the likelihood that the auditor will act in favor of the client (Tepalagul & Lin, 2015). Zhou and Zu (2012) examined the effect of the Asian financial crisis in 1997 on the relationship between client importance and auditor independence. They find that auditors are less incentivized to compromise their independence during a financial crisis. Zhou and Zu (2012) explain this finding by the fact that the government improved regulation and supervision policies. Due to the stricter regulations, the litigation risks for auditors increases significantly during a financial crisis. Next to the risk of higher litigation costs, there is also a higher risk of reputation damage.
Reynolds and Francis (2001) examine the influence of large clients on office-level auditor reporting decisions. They use data from 1996, and just look at Big 51 auditors at local-office level. They measure audit quality by using
17 conducted at local-office level, because the economic consequences of losing a client are bigger for local-offices, than for national firms (Reynolds & Francis, 2001). They conclude that auditor’s independence is not impaired, and even found a little evidence that Big 5 firms report more conservative to their large clients. This research is also done in a sample of firms audited by non-Big 5 firms. Hunt and Lulseged (2007) conclude the same as Reynolds and Francis (2001). Non-Big 5 firms do not allow more earnings management at their larger clients compared to their smaller clients. In addition, they also find evidence that non-Big 5 firms are more conservative to their large clients compared to their small clients. In both researches, the explanation for the findings is given by the fact that auditors are very careful with possible reputation loss. Protecting their reputation is important, and more valuable than the economic fee. The risk of reputation loss brings us to hypothesis 2a, where no association is expected between client importance and earnings management in a pre-crisis setting:
H2a: There is no association between client importance at local-office level and earnings management in a pre-crisis setting.
Auditors are under greater scrutiny since the financial crisis (Xu et al., 2013), so their concerns of litigation and reputation loss will be higher. When auditors suffer reputation loss, clients will leave them and it will be harder to attract new clients, so their revenues will decline. During a crisis period, the supervision is of a higher level, and the regulations are stricter. Combined with the findings of Reynolds and Francis (2001) and Hunt and Lulseged (2007) about the higher level of conservatism of both Big 42 as non-Big 4 audit firms for bigger clients, I
expect that the level of conservatism will increase during the financial crisis. This leads to hypothesis 2b, which expects that auditors will not compromise their independence under influence of client importance and will be more conservative to their clients, because of concerns about litigation and reputation loss, which results in client loss:
H2b: There is a negative association between client importance at local-office level and earnings management in a financial crisis setting.
2.4.2 Client importance and going-concern opinions
Another way of measuring audit quality is the propensity of the auditor to issue a going-concern opinion. Auditors should issue a going-concern opinion when the client’s financial condition is such that there is doubt about the firm’s ability to continue as a going-concern. This means there is a doubt about the firm’s ability to meet its obligations in the future. Reynolds and Francis (2001) do not only use absolute discretionary accruals from the Modified Jones Model to measure auditor independence, but also the association between client importance and the propensity of the auditor in issuing a going-concern opinion. They measure client importance by the part of the client’s revenues compared to all clients’ revenues. They find a positive association, which implies that important clients are associated with a higher propensity of issuing a going-concern audit opinion. As mentioned before, they find a negative association between client importance and discretionary accruals.
DeFond et al. (2002) examine whether audit and non-audit fees affect auditor independence. To measure auditor independence they use the propensity to issue a going-concern opinion in financially distressed companies (p.1248). Financially distressed firms are firms which either report negative earnings, or negative cash flows from operational activities. Their sample was taken in the US in 2000. They find no association, so they conclude there is no association between both audit and non-audit services and the propensity of issuing a going-concern opinion. Another research that is examining the association between audit and non-audit services and auditor independence by looking at going-concern opinions is the research of Callaghan et al. (2009). They examined firms that went bankrupt in the US from 2001 until 2005. They looked whether these firms have had a going-concern audit opinion before they went bankrupt. They did not find an association between both audit and non-audit fees and the issue of going-concern opinions.
19 Geiger and Rama (2003) examined the association between audit and non-audit fees and the firm’s likelihood of receiving a going-concern opinion. They chose financially distressed firms in the manufacturing industry in the U.S. in 2000 and 2001. They could not find any association between the likelihood of receiving a going-concern opinion and the magnitude of non-audit fees. They found a significant positive association between the magnitude of audit fees and the firm’s likelihood of receiving a going-concern opinion. This implies that audit clients that generate high revenues, have a higher chance of receiving a going-concern opinion. The findings of Geiger and Rama (2003) are supported by the findings of Li (2009). Li (2009) examines whether auditor independence is influenced by client importance at local office level. The research is a comparison between the pre-SOx period and the post-SOx period in the US. Li (2009) measures auditor independence by examining going-concern opinions. Client importance is measured by the ratio of audit, non-audit and total fees paid to the audit office compared to all revenues of the audit office. In the pre-SOx period, there is no association found between the paid fees and the propensity of issuing a going-concern opinion. In the post-SOx period Li (2009) found a positive association. This is why Li (2009) concludes that the market concerns about the lack of independence of audit firms is not supported by these findings. In a pre-crisis setting, this economic dependency on the client is relatively low. This is why is expected that in a pre-crisis setting auditors are not explicitly willing to compromise their independence and so on will evaluate the issue of a going-concern opinion at a more honest manner. This leads to hypothesis 3a:
H3a: There is no association between client importance at local-office level and the propensity of issuing a going-concern opinion in a pre-financial crisis setting.
While developing hypothesis 1b, the economic dependency during a financial crisis was mentioned. When an audit firm is exposed to a higher risk of getting in a financially distressed situation, there is a higher possibility that the audit firm will be economically dependent on certain clients. When the audit firm is more dependent on a certain client, the possibility exists that the auditor will be less
critical in the process of issuing a going-concern opinion, and so on will compromise the independence. This explains hypothesis 3b:
H3b: There is a negative association between client importance at local-office level and the propensity of issuing a going-concern opinion in a financial crisis setting.
As described in the development of hypotheses 2a and 2b, a financial crisis triggers governments to implement stricter regulations and supervision policies. Auditors are under increased scrutiny during a financial crisis, and the litigation risk and reputation loss risk increases (Xu et al., 2013). Xu et al. (2013) examined the response of auditors to the financial crisis in Australia. They found that auditors are more willing to issue a going-concern opinion during a financial crisis, than before a financial crisis. This is explained by the higher concerns of litigation and possible reputation loss. This leads to hypotheses 4a and 4b:
H4a: There is no association between client importance at local-office level and the propensity of issuing a going-concern opinion in a pre-financial crisis setting.
H4b: There is a positive association between client importance at local-office level and the propensity of issuing a going-concern opinion during the financial crisis.
2.5 Oversight
In table 1, an oversight is presented which summarizes the main papers in the field of audit quality and client importance.
Table 1: Summary of prior research on audit quality
Table 1
Summary of prior research on audit quality
Client Importance Measure and Association:
Author(s) Sample Area Sample year(s) Proxy of audit
quality
Analysis level Audit Fees Non-audit
fees
Total Fees Clients Size
Chung & Kallapur (2003)
U.S.A. 2001 Discretionary
Accruals
National N/T N/S N/S N/T
Office N/T N/S N/S N/T
Frankel et al. (2002) U.S.A. 2001 Discretionary Accruals
National - + N/S N/T
Sharma et al. (2011) New Zealand 2004-2005 Discretionary accruals Office N/T + + N/T Reynolds et al. (2004) U.S.A. 2001 Discretionary Accruals National N/S N/S N/S N/T Ashbaugh et al. (2003) U.S.A. 2001 Discretionary Accruals National N/S N/S N/S N/T
Zhou & Zu (2012 Six Asian countries
1994-2001 Qualified auditor opinions
National N/T N/T N/T Pre-crisis: +
Post-crisis: -
Reynolds & Francis (2001) U.S.A. (Only Big 5) 1996 Discretionary Accruals Office N/T N/T N/T - Going-concern Opinions N/T N/T N/T +
Hunt & Lulseged (2007) U.S.A. (Only non-Big 5) 2001-2003 Discretionary Accruals Office N/S N/S N/S - National N/S N/S N/S N/S
Table 1 (continued)
Summary of prior research on audit quality
Client Importance Measure and Association:
Author(s) Sample Area Sample year(s) Proxy of audit
quality
Analysis level Audit Fees Non-audit
fees
Total Fees Clients Size
DeFond et al. (2002) U.S.A. 2000 Going-concern opinions National N/S N/S N/S N/T Callaghan et al. (2009) U.S.A. 2001-2005 Going-concern opinions National N/S N/S N/T N/T
Geiger & Rama (2003) U.S.A. 2000-2001 Going-concern opinions National + N/S N/T N/T Li (2009) U.S.A. 2001 vs. 2003 Going-concern opinions Office Pre-SOx: N/S Post-SOx: + N/S N/S N/T
Xu et al. (2013) Australia 2005-2009 Going-concern opinion
National N/T N/T N/T +
N/T = Not tested
N/S = No significant association found with audit quality + = Significant positive association found with audit quality - = Significant negative association found with audit quality
3. Research methodology
In this chapter the research methodology to examine whether client importance at local-office level affects audit quality will be described. The research methodology used in this research is based on database research. First the research model and sample selection methods will be described. This will be followed by two subsections: one about the sample of the discretionary accruals model and one about the sample of the going-concern model. These subsections will be followed by two sections, where both models are described.
3.1 Research model and sample selection
There are several measures in prior literature to research audit quality. The most common measures are discretionary accruals (e.g. Reynolds & Francis, 2001; Frankel et al., 2002; Ashbaugh et al., 2003; Reynolds et al., 2004; Hunt & Lulseged, 2007; Chung & Kallapur, 2003; Sharma et al., 2011) and going-concern audit opinions (e.g. DeFond et al., 2002; Callaghan et al., 2009; Li, 2009; Reynolds & Francis, 2001; Zhou & Zu, 2012; Xu et al., 2013; Geiger & Rama, 2003). There are also other measures in prior literature, but they are less used than the two most common used measures. Some researchers measure audit quality by looking at restatements (e.g. Stanley & DeZoort, 2007; Kinney et al., 2004) or for example by looking at conservatism in reported earnings (e.g. Jenkins & Velury, 2008; Ruddock et al., 2006). There are lots of other measures identified, but the amounts of different researches that use these measures are low (Tepalagul & Lin, 2015). This is why in this research is chosen to use the (1) discretionary accrual model and (2) the propensity of an auditor to issue a going-concern opinion for distressed clients to examine audit quality. A firm is in a financially distressed situation when it reports a negative net income and/or negative cash flows from operational activities (Li, 2009; Reynolds & Francis, 2001; DeFond et al., 2002). Client importance is measured as the ratio of the client’s audit, non-audit and total fees compared to the total fees received of all clients of the office, according to different researches in prior literature (e.g. Li, 2009; Sharma et al., 2011; Chung & Kallapur, 2003; Hunt & Lulseged, 2007). This is a very clear
manner of measuring client importance. Besides that, Li (2009) also measures a change over time, namely the influence of client importance on auditor independence in a pre-SOx and a post-SOx setting. In this research there is also a change over time measured, namely before the financial crisis started, and during the financial crisis.
In total, four samples are used for this research: two samples in the 2004-2007 timeframe and two samples in the 2009-2015 timeframe. Two models are used to examine the influence of client importance on audit quality: (1) the discretionary accruals model and (2) the going-concern opinion model. For both models a different sample of company codes is used. The sample of model one, the discretionary accruals model, is based on S&P500 firms. The sample of model two, the going-concern model, is based on S&P1500 firms. The use of different samples is because of different reasons. First of all, the sample of the going-concern model consists of only financially distressed firms. Financially distressed firms are firms which either report negative earnings, or a negative cash flow from operational activities. Using a sample based on S&P500 firms would limit the sample too much, because just a small portion of these firms is in a financially distressed situation. On the other hand, for the discretionary accruals model there was no need to use a larger basis of firms than the S&P500 firms basis. Former research using this model cannot find differences in results when researchers use S&P500 firms in their sample (e.g. Callaway Dee et al., 2006; Platikanova, 2008) and S&P1500 or even all available firms in their sample (e.g. Frankel et al., 2002; Reynolds et al., 2004; Larcker & Richardson, 2004; Chung & Kallapur, 2003). The extra amount of data while using S&P1500 firms would be hard to process, while there is a very high probability that the results will not differ. In the following subsections the samples and the sample selection process for both model one as model two will be described.
3.1.1 Sample discretionary accruals model
In this subsection the sample of the first model, the discretionary accruals model, will be described. As mentioned in the former section, the sample of the discretionary accruals model consists of S&P500 firms. All financial company
25 data about location and fees was gathered from the WRDS/AuditAnalytics database. The data reduction process is illustrated in Table 2, and will be described in this section. The first sample consists of the years 2004 until 2007. Gathering financial data for the discretionary accruals model in Compustat led to 3297 unique firm observations during this period. Firstly 958 financial institutions were deleted, because these firms have unique procedures, which make discretionary accruals not meaningful to them (Chung and Kallapur, 2003, p.940; Dechow et al., 1995). The financial institutions are the firms categorized by SIC Codes 6000 – 6999. After merging the financial data with audit fee and location data in Microsoft Excel, 585 observations were deleted because the auditor data was inaccurate. Both datasets were merged by creating a unique code per firm per year, existing of the Central Index Key (CIK) and the fiscal year. As third, all data was checked for missing financial data to make the right calculations. In this step 526 observations were deleted. Most of these observations were deleted because data about cash flows of operations was missing to calculate the variable OCF. Finally the sample was checked whether all observations were complete for all four years. This means that a firm could only be included in the sample if data for all four sample years was available. In this final step 180 observations were deleted. This made the final sample for period one 1048 firm observations, which is 262 observations per year.
The second sample includes data about S&P500 firms in the years 2009 until 2015. The data reduction process was the same as the process in the first period. Gathering financial data for the discretionary accruals model in Compustat led to 5191 observations for this period. Firstly 1510 financial institutions were deleted. Secondly 1065 observations were missing essential audit fee and/or location data. A note has to be made that of these 1065 deleted observations 354 deleted observations are observations of 2015. This high proportion is because the AuditAnalytics database is less up to date than the Compustat database. This results in a low availability of 2015 audit fee and audit location data. As third, 190 observations were missing essential financial data. Lastly, 293 observations were deleted because the observations were not complete for all years. 2015 is excluded in this completeness test, because the high amount of deleted observations in 2015 caused by the lack of available
auditor data about this year. This made the final sample of 2133 observations, which consists of 350 observations per year for the period 2009 until 2014 and 33 observations for the year 2015.
After the former reduction process, the data of both samples is uploaded in STATA. In STATA all variables, which are used in the regression, are winsorized. This means that the highest and lowest one percent of the values of all observations was changed to the value of the first and 99th percentile. This is done to control for the effect of outliers. After winsorizing, the variables are tested for linearity, normality, homoscedasticity and multi-collinearity. According to Field (2013, p.401) the regression cannot be used when one of these four requirements is not met. Making scatterplots does the linearity test. Linearity is present when the scatterplot displays that all data is well balanced and no excessive outliers are visible. No issues were shown, so linearity is assumed. To test for normality, the histograms are checked. When the data is normally distributed, the histogram should look like a bell shaped curve. Dummy variables are excluded for this test, because only two values3 are available. This
is why dummy variables will never display a normal distribution. For all other variables, the bell shaped curve is visible, so there are no normality issues either. As third, the sample is tested for homoscedasticity. Homoscedasticity exists when all variables have the same finite variance (Field, 2013). Three tests are used to test for homoscedasticity: (1) residual versus fit plot, (2) Cameron & Trivedi’s decomposition of M-test and (3) Breusch-Pagan / Cook-Weisberg test. The residual versus fit plot shows that all residual values are displayed in the same direction, which is evidence for the existence of homoscedasticity. Also the Cameron & Trivedi’s decomposition of M-test and the Breusch-Pagan / Cook-Weisberg test did not find any issues. This is why homoscedasticity is assumed. Lastly, the variables were tested for collinearity. Testing for multi-collinearity shows whether variables are highly correlated. Using a Variance Inflation Factor (VIF) Value Test tests the presence of multi-collinearity. The assumption is made that a VIF-Value above 10 means that multi-collinearity is present. The value of 10 is chosen because this value is also used by multiple former researches (e.g. Chen et al., 2010; Marquardt, 1980; Li, 2009). The
VIF-27 Values tables are disclosed in Appendix 1. The highest VIF-Value is 2.67 and the lowest VIF-Value is 1.02 for all 3 regressions equations, so the assumption is made that multi-collinearity is not present.
Table 2: Data reduction process model 1.
Table 2
Data reduction process model 1 Discretionary Accruals Model
Period 1 2004-2007
Period 2 2009-2015
Unique Firm Observations 3297 5191
Financial Institutions (958) (1510)
Incomplete Auditor Data (585) (1065)*
Incomplete Financial Information (526) (190)
Incomplete Years (180) (293)
Final Sample: 1048 2133**
* = Of the deleted auditor observations were 711 observations in 2009-2014 and 354 observations in 2015 ** = Distribution of the sample is as follows:
- 2009-2014: 350 observations per year - 2015: 33 observations
3.1.2 Sample Going-Concern Model
In this subsection the sample of the second model, the going-concern model, will be described. As mentioned in section 3.1, the sample of this model consists of S&P1500 firms. Just like the discretionary accruals model, all financial data is gathered from the WRDS/Compustat database and all audit fee and location data is gathered from the WRDS/AuditAnalytics database. The data reduction process is illustrated in table 3, and the process will be described in this section.
The first sample consists of S&P1500 firms in the period from 2004 until 2007. Gathering necessary financial data in the Compustat database resulted in 10986 unique firm observations. First the financial data was merged with the available auditor location and fee data in Microsoft Excel, by generating unique codes existing of the Central Index Key (CIK) and fiscal year. After merging these two datasets 3318 observations were deleted because of missing audit fee and/or location data. Secondly, all financial data necessary to make the calculations for the variables was checked. This resulted in 1937 deleted
observations, merely because of the lack of data for the variable LIQUIDITY. The last step of the data reduction process was to check whether all observations were complete for all four years. This means that a firm observation could only be included in the final sample, when the firm data was observed for all four years. This resulted in 1045 deleted observations. This made the final sample of 4596 firm observations, which are 1149 observations per year. These 4596 observations consist of 751 financially distressed firm observations. The amount of financially distressed per year are displayed in figure 1.
The second sample consists of S&P1500 firms in the period 2009 until 2015. Gathering financial data in Compustat led to 15909 unique firm observations. Merging this dataset with the audit fee and location data from AuditAnalytics resulted in 5554 deleted observations. A note has to be made that of these 5554 deleted observations 4484 observations were in the period from 2009 until 2014 and 1070 observations were in 2015. This is caused by the lower recent data availability in the AuditAnalytics database, as already mentioned in the former section. After the merge of both datasets, all financial data necessary for the calculation of variables was checked. This resulted in 3023 deleted observations. The final step of the reduction process was the year-completeness check. This resulted in 1440 deleted observations. 2015 is excluded in this completeness test, because of the high amount of deleted observations in this year caused by the low auditor data availability. These steps resulted in the final sample of 5892 observations, which consists of 968 observations per year for the period from 2009 until 2014 and 84 observations for 2015. These 5892 observations consist of 1028 financially distressed firm observations. The amount of financially distressed per year are displayed in figure 1.
After the former reduction process, the data is uploaded in STATA. Just as the in the discretionary accruals model, all variables which are used in the regression are winsorized at the first and last percent. After winsorizing, the variables are tested for linearity, normality, homoscedasticity and multi-collinearity as well. Making scatterplots does the linearity test. No issues were shown, so linearity is assumed. To test for normality, the histograms are
29 shaped curve, so there are no normality issues either. As third, the sample is tested for homoscedasticity. Three tests are used to test for homoscedasticity: (1) residual versus fit plot, (2) Cameron & Trivedi’s decomposition of M-test and (3) Breusch-Pagan / Cook-Weisberg test. No issues were found, so homoscedasticity is assumed. Lastly, the variables are tested for multi-collinearity. Testing for multi-collinearity shows whether variables are highly correlated. The Variance Inflation Factors (VIF) Values are checked per regression and displayed in appendix 2. In the three tables, the highest VIF value is 2.24, which is far below the threshold of 10 (Chen et al., 2010; Marquardt, 1980; Li, 2009). This is why is assumed that multi-collinearity is not present.
Table 3: Data reduction process model 2.
Table 3
Data reduction process model 2 Going-Concern opinion Model
Period 1 2004-2007
Period 2 2009-2015
Unique Firm observations 10896 15909
Auditor Data (3318) (5554)*
Financial Data (1937) (3023)
Incomplete Years (1045) (1440)
Sample Firms: 4596 5892**
Financially Distressed firms: 751 1028
* = Of the deleted auditor observations were 4484 observations in 2009-2014 and 1070 observations in 2015
** = Distribution of the sample is as follows: - 2009-2014: 968 observations per year - 2015: 84 observations
Figure 1: Financially distressed firms per year
3.2 Discretionary accruals model
For hypotheses 1a, 1b, 2a and 2b, audit quality is measured by the discretionary accruals. Discretionary accruals are the part of the total accruals that can be manipulated by the management. This is a manner for the management to manage earnings upwards or downwards, and so on create an unfair view of the financial statements for users of these financial statements. The most-common used model to investigate earnings management is the Modified Jones model, initially introduced by Jones (1991). Dechow et al. (1995) conclude that the Modified Jones model is the most powerful model in detecting earnings management. Reynolds and Francis (2001) and Chung and Kallapur (2003) also use the Modified Jones model in their researches. These researches are leading researches in the audit quality field, so this is why in this research is also chosen to use the Modified Jones model to investigate discretionary accruals. The model is as follows:
Equation 1: Total Accruals
TAt = (∆CAt - ∆CLt - ∆Casht + ∆STDt – Dept) / At-1 189 175 176 186 232 133 129 163 163 176 8 0 50 100 150 200 250 2004 2005 2006 2007 2009 2010 2011 2012 2013 2014 2015
31 With:
Table 4: Total Accruals Variables
Table 4
Total Accruals Variables
TA Total accruals
∆CA Change in current assets
∆CL Change in current liabilities
∆Cash Change in cash and cash equivalents
∆STD Change in debt included in current liabilities Dep Depreciation and amortization expense
A Total Assets
Total accruals consists of discretionary accruals and non-discretionary accruals:
Equation 2: Total Accruals (2)
TA = DA + NDA
After calculating the TA, a cross-sectional regression analysis is made of the groups per industry, to estimate the values of the industry specific parameters. The firms are classified per industry by using the SIC code classification as illustrated in table 6. The following model is used:
Equation 3: NDA regression
NDAt = a1 (1/At-1)+ a2 ((∆REVt - ∆RECt)/At-1) + a3 PPEt / At-1 With:
Table 5: NDA regression variables
Table 5
NDA regression variables
NDAt Non-discretionary accruals current fiscal year
At-1 Total assets previous fiscal year
∆REVt Revenues current fiscal year (t) minus revenues previous fiscal
year (t-1)
∆RECt Net receivables current fiscal year (t) minus net receivables
previous fiscal year (t-1)
PPEt Property, plant and equipment in current fiscal year (t)
Table 6: SIC Code Table
Table 6
SIC* Code Table
Observations Range of SIC
Codes
Industry Type Sample 1 Sample 2 Total
0100 – 0999 Agriculture, Forestry and Fishing 0 0 0
1000 – 1499 Mining 88 180 268
1500 – 1799 Construction 4 19 23
2000 – 3999 Manufacturing 464 977 1441
4000 – 4999 Transportation, Communications, Electric, Gas and Sanitary services
180 380 560 5000 – 5199 Wholesale Trade 28 49 77 5200 – 5999 Retail Trade 112 224 336 6000 – 6999 Financial Institutions 0 0 0 7000 – 8999 Services 172 304 476 9100 – 9729 Public Administration 0 0 0 9900 – 9999 Nonclassifiable 0 0 0 1048 2133 3181
* Standard Industrial Classification (SIC) codes are gathered from the North Carolina State University, at: https://onece.ncsu.edu/mckimmon/divisionUnits/ceus/sicCodePickList
To test the possible association between the discretionary accruals and client importance, the following OLS (Ordinary Least Squares) regression model is used:
Equation 4: Discretionary Accruals regression model
|DA| = 0 + 1 TFR / AFR / NAFR+ 2 FIRMSIZE +3 OCF + 4 DOCF+ 5 BIG4
33 With:
Table 7: Discretionary Accruals regression model variables
Table 7
Discretionary Accruals regression model variables |DA| Absolute value of discretionary accruals
TFR Total Fee Ratio = Total client fees / Local office revenues AFR Audit Fee Ratio = Total audit fees / Local office revenues NAFR Non-Audit Fee Ratio = Non-Audit Fees / Local office revenues FIRMSIZE Natural Logarithm of the firm’s total assets
OCF Operating cash flow at the end of the year / total assets year (t-1) DOCF Dummy Variable for OCF. 1 for OCF values equal to zero or above zero, 0
for OCF values under zero
BIG4 Dummy Variable for Big 4 audit firm. 1 for Big 4 audit firms, 0 for non-Big 4 audit firms.
PYROA Net income year (t-1) / Total assets year (t-2)
DPYROA Dummy Variable for PYROA. 1 for PYROA equal to zero or above zero, 0 for OCF under zero.
GROWTH (Assets (t) – Assets (t-1)) / Assets (t-1) LEVERAGE Total Debt (t) / Total Assets (t)
The dependent variable is |DA|, which is the absolute value of discretionary accruals. This is the absolute value of the residual from equation 3, the NDA regression. The absolute value of the discretionary accruals is used to measure both upwards and downwards earnings management (Chung & Kallapur, 2003; Becker et al., 1998). TFR, AFR and NAFR are the independent or test variables. TFR is the Total Fee Ratio, which is the measure for client importance. This ratio is calculated as the total fees (audit + non-audit fees) divided by the total local office revenues according to Li (2009).The total fees exist of audit fees and non-audit fees. This is why AFR, the non-audit fee ratio, is also used to measure client importance, as well as NAFR, the non-audit fee ratio.
All other variables are control variables. FIRMSIZE is the natural logarithm of the firm’s total assets. It is included because according to prior literature relatively large firms will probably have systematically smaller accruals (Reynolds & Francis, 2000; Reynolds et al., 2004). BIG4 is a dummy variable which equals 1 if the company is audited by a Big 4 auditor, and which equals 0 if not. This dummy variable is included, because prior literature
suggests that Big 4 auditors are less likely to allow earnings management (e.g. Frankel et al., 2002; Francis et al., 1999; DeFond & Jiambalvo, 1994; Becker et al., 1998). OCF and PYROA are included, because discretionary accruals are dependent on these variables (Dechow et al., 1995; Chung & Kallapur, 2003). Following the research of Chung and Kallapur (2003), dummy variables for OCF and PYROA are created. These dummy variables are created, because the dependent variable is the absolute value of discretionary accruals. These dummy variables will indicate whether the OCF and PYROA value is positive or negative. For a positive value of OCF and/or PYROA, the dummy variables will be equal to 1. If the OCF and/or PYROA have a negative value, the dummy variable will be 0. Another control variable is GROWTH, which is the change in assets from the last fiscal (t-1) year to the current fiscal year (t) divided by the assets from the last fiscal year. This variable is added to the model, because according to Reynolds et al. (2004), previous research has shown that discretionary accrual models generally do not work well for extreme-performance firms. The last variable in the model is LEVERAGE. This is the debt-to-asset ratio, and this variable is included because according to prior literature, firms with higher debt levels are more incentivized to use discretionary accruals to manage earnings upwards when firms are about to violate debt covenants (e.g. Reynolds & Francis, 2001; Frankel et al., 2002; Chung & Kallapur, 2003; Reynolds et al., 2004; Defond & Jiambalvo, 1994).
3.3 Going-concern opinion model
For hypotheses 3a, 3b, 4a and 4b, audit quality is measured by the propensity of the auditor to issue a going-concern audit opinion. Prior research using this measure, focuses on financially distressed firms (e.g. Reynolds & Francis, 2001; Defond et al., 2002; Li, 2009) because these firms are more likely to receive a going-concern opinion. As earlier mentioned, client importance is measured as the ratio of the client’s audit, non-audit and total fees compared to the total fees received of all clients of the office. This measure is from the paper of Li (2009), and the model in that research will be the basis to investigate the effect of the propensity of issuing a going-concern opinion in this paper as well. The model
35
Equation 5: Going-concern opinion model logistic regression:
GC = 0 + 1 TFR / AFR / NAFR + 2 SALES + 3 ROA + 4 LEVERAGE
+ 5 LIQUIDITY +6 CHGDT + 7 PRLOSS + 8 PRNOCF + 9 DELAY
+ 10 NEWDEBT
With:
Table 8: Going-concern opinion model regression variables
Table 8
Going-concern opinion model regression variables
GC Dummy Variable for Going-Concern. 1 for companies receiving a going-concern opinion, 0 if not.
TFR Total Fee Ratio = Total client fees / Local office revenues AFR Audit Fee Ratio = Total audit fees / Local office revenues NAFR Non-Audit Fee Ratio = Non-Audit Fees / Local office revenues SALES Natural logarithm of client’s total sales at the end of the year. ROA Net Income (t) / Total Assets (t)
LEVERAGE Total debt (t)/ Total Assets (t)
LIQUIDITY Total current assets (t) / Total current liabilities (t)
CHGDT The change in long-term debt divided by total assets, from year (t-1) to year (t) PRLOSS Dummy Variable for losses in previous year. 1 for negative net income in (t-1), 0 if
not.
PRNOCF Dummy Variable for negative operation cash flows (OCF) in previous year. 1 for negative OCF in (t-1), 0 if not.
DELAY Number of days between fiscal year-end and the auditor’s report signing date NEWDEBT Dummy Variable for debt issuance. 1 if the company issues new debt, 0 if not.
TFR, AFR and NAFR are the test (independent) variables for client importance. All other variables are control variables4, which can influence the auditor’s
propensity of issuing a going-concern opinion. According to Li (2009) these
4 For the discretionary accruals model and the going-concern opinion model, different control variables are used.
This is done because both models contain different dependent variables. Reynolds and Francis (2001) also examined both models in their research, and they also used different control variables. This is done, because different factors could influence the absolute value of the discretionary accruals and the propensity of the auditor to issue a going-concern opinion. I focused on leading papers in prior literature per model. For the discretionary accruals model I chronologically examined leading papers about discretionary accruals (Dechow et al., 1995; Reynolds & Francis, 2001; Frankel et al., 2002; Chung & Kallapur, 2003; Reynolds et al., 2004) and chose the most used and as best described control variables which could influence the absolute value of the discretionary accruals. For the going-concern model I also examined leading papers about going-concern opinions (Reynolds & Francis, 2001; Defond et al., 2002; Geiger & Rama, 2003; Callaghan et al., 2009; Li, 2009) and chose the most used and as best described variables which could influence the propensity of the auditor to issue a going-concern opinion.
control variables are: client size (SALES), the extent of financial distress (ROA, LEVERAGE, CHGDT, LIQUIDITY, PRLOSS and PRNOCF), audit reporting lag (DELAY) and client new financing (DELAY)(p.210).
SALES is included because prior research found a negative relation between company size and the likelihood of receiving a going-concern opinion (Li, 2009; Reynolds & Francis, 2001). An explanation for this is because auditors do not expect big clients to go bankrupt, because of the availability of resources to prevent a bankruptcy. (McKeown et al., 1991; Mutchler et al., 1997). Financial distress variables are included for multiple reasons. First of all, a client’s financial condition is expressed in the company’s ROA (Li, 2009) and LEVERAGE (Li, 2009; DeFond et al., 2002). LIQUIDITY is used to control the companies’ liquidity risk (Li, 2009). CHGDT is included, because more debt is associated with a higher risk of violations (Mutchler et al., 1997; Reynolds & Francis, 2001). PRLOSS and PRNOCF are included because firms with multiple negative cash flows or negative incomes are more likely to fail (Li, 2009, p.210). DELAY is included because prior research finds that going-concern firms are associated with longer reporting lags (e.g. McKeown et al., 1991; Mutchler et al. 1997; Defond et al., 2002). Finally, NEWDEBT is included because new financing can increase liquidity, and so reduce the bankruptcy risks (Li, 2009; Geiger & Rama; 2003). Besides that, the ability to borrow new debt can be seen as a trust from the loan provider about the firm’s capability to meet loan requirements (Behn et al., 2001).
