The evolution of the long-form auditor’s report: The homogenization of the Key Audit Matters and the effect of auditor change, audit quality & financial distress. Abstract:

(1)

The evolution of the long-form auditor’s report: The homogenization of the Key

Audit Matters and the effect of auditor change, audit quality & financial distress.

Author: Maarten Bolk

Student number: S3275116

Address: Postwagendrift 18

Postal code/city: 3436 AZ Nieuwegein

Phone number: 06-53955702

E-mail: maartenbolk@hotmail.com

Supervisor: prof. dr. D.A. de Waard

Second grader: dr. R.C. Trapp

Abstract:

British and Dutch regulators are requiring a long-form auditor’s report in response to the increased criticism regarding the auditor’s work. The long-form auditor’s report requires auditors to report on, amongst other aspects, Key Audit Matters (KAMs). This thesis attempts to investigate how the KAMs are evolving and how the KAMs are affected by auditor change, audit quality and the client’s risk for financial distress. The isomorphism theory is the underlying theory behind the expectations formed in the hypotheses, as a trend of homogenization is expected. Overall, the results of this thesis indicate that isomorphism occurs to some extent regarding the KAMs in the auditor’s report. Auditors describe the KAMs similarly, in a way that the number of words used for a KAM is increasingly uniform. Furthermore, it seems like an auditor change has no impact on the KAMs. Suggesting that a newly appointed auditor does not change his reporting on KAMs relative to his predecessor. For the analysis regarding audit quality and financial distress risk the main findings are (1) lower audit quality is related to a greater number of KAMs, (2) the coefficient of audit quality is decreasing over the years, indicating an isomorphic trend, (3) auditors report more KAMs with respect to audits on clients with greater risk for financial distress and (4) the coefficient of financial distress risk is increasing over the years, indicating that auditors increasingly perceive that additional disclosure about the audit regarding financially distressed clients is demanded by financial statement users. Keywords: auditor’s report; key audit matters; isomorphism; auditor change; audit quality; discretionary accruals; financial distress.

(2)

1. Introduction

Around 1900 the signature of an auditor or a stamp from the auditor was perceived sufficient by the users of the financial statements. This developed slowly into a statement from the auditor that the annual report had been audited and deemed as a truthful representation (Chandler & Edwards, 1996). However, the past financial crises triggered an increase in the criticism about the work of the auditor (Sikka, 2009). Users of the annual report, such as investors, demanded for more information about the auditor's' work and this resulted into the need for more transparency about the audit procedures and choices (IFAC, 2012; Jones, 2013; FRC, 2015). One of the responses by the Financial Reporting Council (FRC) to these concerns about the auditor’s work was the introduction of new requirements for the auditor’s report, which went in effect for periods commencing on or after October 2012. These requirements resulted into the so-called long-form auditor’s report. From this moment on auditors in the UK are required, in addition to the audit opinion (e.g. qualified/unqualified), to report on the materiality, audit scope, key audit matters (KAMs) and the audit approach, the communication with the board of directors and the selection of the auditor in the auditor’s report for audits regarding stock-listed firms (FRC, 2015). Shortly after the FRC, the Dutch Koninklijke Nederlandse Beroepsorganisatie van Accountants (NBA) followed with similar requirements (NBA, 2014). From 2014, the NBA requires Dutch auditors to report on the aforementioned aspects in a long-form auditor’s report for audits regarding Public Interest Entities (PIEs). The UK and NL are the first to adopt these requirements, therefore this thesis’ research is concentrated on British and Dutch firms. The analysis in this thesis focuses on the KAM reporting, whereas KAM disclosure is described as a disclosure by the auditor that “relates to accounts or disclosures that are material to the financial statements, and involved especially challenging, subjective, or complex auditor judgment”. (PCAOB, June 1, 2017, p. 12)

The auditor’s report has increasingly been a subject of research in the last decade. Humphrey, Loft, & Woods (2009) argued that the standard auditor’s report contains little about what work the auditor has done and his findings. Furthermore, the content of the auditor’s report is commonly not read by analysts (Coram, Mock, Turner, & Gray, 2011). Research about auditors’ reporting frequently focuses on the audit expectation gap between what the public expects from the auditor and how the auditor perceives his role (e.g. Humphrey, Moizer, & Turley, 1992; Porter, 1993; Innes, J., Brown, T., Hatherly, D., 1997; Gay, Schelluch, & Baines, 1998). More recent studies investigated whether expanding the auditor’s report results in a smaller audit expectation gap. Chong and Pflugrath (2008) found that an expanded auditor’s report has little impact on shareholders perceptions. Gold, Gronewold, & Pott (2012) found in an experiment that the longer auditor’s report (as required by ISA 700) did not result in a smaller audit expectation gap for financial statement users. Boolaky and Quick (2016) found no significant effect of a long-form auditor’s report (including materiality and KAMs) on the audit expectation gap. They also argue that “standard setters should carefully analyze the effect of additional information before making decisions on expanding the content of the audit report. Such expansions are not

(4)

necessarily perceived as useful by stakeholders.” (Boolaky & Quick, 2016, p. 1). Lennox, Schmidt and Thompson (2017) investigated the short-window and long-window market reactions to measure investors’ response to KAMs. The study shows that KAMs are not perceived as incrementally useful by investors, as investors were already informed about the audit risks (KAMs) before the auditor started disclosing them (Lennox et al., 2017).

However, Christensen et al. (2014) found in their study that nonprofessional investors do react to a KAM paragraph in an auditor’s report. A limitation in their study is that they focus on an initial reaction to a single KAM paragraph. They argue that subsequent reactions to multiple paragraphs could be different, especially when these paragraphs are standardized (Christensen et al., 2014). Additionally, Reid, Carcello, Li and Neal (2015) present results indicating that investors do react to the new long-form auditor’s report. They investigated the abnormal trading volume before and after the implementation of the mandatory long-form auditor’s report in the UK and found a significant increase of abnormal trading volume. The experimental study of Doxey (2014) also suggest that investors do value the long-form auditor’s report with its paragraph on KAMs, the results show that investors find that the extended disclosure on KAMs increases the transparency and value-relevance of the auditor’s report. In an eye-tracking study by Sirois, Bédard & Bera (2016) it is investigated how additional information in the auditor’s report affects how users perceive and navigate information in the financial statements. They find that users (graduate students) increase their attention on matters which are mentioned in the auditor’s report. Specifically, when one KAM is presented users tend to pay more attention to financial information related to this specific KAM and tend to pay less attention to other financial statement information. This suggests that the long-form auditor’s report is perceived as value-relevant by users. However, the increased attention to highlighted matters is mitigated when the auditor reports on more matters (Sirois et al., 2016).

This thesis investigates how the new long-form auditor’s reports, and more specifically the KAM disclosures, are evolving over time and which factors are determining the KAMs. A survey by the ICAEW showed that shareholders prefer a tailored auditor’s report over a standardized auditor’s report (ICAEW, 2007). The short-form audit report was described by financial statement users as a pass or fail boilerplate model. The purpose of the long-form auditor’s report is to have a higher information value for users of the financials statement by providing transparency and being company-specific, and consequently decrease the audit expectation gap between expected and actual performance of the auditor. The KAM disclosures can potentially be valuable to financial statement users. Auditors have access to private information and their disclosures can be viewed as more credible than managements’ disclosures, due to the requirement of auditor independence (Christensen, Glover and Wolfe, 2014). However, if the new auditor’s report will evolve slowly to a homogeneous version of the auditor’s report which is just longer than the short-form auditor’s report (hence it becomes again a boilerplate pass/fail model), the value adding goal of the long-form auditor’s report can be endangered. For example,

(5)

in the first year of the new auditor’s report in the UK Sarah Deans saw that several auditor’s reports highlighted the same KAMs and believed that these standard disclosures are not helpful to investors (Norris, 2014). Another example, in January 2017 Rolls Royce was fined for 671 million for bribing to secure contracts around the world (Evans, Pegg & Watt, 2017). Following the fine, the audit of Rolls Royce’s was being probed by the FRC (Tovey, 2017). However, in Rolls Royce’s auditor’s report bribery and corruption was mentioned as a KAM and how the auditor dealt with the risk. These examples give an indication that the long-form auditor’s report might just be a standardized report which is just longer than the previous short-form auditor’s report. The main research question that guides this thesis is as follows.

RQ: To what extent are the KAM disclosures in the long-form auditor’s report evolving towards a homogeneous report for listed companies in the UK and NL?

The results of prior studies are mixed. Therefore based on prior research, it can be concluded that the long-form auditor’s report does not necessarily add more information value to the auditor’s report. This thesis contributes to the existing literature by investigating how the long-form auditor’s reports in the UK and NL are evolving over time from the book year of regulatory requirement for the long-form auditor’s report (UK: 2013, NL: 2014) up to and including book year 2016. Additionally, this thesis deepens this analysis by investigating two determinants of KAMs, namely audit quality and financial distress, and how these are evolving during the research period. Where the majority of the prior literature mainly studied the auditor’s report in a qualitative way (e.g. surveys and experiments) or studied a one-year period, this thesis takes on a quantitative approach for a multi-year period. In response to prior studies, this thesis shows how the long-form auditor’s report is evolving in the UK and NL. This thesis seeks to answer whether the long-form auditor’s report, and the KAMs in particular, is evolving towards a standardized and homogeneous auditor’s report or, if it is indeed a tailored and more informative auditor’s report.

Theories that might explain how the reporting on KAMs evolves are the theory of institutional isomorphism (hereafter: isomorphism) (Dimaggio & Powell, 1983) and the legitimacy theory (Dowling & Pfeffer, 1975). The guiding research question builds upon the isomorphism theory. This thesis contributes to the isomorphism research, by investigating whether isomorphism occurs and how this impacts the long-form auditor’s report. If isomorphism occurs, the risk arises that the auditor’s report evolves into a homogeneous version. The main goal of the long-form auditor’s report, closing the audit expectation gap, is then neglected. Lastly, this thesis should be relevant for regulators, financial statement users and auditors. The UK and NL are among the first countries introducing the mandatory long-form auditor’s report, thus this thesis can provide comprehension for regulators in other countries which want to adopt the long-form auditor’s report in the future. Furthermore, an understanding of how the long-form auditor’s reports evolve can give insight into whether the long-form auditor’s report is valuable for financial statement users or not. This thesis also provides comprehension to financial statement

(6)

users about how the KAMs are affected by the audit quality and a client’s financial distress risk. Additionally, this thesis should make auditors aware of the trends regarding the KAMs. Auditors should be challenged to determine and describe KAMs in client-specific and transparent manner, as existing literature provided that a standardized version of the auditor’s report is not what the users desire (e.g. ICAEW, 2007; Church, Davis & McCracken, 2008; Humphrey et al., 2009; Coram et al., 2011; Gray, Turner, Coram & Mock, 2011; Reid et al., 2015).

This thesis proceeds as follows, the next section will elaborate the prior literature and the contribution of this thesis to it, additionally building upon the theoretical background the hypotheses will be formulated. The research methods will be elaborated in section three. The results and conclusion will be presented in section four and five respectively.

(7)

2. Theoretical background

In this section the theoretical background will be provided. Building upon this theoretical background, four hypotheses will be developed. The legitimacy theory and isomorphism theory are theories that are used for the hypotheses. The first three hypotheses (H1a up to and including H3d) are of a more explorative nature and investigate how the KAM disclosures are evolving over the four-year period. The fourth hypotheses (H4a up to and including H4f) deepens the analysis of the KAM disclosures and investigates if the KAMs are tailored towards audit quality and financial distress risk, along with an analysis of how the explanatory power of these two independent variables is evolving over time.

2.1 Legitimacy theory

A theory that might explain how the reporting on KAMs will evolve is the legitimacy theory. Legitimacy theory is a theory widely used in the area of voluntarily reporting (e.g. Deegan, 2002; O’Donovan, 2002; Hedberg & Von Malmborg, 2003; Bebbington, Larrinaga & Moneva, 2008; An, Davey & Eggleton, 2011; Van Bommel, 2014). Dowling and Pfeffer (1975) describe legitimacy as the congruence between the social values of the organization’s activities and the norm of acceptable behavior implied by its external environment. Suchman (1995) gives the following definition of legitimacy: “Legitimacy is a generalized perception or assumption that the actions of an entity are desirable, proper, or appropriate within some socially constructed system of norms, values, beliefs, and definitions.”. Concluding from the two definitions, legitimacy can be described as the acceptance of an actor (e.g. firm or auditor) by the general public. Firstly, following the legitimacy theory one would expect that a mandatory long-form auditor’s report would not be required. That is, if society calls for more transparency in the auditor’s work, it is expected that the auditor would enhance his reporting in order to increase and retain his legitimacy. You would expect this especially during the last decade, as the legitimacy of the auditor’s profession is increasingly being criticized and under pressure (as mentioned before in section one). In section one it was also mentioned that a tailored auditor’s report would be of more value to investors than a standardized auditor’s report, therefore to enhance his legitimacy the auditor would be inclined to report valuable and company-specific information in his report.

2.2 Institutional Isomorphism

The evolution of the KAM report might also be explained by the theory of isomorphism, which was developed in the organizational behavior literature. Isomorphism seeks to explain why organizations are becoming more homogeneous relative to comparable organizations in its environment. Dimaggio and Powell (1983) argue that change seems less driven by competition, instead organizations (in attempt to change) are becoming more similar without necessarily becoming more efficient. Following the theory of isomorphism, three forms of isomorphism can be identified. Firstly, coercive isomorphism can occur forcefully by regulation. Next, mimetic

(8)

isomorphism occurs when one organization perceives another more successful. Lastly, normative isomorphism occurs when there is a norm created within a profession. Empirically the three forms can blend, however the three forms derive from different conditions (Dimaggio & Powell, 1983). The British and Dutch legislation requires auditors of PIEs to report on additional content, such as KAMs and the materiality. It can be concluded that coercive isomorphism occurs with respect to the long-form auditor’s report, since auditors are now adopting the new requirements for their audits of PIEs. What is of interest in this thesis is how the long-form auditor’s report is evolving from the moment of legislation. In this thesis, the expectation is that the long-form auditor’s reports are growing to be similar, in a way that auditors are reporting on KAMs similarly. This might be due to that one audit firms perceives another as more successful, therefore imitating this audit firm as a response to uncertainty (mimetic isomorphism). Furthermore, the audit profession is deeply professionalized, therefore the homogenization might also be embedded within the profession (normative isomorphism). Normative isomorphism occurs primarily due to formal education and the elaboration of professional networks (Dimaggio & Powell, 1983). The expectation in this thesis is that the long-form auditor’s report is evolving into a standardized and homogeneous report, due to the three mechanisms of isomorphism. This contradicts the expectation, which one could formulate upon the legitimacy theory, that the auditor would be inclined to tailor his KAM disclosures towards the client to enhance his legitimacy. To strengthen the expectation that homogenization due to isomorphic forces occurs (instead of tailored reports to enhance legitimacy), some studies argue that isomorphic behavior legitimizes (e.g. Meyer & Rowan, 1977; Galaskiewicz & Wasserman, 1989; Deephouse, 1996). That is, organizations that are similar to other organizations are perceived more legitimate by regulators and the general public than organizations which deviate from normal behavior (Deephouse, 1996). Organizations deviating from normal behavior violate societal or legal expectations, therefore they are prone to challenges of legitimacy and may be regarded as unacceptable (Stjernberg & Philips, 1993; Westphal, Gulati & Shortell, 1997; Deephouse & Carter, 2005). Based on this, it is expected that the long-form auditor’s reports are growing to be similar, due to isomorphic forces. The aforementioned isomorphism theory results in the expectation that the reporting on KAMs is evolving to be similar and standard. The focus of this thesis will be on the number of KAMs and the average number of words per KAM. The first two hypotheses expect that auditors will be reporting increasingly on the same amount of KAMs and will use increasingly the same amount of words to describe a KAM.

H1a: The number of KAMs are growing towards each other, resulting in a declining deviation from the mean.

H1b: The average number of words per KAM are growing towards each other, resulting in a declining deviation from the mean.

(9)

2.3 Homogenization within the audit firm

In this thesis, the homogenization of the auditor’s report is expected to be more present for clients with the same audit firm. The expectation is that internal education, the rules and procedures within a firm and the influence of the professional practice office intensifies the homogeneous process for the KAM reporting within the audit firm. Consequently the hypothesis is that the reporting on KAMs for clients with the same audit firm are growing to be more homogeneous than clients with different audit firms.

H2a: The number of KAMs for clients with the same audit firm are growing towards each other to a greater extent, resulting in a greater declining deviation from the mean.

H2b: The average number of words per KAM for clients with the same audit firm are growing towards each other to a greater extent, resulting in a greater declining deviation from the mean.

2.4 Auditor change

Due to new legislation which requires mandatory audit firm rotation, a great deal of firms changed auditors during the four-year period of this thesis and this creates opportunity for this thesis to analyze the effect of the auditor change on the KAMs. If H1a and H1b are true and the long-form auditor’s report does evolve into a standardized version, an auditor change should have no impact on the KAM disclosures. Since the reporting on KAM is expected to be homogeneous. However, following H2a and H2b, one would expect that a change would impact the reporting on KAMs. If the homogenization is stronger within the audit firms than the homogenization within the audit industry overall, then it is expected that the KAMs will be different in the first year of the auditor change. After the auditor change the KAMs will homogenize again towards the new audit firm’s mean. It is expected that this impact of the auditor change on the KAMs will be significant in the early years (2013-2014) and will lose its significance when the homogenization within the audit industry will increase after the initial years, due to isomorphic forces. Another explanation for the first year difference can be the lack of knowledge of the incoming auditor for the specific client. Results of prior studies show that an auditor change (mandatory or voluntarily) results in lower audit quality due to lack of new-client-specific knowledge (e.g. Johnson, Khurana & Reynolds, 2002; Cameran, Francis, Marra & Pettinicchio, 2013; Francis, Hunter, Robinson, Robinson & Yuan 2017). The expectation is that the auditor change results in a decrease of the KAM disclosure size, as a result of the lack of client-specific knowledge. The third hypothesis is that an auditor change has a significant negative effect on the number of KAMs and the average number of words per KAM in the first year of auditor change and the significance is expected to decrease as the homogenization intensifies, due to increasing isomorphic forces.

H3a: An auditor change negatively impacts the number of KAMs in the first year after

the change in the initial years of the long-form auditor’s report.

H3b: An auditor change negatively impacts the average number of words per KAM in the first year after the change in the initial years of the long-form auditor’s report.

(10)

H3c: An auditor change negatively impacts the number of KAMs in the first year after the change to a lesser extent in later years.

H3d: An auditor change negatively impacts the average number of words per KAM in the first year after the change to a lesser extent in later years.

2.5 Explanatory elements related to KAMs

The first hypotheses are more of an explorative nature and give an indication of how the KAM disclosures evolve during the four-year period (2013-2016). To deepen the analysis of the long-form auditor’s report, and more specifically the KAMs, this thesis makes an attempt to investigate whether the KAMs are company-specific or more standardized. Prior research indicates that users of financial statements prefer a tailored auditor’s report over a standardized version (e.g. ICAEW, 2007; Church, Davis & McCracken, 2008; Humphrey et al., 2009; Coram et al., 2011; Gray, Turner, Coram & Mock, 2011; Reid et al., 2015). Therefore, if it turns out that the KAMs are not tailored to the client, the added value of the KAMs can be questioned. To further investigate whether homogenization occurs or not, this thesis investigates how audit quality affects the KAM disclosures and whether audits of more risky clients (defined by the proxy financial distress) also have more KAMs in the auditor’s report. If auditors indeed tailor the KAM disclosures towards the audit quality and disclose more in the auditor’s report in response to the higher risk associated with the audit for the client, then this gives an evident indication that the auditor is tailoring his disclosure in the auditor’s report. In accordance with the first three hypotheses, it is expected that the explanatory power of the audit quality and financial distress risk loses its effect as the isomorphic forces increases.

2.5.1 Audit quality

Following prior literature, accounting accrual measures are used as proxy for audit quality. This is a proxy that is numerously used in studies on earnings quality (Jones, 1991; Dechow & Dichev, 2002; Francis, LaFond, Olsson & Schipper, 2005; Dechow & Schrand 2010) and various studies also use earnings quality as proxy for audit quality (Johnson et al., 2002; Frankel, Johnson & Nelson, 2002; Myers, Myers & Omer, 2003; Kwon, Lim & Simnett, 2010; Francis, 2011). To further support the argument that accounting accrual measures are a good proxy for audit quality, previous studies show that lower accounting accrual levels are associated with

higher auditor conservatism, and auditor conservatism is seen as higher audit quality(Becker et

al. 1998; Francis et al. 1999; Francis and Krishnan 1999). On top of that, previous studies show that higher accounting accrual levels are associated with proxies for failing audit quality, such as audit reporting failures and the probability of auditor litigation (Heninger, 2001; Geiger & Raghunandan, 2002; Myers et al., 2003). It is expected that higher audit quality could result in the detection of more audit risks and thus the auditor is capable to report more on KAMs. Additionally, an auditor with high audit quality might be more willing to report more KAMs, because he can show that he did a good job. However, auditors could also report more on KAMs to signal higher audit quality, to camouflage their lower audit quality to prevent litigation costs

(11)

(Kachelmeier, Schmidt & Valentine, 2017). The expected relationship between audit quality and the KAM disclosures is unclear, therefore the hypothesis is neutrally formulated. Additionally, it is hypothesized that the explanatory power of the audit quality for the KAM disclosure decreases over the years as the isomorphic forces increase..

H4a: A higher or lower number of KAMs are disclosed for audits with higher or lower audit quality.

H4b: A higher or lower average number of words per KAM are disclosed for audits with higher or lower audit quality.

H4c: The explanatory power of audit quality for the KAMs decreases over time. 2.5.2 Financial distress

Client’s business risk is defined as “the risk that the client’s economic condition will deteriorate in either short-term or long-term” (Johnstone, 2000, p.3), in this thesis financial distress proxies for client’s business risk. Audit risk is defined as the risk that the auditor may unknowingly fail to appropriately modify the opinion on financial statements that are materially misstated (Colbert, Luehlfing & Alderman, 1996; AICPA, 2006). Financial distress initially only affects the client’s business risk, thus it is not an audit risk. However, audit risk is affected by the client’s business risk, as the auditor’s evaluates client’s business risks to assess the audit risk (Jones & Raghunandan, 1998; Johnstone, 2000). Concluding, audits for clients which are financially distressed bare a higher audit risk. One would expect more KAMs and a more detailed description of the KAMs (thus more words) for riskier clients, if the KAM disclosures are indeed tailored, as more risk results in more risk disclosures (KAMs). Additionally, auditors may disclose more than justified for clients with a higher level of risk to signal a higher audit quality to prevent litigation costs (Kachelmeier et al., 2017). In this thesis it is hypothesized that the auditor does tailor the KAMs in his report towards the risk associated with the audit of his client. However, building upon the isomorphism theory it is expected that homogenization in respect to the KAMs occurs due to isomorphic forces. It is hypothesized that the explanatory power of the financial distress risk for the KAM disclosure size decreases over the years as the isomorphic forces increase.

H4d: A higher number of KAMs are disclosed for audits on clients with a greater financial distress risk.

H4e: A higher average number of words per KAM are disclosed for audits on clients with a greater financial distress risk.

(12)

3. Research methods & Descriptive statistics

In this section the research methods will be elaborated. Firstly, the sample and data collection will be explained in section 3.1. The descriptive statistics are shortly elaborated in section 3.2. Control variables are given in section 3.3. Section 3.4 elaborates the hypothesis testing of H1a, H1b, H2a and H2b, namely the homogenization of the KAM disclosures. Section 3.5 focuses on the hypothesis testing for H3a, H3b, H3c and H3d (auditor change). The methods regarding the hypothesis testing for the two determinants, audit quality and financial distress risk (H4a until H4f), are explained in section 3.6. Lastly, section 3.7 will elaborate the interviews which are held to validate the results of this thesis.

3.1. Sample

This thesis uses data of auditor’s reports for British stock-listed firms and Dutch stock-listed firms. The selected firms represent a substantial portion of the British and Dutch firms that require the long-form auditor’s report. For the UK there are roughly 2.200 stock-listed firms, however only 250 British firms are selected in thesis, as these firms represent the most leading group of firms that require the mandatory long-form auditor’s report in the UK. For the Dutch firms which require the mandatory long-form auditor’s report, only some financial firms are excluded which are PIEs without being stock-listed. Therefore, the sample covers the leading group of firms that require the new long-form auditor’s report for the UK and NL. Data for British stock-listed firms is collected for financial years 2013, 2014, 2015 and 2016. For Dutch stock-listed firms the data spans over 2014, 2015 and 2016, because the long-form auditor’s report was mandatory one year later than in the UK. However, 23 Dutch firms that voluntarily adopted the long-form auditor’s report in 2013 are included in the sample. The data collection for the KAMs is part of a larger research and has been hand collected by multiple students. The database consists of the FTSE100 and the first 150 of the FTSE250 firms for the UK and the Dutch listed companies. This results in a sample of 325 firms, 273 over four years (2013-2016) and 52 over three years (2014-2016). After excluding missing data for the KAMs, merged firms and bankrupt firms, the entire sample consists of 1.075 firm-year observations. The number of KAMs and number of words in the KAMs are determined by hand collecting it for every auditor’s report and storing it in the database. The average number of words per KAM for firm j is calculated by dividing the total number of words in the KAM disclosure by the number of KAMs for firm j.

(13)

Table 1

Descriptive statistics variables included in models Panel A: Descriptive statistics variables included in models

Variables N Mean St. Dev. Min Max Median

NoKAMs 1.075 3.86 1.53 1 9 4 AvgWords 1.075 299 142 14 1.139 280 AUDITQ 729 0.035 0.029 0ª 0.093 0.026 DISTRESS 674 0.572 0.254 0.001 0.898 0.613 SIZE 731 8.03 1.59 -0.477 12.93 7.94 CMPLX 730 4.45 1.26 0 7.89 4.51 ROA 731 0.588 0.133 -1.93 2.14 0.055

ª The minimal value is very close to 0.

Panel B: Correlation table variables included in models

N oK A M s A vg W or ds A U D IT Q D IS T R E SS SI Z E C M PL X R O A NoKAMs 1.00 AvgWords -0.161* 1.00 AUDITQ -0.029 -0.023 1.00 DISTRESS 0.286* 0.005 -0.048 1.00 SIZE 0.444* 0.011 -0.165* 0.182* 1.00 CMPLX 0.326* 0.004 -0.175* 0.158* 0.594* 1.00 ROA -0.177* 0.044 0.098* -0.189* -0.105* -0.116* 1.00 *p < 0.1 3.2 Descriptive statistics

Table 1 panel A documents the descriptive statistics for the variables included in the models of this thesis. It is important to keep in mind the possibility of multicollinearity. Panel B documents the correlation matrix. No real problems regarding multicollinearity are identified in the correlation matrix, which is supported by a non-reported VIF test. Variables are winsorized where necessary, to limit the effect of spurious outliers. An overview of all the dependent and independent variables used in this thesis is provided in appendix A. Table 2 documents the descriptive statistics breakdown for the number of KAMs and the average number of words per KAM. Panel D provides a graph showing the yearly means of the number of KAMs (NoKAMs) and the average number of words per KAM (AvgWords) for each audit firm separately. The graph of NoKAMs shows that the means of the audit firms are not growing to the overall mean. However, the graph of AvgWords shows that the means of the audit firms are growing towards the overall mean. This suggests that a trend of homogenization is visible for the AvgWords.

(14)

Table 2

Descriptive statistics breakdown NoKAMs & AvgWords

Panel A: Descriptive statistics per SIC1

SIC Observations % Mean NoKAMs* Mean AvgWords**

1 105 10.5% 3.83 320 2 146 14.6% 4.10 268 3 126 12.6% 4.02 287 4 90 9.0% 4.97 249 5 114 11.4% 3.87 289 6 264 26.3% 3.42 333 7 107 10.7% 3.78 295 8 44 4.4% 3.91 310 9 6 0.6% 4.33 299 Total 1.002 100%

Panel B: Descriptive statistics per audit firm2

Audit Firm Observations % Mean NoKAMs* Mean AvgWords**

Deloitte 247 23.4% 3.91 244

EY 177 16.8% 4.00 269

KPMG 284 26.9% 3.19 331

PwC 348 33.0% 4.30 330

Total 1.056 100%

Panel C: Descriptive statistics per year

NoKAMs

Year Observations Mean* Standard deviation*

2013 168 3.99 1.46 2014 296 4,00 1,51 2015 306 3,80 1,54 2016 305 3,71 1,56 Total 1.075 AvgWords

Year Observations Mean** Standard deviation**

2013 168 175 103

2014 296 258 119

2015 306 334 122

2016 305 370 141

Total 1.075

1 _{73 firm-year observations with missing SIC data} 2_{19 firm-year observations with non-Big Four firm}

*Rounded off at two decimals **Rounded off at rounded numbers

(15)

3.3 Control variables

Relating to the statistical tests in section 3.4 and 3.5 a number of control variables are utilized for the regression models. Following prior literature (Hay, Knechel & Wong, 2006; Cameran et al., 2013; Bédard, Gonthier-Besacier & Schatt, 2014; Lennox et al., 2017; Morais and Pinto 2017)), the control variables are firm size (SIZE), firm complexity (CMPLX), profitability (ROA), industry (INDS_FE), year (YEAR_FE) and country (COUNTRY_FE). SIZE is measured as the natural logarithm of total assets to avoid problems of scale, Morais and Pinto (2017) report a significant relation between the number of KAMs and firm size. CMPLX is a control variable, because the KAM disclosures could be affected by the firm complexity. Results of the study by Lennox et al. (2017) support this argument, as the results show that a greater firm complexity relates to a higher number of KAMs. Following Lennox et al. (2017), the natural logarithm of the number of subsidiaries will proxy for firm complexity. Profitability (as measured by the return on assets) is a control variable, because previous literature provide evidence that profitability is associated with the qualification of an audit opinion (e.g. Laitinen & Laitinen, 1998; Johl, Jubb & Houghton, 2007). To take into account the expected differences across industries and years, industry fixed effects (INDS_FE) and year fixed effects (YEAR_FE) are employed. Industries are classified on a one-digit SIC code. The use of one digit SIC code is based on the consideration of defining industry groups narrowly enough to capture the industry effect, while having enough firms in each industry. Country fixed effects (COUNTRY_FE) are employed to control for country differences between the NL and UK.

Table 2 (continued)

Panel D: Means for NoKAMs and AvgWords per audit firm line charts.

(16)

3.4 Homogenization of the KAM disclosures

To test for homogenization the KAMs will be analyzed over the years. A significant decrease in the standard deviation of the number of KAMs (SDNoKAMs) and of the average number of words per KAM (SDAvgWords) provides an indication that homogenization occurs. For each client industry separately (to control for client industry differences) it will be analyzed whether the standard deviation is decreasing from 2013 till 2016. A decrease of standard deviation implies that the number of KAMs and the average number of words per KAM are growing towards the mean of the client industry, thus the size of the auditor’s reports are growing to be similar. The yearly standard deviations are scaled to the yearly means, to control for the mean increases during the years. As documented in the descriptive statistics in table 2 panel C, the mean of the NoKAMs slightly decreased over the years and the AvgWords greatly increased. To see if the homogenization is increasing on a significant level, the yearly scaled SDNoKAMs and the yearly scaled SDAvgWords will be regressed on Time. Time regression has been used similarly to investigate whether a decrease/increase is significant or not (e.g. Francis & Schipper, 1999; Lev & Zarowin, 1999). For the time regression per industry the sample size is 1.002 firm-year observations and for time regression per audit firm the sample size is 1.056 firm-firm-year observations, due to missing values and non-Big Four audit firms (see table 2). The formulations

for the time regressions are as follows1:

𝑆𝐷𝑁𝑜𝐾𝐴𝑀𝑠 = 𝛼 + 𝛽 𝑇𝑖𝑚𝑒 + ɛ (1)

𝑆𝐷𝐴𝑣𝑔𝑊𝑜𝑟𝑑𝑠 = 𝛼 + 𝛽 𝑇𝑖𝑚𝑒 + ɛ (2)

Where Time refers to the sample year, 1 till 4 corresponds to the years 2013 till 2016.

To extend the analysis of the KAMs, it will be tested whether the mean of the number of KAMs and the mean of the average number of words per KAM differs significantly across audit firms and industries a two-way ANOVA will be performed. The ANOVA will be performed yearly and two times, once with the number of KAMs as dependent variable and once with the average number of words per KAM as dependent variable. The audit firm and the industry are the independent variables for the ANOVA. ANOVA has been used similarly to measure uniformity between audit firms in issuing qualified opinions (Warren, 1975, 1980). Warren (1975) transforms his dependent variable to an arcsine transformation to better fit the assumptions of the ANOVA, because his dependent variable consists of proportions. In this thesis the NoKAMs and AvgWords are log-transformed in the ANOVA to better fit the assumptions, as in this thesis the dependent variables consist of count data (Bartlett, 1947; McDonald, 2009).

(17)

3.5 Auditor change

During the four financial years which are part of this research 75 firms changed of audit firm, these changes were mainly due to legislation that requires mandatory audit firm rotation. The impact of the audit firm change on the KAMs can be measured by analyzing whether after the year of auditor change the size of the KAM disclosure (year t) are significantly different compared to year t-1. KAMs were not disclosed in the year t-1 for the financial year 2013, and for some firms 2014. These years for the particular firms are excluded for the hypothesis testing on auditor changes, as the impact of the auditor change on the KAMs cannot be measured. This results in an analysis over three periods, namely an auditor change between one of the following periods: 2013-2014, 2014-2015 and 2015-2016. Where the auditor change in the 2013-2014 period will consists out of solely UK firms and the periods 2014-2015 & 2015-2016 will consist out of both UK and NL firms.

3.5.1 Difference-in-difference model

To test whether the auditor change significantly impacts the auditor’s report, a difference-in-difference (DID) research design is conducted. The auditor change group (treatment group) will be compared to the no change group (control group). For each of three periods the effect of the auditor change will be estimated separately. It is expected that during the initial years the auditor change will significantly impact the KAMs (H3a & H3b). Following upon that, it will be analyzed whether the homogenization of the KAM disclosures intensifies over the years, consequently the significance of the effect of the auditor change decreases over the years (H3c & H3d). The sample size for the models for the effect of an auditor change is 997, due to missing values regarding the control variables. Following partially the DID design of Francis et al. (2017) the model is formulated as follows:

𝑁𝑜𝐾𝐴𝑀𝑠, = 𝛼 + 𝛽 𝐶𝐻𝐴𝑁𝐺𝐸 + 𝛽 𝑃𝑂𝑆𝑇 + 𝛽 (𝐶𝐻𝐴𝑁𝐺𝐸 ∗ 𝑃𝑂𝑆𝑇) , + 𝛽 𝑆𝐼𝑍𝐸 + 𝛽 𝐶𝑀𝑃𝐿𝑋 + 𝛽 𝑅𝑂𝐴 + 𝐼𝑁𝐷𝑆_𝐹𝐸 + 𝐶𝑂𝑈𝑁𝑇𝑅𝑌_𝐹𝐸 + ɛ (3) 𝐴𝑣𝑔𝑊𝑜𝑟𝑑𝑠 , = 𝛼 + 𝛽 𝐶𝐻𝐴𝑁𝐺𝐸 + 𝛽 𝑃𝑂𝑆𝑇 + 𝛽 (𝐶𝐻𝐴𝑁𝐺𝐸 ∗ 𝑃𝑂𝑆𝑇) , + 𝛽 𝑆𝐼𝑍𝐸 + 𝛽 𝐶𝑀𝑃𝐿𝑋 + 𝛽 𝑅𝑂𝐴 + 𝐼𝑁𝐷𝑆_𝐹𝐸 + 𝐶𝑂𝑈𝑁𝑇𝑅𝑌_𝐹𝐸 + ɛ (4)

Where CHANGEj is valued at 1 if firm j changed auditors in year t and 0 if not. POSTt is valued

at 1 for year t (after change) and 0 for year t-1 (before change). The variable of primary interest

is (CHANGE * POST)j,t, which captures the net impact of the auditor change on NoKAMsj,t and

AvgWordsj,t in the post-change period for auditor change firms compared to their matched

(18)

3.5.2 Robustness

It is recognized that the two dependent variables consist of count data. For count data a Poisson regression model would be more appropriate. With Ordinary Lease Square (OLS) regression the expected value can take any value, positive or negative and integer or fractional, but counts cannot be negative or fractional and count data cannot take any value (e.g. the number of KAMs are not likely to exceed 10 and the average number of words are not likely to exceed 1.000). Morais and Pinto (2017) also estimated the number of KAMs using both OLS and Poisson regression, and find no differences in results. However, Poisson regression assumes (amongst other assumptions) that the mean equals the variance. This assumption is not met, as the dependent variables NoKAMs and AvgWords do not have a mean equal to the variance (as displayed in the descriptive statistics in table 1). Therefore, a negative binomial regression, which also assumes non-negative and non-infinite data, would be more appropriate than the Poisson regression. To make sure the results are robust equations 3 and 4 are re-estimated twice. First, using an OLS regression with log-transformed dependent variables, and a second time using a negative binomial regression.

3.6 Determinants of the KAM disclosures

To deepen the analysis of the KAM disclosures an OLS regression is performed. In this thesis the focus will be on the variables audit quality (AUDITQ) and financial distress risk (DISTRESS) as independent variables. The dependent variables are the number of KAMs (NoKAMs) and the average words per KAM (AvgWords). Financial data related to the sample is required for the hypothesis testing for H4a up to and including H4f. The financial data is retrieved from the database COMPUSTAT. For the statistical tests in this section, firms operating in the financial industry (SIC code 6) are excluded. This results in a decrease of sample size of 264 firm-year observations. This is common in existing literature regarding audit quality and financial distress, as it could adversely affect the results. On top of that, the required data is less available for financial firms. Additional to the exclusion of financial firms, some observations are excluded due to missing values concerning data which was required for the calculation of audit quality and financial distress risk. The remaining sample sizes are presented in the table that presents the results (table 5) in the next chapter. The regression models are formulated as follows:

𝑁𝑜𝐾𝐴𝑀𝑠 = 𝛼 + 𝛽 𝐴𝑈𝐷𝐼𝑇𝑄 + 𝛽 𝑆𝐼𝑍𝐸 + 𝛽 𝐶𝑀𝑃𝐿𝑋 + 𝛽 𝑅𝑂𝐴 + 𝐼𝑁𝐷𝑆_𝐹𝐸 + 𝑌𝐸𝐴𝑅_𝐹𝐸 + 𝐶𝑂𝑈𝑁𝑇𝑅𝑌_𝐹𝐸 + ɛ (5) 𝐴𝑣𝑔𝑊𝑜𝑟𝑑𝑠 = 𝛼 + 𝛽 𝐴𝑈𝐷𝐼𝑇𝑄 + 𝛽 𝑆𝐼𝑍𝐸 + 𝛽 𝐶𝑀𝑃𝐿𝑋 + 𝛽 𝑅𝑂𝐴 + 𝐼𝑁𝐷𝑆_𝐹𝐸 + 𝑌𝐸𝐴𝑅_𝐹𝐸 + 𝐶𝑂𝑈𝑁𝑇𝑅𝑌_𝐹𝐸 + ɛ (6) 𝑁𝑜𝐾𝐴𝑀𝑠 = 𝛼 + 𝛽 𝐷𝐼𝑆𝑇𝑅𝐸𝑆𝑆 + 𝛽 𝑆𝐼𝑍𝐸 + 𝛽 𝐶𝑀𝑃𝐿𝑋 + 𝛽 𝑅𝑂𝐴 + 𝐼𝑁𝐷𝑆_𝐹𝐸 + 𝑌𝐸𝐴𝑅_𝐹𝐸 + 𝐶𝑂𝑈𝑁𝑇𝑅𝑌_𝐹𝐸 + ɛ (7)

(19)

𝐴𝑣𝑔𝑊𝑜𝑟𝑑𝑠 = 𝛼 + 𝛽 𝐷𝐼𝑆𝑇𝑅𝐸𝑆𝑆 + 𝛽 𝑆𝐼𝑍𝐸 + 𝛽 𝐶𝑀𝑃𝐿𝑋 + 𝛽 𝑅𝑂𝐴 + 𝐼𝑁𝐷𝑆_𝐹𝐸 + 𝑌𝐸𝐴𝑅_𝐹𝐸 + 𝐶𝑂𝑈𝑁𝑇𝑅𝑌_𝐹𝐸 + ɛ

(8)

Additionally, both audit quality and financial distress risk are combined in the following models: 𝑁𝑜𝐾𝐴𝑀𝑠 = 𝛼 + 𝛽 𝐴𝑈𝐷𝐼𝑇𝑄 + 𝛽 𝐷𝐼𝑆𝑇𝑅𝐸𝑆𝑆 + 𝛽 𝑆𝐼𝑍𝐸 + 𝛽 𝐶𝑀𝑃𝐿𝑋 + 𝛽 𝑅𝑂𝐴 + 𝐼𝑁𝐷𝑆_𝐹𝐸 + 𝑌𝐸𝐴𝑅_𝐹𝐸 + 𝐶𝑂𝑈𝑁𝑇𝑅𝑌_𝐹𝐸 + ɛ (9) 𝐴𝑣𝑔𝑊𝑜𝑟𝑑𝑠 = 𝛼 + 𝛽 𝐴𝑈𝐷𝐼𝑇𝑄 + 𝛽 𝐷𝐼𝑆𝑇𝑅𝐸𝑆𝑆 + 𝛽 𝑆𝐼𝑍𝐸 + 𝛽 𝐶𝑀𝑃𝐿𝑋 + 𝛽 𝑅𝑂𝐴 + 𝐼𝑁𝐷𝑆_𝐹𝐸 + 𝑌𝐸𝐴𝑅_𝐹𝐸 + 𝐶𝑂𝑈𝑁𝑇𝑅𝑌_𝐹𝐸 + ɛ (10)

The regressions will also be performed for each year separately without the year fixed-effects, to analyze whether the explanatory power of the two independent variables decreases over time. The yearly slope coefficients of models 9 and 10 are regressed on time to test for significance. The models for the time regression are formulated as follows:

𝛽_𝐴𝑈𝐷𝐼𝑇𝑄 = 𝛼 + 𝛽1𝑇𝑖𝑚𝑒 + ɛ (11)

𝛽_𝐷𝐼𝑆𝑇𝑅𝐸𝑆𝑆 = 𝛼 + 𝛽1𝑇𝑖𝑚𝑒 + ɛ (12)

Where Time refers to the sample year, 1 till 4 corresponds to the years 2013 till 2016. 3.6.1 Audit quality

As acknowledged in section 2.5, earnings quality is used as a proxy for audit quality. Prior literature provides several models for calculating earnings quality. Two notable models are extensively used in the existing literature. These are the abnormal accruals model by Jones (1991) and the accrual estimation errors model by Dechow and Dichev (2002). The model by Dechow and Dichev requires lead and lag cash flows, thus in order to calculate the audit quality for 2016 financial data for 2017 is required. At the moment of research this data is not yet available. Therefore, choosing for the Dechow and Dichev model would result in a decrease in sample size and the trend analysis for the KAMs would be shortened by one year. The model by Jones does not require financial data for year t+1. To have a stronger trend analysis (2013-2016 instead of 2013-2015), the model by Jones is chosen for the calculation for the audit quality proxy. The Jones model has been modified by Dechow, Sloan & Sweeney (1995) by adjusting the change in revenues for the change in receivables. This modified Jones model has been widely used in existing literature and it is believed that this modified Jones model better captures the discretionary accruals than the original Jones model. In this thesis audit quality is measured by the absolute value of the discretionary accruals, which measures the combined effect income-decreasing and income-increasing discretionary accruals (Frankel, Johnson & Nelson, 2002). In the modified Jones model the discretionary accruals are estimated as follows:

(20)

Where TACCt are total accruals in year t and NDACCt are the nondiscretionary accruals in year t. The nondiscretionary accruals are estimated with the following model:

𝑁𝐷𝐴𝐶𝐶 = 𝛼 1 𝑇𝐴 + 𝛼 (∆𝑅𝐸𝑉 − ∆𝑅𝐸𝐶 ) 𝑇𝐴 + 𝛼 𝑃𝑃𝐸 𝑇𝐴 (14)

where TAt-1 are the total assets in year t-1, ∆REVt are the revenues in year t minus the revenues in

year t-1, ∆RECt are the receivables in year t minus the receivables in year t-1 and PPE is the

gross property plant and equipment in year t.

The firm’s specific coefficients (α1, α2 & α3) in equation 14 are generated by using the following

model: 𝑇𝐴𝐶𝐶 = 𝛾 1 𝑇𝐴 + 𝛾 (∆𝑅𝐸𝑉 − ∆𝑅𝐸𝐶 ) 𝑇𝐴 + 𝛾 𝑃𝑃𝐸 𝑇𝐴 + 𝜀 (15)

Where γ1, γ2 and γ3 denote the OLS estimates of α1, α2 and α3. The total accruals are estimated with the following model:

𝑇𝐴𝐶𝐶 =∆𝐶𝐴 − ∆𝐶𝐿 − ∆𝐶𝑎𝑠ℎ + ∆𝑆𝑇𝐷 − 𝐷𝐸𝑃

𝑇𝐴

(16)

Where ΔCAt is the change in current assets between year t-1 and year t, ΔCLt is the change in

current liabilities between year t-1 and year t, ΔCasht is the change in cash between year t-1 and

year t, ΔSTDEBTt is the change in debt included in current liabilities between year t-1 and year t,

DEPt is the depreciation and amortization expense in year t.

3.6.2 Financial distress

Ohlson’s O-score (Ohlson, 1980) is used to proxy for financial distress risk. O-score is a widely used and accepted measure of financial distress by the existing literature (e.g. Dichev, 1998; Griffin & Lemmon, 2002). Ohlson (1980) reports a predictive accuracy of 96,3%, which is higher than the concurrent Altman Z-score (Altman, 1968). The model for calculating financial distress risk based on Ohlson’s O-score is as follows:

DISTRESS = –1,32 – 0,407log(TAj,t) + 6,03(TLj,t/TAj,t) – 1,43(WCj,t/TAj,t)

+ 0,0757(CLj,t/CAj,t) – 1,72Xj,t – 2,37 (NIj,t/TAj,t) – 1,83(FFOj,t/TLj,t)

+ 0,285Yj,t – 0,521((NIj,t - NIj,t-1)/(|NIj,t|+|NIj,t-1|))

(17)

Where DISTRESS is the probability for financial distress (O-score), TAj,t is firm j’s total assets in

year t, TLj,t is firm j’s total liabilities in year t, WCj,t is firm j’s working capital for year t, CLj,t is

(21)

TLj,t > TAj,t for firm j in year t, 0 otherwise, NIj,t is firm j’s net income in year t, FFOj,t is firm j’s

funds from operations in year t, Yj,t is 1 if a net loss for the last two years, 0 otherwise for firm j

in year t.

3.6.3 Robustness

To enhance the robustness of the results, a number of additional tests are performed. For the results concerning the financial distress risk a replacement proxy will be used to enhance robustness. Besides the O-score, the Altman Z-score (Altman, 1968) is a widely used proxy for financial distress. This thesis will utilize the revised score (Altman, 2000), as this revised Z”-score is better suited for non-manufacturing firms and the sample in this thesis exists for the bigger part of non-manufacturing firms. Equation 7, 8, 9 and 10 are performed another time with the Z”-score as replacement for the O-score, which enhances the results of this thesis results. Furthermore, as stated in section 3.5.2, the dependent variables NoKAMs and AvgWords consists of count data. Thus, similar for the regressions on auditor change, equations 5, 6, 7, 8, 9 and 10 are re-estimated twice. First, using an OLS regression with log-transformed dependent variables, and a second time using a negative binomial regression.

3.7 Validation of results

This research is performed partly in commission of a Big-Four firm, as it is part of a master thesis internship. This provides the researcher access to auditors within this audit firm for interviews. Three in-depth interviews are held with experienced partners of this audit firm to discuss the results of this thesis. All three partners are signing partner for one or multiple Dutch PIEs. The interviews were held in a semi-structured way. Some questions were sent beforehand to the partners. These questions guided the interviews, but there was also room for probing questions based on the answers provided by the partners. The questions that guided the interviews are provided in appendix C. The interviews give additional insight into the process of creating the KAM report and whether the results are recognized in practice. Three interviews are perceived as sufficient, as the goal of the interviews is to validate the results and not to produce results. It is expected that more than three interviews will not produce a different outcome than the three interviews which are held. The outcome of the three interviews will be compared with each other to verify the information.

(22)

4. Results

In this chapter the results of this thesis are presented. Section 4.1 will elaborate the results the time regressions regarding the yearly standard deviations and the results of the ANOVAs. Section 4.2 will present the results with respect to the effect of an auditor change on the KAMs. In section 4.3 the results of the regressions which include audit quality and financial distress risk will be given. Lastly, the outcome of the interviews with three partners of a Big-Four firm is given in section 4.4.

4.1 Homogenization of the KAM disclosures

Table 3 documents the results of the time regressions regarding the yearly standard deviations. Panel A documents that the standard deviation for the number of KAMs has significantly increased from 2013 until 2016 (P < 0.01). This implies that the auditors are increasingly deviating from the mean for the number of KAMs, while a significant decrease was expected. For the average words used per KAM the standard deviation has significantly decreased from 2013 until 2016 (P < 0,1). This implies that auditors are increasingly using the same amount of words to describe their KAMs, conform the expectations of this thesis. However, if we control for the industry differences by analyzing the standard deviations over the years per industry documented in panel B, the results differ. The yearly increase of the standard deviation of the number of KAMs is still present, however no longer statistically significant. For the average words used per KAM the yearly decrease of the standard deviation is now significant on a higher level (P < 0.01). Time regressions of the standard deviations for each audit firm separately are documented in panel C. The results in panel C show that two audit firms are increasingly deviating from the mean and the other audit firms are decreasingly deviating from the mean. Only the yearly increase of the standard deviation of the number of KAMs for Deloitte (P < 0.05) and PwC (P < 0.01) are significant. This implies that these two firms are increasingly using a KAM report with a number of KAMs that deviates from the mean, while the other two, EY and KPMG, are remaining roughly on the same standard deviation.

The results of the yearly two-way ANOVA are documented in table 3 panel D. For the log-transformed number of KAMs the interaction term is significant (P < 0.05) in 2013, after the first year the interaction term is no longer significant. This implies that in the first year there is a difference between audit firms considering the industries with respect to the number of KAMs reported. From 2014 until 2016 there is no longer a difference between the audit firms considering the industries regarding the number of KAMs that are disclosed. For the log-transformed average words used per KAM the same trend is visible. In the first two years the interaction terms are significant (2013: P < 0.01 & 2014: P < 0.1), in 2015 and 2016 the interaction terms are no longer significant. On top of that a clear decrease in the F-values is visible for the average words per KAM, indicating that KAMs consists increasingly of the same amount of words compared to other audit firms. These results complement the time regression results in panel A and B, which also indicate a trend of homogenization.

(23)

Table 3

Tests on the homogenization of the KAM disclosures

Panel A: Time regression on the yearly standard deviations of the number of KAMs (1) and the yearly standard deviations of the average words per KAM (2).

(Eq1) (Eq2) SDNoKAMs SDAvgWords Year 0.018*** -0.072* (0.002) (0.023) Constant 0.346*** 0.630*** (0.007) (0.056)

Panel B: Time regression on the yearly standard deviations per industry of the number of KAMs (1) and the yearly standard deviations per industry of the average words per KAM (2). (Eq1) (Eq2) SDNoKAMs SDAvgWords Year 0.019 -0.063*** (0.018) (0.018) Constant 0.337*** 0.518*** (0.044) (0.054)

Panel C: Time regression on the yearly standard deviations of the number of KAMs (1) and the yearly standard deviations of the average words per KAM (2) for each audit firm separately. (Eq1) (Eq2) SDNoKAMs SDAvgWords Deloitte Year 0.031** 0.003 EY Year -0.020 -0.091 KPMG Year -0.002 0.013 PwC Year 0.046*** -0.041

The standard deviations of the number of KAMs and the average number of words per KAM are the dependent variables. The results of the time regression show whether the standard deviations are increasing or decreasing over time. Robust standard errors are shown in parentheses.

*** p<0.01, ** p<0.05, * p<0.1

(24)

4.2 Auditor change

The results of the DID model are documented in table 4. These results show the effect of an auditor change on the NoKAMs and the AvgWords, also the results are presented for each change period separately. Table 4 panel A documents a yearly decrease of the coefficient of the CHANGE x POST interaction term, however the interaction variable is and remains insignificant. This implies that the KAM disclosure does not change in size after receiving a new audit firm as auditor. Furthermore, panel B documents a significant decrease in the average words per KAM after an auditor change in the 2013-2014 period. Implying that if a client changes auditors, the new auditor uses 91 words less in his KAMs than his predecessor. However, in the 2013-2014 only six auditor changes were observed, also the result does not hold when performing the DID model with a log-transformed AvgWords or with a negative binomial regression (See appendix B table 1 and 2, respectively). In the two following periods the audit change effect is insignificant, implying that a change of auditor has no effect on the average words per KAM.

Table 3 (continued) Panel D: Yearly two-way ANOVA results

logNoKAMs logAvgWords df F-Values df F-Values 2013 AUDIT_FIRM 3 4.07*** 3 35.04*** SIC 7 1.89* 7 3.19*** AUDIT_FIRM x SIC 20 1.85** 20 3.46*** Residual 125 125 2014 AUDIT_FIRM 3 3.66** 3 12.67*** SIC 8 2.75*** 8 3.03*** AUDIT_FIRM x SIC 22 0.60 22 1.56* Residual 237 237 2015 AUDIT_FIRM 3 2.38* 3 10.19*** SIC 8 4.46*** 8 2.12** AUDIT_FIRM x SIC 22 0.83 22 1.24 Residual 248 248 2016 AUDIT_FIRM 3 5.69*** 3 3.30** SIC 8 2.26** 8 1.71* AUDIT_FIRM x SIC 22 0.79 22 0.71 Residual 243 243

For the two-way ANOVA the dependent variables are the log-transformations of the number of KAMs and the average number of Words per KAM. The interaction terms are of primary interest.

(25)

Table 4

Tests on the effect of auditor change

Panel A: DID model of the effect of auditor changes on the NoKAMs.

(Eq3) (Eq3) (Eq3)

2013-2014 2014-2015 2015-2016 CHANGE -0.436 -0.233 -0.362 (0.369) (0.255) (0.239) POST 0.005 -0.233* -0.250* (0.192) (0.135) (0.140) CHANGE x POST 0.700 0.554 0.126 (0.594) (0.370) (0.347) SIZE 0.260*** 0.343*** 0.369*** (0.100) (0.058) (0.056) CMPLX 0.085 0.027 0.145** (0.107) (0.064) (0.067) ROA -3.236*** -3.754*** -1.333 (1.104) (0.868) (0.853) Constant 1.906** 1.167 0.636 (0.779) (0.728) (0.949) Adjusted R-squared 0.193 0.244 0.247 Observations 192 400 405 Observed changes 6 33 36

Country fixed-effects Yes Yes Yes

Industry fixed-effects Yes Yes Yes

Panel B: DID model effect of auditor changes on the AvgWords.

(Eq4) (Eq4) (Eq4)

2013-2014 2014-2015 2015-2016 CHANGE -0.588 7.527 -9.616 (42.127) (16.475) (15.173) POST 110.911*** 70.420*** 25.563** (14.250) (10.829) (11.679) CHANGE x POST -91.319** 5.811 15.918 (44.475) (23.722) (26.142) SIZE -8.245 -4.286 3.747 (7.380) (3.934) (4.424) CMPLX 5.516 3.846 -4.166 (7.366) (4.560) (5.409) ROA -40.780 -70.419 -3.347 (72.169) (68.659) (47.779)

(26)

4.3 Determinants of the KAM disclosures

Table 5 documents the results of the tests on the effect of audit quality and financial distress risk on the KAMs. Panel A shows significant coefficients of AUDITQ and DISTRESS for the regressions with the dependent variable NoKAMs, for the separate models and the combined model. AUDITQ has a significant positive coefficient (P < 0.05), however this should be reversely interpreted. Audit quality is measured by the discretionary accruals, thus higher AUDITQ in fact means lower audit quality. Therefore, panel A implies that lower audit quality is associated with more KAMs. For DISTRESS the coefficient is significantly positive (P < 0,01). This indicates that more risk for financial distress (measured with O-score) is associated with more KAMs. For the regressions on AvgWords none of variables are significant. This implies that there is no relation between the average number of words used for the KAMs and audit quality or financial distress risk. Table 6 extends the tests on the effect of audit quality and financial distress risk, by documenting the results of the yearly models. Panel A documents the coefficients of AUDITQ and DISTRESS of the yearly regressions models. The determining power of AUDITQ on NoKAMs is, after a small increase in 2014, decreasing over time. For DISTRESS on NoKAMs a small increase is documented as trend. For the tests on AvgWords both variables remain insignificant on a yearly basis. The yearly coefficient of AUDITQ on AvgWords has no clear trend and for the yearly DISTRESS a decreasing trend is visible. The results of the time regressions are presented in panel B. A significant decrease of the coefficient AUDITQ on NoKAMs is documented, implying that over the four-year period the number of KAMs are decreasingly determined based on the level of audit quality. Moreover, a significant increase of the coefficient DISTRESS on NoKAMs is documented. This implies that the number of KAMs are increasingly determined based on the client’s level of financial distress risk.

Table 4 (continued)

(Eq4) (Eq4) (Eq4)

2013-2014 2014-2015 2015-2016 Constant 198.486*** 305.284*** 374.305*** (47.920) (33.667) (35.434) Observations 192 400 405 Observed changes 6 33 36 Adjusted R-squared 0.251 0.187 0.148

Country fixed-effects Yes Yes Yes

Industry fixed-effects Yes Yes Yes

The primary variable of interest is the interaction term CHANGE x POST which documents the increase or decrease after an auditor change, relative to the no change firms. Fixed-effects for country and industry are applied for all models in this table. Robust standard errors are shown in parentheses.

(27)

T ab le 5 T es ts o n th e ef fe ct o f au di t q ua li ty a n d fi na n ci al d is tr es s ri sk P an el A : T es ts o n th e ef fe ct o f au di t q ua li ty a nd f in an ci al d is tr es s ri sk o n th e N oK A M s an d A vg W or ds (E q5 ) (E q6 ) (E q9 ) (E q7 ) (E q8 ) (E q1 0) N oK A M s N oK A M s N oK A M s A vg W or ds A vg W or ds A vg W or ds A U D IT Q 3. 52 3* * 3. 50 7* * -3 0. 43 1 -2 0. 93 4 (1 .6 60 ) (1 .7 28 ) (1 52 .2 74 ) (1 62 .6 16 ) D IS T R E S S 0. 98 6* ** 0. 98 9* ** 21 .5 74 21 .8 62 (0 .2 12 ) (0 .2 13 ) (1 8. 03 7) (1 8. 19 3) SI Z E 0. 34 8* ** 0. 28 7* ** 0. 29 0* ** -1 .3 95 -1 .5 30 -1 .7 87 (0 .0 47 ) (0 .0 43 ) (0 .0 43 ) (3 .1 71 ) (3 .4 06 ) (3 .4 09 ) C M P L X 0. 10 9* * 0. 10 9* * 0. 12 5* ** -0 .3 58 0. 64 8 0. 93 3 (0 .0 49 ) (0 .0 48 ) (0 .0 48 ) (3 .7 05 ) (4 .0 92 ) (4 .1 34 ) R O A -2 .0 20 ** -1 .2 55 * -1 .3 04 ** 11 .0 70 11 .6 07 13 .8 03 (0 .9 71 ) (0 .6 43 ) (0 .6 63 ) (4 4. 07 4) (4 3. 48 7) (4 2. 70 5) C on st an t 3. 69 6* 0. 52 8 0. 42 1 48 7. 42 2* ** 19 7. 90 3* ** 19 8. 36 1* ** (2 .1 56 ) (0 .7 89 ) (0 .7 88 ) (9 8. 77 6) (3 8. 36 5) (3 8. 48 6) O bs er va tio ns 72 8 67 4 67 2 72 8 67 4 67 2 A dj us te d R -s qu ar ed 0. 25 1 0. 25 7 0. 26 3 0. 32 1 0. 31 5 0. 31 3 C ou nt ry f ix ed -e ff ec ts Y es Y es Y es Y es Y es Y es Y ea r fi xe d-ef fe ct s Y es Y es Y es Y es Y es Y es In du st ry f ix ed -e ff ec ts Y es Y es Y es Y es Y es Y es Fi xe d-ef fe ct s fo r co un tr y, y ea r an d in du st ry a re a pp lie d fo r al l m od el s in th is ta bl e. R ob us t s ta nd ar d er ro rs a re s ho w n in p ar en th es es . ** * p< 0. 01 , * * p< 0. 05 , * p < 0. 1

(28)

The results remain unchanged when performing the models with a log-transformation of NoKAMs and AvgWords and using a negative binomial regression, as documented in appendix B table 3 and 4 respectively. One should note that in the combined model (Eq9) audit quality is insignificantly (P = 0.106) related to the log-transformed NoKAMs. However, as the P-value is close to being below 0.1 and the relation remains significant (P < 0.1) when performing the

Table 6

Trend analysis of the effect of audit quality and financial distress risk

Panel A: Yearly tests on the effect of audit quality and financial distress risk on the NoKAMs and AvgWords using combined models. Only the coefficients for AUDITQ and DISTRESS are presented, as these are relevant for the time regression.

2013 2014 2015 2016 NoKAMs AUDITQ 9.443 9.598** 2.350 -0.017 (1.94) (3.14) (0.72) (-0.01) DISTRESS 0.520 1.014** 0.901* 1.171** (1.11) (2.43) (2.11) (2.83) Observations 102 184 190 196 AvgWords AUDITQ 104.0 -416.9 304.0 142.6 (0.36) (-1.60) (0.96) (0.37) DISTRESS 56.75 6.194 8.329 -4.739 (1.10) (0.20) (0.24) (-0.12) Observations 102 184 190 196

Panel B: Time regressions on the yearly coefficients of AUDITQ and DISTRESS for NoKAMs and AvgWords.

NoKAMs NoKAMs AvgWords AvgWords

AUDITQ DISTRESS AUDITQ DISTRESS

Year -3.563** 0.184* 83.670 -18.233

(0.656) (0.055) (106.547) (6.590)

Constant 14.251** 0.441 -175.750 62.217

(2.481) (0.208) (400.575) (23.131)

The yearly coefficients are received by performing equation 9 and 10 on a yearly basis. Only the two main variables are presented in this table, as the trend of the control variables is not of interest here. After that the yearly coefficients are regressed on time, these results are documented in panel B. Robust standard errors are shown in parentheses.

The evolution of the long-form auditor’s report: The homogenization of the Key Audit Matters and the effect of auditor change, audit quality & financial distress. Abstract: