Audit fees and auditor’s reports under the new auditing standards:
the roles of corporate boards and CEOs
University of Groningen Research Master Thesis Author: Yueyao Ding Student number: S3015920 Supervisor: Dr. Reggy Hooghiemstra
Co-assessor: Dr. Vlad Porumb
Abstract
New auditing standards implemented since 2013 in the UK require auditor’s reports disclosed in an extended form. This reform is intended to make traditional standardized auditor’s reports more firm-specific with relevant information useful to report users. Drawing on the agency theory, this research explores how auditors, boards of directors and CEO contribute to better quality of auditor’s reports. A sample of FTSE 100 constituent firms for the period 2013 - 2015 was adopted. Results reveal that higher audit fees encourage auditors to issue more informative auditor’s reports but not necessarily to issue more readable reports. Neither corporate boards nor CEOs present significant moderation influence on the association between audit fees and quality of audit reports. Research findings provide more insights into improving quality of auditor’s reports and call for continuous attention to the implementation of the new auditing standards.
2
1. Introduction
Auditor’s reports issued by external auditors contain auditors’ opinions about financial statements of auditees. An unqualified opinion validates that financial statements provide “true and fair” view and are free from “material misstatements” whereas other opinions basically suggest that financial statements fail to do so. Investors search information from auditor’s reports for their investment decisions and listed firms rely on auditor’s reports with an unqualified opinion to attract more investors. However, traditional auditor’s reports have been criticized as a standard pass or fail model which provides limited information for report users in capital markets (Church et al., 2008; Mock et al., 2013). Scholars, professionals and regulators have been exploring how to enhance the quality of auditor’s reports.
Since 2013, a new auditing standard requiring an extended auditor’s report has been taken into effect in the UK. According to current auditing standards and regulations in most markets around the globe, a typical auditor’s report summarizes auditors’ overall opinion in a few standardized sentences which state whether financial statements “provide true and fair value” or not (Institute of Chartered Accountants in England and Wales (ICAEW), 2016). However, the new standard in the UK requires that “the most significant risks of material misstatement”, “the application of materiality” and “the scope of the audit” should be disclosed (Gutierrez et al., 2016). This requirements of extended disclosure mean that a binary model with a brief answer of “pass or fail” is no longer sufficient.
In 2015, the International Auditing and Assurance Standards Board (IAASB) revised ISA (International Standard on Auditing) 700 and issued ISA 701, likewise requiring a more extensive disclosure in auditor’s reports. ISA 701 raised the concept of Key Audit Matters (hereafter, KAMs) and defines it as audit matter which external auditors think “were of most significance in the audit of the financial statements of the current period” (IAASB, 2015; PricewaterhouseCoopers (PwC), 2015). This shares the same concept with “the most significant risks of material misstatement” implemented in the UK in 2013 (Lawson et al., 2017).
3
audit fees. Paragraphs describing KAMs1 detected in the audit work did influence how report users make decisions, while this influence seem to vary among different types of report users (Christensen et al., 2014; Köhler et al., 2016). However, studies on auditor’s reports under the recent auditing standard reform are far from sufficient. It remains unclear how quality of auditor’s reports, the main target of recent auditing standards, is influenced (PwC, 2015). Since this recent reform is mainly aimed at improving information value and readability of auditor’s reports (PwC, 2015; Köhler et al., 2016), this research will explore key factors that influence informativeness and readability of auditor’s reports. The agency theory will be employed to investigate interaction between auditors and corporate boards as well as between auditors and CEOs in delivering auditor’s reports.
Relying on a sample of FTSE 100 constituents during the fiscal year 2013 to 2015, results support that higher audit fees lead to more informative auditor’s reports but do no influence the readability. Meanwhile, deviated from our expectation, boards of directors and CEOs in audited firms do not influence the relationship between audit fees and quality of auditor’s reports. even though hypotheses formed in this research are not fully supported, this paper contributes to future research for several reasons.
First, unlike many current studies which focus only on auditors, shareholders or managers (eg. Dao and Pham, 2014; Beasley and Petroni, 2001), this paper clarifies two main agency relationships related to auditor’s reports and explore interactions of different players. This investigation in the process of forming auditor’s reports is important because the new standards address that auditors should communicate with “those charged with governance” in audited firms to determine KAMs (IAASB, 2015). Insignificant moderation effects of board strength and CEO power remind us that both corporate boards and CEOs should be included in one model and their interaction should not be overlooked. This is in line with previous findings of interaction effects between two parties (eg. Westphal and Zajac, 1995; Lisic et al., 2016), calling for further exploration on how to capture their interaction more precisely.
Second, this research focuses on informativess and readability of auditor’s reports; higher
1Christensen et al. (2014) focus on critical audit matters (hereafter, CAMs) proposed by the Public Company Accounting Oversight
4
audit fees turn out to make auditor’s reports more informative but not necessarily more readable. While many current studies measure audit quality or quality of auditor’s reports by presence of misstatement or discretionary accruals (Knechel et al., 2012; Knechel and Sharma, 2012), this research provides direct evidence how informativess and readability are influenced, the main concern of the recent auditing standards to enhance transparency and readability of reporting (PwC, 2015; Köhler et al., 2016). Differentiated results in informativeness and readability remind us that measuring quality of reporting should not be limited in one or two indicators. Furthermore, insignificant results of readability implies that more informative reports may not be more readable so facilitating communications with report users could be more challenging than closing the information gap between them (Lehavy et al., 2011; Mock et al., 2012). This demonstrates that improving readability of auditor’s reports is a critical step to enhance overall quality of reporting.
Third, this research selects instrumental variables to alleviate risks of endogeneity and tests results further validate main findings. Potential endogeneity is discovered in models of informativeness, consistent with Markelevich and Rosner (2013) that audit fees and reporting quality are likely to be simultaneously determined. Although this research could not adopt more possible instruments suggested by previous studies due to limited data availability (eg. Ireland and Lennox, 2002; Geiger and Rama, 2003), it provides evidence that informativeness is more likely to be endogenous than readability in the association with audit fees. It also provides more insights into selecting an effective instrument which is critical to reach solid conclusions in accounting research (Larcker and Rusticus, 2010).
5
2. Theoretical backgrounds and hypotheses 2.1 Extended auditor’s reports
2.1.1 Recent changes in auditing standards and Key Audit Matters
Despite that auditor’s reports play important roles in capital markets, they have long been criticized for their template form and standardized wording (Gutierrez et al., 2016). Traditional auditor’s reports present a pass or fail model which answers whether the report is unqualified (pass) or qualified (fail) but does not provide further information about specific issues related to audited firms (Church et al., 2008; Mock et al., 2013); these reports have been increasingly questioned since the 2008 financial crisis (ICAEW, 2016). Report users desire more relevant information than a simple “pass” or “fail” comment for decision-making purposes.
Responding to changes in the market, regulators and standard setters have revised existing standards and released new standards in the last few years. One of the most recent changes in auditing standards occurred in the UK. In 2013, the Financial Reporting Council (FRC) issued ISA
(UK and Ireland) 700- The independent auditor’s report on financial statements (Revised June 2013) (FRC, 2013). This revised version requires an extended form of auditor’s reports compared
to that of traditional auditor’s reports: it requires auditors of all premium-listed companies on the London Stock Exchange (LSE) to disclose “the most significant risks of material misstatement, the application of materiality and the audit’s scope” (Gutierrez et al., 2016). FRC (2016) explains that the “significant risks” refer to “risks of material misstatement that were identified by the auditor, and which had the greatest impact on the audit strategy, resources required and the work of the engagement” and that the scope includes “how it responded to the risks of material misstatement and the application of materiality”.
In January 2015, the International Auditing and Assurance Standards Board (IAASB) revised
ISA 700- Forming an Opinion and Reporting on Financial Statements and released ISA 701- Communicating Key Audit Matters in the Independent Auditor’s Report (IAASB, 2015; ICAEW,
6
governance” (IAASB, 2015). The definition of KAM, which share the same thought as “the most significant risks of material misstatement” by FRC, is regarded as the most important change in this reform of auditor’s reporting (PwC, 2015; Köhler et al., 2016). This reform makes a binary auditor’s report not sufficient anymore in relevant countries after its implementation. Instead, an auditor’s report with the extended form is expected to describe what were the most significant risks of material misstatement during auditing, why they were significant risks, how auditors dealt with them and what auditors found out (IAASB, 2015). These changes are expected to provide more specific insights of audited firms, enhance transparency of the audit and improve readability of auditor’s reports (PwC, 2015).
2.1.2 Audit quality and the quality of auditor’s reports
When regulators and auditor’s report users expect extended auditor’s reports to provide more specific insights of auditees, enhance transparency and improve readability (PwC, 2015; Köhler et al., 2016), their expectations are connected to audit quality. While there is no single definition of audit quality, a commonly referred definition from DeAngelo (1981) includes two main components: “the likelihood that an auditor discovers existing misstatement” and “appropriately acts on the discovery” (Knechel et al., 2012). These two components are summarized as the “competence” to detect a breach and the “independence” to report the breach for an auditor (FRC, 2008; Abbott et al., 2016).
FRC (2008) identifies five drivers of audit quality, which include the culture within an audit firm, skills and personal qualities of auditors, effectiveness of audit process, reliability and usefulness of reporting and other factors that beyond auditors’ control. Francis (2011) points out that audit quality lies in every unit of audit, ranging from audit inputs and process to audit environment and outcomes. Knechel et al. (2012) summarize indicators of audit quality and view an auditor’s report as a signal of audit quality. Hence, when report users complain that the binary model and standardized wording contain too many technical terms to read thoroughly or provide insufficient information for investment decisions, they express dissatisfaction with the quality of auditor’s reports, an audit outcome which accounts for an important aspect of audit quality.
7
conservative with disclosure (Afterman, 2016; Lennox et al., 2017; Motahary and Emami, 2016). To have a more in-depth understanding of the two viewpoints and a clearer view of actual outcomes of this reform in auditing standards, following paragraphs will review relevant research on auditor’s reports, particularly reports under the recent auditing standards reform.
2.1.3 Traditional auditor’s reports and extended auditor’s reports under the recent auditing standards reform
Auditor’s reports and quality of auditor’s reports have attracted continuous attention from the business and the academia. The information content of auditor’s reports is one of the most prevalent research topic. Researchers point out that an information gap exists between “what users desire” from auditor’s reports and “what is available” to report users in auditor’s reports; many informative items that report users expected are poorly disclosed (Asare and Wright, 2012; Mock et al., 2013). Bamber and Stratton (1997) test whether an explanatory paragraph of uncertainty modification is redundant when all required financial statements have been disclosed according to the generally accepted accounting principles (GAAP). Their findings reveal that bank loan officers are less willing to grant a loan and more likely to require higher loan interests premium to a firm which includes an uncertainty modified paragraph in the auditor’s report , supporting the information value of this additional paragraph. However, in an Australian context, Bessella et al. (2003) find that a modified paragraph in the auditor’s report in the presence of going concern uncertainty does not present additional information value, since this paragraph does not significantly influence risk assessment or loan grant of report users.
Even though not every additional disclosure necessarily make a difference to the decision making of report users, shareholders would like to have more extensive disclosure of firm-specific information especially difficulties and risks that firms are faced with (ICAEW, 2007). Mock et al. (2013) summarize previous studies and agree that report users prefer more informative disclosure about “the auditor, the audit and the financial statements”. Therefore, what information is disclosed and how much information is available closely influence how investors react to auditor’s reports, composing a critical aspects of quality of auditor’s reports.
8
reports are helpful to report users, Audit Quality Forum reveal that many shareholders think that reports should be easier to read in order to make effective use of auditor’s reports (ICAEW, 2007). This suggests that auditor’s reports which are too technical to read impair the information value perceived by report users. In addition, the communication effectiveness of information in auditor’s reports vary from professional users such as institutional investors and unprofessional or less sophisticated users such as private investors without relevant knowledge backgrounds (Gray et al., 2011). Unprofessional report users experience more difficulty in obtaining relevant information (Bedard et al., 2012). Therefore, improving readability of auditor’s reports is another important driver to enhance quality of auditor’s reports, especially to the massive unprofessional users in capital markets.
The recent reform in auditing standards made a further step to satisfy demands of more informative and readable auditor’s reports (PwC, 2015; ICAEW, 2016). Although the recent reform has not been extensively studied owing to its very recent implementation, a few studies have provided some latest insights (eg. Gutierrez et al., 2016; Cordos and Fülöpa, 2015 and Christensen et al. 2014).
Gutierrez et al. (2016) investigate information contents of extended auditor’s reports in the UK. They find that higher audit fees are related to longer auditor’s reports, with more risks identified and disclosed and longer text to describe each risk on average, whereas readability of auditor’s reports was ignored. Comparing reports issued two years before and after issuing the new standards, Gutierrez et al. (2016) note that audit fees did not change as much as expected and investors’ reaction did not change significantly with issuing of extended auditor’s reports. These findings increase concerns whether the new auditing standards could make a difference.
9
this additional paragraph describes how the risk identified in the KAM paragraph was audited and explicitly assures that no material modification is necessary in reporting of this risk. The authors find that participants who read the forth type of auditor’s reports are more likely to invest in the audited firm compared to those who read the second type of auditor’s reports. This means that the influence of the KAM paragraph is reduced in the presence of a resolution paragraph. Therefore, different from Gutierrez et al. (2016), Christensen et al. (2014) validate potential added value of KAM disclosure required by the new auditing standards.
Köhler et al. (2016) perform another study of the information value of KAM disclosure on goodwill impairment. The authors extend the previous two studies by findings on different types of report users. They find that professional report users tend to evaluate financial situations of the auditee as more negative whereas non-professional users do not show significant reaction. They explain that non-professional users are unable to fully understand information communicated in reports.
These recent studies provide some insights but inconsistent evidence of the recent auditing standards reform. Köhler et al. (2016) argue that non-professional report users have lower capability to fully understand auditor’s reports, which extends Christensen et al. (2014) and Gutierrez et al. (2016) by looking beyond the informativeness of reports. However, readability, the other critical aspect of quality of auditor’s reports still not stands out. Furthermore, these studies do not answer what lead to reports which are not informative or less readable. To provide broader insights into the recent auditing standards, following paragraphs will use agency theory to study quality of auditor’s reports.
2.2 Agency theory and the quality of auditor’s reports
Agency theory is one of the most discussed theories in business research. Berle and Means (1932) regard a firm as “a nexus of contracts between principals and agents”. The agency problem occurs when ownership and control are separated: i.e., an agent “perform some service” on the behalf of a principal “which involves delegating some decision-making authority to the agent” (Jensen and Meckling, 1976). Since both the principal and the agent are “utility maximizers”, they have different interests and the principal may hardly know how the agent is doing due to asymmetric information (Eisenhardt, 1989).
10
1989). To ensure that their interests are not invaded by managers, shareholders want to know the actual performance of managers; on the other hand, managers want to prove that they fulfil their responsibilities; external auditors as an independent monitoring of managers are thereby demanded (Watts and Zimmerman, 1983; Melumad and Thoman, 1990; Baiman et al., 1987). More specifically, since it is difficult for shareholders to gain direct and clear information about the work done by managers, external auditors are expected to verify the information provided by managers and disclose opinions of the performance of managers in auditor’s reports (Watts and Zimmerman, 1983). This means an auditor’s report serves as a tool to mitigate information asymmetry between shareholders and managers. Therefore, what information is disclosed and how the information is disclosed indicate the quality of auditor’s reports and the work done by auditors (Knechel et al., 2012). The disclosure of auditor’s reports is closely related to how much the agency problem between shareholders and managers could be mitigated.
While recognizing the agency relationship between shareholders and managers, we should also realize another agency relationship when discussing auditors and auditor’s reports. As the third party between shareholders and managers, auditors have interests which do not necessarily match those of shareholders or managers. Antle (1982) explains that many studies ignore auditors’ own incentives. He thinks that an external auditor should be modelled as “an expected utility maximizer” and treated as an “economic agent” in the shareholder-manager agency relationship. Khurana and Raman (2004) agree that managers are required to disclose financial statements to shareholders so managers purchase audit services from auditors to prove the reliability of financial statements. Managers expect an auditor to disclose a report which is in favour of their own benefits, normally an unmodified report verifying that managers performed well in the last audit period. However, such a report may not maximize auditors’ interests, because what managers could gain from this report, including performance-based compensations and lower capital costs, probably do not benefit auditors, especially because that auditors have to take litigation responsibility for the auditor’s report . From this perspective, the disclosure of auditor’s reports is also influenced by the relationship between managers and auditors.
11
(Köhler et al., 2016), which is determined by auditors and shaped by interactions with shareholders and managers. Therefore, this research will investigate how the quality of auditor’s reports, the main target of the recent auditing reform, is influenced by auditors, shareholders and managers. Since closing information gaps and communication gaps is a key driver to better quality of auditor’s reports, this research will focus on informativeness and readability of auditor’s reports. The following sections will discuss the research question in more details and form hypotheses.
2.3 Audit fees and the quality of auditor’s reports
While listed firms pay external auditors for their audit service to verify financial information provided to shareholders and other stakeholders in capital markets (Depuch and Simunic, 1982; Khurana and Raman, 2004), audit firms largely rely on audit fees to support audit work and daily operation. Many studies contend that audit fees link the interests of external auditors and those of managers from the audited firm by creating an economic bonding (eg. Asthana and Boone, 2012; FRC, 2006; Frankel et al., 2002). If an auditor detects certain risks that managers are not willing to disclose and even try to conceal but the auditor insists on disclosure, this auditor will probably annoy managers and run the risk of being replaced by another auditor next time. Therefore, to keep their clients and a main income source, auditors tend to disclose auditor’s reports in the way that managers prefer, thus causing threat to the independence and credibility of information conveyed in auditor’s reports (Baiman et al., 1991; FRC, 2006). In brief, this argument thinks that the economic bonding helps to align the interests of auditors to those of managers, casting doubts on audit independence and hence the quality of auditor’s reports. In this view, higher audit fees impair audit independence to a greater degree and ultimately results in lower quality of auditor’s reports.
12
found to misbehave in audit, the reputational damage to this single auditor could spread to other auditors in this firm, cause reputation loss to the whole audit firm and make it more difficult to keep current clients and attract new clients (He et al., 2016). Therefore, auditors have incentives to not only behave themselves but also monitor peer auditors. The possibility of expanded reputation loss and peer pressure deviate the interests of auditors even further from interests of managers, making it more difficult for auditors to flatter managers with auditor’s reports of lower quality. From this perspective, the litigation risk and reputation risk motivate auditors to behave themselves and ensure good quality of auditor’s reports.
In addition, in a market where listed firms and auditors are free to choose agents and clients respectively, a certain amount of audit fee is reached by agreement of the listed auditee and the audit firm. From the demand side, the listed firm is willing to pay higher audit fees because of stronger demand of high-quality audit services (Mitra et a., 2007; Bliss, 2011), which facilitates a better quality of the auditor’s report. Besides, higher audit fees provide auditors with sufficient budget and resources to work; after all, taking all necessary auditing procedures become difficult if auditors have only very limited budget (FRC, 2006). From the supply side, the audit firm asks for higher audit fees because it assesses that more audit efforts are needed and it is likely to take higher audit risks of this auditee (Bell et al., 2001; Mitra et al., 2007; Bliss, 2011). Moreover, audit firms with big brands such as “Big 4” have well-recognized reputation of better audit services and usually gain higher audit fees (Basioudis and Francis, 2007). In turn, “Big 4” care more about their reputation, since once their reputation is damaged, they have much more to lose than other audit firms which do not have as many clients (DeAngelo, 1981; Reynolds and Francis, 2000). Arthur Andersen for example, which used to be one of the “Big 5” audit firms, went bankrupt in a short period after the Enron scandal. Therefore, in either case, higher audit fees is related to higher audit quality, hence auditor’s reports tend to be more informative and more readable.
Therefore, although higher audit fees cause stronger economic bonding which may threaten audit independence, auditors are faced with considerable litigation risks and reputation risks. In addition, higher audit fees facilitate sufficient audit efforts and audit firms which earn higher audit fees tend to bear higher risks. Thus, this research expects a positive influence of audit fees on the quality of auditor’s reports.
13
2.4 Boards of directors and the quality of auditor’s reports
Emphasizing the agency relationship between shareholders and managers, agency theorists contend that effective monitoring of managers is necessary to mitigate the agency problem (Shleifer and Vishny, 1997; Zahra and Pearce, 1989). Standing for interests of shareholders, boards of directors2 occupy the uppermost positions in listed firms and play a vital role in corporate governance: providing strategic advice, introducing network resources, and most importantly, monitoring the management team (Zahra and Pearce, 1989; Eisenhardt, 1989; Van den Berghe and Levrau, 2004). The corporate board is one of the most central mechanisms of internal monitoring (Daily et al., 2003). When communicating with capital markets in corporate financial reports, boards of directors are supposed to protect interests of shareholders and secure the quality of communication. Therefore, how effective a corporate board fulfil its responsibilities is closely related to the quality of audit work and auditor’s reports.
Current research recognizes that boards of directors contribute to corporate reporting from different aspects and propose an integrated concept of board strength (eg. Hoitash et al., 2009; Hooghiemstra, 2012). Hoitash et al. (2009) take the size, independence and meetings of a board and tenures and external positions of directors into consideration; the authors conclude that a stronger board leads to higher internal control effectiveness and higher reporting quality. Hooghiemstra (2012) employs a similar index including the size, independence and meetings of a board and the number of committees within a board in forming board strength; the finding suggests that a stronger board facilitates more informative disclosure.
Other studies which focus on one or two aspects of board characteristics provide additional support to the influence of board strength in corporate reporting. Carcello et al. (2002) think that a board with higher percentage of outside directors and hold more frequent meetings demands higher quality of audit; this board works more diligently to assure that audit firms put sufficient effort and make effective use of audit fees in auditing. Beasley and Petroni (2001) agree with the influence of outside directors; they think that a corporate board with more outside directors are more likely to keep its independence from the management team and choose audit firms with big brand and more industry expertise; these audit firms normally have more relevant audit
2 Since this research will take listed firm in LSE as samples, a board of directors in this research refers to the one-tier board under
14
experiences and better reputation of high audit quality. Moreover, when external auditors negotiate with the management team regarding specific reporting issues, the existence of an audit committee in the board facilitates external auditors to make their own professional judgement rather than compromise to pressure from the management team (Carcello and Neal, 2003; Badolato et al., 2014). Apart from audit committee, other committees are common in a corporate board, such as compensation committee and nomination committee. These committees take responsible for different aspects of corporate strategies and drive most important board decisions (Kesner, 1998). The recent auditing standard reform explicitly addresses that before disclosing KAMS, auditors should communicate with those in governance of the auditee to determine what to include in the auditor’s report (FRC, 2013). In this context, board strength complements the audit work of external auditors and is supposed to secure the quality of auditor’s reports. Therefore, I expect that a strong board facilitates to the positive influence of audit fees on auditor’s reports.
Hypothesis 2: the strength of corporate boards strengthens the positive influence of audit fees on the quality of auditor’s reports; when a board is stronger, higher audit fees are more likely to result in a more informative and more readable auditor’s report
2.5 CEO and the quality of auditor’s reports
As the top-level manager who takes charge of the overall operation and performance of a listed firm, an CEO is hired to deliver decent financial performance of listed firms (Friedman, 2014). If CEOs fail to deliver the expected result, they will receive much less compensation and will even be dismissed; disappointing performance also limits their access to the next job opportunity. Therefore, CEOs are highly motivated to deliver decent performance and even to manipulate financial results to make their performance “appear” more decent (Bergstresser and Philippon, 2006).
15
During daily operation and management of listed firms, CEOs as agents of shareholders have advantages over boards of directors especially in information that CEOs are not willing to share (Armstrong et al, 2010). Information asymmetry makes it difficult for directors to monitor CEOs behaviours and impairs board effectiveness further as CEOs gain more advantages in audited firms. When investigating audit committee in a board, the committee responsible for securing audit quality, Lisic et al. (2016) find that an CEO who takes the chairman position in a board or has higher relative compensation decreases the quality of information that audit committee depends on to monitor the CEO. In this context, the increasing power of an CEO means the relative decreasing strength of a corporate board. Lin et al. (2014) provide similar arguments that an CEO who takes the chairman position and has served the listed firm for longer period is associated with lower quality of internal control. In this case, the corporate board is less likely to monitor CEO behaviour effectively and external auditors experience more difficulty in collecting audit information.
Therefore, the presence of a powerful CEO impairs the quality of audit process, decrease monitoring effect of corporate boards and make the audited firm exposed to more internal control weaknesses. Based on these concerns, I expect that powerful CEOs impairs the positive influence of high audit fees.
16
3. Methodology
3.1 Samples and sources of data
The auditing standards updated in the UK in 2013 require all premium-listed firms on LSE to provide the extended auditor’s reports, which went into effect for “reports issued after Oct 1st, 2013” (PwC, 2015). By the time of collecting data for this research, annual reports of the year 2016 had not yet been available for all premium-listed firms. Therefore, in this research analyzed annual reports from 2013 and 20153. Considering time limitation and data availability, I focused on Financial Times Stock Exchange (FTSE) 100 Index constituents composed of the 100 most highly capitalized blue-chip companies listed on LSE”4. The list of FTSE 100 constituents in the version of December, 2015. Then, the first-year observations for firms listed during 2013 and 2015 were deleted to avoid potential bias caused by their IPO activities. In addition, the observation of Tui Travel in 2015 was deleted because this firm experienced major restructuring in this fiscal year and only data for few variables for this observation was available. This process led to a sample of 289 observations from 98 firms.
The data of this sample relies on several sources. Data of board characteristics and CEO features were mainly downloaded from BoardEx. Board data not available in BoardEx, general firm-level data and audit fees came from DataStream. Data at the level of auditor’s reports was shared by Professor Dick de Waard and his colleagues; they hand collected data from official annual reports and analyzed effectiveness of auditor’s reports online5. In addition, data not available from any the three sources was hand collected from annual reports by the author.
3.2 Dependent variables and moderators
Dependent variable: informativeness of auditor’s reports. Although the definition of KAM
was not officially taken into effect in auditor’s reports until 2015, “the most significant risks of material misstatement” which was proposed earlier in the UK is in fact consistent with the
3 Firms in this sample have different ending dates of a fiscal year and different timing period of financial statements. For example,
for two reports both issued as 2013 reports, one firm may have a fiscal year from April 2012 to March 2013 while the other may consider a fiscal year from January 2013 to December 2013. To keep consistent timing and improve comparability of reporting, this research treated all reports with a fiscal year ending before July 1st of Year t as reports of Year t-1 and reports with a fiscal year
ending on or after July 1st of Year t as reports of Year t. Therefore, reports of 2015 in my sample include some reports which were
issued as reports of 2016.
4 Official explanation of FTSE 100 is found on http://www.ftse.com/products/indices/uk .
5 The website Tests Document Readability https://www.online-utility.org/english/readability_test_and_improve.jsp was used in
17
definition of KAM (Lawson et al., 2017; Afterman, 2016). To simplify the description and keep consistent wording, this research treats the risks disclosed in the auditing standards change in the UK as KAM too (Köhler et al., 2016). Therefore, the number of KAM presents how many risks that external auditors and audited firms agreed to disclose. This part of disclosure is the key component in the recent auditing standards reports that differentiates the extended report to a traditional report (Gutierrez et al., 2016; Christensen et al., 2014). Therefore, the number of KAM is a compelling indicator of information value contained in auditor’s reports (Lennox et al., 2017).
Dependent variable: readability of auditor’s reports. Traditional auditor’s reports have been
criticized to create communication gaps between reports providers and reports users (Pound, 1981; IAASB, 2011; Mock et al., 2013). Although institutional investors think professional auditor’s reports are easy to read, a large percentage of private investors especially those with insufficient relevant knowledge often have difficulty understanding such reports (Köhler et al., 2016). If an auditor’s report cannot be properly understood by investors from the capital market, its quality is questioned. Therefore, this research uses readability to indicate how well an auditor’s report could be understood. The Gunning Fog Index (hereafter FOG) developed by Gunning (1952) is widely used in social science research (eg. Li, 2008; Lehavy et al., 2011;Ajina et al., 2016). It estimates “the number of years of formal education a reader with average intelligence” needs to understand the text on the first reading, meaning that the easier a report is, the lower its FOG index becomes (Ajina et al., 2016).
Moderator: board strength. Based on Hoitash et al. (2009) and Hooghiemstra (2012), this
18
usually busy with positions in different fields to concentrate on corporate issues and work together. More frequent meetings of a board make directors to devote more time into this firm. Hence, I use number of board meetings during a fiscal year to measure board diligence, a dummy valued as 1 if the number of meetings is more than the median level of the sample and valued as 0 otherwise. Fourth, committees within a corporate board focus on different corporate strategies. In addition to an audit committee responsible for securing the quality of audit work (Badolato et al., 2014; Cohen et al., 2013), common committees such as compensation committee and nomination committee work directly related to executive compensations and appointment. Hence, more specialized committee facilitates a board as a whole to fulfill its responsibilities so I take the number of committees into board strength as well: this dummy is valued as 1 if the number of committees is more than the median level of the sample and valued as 0 otherwise. Therefore, board strength is a score ranging from 0 to 4 and the higher score refers to a stronger board.
Moderator: CEO power. This is an overall indicator including tenures of CEO in the audited
firm, the board and the position of CEO6. According to Magee and Galinsky (2008), the central concept of power is the capability to control valuable resources and influence other people. Hence, what positions a CEO has and how long the CEO has served in these positions largely determines the value of resources and the amount of influence this person occupies. Meanwhile, the tenure of CEOs is well recognized to their power. Westphal and Zajac (1995) agree that serving as a CEO for longer period make the person more familiar corporate operation which allows the CEO to gain more resources from the firm; moreover, long tenure helps a CEO to build up personal authority within organizations. Therefore, this research employs tenures to indicate power. Considering that a CEO plays more than one role in the firm, this research describes personal tenure at different level: tenure as the CEO, tenure in the board of directors and tenure in the firm. Similar to the score of board strength, each aspect of tenure is a dummy, valued as 1 if the number if above the median level of the sample and valued as 0 otherwise. The CEO power is hence the sum of the three aspects, ranging from 0 to 3.
6 CEO duality, the combination of roles as CEO and chairman of the corporate board is a widely used measure (eg. Lisic et al.,
19
3.3 Control variables
Control variables include a series of variables that possibly confound the relationship between independent variables and dependent variables. Based on previous studies and data availability, I take the following factors into consideration.
Firm size. A larger-sized firm tend to have more complex business activities which are
possible to influence the risk assessment of the auditee and the amount of auditing effort (Bell et al., 2001). Following previous studies in accounting and auditing (eg. Hoitash et al.,2009; Hooghiemstra , 2012; Knechel and Sharma, 2012), this research uses Total Assets to describe firm size.
Profitability. A higher profitability is likely to indicate healthier operation and less risky
financial conditions thus influences audit risks and reporting quality (Markelevich and Rosner, 2013; Hoitash et al., 2009). Return on assets (ROA), a widely used indicator of firm profitability (eg. Markelevich and Rosner, 2013; Knechel and Sharma, 2012), is thus controlled in estimating the quality of auditor’s reports.
Leverage. A higher level of leverage provides firms with more financing access but it also
causes more pressure from creditors thus may increase audit risks and effort (Markelevich and Rosner, 2013; Knechel and Sharma, 2012). Hence, the ratio of total debt to total assets is controlled to include potential influence of leverage (Markelevich and Rosner, 2013; Knechel and Sharma, 2012).
Industry. Industry of the auditee is prevalently controlled in previous studies on account that
industry features may influence firm risks and disclosure (eg. Hooghiemstra, 2012). In particular, the financial industry is regarded to experience higher operation risks and different accounting practices compared to other industries thus usually eliminated or treated separately (eg. Knechel and Sharma, 2012; Bell et al., 2001). Therefore, this research controls the industry effect by a dummy variable which equals to 1 if the observation is in financial industry and 0 otherwise.
Time dummies. Potential influence of time is commonly controlled especially when a
substantial change occurred in the market (eg. Hooghiemstra , 2012; Knechel and Sharma, 2012). In this research, since the new auditing standards in the UK were taken into effect in 2013, the fiscal year 20137 was the first time when the overwhelming majority of listed firms issued the
7 The fiscal year here is consistent with previous description how this research divided timing of a fiscal year. Therefore, the
20
extended auditor’s reports. In addition, because the new auditing standards do not provide details such as what risks should be disclosed and how to describe these risks, issuing extended auditor’s reports were a new attempt and exploration for auditors; after the first year of implementation, auditors and auditees learned from all published reports and were able to adjust their disclosure accordingly. Therefore, two dummy variables, year 1 and year 2 are adopted to distinguish three different years of practices.
21 Table 1. Variables description
Variable Indicator Description
KAM
the number of Key audit matters disclosed in an auditor’s report
Before Key audit matter was officially defined, I counted the number of "the most significant risks of material misstatement" disclosed in auditor’s reports.
FOG the Gunning Fog Index
A widely used indicator of readability of text; measures “the number of years of formal education a reader with average intelligence” needs to understand the text on the first reading (Ajina et al., 2016).
AF
audit fees paid from audited firms to external auditors (audit firms)
The audit fees of a certain fiscal year disclosed in annual reports; the unit of numbers is Pound.
BS the strength of a board of directors
The strength of a corporate board includes four aspects of board characteristics: board size, board independence, the number of board meetings and the number of committees established in the board. This is an overall score which sums up scores of four aspects, ranging from 0 to 4.
CP the power of CEO
The power of CEO includes three aspects of CEO features: tenure as the CEO, tenure in the corporate board and tenure in the firm. This is an overall score which sums up scores of three aspects, ranging from 0 to 3.
TA total assets The value of total assets; the unit of numbers is thousand-Pounds.
ROA return on assets The ratio of net income to average total assets, measuring the profitability of a firm.
Leverage debt assets ratio The ratio of total debt to total assets.
Industry financial industry or not
This dummy variable equals 1 if the firm is in financial industry and 0 otherwise. The classification of industry is based on FTSE sector code available in Datastream.
yr1, yr2 year dummies
22
4. Results analyses
4.1 Descriptive statistics and correlation analyses
Table 2 provides description statistics of dependent variables and all explanatory variables except for industry and time dummies. Panel A describes the total sample while Panel B and Panel C describes several subsamples based on time and industry.
According to Panel A, audit fees experience substantial variance, with the max value more than 300 times of the minimum value. FOG Index ranges from around 16 which corresponds to college senior level of education to around 26 which implies text is too difficult to read even for a college graduate; and on average, auditor’s reports are very difficult even for college graduates (with a mean value of 21.067). The disclosure of KAM also varies substantially, from zero KAM described up to 10 KAMs specified. The mean value of KAM (4.388) is greater than that found by Lennox et al. (2017) (3.83) and Gutierrez et al. (2016) (3.928). Since Lennox et al. (2017) adopt a sample of all premium listed firms on LSE and Gutierrez et al. (2016) cover all non-financial premium listed firms on LSE while this research uses a samples of FTSE 100 constituents, a greater mean of KAM to some extent supports that larger-sized firms tend to disclose more KAMs.
The mean of board strength is 1.668, suggesting that the overall strength of corporate boards is relatively low (considering that the max value is 4). Meanwhile, the mean of CEO power is 1.841, suggesting that in CEOs in the sample are relatively powerful at an overall level. However, since the standard deviation of CEO power is 1.174, we can tell that CEO power in different firms is likely to differ substantially.
Panel B presents description of variables by time. The minimum audit fees increased in 2014 and the maximum fees remained stable over the three years. The mean value and maximum value of FOG decreased in 2014 and then stabled in 2015 while the minimum value decreased by time. These changes suggest that auditors possibly learned how to make reports more understandable as they gained more experience. Similarly, the mean value of KAM decreased by time. To some extent, this is consistent with some concern about the new standards that auditors may become more cautious in disclosing risks and this change trend implies that auditors tend to adjust following reporting after learning what their auditor peers disclosed (Afterman, 2016; ; Lennox et al., 2017; Motahary and Emami, 2016).
23
audit fees and more difficult auditor’s reports to read. While the variance of KAM disclosure is very close in financial firms and non-financial firms (standard deviation is 1.722 and 1.713 respectively), financial firms on average disclose fewer KAMs compared to non-financial firms. This difference in the mean of KAM is consistent with the finding of Lennox et al. (2017) which describe that financial firms disclosed fewer KAMs than other firms did. Besides, board of directors in financial firms are stronger than non-financial firms on average (their mean value are 1.736 and 1.645 respectively) while CEOs in financial firms are less powerful on average (1.736 compared to 1.876).
The above descriptive statistics illustrate that variables in general have considerable variance. To avoid potential disturbance of extreme values, I winsorized all continuous variables at 1% level
Table 2. Descriptive statistics
A- Full sample Observation Mean Std. Dev. Min Max
AF 288 5163755 6675026 94000 35000000 FOG 284 21.067 1.524 15.820 25.770 KAM 289 4.388 1.719 0.000 10.000 BS 289 1.668 0.972 0.000 4.000 CP 289 1.841 1.174 0.000 3.000 TA 289 89700000 252000000 333495 1690000000 ROA 289 7.158 7.375 -17.960 51.020 Leverage 287 22.910 16.633 0.000 77.110
B Year 1 Observation Mean Std. Dev. Min Max
AF 94 4924914 6408299 94000 35000000 FOG 93 21.449 1.613 18.420 25.770 KAM 94 4.500 1.740 0.000 10.000 BS 94 1.553 0.923 0.000 4.000 CP 94 1.926 1.138 0.000 3.000 TA 94 90700000 257000000 333495 1610000000 ROA 94 7.774 7.746 -4.590 51.020 Leverage 93 22.141 16.406 0.000 71.100
B Year 2 Observation Mean Std. Dev. Min Max
24
B Year 3 Observation Mean Std. Dev. Min Max
AF 97 5400501 6997771 136000 34600000 FOG 96 20.964 1.591 15.820 24.840 KAM 97 4.216 1.769 0.000 9.000 BS 97 1.711 1.010 0.000 4.000 CP 97 1.753 1.225 0.000 3.000 TA 97 87200000 239000000 848803 1630000000 ROA 97 6.407 7.443 -17.960 39.930 Leverage 96 23.679 16.650 0.000 64.670
C financial Observation Mean Std. Dev. Min Max
AF 72 6075779 8497725 94000 35000000 FOG 71 21.158 1.533 15.820 23.900 KAM 72 4.139 1.722 1.000 9.000 BS 72 1.736 1.233 0.000 4.000 CP 72 1.736 1.245 0.000 3.000 TA 72 297000000 443000000 522844 1690000000 ROA 72 5.179 7.129 -1.150 32.510 Leverage 70 14.221 16.925 0.000 71.100
C non-financial Observation Mean Std. Dev. Min Max
AF 216 4859747 5937938 203000 34600000 FOG 213 21.036 1.523 18.200 25.770 KAM 217 4.470 1.713 0.000 10.000 BS 217 1.645 0.870 0.000 3.000 CP 217 1.876 1.150 0.000 3.000 TA 217 21000000 37000000 333495 223000000 ROA 217 7.815 7.353 -17.960 51.020 Leverage 217 25.713 15.571 0.000 77.110
25 Table 3. Pearson correlations
AF BS CP TA ROA Leverage Industry
AF 1 BS 0.240*** 1 CP -0.117** -0.187*** 1 TA 0.483*** 0.209*** -0.176*** 1 ROA -0.276*** -0.137** 0.228*** -0.285*** 1 Leverage -0.017 -0.118** 0.026 -0.155*** 0.150 1 Industry 0.079 0.041 -0.052 0.474*** -0.155*** -0.297*** 1 * p<0.1; ** p<0.05; *** p<0.01
Notes: correlations with year dummies are omitted in this table, following the practice of Hooghiemstra (2012). All correlations with year dummies are lower than 0.1 and insignificant.
4.2 OLS regressions and hypotheses testing
This research applied pooled Ordinary Least Square (OLS) regressions in Stata 14 to analyze all four models. OLS regressions were conducted with the “robust” option to avoid the risk of heteroscedasticity. Besides, the variance of inflation factors (VIF)8 was checked for every model. Results show that the mean VIF in each model is below 1.5 and the maximum value of VIF is below 2.0. This verifies that multicollinearity is not a problem in all these models. Table 4 presents results of four models and tests three hypotheses.
The second column of Table 4 presents the result of Model 1. The coefficient of AF is significantly positive (p<0.01) while the coefficient of BS*AF is insignificantly negative. The coefficient of TA is significantly positive (p<0.05) whereas the coefficient is very minimal. ROA turns out negatively related to the number of KAM (p<0.05), suggesting that firms with higher profitability disclosed fewer KAMs. The coefficient of industry is significantly negative (p<0.01), meaning that financial firms tended to disclose fewer KAMs. The time dummy yr1 is significantly (p<0.1) and positively related to the number of KAMs, suggesting that reports in the fiscal year of 2013 were likely to disclose more KAMs while reports in the fiscal year of 2014 and 2015 were likely to disclose fewer KAMs. This finding is consistent with what we found in description that the average number of KAMs decreased with time from 2013 to 2015, supporting that auditors learn from published auditor’s reports and cautiously adjusting their reporting. The third column exhibits the result of Model 2. Similar to the result of Model 1, the coefficient of AF is significantly positive (p<0.01) while the coefficient of CP*AF is insignificantly negative. The coefficient of
26
TA is remains significantly positive and minimal (p<0.05). Coefficients of ROA and industry remain significantly negative (p<0.05) and the coefficient of yr1 remains significantly positive (p<0.1) . Results of Model 3 and Model 4 are presented in the two columns on the right. However, AF does not show significant influence in either of the two models. Moreover, the moderation terms of board strength and CEO power are insignificant. Except that the time dummy yr1 remains positive and significant (p<0.1), the rest of control variables are not significant any more.
Table 4. OLS regressions
Model 1 Model 2 Model 3 Model 4
AF 0.511 0.487 0.008 0.085 (0.104)*** (0.097)*** (0.096) (0.094) AF*BS -0.081 0.069 (0.078) (0.079) BS 0.082 0.191 (0.101) (0.093)** AF*CP -0.013 0.028 (0.079) (0.084) CP -0.115 -0.157 (0.104) (0.094)* TA 0.000 0.000 -0.000 -0.000 (0.000)** (0.000)** (0.000) (0.000) ROA -0.202 -0.189 -0.008 0.023 (0.098)** (0.098)* (0.093) (0.093) Leverage 0.010 -0.005 0.062 0.040 (0.087) (0.085) (0.088) (0.089) Industry -0.739 -0.784 0.173 0.197 (0.220)*** (0.212)*** (0.239) (0.239) yr1 0.443 0.440 0.448 0.446 (0.232)* (0.228)* (0.230)* (0.230)* yr2 0.335 0.349 -0.191 -0.184 (0.220) (0.220) (0.206) (0.207) _cons 4.249 4.245 20.949 20.947 (0.172)*** (0.171)*** (0.167)*** (0.169)*** R2 0.19 0.19 0.05 0.04 N 286 286 281 281 * p<0.1; ** p<0.05; *** p<0.01
Notes: standard errors are reported parentheses above and corrected for heteroskedasticity.
27
differentiation in two aspects of auditor’s reports is reasonable if we recognize that identifying and disclosing KAMs require sufficient auditing efforts, which in turn leave less chance for auditors to write reports in “plain language” (Lehavy et al., 2011).
Hypothesis 2 expects a positive moderation effect of board strength on the association between audit fees and quality of auditor’s reports and hypothesis 3 expects a negative moderation effect of CEO power on this association. However, Table 4 presents that either the interaction between audit fees and board strength or the interaction between audit fees and CEO power is significant (p>0.1 from Model 1 to Model 4). Although Graph 1 in Appendix 29 shows an intersection of two lines, it could not provide solid evidence for the moderation effect of board strength because the coefficient of the interaction term is insignificant. Similarly, moderation effects in other models are not supported either. Therefore, hypothesis 2 and hypothesis 3 are not supported. Nevertheless, this does not mean that board of directors or CEOs are not relevant in the relationship between audit fees and quality of auditor’s reports. Instead, this unexpected result could be explained by the interaction between corporate boards and CEOs. For example, although directors take the responsibility for monitoring CEOs, powerful CEOs are able to influence director selection (Westphal and Zajac, 1995); increasing CEO power weakens work effectiveness of corporate boards (Lisic et al., 2016). Thus, the moderation effect of board strength is likely to decreases with increasing CEO power while the moderation effect of CEOs may decrease with increasing board monitoring. Therefore, such mutual interaction is likely to dominate and mask the effect of either single factor.
4.3 Endogeneity tests and instrument variables 4.3.1 Potential risk of endogeneity
Although risks of multicollinearity and heteroscedasticity have been controlled, another risk prevalent in accounting and management research is endogeneity. One source of endogeneity is omitted variable. As part of audit quality, quality of auditor’s reports may be influenced by factors other than explanatory variables included in my models, factors in the context, inputs and the process of auditing (Knechel et al, 2013). Since data source is limited in this research, omitted variable is possible in the models.
28
The other source of endogeneity mainly lies in simultaneity (Chen, 2010). As discussed in the theoretical section, higher audit fees facilitate audit work and audit firms which charge higher audit fees work harder to protect their reputation, thus audit quality and quality of auditor’s reports will be improved. However, it is also conceivable that audit fees as well as the informativeness and readability of auditor’s reports are simultaneously determined. When a firm is experiencing more financial or operational pressure, this firm tends to have more risks to be disclosed as KAMs, resulting in more audit efforts and higher audit fees (Ireland and Lennox, 2002; Markelevich and Rosner, 2013), thus audit fees and informativeness of auditor’s reports are positively associated. In addition, disclosing a readable report is not easy 10 and disclosing a report for an auditee with higher risks and complex business activities requires even more audit efforts (Lehavy et al., 2011), thus higher audit fees are demanded. Then, specific firm characteristics may influence audit fees and readability of reports simultaneously. Therefore, audit fees are likely to be endogeneous in my models.
To explore risks of endogeneity caused by omitted variables and simultaneity and verify research findings, I adopted the method of instruments and two-stage least square regressions (2SLS), a commonly recognized method in accounting research (Larcker and Rusticus, 2010).
4.3.2 Selection of instrument variables
A proper instrument variable should be closely related to the endogenous explanatory variable and uncorrelated to the error term in the original model (Chen, 2010; Larcker and Rusticus, 2010). Thus, non-audit fees and the lagged audit fees were selected as instruments.
Markelevich and Rosner (2013) synthesize several theories regarding the relationship between audit fees and audit quality. Apart from views of the economic bonding, audit effort and audit risks that we have discussed in the theoretical section, they recognize the knowledge spillover effects between audit services and non-audit services. They think that knowledge gained from non-audit services is likely to be transferred to audit work and influences the amount of audit effort and audit outcomes. Knechel and Sharma (2012) point out that similar to audit services, non-audit services also require resources including professional auditors. They explain that if external auditors already gain knowledge about their clients such as structure, management teams and major business activities in non-audit services, auditing work will not take as much time and efforts thus
10 Descriptive statistics support this possibility: the mean value of FOG in our sample is around 21, meaning that on average,
29
audit fees demanded to ensure sufficient auditing work will decrease. Hence, non-audit fees are close related to audit fees. Following Markelevich and Rosner (2013) and Knechel and Sharma (2012), non-audit services fees reflect different firm characteristics including financial risks, business complexity and firm growth. Accordingly, using non-audit fees as an instrument variable of audit fees is possible to capture potential influence of omitted variables. In addition, the informativeness and readability of auditor’s reports are not likely to influence how much non-audit fees that listed firms have to pay. Thus non-audit fees could serve as an instrument of audit fees.
Another commonly used instrument in endogeneity problem is the lagged endogenous variable (Larcker and Rusticus, 2010). The overall level of audit fees that an auditee pay to its external auditor usually remains stable unless it experiences substantial changes in operations and management. For example, a FTSE 100 constituent firm is not likely to reduce its audit fees to the same amount as that of a recently established firm within a short period. Results of descriptive statistics also demonstrate that audit fees fluctuate around a relatively stable level in similar ranges over time. Therefore, audit fees paid in the last fiscal year is highly relevant to audit fees this year. In addition, the “exogenous part” of audit fees which is not jointly determined with quality of auditor’s reports tends to “persist over time” (Larcker and Rusticus, 2010). For instance, external auditors of a listed firm during a certain period normally come from the same audit firm and listed firms are not likely to switch audit firms suddenly. Therefore, characteristics of external auditors, including their professional expertise and protection of their reputation remain stable over time. On the other hand, the “endogenous part” which may be jointly determined with quality of auditor’s reports including financial risks and operational risks are less likely to persist for a long time, especially for FTSE 100 constituents which are among the most well-performed firms listed on LSE. Moreover, the informativeness and readability of auditor’s reports this year are unlikely to influence audit fees paid last year. Therefore, the lagged audit fee is a reasonable choice of instrument.
4.3.3 2SLS regressions and analyses
Based on previous discussion, I adopted the non-audit fees (noted as NAF) and the lagged audit fees (noted as LAF) as instrument variables for the audit fees. Following Larcker and Rusticus (2010), two-stage least square (2SLS) regressions were conducted for all four models.
30
endogenous explanatory variable, the minimum eigenvalue statistic is reflected in the first-stage F statistics (p<0.05), which is far beyond the critical value of 19.93 from Model 1 to Model 4; the relevance of instruments in all specifications are thus confirmed (Stock and Yogo, 2005; Chen, 2010). In addition, since the Sargan’s chi-square is insignificant at the conventional level (p>0.05), the exogeneity of instruments are validated 11(Chen, 2010; Hollander and Verriest, 2016). An overview of first-stage regressions is available in Table 1 in Appendix 3.
Table 5. 2SLS regressions
Model 1 Model 2 Model 3 Model 4
Instrumented variable
AF 0.472*** 0.446*** 0.041 0.138
(0.122) (0.123) (0.116) (0.128)
Weak instruments test
partial R-square 0.9605 0.9611 0.9605 0.9610
F-statistics 566.703 556.249 568.194 551.314
p<0.05 p<0.05 p<0.05 p<0.05
Instrument exogeneity test
Sargan's chi-square 0.702 1.057 3.061 2.894
p=0.4021 p=0.3040 p=0.0612 p=0.0889
Test of endogeneity
Durbin's chi-square 2.908 4.715 0.205 0.657
p=0.0882 p=0.0299 p=0.6507 p=0.4176
control variables included included included included
number of observations 186 186 182 182
R-square 0.2113 0.2210 0.0411 0.0320
* p<0.1; ** p<0.05; *** p<0.01
Note: standard errors are disclosed in parentheses in the fourth row, corrected for heteoskedasticity.
The next step compares estimators of the previous OLS regression and the 2SLS regression. Durbin’s chi-square is insignificant (p>0.05) in Model 1, Model 3 and Model 4 so the null hypothesis that the instrumented variable is exogenous cannot be rejected, which means that audit fees are exogenous in these models (Hollander and Verriest, 2016). Therefore, the instrument variable approach is not necessary for Model 1, Model 3 and Model 4; estimators of OLS are
11For Model 3 and Model 4, the p value of Sargan’s chi-square is only slightly more than 0.05 (<0.1), which suggests that the
31
preferred in the three models12 while estimators of 2SLS regression are preferred in Model 2 (Hollander and Verriest, 2016).
Table 6 reports estimators of 2SLS regressions of Model 1 and Model 2. Estimators of 2SLS regressions continue to support main findings in OLS regressions that audit fees are positively associated with the informativeness of auditor’s reports and that moderation effect of corporate boards and CEOs is insignificant. In addition, coefficients of ROA and industry remain negative (p<0.1 and p<0.05). The dummy variable yr1 is omitted since the lagged audit fee is adopted and the coefficient of yr2 remains positive (p<0.05), consistent with the result in OLS regression. Similarly, estimators of Model 1 stay in line with results of OLS regression. Therefore, even if we think audit fees are endogeneous in Model 1, main findings of OLS are validated.
Table 6. Results of first-stage in 2SLS
Model 1 Model 2
KAM Coef. Std. Err. P>z Coef. Std. Err. P>z
AF 0.472 0.122 0.000 0.446 0.123 0.000 AF*BS -0.094 0.094 0.318 BS 0.074 0.110 0.498 AF*CP 0.015 0.098 0.880 CP -0.210 0.124 0.090 TA 0.000 0.000 0.237 0.000 0.000 0.312 ROA -0.263 0.113 0.019 -0.216 0.120 0.070 Leverage 0.024 0.104 0.817 0.008 0.103 0.937 Industry -0.783 0.268 0.003 -0.827 0.253 0.001
yr1 0.000 (omitted) 0.000 (omitted)
yr2 0.444 0.212 0.036 0.472 0.210 0.024
_cons 4.311 0.178 0.000 4.300 0.178 0.000
In Model 3 and Model 4, since their Sargan’s chi-square is insignificant when the significance level is 0.05 but turns significant when the significance level is defaulted as 0.1 (0.05<p <0.1), exogeneity of instruments seems not strongly supported. To further investigate research results, two instruments are tried separately in Model 3. When NAF is used as the instrument, although the first-stage F-statistics is greater than 10, a rule of thumb to determine weak instrument (Chen,
12 The Durbin’s chi-square in Model 1 is significant when α=0.1 although it is insignificant when α=0.05. This result to some extent
32
2010), it is smaller than the critical value (14.59<16.38) (Stock and Yogo, 2005) and the partial R-square is relatively small (0.0946). However, when LAF is used as the instrument, the first-stage F-statistics is greater than the critical value (1216.05>16.38) and the partial R-square is larger (0.9598). Since we cannot test whether NAF or LAF is exogenous when only one of the them is used as instrument, LAF is a stronger instrument thus is chosen as the instrument to test the endogeneity risk in Model 3. The value of Durbin's chi square is 0.003 (p>0.05), implying that audit fees are exogenous in Model 3 and estimators of the OLS regression should be preferred. Similar procedures were taken in Model 4 and results continue to support the OLS regression (p>0.05). Table 7 summarizes results of additional analyses of Model 3 and Model 4.
Table 7. Additional tests: LAF as the only instrument
Model 3 Model 4
FOG Coef. Std. Err. P>z Coef. Std. Err. P>z
AF 0.025 0.116 0.828 0.138 0.128 0.278 AF*BS 0.108 0.090 0.231 BS 0.200 0.110 0.069 AF*CP 0.053 0.102 0.603 CP -0.180 0.112 0.107 TA 0.000 0.000 0.760 0.000 0.000 0.906 ROA -0.007 0.113 0.948 0.072 0.120 0.551 Leverage 0.075 0.104 0.468 0.045 0.107 0.674 Industry 0.179 0.292 0.539 0.291 0.301 0.333
yr1 0.000 (omitted) 0.000 (omitted)
yr2 -0.215 0.208 0.300 -0.216 0.212 0.308
_cons 20.937 0.171 0.000 20.938 0.175 0.000
number of observations 182 182
R-square 0.0383 0.0301
weak instrument test
F-statistics 1216.05 551.314
p<0.05 p<0.05
Test of endogeneity
Durbin's chi-square 0.0028 0.6572
p=0.9577 p=0.4176
