The impact of firm-level, office-level and joint-level industry specialization on audit quality: Evidence from the German audit market

(1)

The impact of firm-level, office-level and joint-level

industry specialization on audit quality: Evidence from

the German audit market

Master thesis MSC Accountancy & Controlling, specialization Accountancy University of Groningen, Faculty of Economics and Business

June 22, 2020 FRANK MEIJER Student number: 3119939 Carneool 41 9207 GE Drachten Tel.: +31 (0)6 34 32 56 18 E-Mail: f.meijer.5@student.rug.nl Supervisor: C. A. Huijgen

Word count (excl. table of contents, references & appendices): 12,455

(2)

2

ABSTRACT

My paper examines the impact of firm-level, office-level and joint-level auditor industry specialization on audit quality in the German audit market. Since the German setting is characterized by several unique peculiarities compared to other environments, this study might increase the understanding of the phenomenon of auditor industry specialization. In order to obtain robust results, I employ several proxies for audit quality and industry specialization. Audit quality is measured based on absolute abnormal discretionary accruals. For industry specialization, I combine several market share-based proxies with ‘seasoning’ variables, which deal with the development process of auditor industry expertise. This way, I create a more complete and reliable measure for the multidimensional construct of industry specialization. Unlike prior research, I find no significant evidence that industry specialized auditors enhance audit quality when pure market share-based measures are used. When the seasoning component is considered, however, I find that seasoned industry specialized auditors with more industry experience provide audits of higher quality than other auditors. This suggests that differences exist within the group of industry specialists regarding their impact on audit quality.

(3)

3

1. INTRODUCTION

Throughout the years, auditors have faced severe scrutiny over several accounting scandals (Jolly, 2019). An example is the bankruptcy of the British construction company Carillion in January 2018, which collapsed with a debt of £1.3 billion after large scale fraud was found in the financial statements (Kollewe, 2018). In response, the UK’s Financial Reporting Council (FRC) started an investigation into KPMG, questioning why KPMG had “signed off the increasingly fantastical figures” in the years before (Jolly, 2019). Next to the scandal of Carillion and KPMG, numerous other cases of questionable audit quality have appeared in the media. Therefore, it is not surprising that a stream of auditing literature has emerged which examines the underlying determinants of audit quality (see Francis, 2004 for an early literature review).

Audit quality can be defined as “the market-assessed joint probability that a given auditor will both (a) discover a breach in the client’s accounting system, and (b) report the breach” (DeAngelo, 1981). This means that auditors are required to be competent and independent in order to provide their clients with high quality audits (Garcia-Blandon & Argiles-Bosch, 2018). Based on the agency perspective, a high quality auditor can play an important role in monitoring earnings management by managers, thereby aligning the interests of shareholders and management and reducing information asymmetries (Anissa, Mukhlasin & Petronila, 2019).

Furthermore, Sun and Liu (2013) assume that auditors have a higher probability to detect earnings management practices and financial misstatements when they are industry specialists. In other words, it is assumed that an audit firm or office can perform higher quality audits when they have more specific knowledge about the particular industry of their audit clients (Balsam, Krishnan & Yang, 2003; DeFond, Francis & Wong, 2000; Reichelt & Wang, 2010).

Jaggi, Gul and Lau (2012) encourage further research regarding the impact of industry specialization on audit quality. They argue that different institutional settings with differing legal environments should be studied in order to obtain a better understanding of the phenomenon of industry specialization. Bedard (2012) agrees and mentions that the institutional context, which largely differs between countries, is so important for the accountability of auditors that additional research in understudied environments is needed.

(5)

5 This study responds to the request by Jaggi et al. (2012) and Bedard (2012) by investigating the impact of firm-level (i.e. national), office-level (i.e. city) and joint-level industry specialization on audit quality in the German audit market. The analysis of the German setting is motivated by its representativeness to other European countries and by its special disclosure requirements (Quick, Schenk, Schmidt & Towara, 2018). Consistent with Gul, Wu and Wang (2013), Porumb, de Jong, Huijgen, Marra and van Dalen (2018) and Garcia-Blandon and Argiles-Bosch (2018), I use the absolute amount of abnormal discretionary accruals as a proxy for audit quality.

This leads to the following research question:

To what extent do firm-level, office-level and joint-level industry specialization have an impact on audit quality?

The German institutional environment is characterized by a two-tier governance system (Quick et al., 2018). Similar to the Chinese setting examined by Gul et al. (2013), both the engagement partner and the review partner are required to sign the audit report (Quick et al., 2018; Porumb et al., 2018). The engagement partner performs the audit engagement with the engagement team, while the review partner reviews the quality of the work done by the engagement partner and the engagement team (Epps & Messier, 2007). In other words, the engagement partner has a more active role in the engagement process, while the review partner acts more passive and is less dependent on the audit client (Epps & Messier, 2007).

The German setting can be distinguished from other markets because it has several unique peculiarities (Quick et al., 2018). For instance, litigation risk for German auditors is limited (Quick et al., 2018). Due to this limited liability, there is less incentive for German auditors to perform high quality audits (Quick et al., 2018; Porumb et al., 2018). However, the German audit market is also known for its high reputational risk, where audit clients are likely to dispose of low quality auditors (Quick et al., 2018). As a consequence of the low litigation risk, German auditors consider the risk of reputational losses as a stronger incentive to provide high quality audits (Quick et al., 2018; Porumb et al., 2018).

(6)

6 By considering the characteristics of the German audit market, as a unique and understudied institutional setting with low litigation and high reputational risk, this study might give additional insight in the relationship between firm-level and office-level industry specialization and audit quality. Although the German setting is quite distinctive in several aspects, it also shows similarities with other European countries (Quick & Warming-Rasmussen, 2009). This means that the findings of this study might be valid for other countries with similar institutional environments as well.

Furthermore, this study contributes to literature assessing the impact of industry specialization on audit quality by examining a unique dataset with German public listed firms. This dataset for the years 1999-2011 is hand-collected by Porumb et al. (2018) and consists of information regarding individual partners, their firms, offices and accounting data. This makes it possible to analyze the influence of industry specialization on audit quality at multiple levels. As such data is scarce and not always publicly available (DeFond & Zhang, 2014; Gul et al., 2013), this study might provide interesting insights into the impact of industry specialization on audit quality.

A challenging aspect of the construct of industry specialization is that it can only be measured indirectly through the use of proxies (Garcia-Blandon & Argiles-Bosch, 2018). Moreover, DeFond and Zhang (2014) mention that there is no consensus on the best measure for industry specialization, while accurate measurement is crucial for reliable and valid results (Audousset-Coulier, Jeny & Jiang, 2016). Most papers tend to determine the multidimensional construct of industry specialization with a pure and single market share-based measure (e.g. Balsam et al., 2003; Reichelt & Wang, 2010). However, this leads to inconsistent classifications of industry expertise as well as validity issues (Audousset-Coulier et al., 2016). In order to obtain robust results, I therefore use multiple market share-based measures for industry specialization simultaneously. In addition, I contribute to existing literature on auditor industry specialization by considering a development or seasoning process which takes the tenure of industry specialists into account (adapted from Gaver & Utke, 2019). This way, I intend to create a more encompassing variable to obtain more valid and reliable results.

(7)

7 My study shows mixed results. Unlike prior studies, I do not find evidence that industry specialist auditors enhance audit quality when pure market share-based measures are used. This result seems to be robust to most of my additional robustness tests. However, when I also consider the effect of seasoning, I find that seasoned industry specialist auditors with more years of experience within an industry significantly improve audit quality at the firm and at the office level and that unseasoned industry specialist auditors with fewer years of experience do not significantly influence audit quality. At the joint level, my results are insignificant, although they prove that unseasoned industry specialist auditors with fewer years of experience seem to have a significant decreasing impact on audit quality. These results seem to indicate that differences exist within the group of specialist auditors, where first year auditors do not and auditors with more experience do enhance audit quality. This is an interesting finding, which might have implications for other literature on industry specialization and audit quality.

The paper will proceed as follows. Section 2 discusses the characteristics of the German institutional setting and the theoretical background, including the hypotheses development. Section 3 discusses the research method. Section 4 shows the results. In section 5, the results from the additional robustness and sensitivity tests are presented. Finally, section 6 provides a discussion in which the conclusions of this paper are drawn.

(8)

8

2. INSTITUTIONAL BACKGROUND, THEORETICAL BACKGROUND

AND HYPOTHESIS DEVELOPMENT

2.1 The German Institutional Setting

The German audit market has several unique peculiarities which distinguish this setting from other auditing environments. It is likely that these specific features lead to different results in comparison to other settings (Quick & Warming-Rasmussen, 2009). The most important characteristics are discussed in more detail below.

The German setting is characterized by a two-tier corporate structure consisting of an executive board and a supervisory board (Krauß, Pronobis & Zülch, 2015; Porumb et al., 2018; Quick et al., 2018). The executive board is responsible for the management of the company, while the supervisory board monitors and appoints the executive board (Quick et al., 2018; Krauß et al., 2015). The supervisory board also appoints the auditor (Quick et al., 2018; Quick & Warming-Rasmussen, 2009). The auditor communicates his findings to the supervisory board in the audit report (Quick et al., 2018; Quick & Warming-Rasmussen, 2009). In effect, the auditor acts as an agent of the supervisory board and indirectly as an agent of the shareholders and other stakeholders (Quick & Warming-Rasmussen, 2009; Watts & Zimmerman, 1983). Consequently, and in contrast to the one-tier system, the risk that the auditor is bonding with the management of the company is mostly mitigated (Porumb et al., 2018; Quick et al., 2018).

Krauß et al. (2015) argue that the segregation of duties between the management and the monitoring function is stronger and more clear in two-tier governance systems than in the one-tier setting and that this has consequences for the demand for audit quality. Weber, Willenborg and Zhang (2008) agree and find that the demand for high quality audits is lower in institutional settings where the degree of board monitoring is already high, as is the case in Germany. In such environments, shareholders seem to be more interested in the monitoring performed by their representatives on the supervisory board instead of in the work performed by the auditor (Weber et al., 2008; Quick & Warming-Rasmussen, 2009). Because investors seem to neglect the auditors’ opinion, German auditors have less incentive to deliver a high quality audit (Weber et al., 2008; Quick & Warming-Rasmussen, 2009).

(9)

9 Another distinctive characteristic of the German audit market is that the litigation risk is relatively low (Quick et al., 2018). According to section 323 of the German Commercial Code, the liability of auditors towards listed audit clients for misconduct is limited to an amount of four million euros (Quick et al., 2018). This is much lower compared to other settings, such as the US and the UK (Porumb et al., 2018). Moreover, the audit client is the only party that can sue the auditor for damages caused by financial misstatements (Porumb et al., 2018) and the client must be able to demonstrate that “the auditor acted intentionally or with reckless disregard for the truth” (Weber et al., 2008). So, it is notoriously difficult for audit clients to make a successful claim (Weber et al., 2008).

However, the German audit market is also characterized by high reputational risk (Quick et al., 2018; Porumb et al., 2018). German audit firms recognize that clients are likely to leave to another firm when they provide low quality audits (Quick et al., 2018). Therefore, the high reputational risk seems to be the most important motivation for German auditors to deliver high quality audits (Quick et al., 2018; Porumb et al., 2018).

Furthermore, Germany differs from other countries when it comes to the path that future auditors must follow in order to obtain the right to practice the audit profession (Porumb et al., 2018). First of all, to become a licensed auditor in Germany, people are required to have a university degree and a minimum experience of three years (Quick et al., 2018; Porumb et al., 2018). On top of this, an auditor should pass all the public accountant exams (Quick et al., 2018).

Finally, in the German setting it is required that the audit report is signed by two auditors – the engagement partner and the review partner (Porumb et al., 2018; Epps & Messier, 2007). While the engagement partner actively performs the audit with the engagement team, the review partner is responsible for reviewing their performance (Epps & Messier, 2007). The review partner basically acts as a quality control instrument by providing a second opinion on the findings of the engagement partner and the engagement team (Epps & Messier, 2007). The main idea behind this is that the additional quality review will help to enhance audit quality (Porumb et al., 2018; Epps & Messier, 2007).

(10)

10

2.2 Theoretical Background

The research on audit quality is part of the stream of auditing literature and can be linked to agency and contracting theory (DeFond & Zhang, 2014; Goodwin & Wu, 2014; Quick et al., 2018). Agency theory suggests that agency conflicts are caused by the separation of ownership and control, which leads to information asymmetry problems between shareholders and management (Jensen & Meckling, 1976; Fama & Jensen, 1983). This so-called moral hazard problem gives management the opportunity to act in their own interests instead of the shareholders’ interests (Jensen & Meckling, 1976; Fama & Jensen, 1983). Contracting theory assumes that accounting-based contracts can reduce agency costs (Jensen & Meckling, 1976). Auditors can provide assurance of the integrity of the accounting numbers from the firms’ financial statements, which makes it possible to use these in contracting (Jensen & Meckling, 1976).

According to Goodwin & Wu (2014), “auditors should, in addition to the generic knowledge required for all audits, command engagement-specific expertise, including familiarity with the client’s accounting and internal control systems, specialized accounting rules, economic activities, contractual arrangements, ownership structure, and the nature of agency conflicts between contracting parties in order to provide effective assurance services”. With this specific knowledge, auditors are better able to identify and detect accounting misstatements in the financial statements or in contracts (Craswell, Francis & Taylor, 1995). In other words, industry specialists can offer higher audit quality to their clients and reduce the clients’ agency costs (Goodwin & Wu, 2014). Furthermore, because agency problems in an industry are often similar to each other, industry specialization could lead to economies of scale and, consequently, to audits that are cheaper to perform (Craswell et al., 1995).

This study will draw on the agency and contracting theory to explain the impact of firm-level and office-level industry specialization on audit quality. It is likely that managers in practice try to manage earnings for their own benefit or to avoid reputational damage at the cost of shareholders (Stein, 2019; Anissa et al., 2019). Agency and contracting theory can explain why industry specialized auditors can be an effective monitoring instrument to reduce earnings management and agency costs, reducing the information asymmetries between shareholders and management (Anissa et al., 2019; Stein, 2019; Goodwin & Wu, 2014; Craswell et al., 1995).

(11)

11

2.3 Hypothesis Development

Research on industry specialization has been a popular topic for years and is considered by auditors, investors and regulators as a determinant of audit quality (Christensen, Glover, Omer & Shelley, 2016). There are many studies that find a positive association between firm-level industry specialization and audit quality (DeFond & Zhang, 2014). For instance, Balsam et al. (2003) find that audit quality is higher for clients who hire audit firms that are industry specialists. This in consistent with Krishnan (2003), who find that audit firms with industry expertise reduce the degree of earnings management more than nonspecialist auditors.

Furthermore, Lim and Tan (2010) add that industry specialized auditors train personnel and invest in technology development in certain industries to maintain and enhance audit quality, which means that industry specialized firms have more possibilities to develop expertise than nonspecialist firms. In addition, industry specialized firms are more likely to be concerned about reputational losses and litigation exposure (Lim & Tan, 2008; Lim & Tan, 2010). Similar to DeAngelo (1981), who found that larger audit firms deliver higher audit quality because they have more audit clients to lose, industry specialists have a higher risk to lose future earnings and audit fees when they perform low quality audits within the industries in which they are specialized (Lim & Tan, 2010). In other words, industry specialized audit firms have greater incentives to perform high quality audits to avoid losing reputation compared to nonspecialized audit firms (Watts & Zimmerman, 1983).

In line with what is mentioned above, I expect that firm-level industry specialization in positively associated with audit quality. By studying the German audit market, which is mostly characterized by high reputation risk, I predict a significant effect of audit firm specialization on the level of audit quality. In other words, my expectation is that abnormal discretionary accruals are lower when the audit firm has industry expertise. This leads to the following hypothesis:

Hypothesis 1: Industry specialized audit firms have a decreasing impact on the level of abnormal discretionary accruals of the audit client.

(12)

12 Although a wide body of research on firm-level industry specialization is already available, studies conducted at the office-level are still relatively limited (DeFond & Zhang, 2014). Ferguson, Francis and Stokes (2003) investigated the Australian setting and find that industry expertise is mostly driven by office-level industry leadership within audit markets that are city-specific. This is in line with Francis, Reichelt and Wang (2005) and Fung, Gul and Krishnan (2012), who find evidence of significant audit fee premia for audit offices that are industry specialists in the US audit market. Basioudis and Francis (2007) also document this effect for the UK setting. Furthermore, Stein (2019) argues that audit clients are more likely to record asset impairments and bad news when the auditor is specialized at the office-level.

With their findings in mind, I also expect to find a positive association between office-level industry specialization and audit quality in the German audit market. This means that I expect that the amount of abnormal discretionary accruals will be lower for industry specialized audit offices. Therefore, my second hypothesis is:

Hypothesis 2: Industry specialized audit offices have a decreasing impact on the level of abnormal discretionary accruals of the audit client.

In addition to the separate effects of firm-level and office-level industry specialization on audit quality, several other studies find a joint impact of firm- and office-level industry specialization on audit quality. For instance, Reichelt and Wang (2010) argue that audit quality is highest when auditors are both firm-level and office-level industry specialists, suggesting the presence of synergies. This is supported by Mohd Kharuddin and Basioudis (2018), who find that audit quality is only affected when an auditor has industry expertise on both the firm-wide and level. More recent studies also tend to agree that there is a joint effect of firm-level and office-level industry specialization on audit quality (Anissa et al., 2019; Miah, 2019).

Therefore, I also expect that there will be a joint impact of firm-level and office-level industry specialization on audit quality. That is, I expect that the amount of abnormal discretionary accruals will be lower when an auditor has industry expertise at the firm- as well as the office-level. This leads to my third hypothesis:

Hypothesis 3: The joint effect of firm-level and office-level industry specialization has a decreasing impact on the level of abnormal discretionary accruals of the audit client.

(13)

13

3. RESEARCH METHOD

3.1 Dependent and Independent Variables

3.1.1 Audit Quality

In my study, I use the amount of abnormal discretionary accruals to proxy for audit quality. This is one of the most widely used audit quality measures (DeFond & Zhang, 2014) and is consistent with other studies on industry specialization (e.g. Balsam et al., 2003; Lim & Tan, 2008; Reichelt & Wang, 2010).

Since managers can perform earnings management both upwards (income-increasing) and downwards (income-decreasing) (Dechow, Hutton, Kim & Sloan, 2012), I consider the absolute value of abnormal accruals. This is because an absolute accruals measure can correct earnings management in both directions (Lennox, Wu & Zhang, 2016). Moreover, accruals reverse at a certain moment and an absolute measure of abnormal accruals can capture this time effect (Lennox et al., 2016).

Accruals are calculated based on the Jones (1991) model, as modified by Dechow, Sloan and Sweeney (1995). Total accruals are estimated in Equation (1):

𝑇𝐴𝑖𝑡 𝐴𝑖𝑡−1

= 𝛽

0

(

1 𝐴𝑖𝑡−1

) + 𝛽

1

(

∆𝑅𝐸𝑉𝑖𝑡−∆𝑅𝐸𝐶𝑖𝑡 𝐴𝑖𝑡−1

) + 𝛽

2

(

𝑃𝑃𝐸𝑖𝑡 𝐴𝑖𝑡−1

) + 𝜀

𝑖𝑡 (1) where:

TA_𝑖𝑡 = total accruals (income before extraordinary items – operating cash flow) for client firm i in year t.

∆𝑅𝐸𝑉𝑖𝑡 = sales of client firm i in year t minus sales of company i in year t – 1.

∆𝑅𝐸𝐶_𝑖𝑡 = receivables of client firm i in year t minus receivables of company i in year t – 1.

𝑃𝑃𝐸𝑖𝑡 = net property, plant and equipment for client firm i in year t. 𝜀_𝑖𝑡 = error term.

𝐴_𝑖𝑡−1 = lagged total assets for client firm i in year t – 1.

In line with Dechow et al. (1995), the expected total accruals are estimated in Equation (2):

𝐸𝑇𝐴𝐶𝐶_𝑖𝑡 𝐴_𝑖𝑡−1

= 𝛽̂

0

(

1 𝐴_𝑖𝑡−1

) + 𝛽̂

1

(

∆𝑅𝐸𝑉_𝑖𝑡−∆𝑅𝐸𝐶_𝑖𝑡 𝐴_𝑖𝑡−1

) + 𝛽̂

2

(

𝑃𝑃𝐸_𝑖𝑡 𝐴_𝑖𝑡−1

)

(2)

(14)

14 where :

ETACC_𝑖𝑡 = expected total accruals (income before extraordinary items – operating cash flow) for client firm i in year t.

𝛽̂₀ to 𝛽̂₂ = estimated coefficients based on Equation (1).

The amount of abnormal discretionary accruals can be computed as the difference between the total accruals and the expected total accruals, the outcomes from Equation (1) and (2):

𝐴𝐵𝑆𝐷𝐴𝐶𝐶

_𝑖𝑡

=

𝑇𝐴𝑖𝑡

𝐴𝑖𝑡−1

−

𝐸𝑇𝐴𝑖𝑡

𝐴𝑖𝑡−1

(3)

where ABSDACC represents the amount of abnormal discretionary accruals for client firm i in year t. In other words, ABSDACC describes the amount of earnings that potentially might have been managed upwards or downwards (Reichelt & Wang, 2010).

Consistent with Porumb et al. (2018), I estimate abnormal accruals per industry based on the Barth industry classification developed by Barth, Beaver and Landsman (1998). I require a minimum of 20 observations per industry to ensure that accruals are estimated with sufficient observations. For potential year effects is controlled by including fixed year effects in the empirical regression models (section 3.2).1

3.1.2 Firm-, Office- and Joint-Level Industry Specialization

This study examines the impact of auditor industry specialization at the firm (i.e. national), office (i.e. city) and joint level. I build on two market share-based industry specialization definitions from Reichelt and Wang (2010) to determine when an auditor can be classified as an industry specialist. Following prior studies (e.g. Balsam et al., 2003), the first definition is concerned with auditor dominance within a specific industry. Following definition 1, I classify an auditor as an industry specialist when the auditor has the largest relative national (city) market share in a specific year and when the market share is at least 10%-points higher than the market share of the second largest competitor in the industry (Reichelt & Wang, 2010).

1_{Accruals are estimated per industry only. This is not in line with Reichelt and Wang (2010), Minutti-Meza (2013)}

and Audousset-Coulier et al. (2016), who estimated accruals per industry and per year. However, these studies use larger databases which contain more observations (10,000+) compared to the database with German firms used for this study (6,304). In order to ensure that sufficient observations remain in the sample, accruals are therefore estimated per industry only.

(15)

15 Following DeAngelo (1981), the second definition assumes that industry expertise should increase when the auditor market share in a specific industry increases as well. This is because it seems likely that auditors with a larger market share have more incentives invest in their specialist industry and provide higher audit quality (Reichelt & Wang, 2010). Therefore, definition 2 classifies an auditor as industry specialist when the auditor has an absolute national (city) market share above 30%- (50%-) points (Reichelt & Wang, 2010).

What should be noted, however, is that Reichelt and Wang (2010) determine calculate auditor market share based on audit fees only. According to Minutti-Meza (2013) and Audousset-Coulier et al. (2016), audit fees only are not a reliable measure: additional market share measures should be used simultaneously. Following their suggestions, I calculate auditor market share as the weighted average of the natural logarithm of total assets and total sales and the number of industry clients:2

𝑀𝐴𝑅𝐾𝐸𝑇𝑆𝐻𝐴𝑅𝐸

_𝑖𝑘

=

∑ 𝑋𝑖𝑗𝑘 𝐽𝑖𝑘 𝑗=1

∑𝐼𝑘_𝑖=1∑𝐽𝑖𝑘_𝑗=1𝑋_𝑖𝑗𝑘 (4)

where:

𝑀𝐴𝑅𝐾𝐸𝑇𝑆𝐻𝐴𝑅𝐸_𝑖𝑘 = auditor market share.

𝑋_𝑖𝑗𝑘 = calculating variable which consists of the natural logarithm of total assets and total sales, and the number of clients in a specific industry.

I = auditor.

J = client firm.

K = industry (based on two-digit SIC codes).

Several studies determine auditor industry specialization solely based on market share, following similar approaches as in Equation (4) (e.g. Lim & Tan, 2008; Reichelt & Wang, 2010; Garcia-Blandon & Argiles-Bosch, 2018). However, there is no consensus on the best proxy for industry specialization (DeFond & Zhang, 2014). Furthermore, Minutti-Meza (2013) and Audousset-Coulier et al. (2016) argue that market share-based proxies on its own are not able to capture all dimensions of auditor industry specialization, which is a multidimensional construct. Adopting this approach might lead to severe validity issues (Audousset-Coulier et al., 2016).

2_{To illustrate the results of the market share calculation, the market share distributions within SIC category 35}

(16)

16 Gaver and Utke (2019) agree that more determinants of industry specialization should be examined. They contend that, in addition to market share, the development or ‘seasoning’ process of industry specialist auditors should be considered to measure the construct of industry specialization more reliably. The ‘seasoning process’ represents the time it takes before an auditor has developed sufficient experience to become an industry specialist and from that point, audit quality is really enhanced (Gaver & Utke, 2019).

Following this suggestion, I combine the market share-based measures introduced by Reichelt and Wang (2010) with the seasoning process proposed by Gaver and Utke (2019) to create a more encompassing variable to measure the multidimensional construct of auditor industry specialization.

3.2. Empirical Model

To estimate the effect of firm-level, office-level and joint-level industry specialization on audit quality, I apply multivariate panel regressions with year fixed effects. I regress the absolute value of abnormal discretionary accruals, which is calculated in Equation (3), on the auditor industry specialization variables introduced in section 3.1.2. Furthermore, I add several variables to control for other determinants affecting abnormal discretionary accruals which are unrelated to the concept of industry specialization. All variables used in the regression models are defined in table 1.3

First, before examining the effect of seasoning, I examine the effect of auditor industry specialization using the market share-based proxies only (Reichelt & Wang 2010).4 This leads to the following regression model (5):

|𝐴𝐵𝑆𝐷𝐴𝐶𝐶_𝑖𝑡| = 𝛽₀+ 𝛽₁𝑁𝐼𝑆𝑃_𝑖𝑡+ 𝛽₂𝐶𝐼𝑆𝑃_𝑖𝑡+ 𝛽₃𝑁𝐶𝐼𝑆𝑃_𝑖𝑡+ 𝛽₄𝑙𝑛𝑆𝐼𝑍𝐸_𝑖𝑡+ 𝛽₅𝑅𝑂𝐴_𝑖𝑡+ 𝛽6𝐶𝐹𝑂𝑖𝑡+ 𝛽7𝐿𝐸𝑉𝑖𝑡+ 𝛽8𝐿𝑂𝑆𝑆𝑖𝑡+ 𝛽9𝑀𝐵𝑖𝑡+ 𝛽10𝑆𝐺𝑅𝑂𝑊𝑇𝐻𝑖𝑡+ 𝛽11𝑙𝑛𝑇𝐸𝑁𝑈𝑅𝐸𝑖𝑡+

𝛽₁₂𝐿𝐼𝑇_𝑖𝑡+ 𝛽₁₃𝐼𝐹𝑅𝑆_𝑖𝑡+ 𝛽₁₄𝐵𝐼𝐺3_𝑖𝑡+ 𝛽₁₅𝑌𝐸𝐴𝑅_{𝐹𝐸 𝑡} (5)

3_{In line with Reichelt and Wang (2010), all continuous variables are winsorized at the 1 and 99 percent level to}

reduce the influence of outliers.

4_{Please refer to section 3.1.2 for further clarification of the market share-based industry specialist definitions 1}

(17)

17 The control variables included in the regression model are widely used in accounting literature (e.g. Reichelt & Wang, 2010; Minutti-Meza, 2013; Gaver & Utke, 2019). For instance, lower levels of abnormal accruals are expected for large firms (lnSIZE), for firms with higher returns on assets (ROA) and higher cash flows from operations (CFO) (Myers, Myers & Omer, 2003; Ashbaugh, LaFond & Mayhew, 2003), longer auditor tenure (lnTENURE) (Gul, Fung & Jaggi, 2009) and a Big 4 auditor (Myers et al., 2003; Ashbaugh et al., 2003).

However, the competitive positions in Germany are different compared to other countries. For example, the lower amount of observations for Deloitte in the German sample are not representative compared to other settings. Therefore, Deloitte cannot be classified as a Big 4 auditor in the German setting.5 This means that I control for KPMG, EY and PWC only (BIG3) because these firms have more observations and are much larger than the other firms represented in the sample. Additionally, I include a dummy variable to control for the IFRS adoption in the EU (and Germany) in 2005, since IFRS is expected to contribute to lower levels of abnormal accruals (Marra, Mazzola & Prencipe, 2011).

Higher levels of abnormal accruals are expected for firms that are highly leveraged (LEV) and report losses (LOSS), which are both incentives for managers to engage in earnings management (Asbaugh et al., 2003). This is also expected for firms that operate in a high-litigation industry (LIT) (Asbaugh et al., 2003) and for firm that have a higher market-to-book (MB) and sales growth ratio (SGROWTH) (Reichelt & Wang, 2010; Carey & Simnett, 2006). Finally, I include year dummies to control for year effects. Consistent with previous studies (e.g. Reichelt & Wang, 2010, Minutti-Meza, 2013), industry fixed effects are not included because I estimate the impact of industry specialization on audit quality per industry.

5_{Deloitte represents (163) on a total of (2,584) observations in my industry specialization sample. This is similar}

to the size of BDO (160), which means that Deloitte can be considered as a second tier auditor instead of a Big 4 auditor. However, the other Big 4 offices KPMG (536), EY (429) and PWC (399) have much more observations, which means that Deloitte is underrepresented in the German sample compared to other countries. Since Deloitte can be considered as a smaller audit firm in the German market, I control for KPMG, EY and PWC as the Big 3 only in my regression analysis.

(18)

18 TABLE 1

Definition of Variables Used in the Regression Models (5) and (6)

Variables Definitions

Dependent Variables

|ABSDACCit| = absolute value of abnormal discretionary accruals for client i at time t,

calculated by Equation (3).

Independent Variables

NISPit = a dummy variable that is assigned 1 if the audit firm is an industry

specialist at the national level for client i at time t, and 0 otherwise.

CISPit = a dummy variable that is assigned 1 if the audit office is an industry

specialist at the city level for client i at time t, and 0 otherwise.

NCISPit = a dummy variable that is assigned 1 if the auditor is an industry

specialist at both national level and city level for client i at time t, and 0 otherwise.

SEASONEDit = a dummy variable that is assigned 1 when the auditor has been classified

as a market leading industry specialist in a given industry for 2 or more consecutive years, and 0 otherwise.

UNSEASONEDit = a dummy variable that is assigned 1 when the auditor is in its first year

of being classified as a market leading industry specialist in a given industry, and 0 otherwise.

Control Variables

lnSIZEit = natural logarithm of total assets for client i at time t.

ROAit = EBIT for client i at time t scaled by lagged total assets for client i at time

t.

CFOit = cash flow from operations for client i at time t scaled by lagged total

assets for client i at time t.

LEVit = total debt for client i at time t scaled by lagged total assets for client i at

time t.

LOSSit = a dummy variable that is assigned 1 if net income is negative for client

i at time t, and 0 otherwise.

MBit = market value of equity for client i at time t / book value of equity for

client i at time t.

SGROWTHit = sales growth for client i at time t: (salest – salest-1) / salest-1.

lnTENUREit = natural logarithm of the number of consecutive years that the auditor

has audited the firm’s financial statements for client i at time t.

LITit = a dummy variable that is assigned 1 if client i operates in a high

litigation industry at time t, (SIC codes 2833-2836, 3570-3577, 3600-3674, 5200-5961, and 7370), and 0 otherwise.

IFRSit = is assigned 1 for client i when applying IFRS or US GAAP at time t,

and 0 otherwise.

BIG3it = a dummy variable that is assigned 1 if the auditor of client i at time t is

a BIG 3 auditor (KPMG, EY or PWC), and 0 otherwise.

YEAR_FEt = dummies for year fixed effects.

(19)

19 Second, I replace the market share-based industry specialist variables with two variables (‘seasoned’ and ‘unseasoned’) to examine the seasoning effect, as proposed by Gaver and Utke (2019). Consistent with their study, I classify an auditor as ‘seasoned’ when the auditor has been classified as an industry market leader for two or more consecutive years and as ‘unseasoned’ when the auditor is in a market leading position within an industry for the first year.6 These variables are coded 1 if the stated condition is applicable, where a coded 0 on both variables means that the auditor is not classified as an industry specialist (Gaver & Utke, 2019). This leads to a revised regression model, which is presented in model (6):

|𝐴𝐵𝑆𝐷𝐴𝐶𝐶_𝑖𝑡| = 𝛽₀+ 𝛽₁𝑆𝐸𝐴𝑆𝑂𝑁𝐸𝐷_𝑖𝑡 + 𝛽₂𝑈𝑁𝑆𝐸𝐴𝑆𝑂𝑁𝐸𝐷_𝑖𝑡 + 𝛽₃𝑙𝑛𝑆𝐼𝑍𝐸_𝑖𝑡 + 𝛽₄𝑅𝑂𝐴_𝑖𝑡+ 𝛽₅𝐶𝐹𝑂_𝑖𝑡+ 𝛽₆𝐿𝐸𝑉_𝑖𝑡+ 𝛽₇𝐿𝑂𝑆𝑆_𝑖𝑡+ 𝛽₈𝑀𝐵_𝑖𝑡+ 𝛽₉𝑆𝐺𝑅𝑂𝑊𝑇𝐻_𝑖𝑡+ 𝛽₁₀𝑙𝑛𝑇𝐸𝑁𝑈𝑅𝐸_𝑖𝑡+

𝛽11𝐿𝐼𝑇𝑖𝑡+ 𝛽12𝐼𝐹𝑅𝑆𝑖𝑡+ 𝛽13𝐵𝐼𝐺3𝑖𝑡+ 𝛽14𝑌𝐸𝐴𝑅_{𝐹𝐸 𝑡} (6)

3.3 Data Source and Sample Selection

As the archival basis for my samples, I use a unique database which is hand collected by Porumb et al. (2018). It includes German public listed firms audited by all types of auditors (Big 4 and non-Big 4) during the 1999-2011 period. The inclusion of both Big 4 and non-Big 4 auditors is consistent with Reichelt and Wang (2010) and Minutti-Meza (2013).7_Financial firms are excluded because of their dissimilar nature compared to firms in other industries.

Porumb et al. (2018) retrieve accounting data from WorldScope. Additionally, the names of audit firms, offices and cities are gathered by a close inspection of the corresponding annual report, which is retrieved from Thomson Research. Furthermore, Porumb et al. (2018) assign an identifier code to the audit firms, offices and cities in which the offices are located. This process provides a sample of 6,304 firm-year observations and contains 259 audit firms, 448 audit offices and 104 cities. In line with Reichelt and Wang (2010), I use this sample to calculate the auditor’s market share at the firm- and office-level per year and per industry (based on two-digit SIC codes).

6_{I determine whether an auditor is a market leader or not by considering the definitions from Reichelt & Wang}

(2010), which are also used for model (5). In model (6), the ISP variables are split up into SEASONED and UNSEASONED to examine the development process of industry expertise.

7_{However, Deloitte cannot be considered as a Big 4 auditor in my sample. Since the number of observations of}

Deloitte significantly differs from the number of observations in other audit markets, I exclude Deloitte from the Big 4 and control for the Big 3 effect (KPMG, EY, PWC) in my multivariate regression analysis only by including a dummy variable in regression models (5) and (6).

(20)

20 Moreover, I use this sample for the analysis of accruals. As described in section 3.1.1, I run the accruals model based on the Barth et al. (1998) industry classification and require a minimum of 20 observations per industry. This requirement leads to a deduction of 3 observations, yielding a final sample of 6,301 observations.

After the accruals analysis, I delete 993 observations with missing values for the variables used in the regression analysis. Furthermore, Francis et al. (2005) mention that too few observation per national-industry-year combinations can lead to measurement errors. Also, it is unlikely that we can speak of a competitive audit market for industry specialization when there is only one company in the industry (Basioudis & Francis, 2007). Therefore, I delete national-industry-year 179 combinations with less than two client and audit firms to ensure that industry specialization is not determined by too few companies (Francis et al., 2005). Following the same rationale, I also deduct city-industry-year 2,545 combinations with less than two observations. This results in a final sample of 2,584 firm-year observations in 33 two-digit SIC industry groups, consisting of 116 audit firms, 217 audit offices and 20 cities. I use this sample to examine the impact of auditor industry specialization on audit quality.

Table 2 provides an overview of the sample selection process. TABLE 2

Sample Selection Process

Panel A: Sample for calculation of market share

German non-financial listed firms with accounting data and auditor data 6,304

Panel B: Sample for analysis of abnormal discretionary accruals

Observations from panel A 6,304

Less: Barth industry groups with less than 20 observations 3

Final sample for analysis of abnormal discretionary accruals 6,301

Panel C: Sample for analysis of industry specialization

Observations from panel B 6,301

Less: observations with missing values 993

Less: SIC2 industry-year combinations with less than two clients and audit firms 179

Less: city-SIC2 industry-year combinations with less than two clients and audit offices 2,545

(21)

21

4. RESULTS

4.1 Descriptive Statistics

Table 3 reports the descriptive statistics of auditor industry specialization. The means and standard deviations of the market share-based8 and seasoning9 definitions of industry specialization are given for the firm, office and joint level. At the firm level, more industry specialists can be identified when the second definition from Reichelt and Wang (2010) is followed (a mean of 13,5% versus 12,7%). This is also true for the second seasoning definition from Gaver and Utke (2019) (9.7% versus 8.4%). The mean of definition 1 is similar to the mean reported by Reichelt and Wang (2010). However, the mean of definition 2 does not match the mean reported in their study: the difference between the means of definition 1 and 2 is much smaller for the German setting.10 This is interesting because it seems that the national industry specialist definitions are more comparable and closer to each other when only the German audit market is analyzed instead of when a cross-country analysis with multiple audit environments is conducted.11

At the office level, more industry specialists can be identified when the first definitions of market share-based industry specialization (28.1% versus 26.4%) and seasoning (17.0% versus 14.9%) are followed. These descriptive statistics are comparable to the statistics found by Reichelt and Wang (2010), Minutti-Meza (2013) and Gaver and Utke (2019). The means of the first and second definition of joint-level industry specialization and seasoning are almost equal to each other (8.1% versus 7.9% and 4.3% versus 4.8%).

8_{Please refer to section 3.1.2 for the market share-based industry specialist definitions as proposed by Reichelt &}

Wang (2010).

9_{Please refer to section 3.2 for the seasoning-based definitions as introduced by Gaver & Utke (2019).}

10_{In their study, Reichelt and Wang (2010) report a mean of 11.6% for definition 1 and a mean of 21.4% for}

definition 2. This is a difference of 9.8%-points, while the descriptive statistics in my study show a difference of only 1.3%-points between the two definitions.

11_{Reichelt and Wang (2010) and Minutti-Meza (2013) use Audit Analytics and Compustat Annual as the main}

databases in their study. These databases consist of observations which come from multiple countries throughout the world.

(22)

22 TABLE 3

Descriptive Statistics of Auditor Industry Specialization

(N = 2,584)

National/City/Joint Level Industry Specialist Mean Std. Dev.

National Level Industry Specialist

Industry Specialist 1 0.127 0.333 Industry Specialist 2 0.135 0.342 Seasoned 1 0.084 0.277 Unseasoned1 0.043 0.203 Seasoned 2 0.097 0.296 Unseasoned 2 0.038 0.192

City Level Industry Specialist

Industry Specialist 1 0.281 0.449 Industry Specialist 2 0.264 0.441 Seasoned 1 0.170 0.375 Unseasoned 1 0.112 0.315 Seasoned 2 0.149 0.356 Unseasoned 2 0.115 0.319

Joint National and City Level Industry Specialist

Industry Specialist 1 0.081 0.272 Industry Specialist 2 0.079 0.269 Seasoned 1 0.043 0.203 Unseasoned 1 0.038 0.191 Seasoned 2 0.048 0.269 Unseasoned 2 0.031 0.174

Table 4, panels A-1 and A-2 report the number of national level industry specialist audit firms in the German audit market when the market share-based definitions 1 and 2 are followed respectively.12 In panels B-1 and B-2, a total of 99 versus 117 industries with industry specialists can be identified over a thirteen-year period, with an annual average of 7.6 versus 9.0. Over this thirteen-year period, KPMG is most regularly classified as national industry specialist with an average of 2.9 versus 3.5 industry-year combinations. PWC (1.9 versus 3.0) and EY (1.6 versus 2.2) follow. Striking is that Deloitte is not classified as industry specialist with an average of 0.1 versus 0.2 industry-year combinations.13 The other audit firms can sometimes be classified as industry specialists with 1.1 versus 2.2 industry-year combinations. Overall, the industry specialist classifications are quite evenly distributed among the audit firms, suggesting that the German market is highly competitive.14

12_{These definitions are introduced by Reichelt and Wang (2010), also see section 3.1.2.}

13_{This demonstrates the underrepresentation of Deloitte in the German audit market for the period 1999-2011}

compared to other settings, meaning that Deloitte has more characteristics of a second tier auditor in Germany.

14_{For brevity, I do not include a distribution of seasoned and unseasoned industry specialist auditors as introduced}

by Gaver and Utke (2019), since their seasoning definitions are partly based on the market share-based definitions from Reichelt and Wang (2010). This means that the distributions partly overlap and are quite similar.

(23)

23 TABLE 4

Distribution of Auditor Industry Specialization

Panel A: National industry specialists by auditor and year (N = 2,584)

A-1: National industry specialists by auditor and year – definition 1 Reichelt & Wang (2010)

Auditors/Years 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Average

KPMG 3 3 3 2 3 3 3 2 2 5 4 2 3 2.9

PWC 1 1 2 2 4 2 1 2 2 1 1 2 4 1.9

EY - - - 1 3 1 2 3 3 2 3 2 1 1.6

Deloitte - - - 1 0.1

Other audit firms - 1 1 1 - - 2 2 2 2 2 1 - 1.1

Total industries 21 22 19 22 23 23 21 20 22 23 21 23 20 21.5

Total national specialists 4 5 6 6 10 6 8 9 9 10 10 7 9 7.6

A-2: National industry specialists by auditor and year – definition 2 Reichelt & Wang (2010)

KPMG 6 5 3 1 3 4 2 2 3 5 4 3 4 3.5

PWC 3 4 5 3 3 1 - 2 4 3 3 3 5 3.0

EY - - - 2 4 3 2 4 2 3 3 3 2 2.2

Deloitte - - - 2 0.2

Other audit firms 3 4 5 2 1 1 2 2 3 2 2 1 - 2.2

Total national specialists 12 13 13 8 11 9 6 10 12 13 12 10 13 9.0

(24)

24 TABLE 4 – continued

Distribution of Auditor Industry Specialization

Panel B: City industry specialists by auditor and year (N = 2,584)

B-1: City industry specialists by auditor and year – definition 1 Reichelt & Wang (2010)

KPMG 7 8 6 6 9 11 9 8 7 8 8 8 5 7.7

PWC 7 7 6 7 9 7 4 4 3 4 6 10 11 6.5

EY 2 3 3 9 7 6 4 5 3 3 5 6 6 4.8

Deloitte - - - 1 - 1 2 2 1 1 1 4 3 1.2

Other audit offices 12 9 7 7 7 6 7 9 9 7 7 6 4 7.5

Total city specialists 28 27 22 30 32 31 26 28 23 23 27 34 33 28.0

B-2: City industry specialists by auditor and year – definition 2 Reichelt & Wang (2010)

KPMG 9 9 8 6 10 12 10 10 9 10 8 8 8 9.0

PWC 6 6 6 8 5 6 5 5 4 4 6 9 7 5.9

EY 3 3 3 6 6 5 3 6 4 6 6 5 4 4.6

Deloitte - 1 - 3 1 2 2 1 2 3 3 6 5 2.2

Other audit offices 18 13 13 10 10 13 9 10 12 8 6 7 7 10.5

(25)

25 Table 4, panels B-1 and B-2 show the number of city industry specialist offices when the market share-based definitions 1 and 2 from Reichelt and Wang (2010) are followed. Panel B-1 reports a total of 364 city industry specialists with an average of 28.0 specialists per year, versus a total of 419 city industry specialists with an average of 32.2 specialists per year in panel B-2. In comparison to the firm-industry-year combinations from panel A, it can be noted that there are more office-level industry specialists. This is likely because there are fewer auditors per two-digit SIC category at the city level, automatically leading to higher market shares for the incumbent audit offices (Reichelt & Wang, 2010). According to the distributions in panels B-1 and B-2, KPMG can be classified as city industry specialist most often with an average of 7.7 versus 9.0 city-industry-year combinations. PWC (6.5 versus 5.9), EY (4.8 versus 4.6) and Deloitte (1.2 versus 2.2) follow. On average, the other audit offices are classified as industry specialists in 7.5 versus 10.5 combinations per year.15

TABLE 5

Descriptive Statistics of Abnormal Discretionary Accruals and Control Variables Used in Multivariate Regression Analysis

(N = 2,584) Variables Mean Std. Dev. 25th %-tile Median 75th %-tile Min. Max.

Abnormal Discretionary Accruals

DACC -0.001 0.121 -0.053 0.003 0.052 -0.478 0.486 |ABSDACC| 0.082 0.089 0.023 0.052 0.102 0.000 0.486 Control Variables lnSIZE 5.158 2.134 3.650 4.867 6.338 0.817 11.523 ROA 0.046 0.179 0.006 0.059 0.112 -2.062 1.038 CFO 0.060 0.211 0.005 0.072 0.140 -2.192 2.554 LEV 0.211 0.281 0.030 0.164 0.310 0.000 6.471 LOSS 0.291 0.454 0.000 0.000 1.000 0.000 1.000 MB 2.224 4.921 1.041 1.612 2.679 -80.271 73.984 SGROWTH 0.125 1.493 -0.058 0.036 0.142 -1.000 59.249 lnTENURE 1.015 0.766 0.000 1.099 1.609 0.000 2.565 LIT 0.125 0.331 0.000 0.000 0.000 0.000 1.000 IFRS 0.741 0.439 0.000 1.000 1.000 0.000 1.000 BIG3 0.527 0.499 0.000 1.000 1.000 0.000 1.000

15_{For panel B-1 (definition 1), 18 non-Big 4 auditors are classified as city industry specialists versus 27 for panel}

B-2 (definition 2). Most of the observations of other industry specialist audit offices can be assigned to BDO (25% versus 23% of 127 versus 157 observations). For the period 1999-2001, Arthur Andersen is also represented in the sample (12% and 11% of total industry specialist combinations in the ‘other audit offices’ category). The remaining 63% versus 66% is distributed among 16 smaller German audit firms.

(26)

26 Table 5 presents the descriptive statistics of the dependent abnormal discretionary accruals variable and the control variables used in the multivariate regression models (5) and (6).16_The values of my variables are consistent with prior literature (e.g. Reichelt & Wang, 2010; Minutti-Meza, 2013; Porumb et al., 2018; Gaver & Utke, 2019), except for the (BIG3) and (lnTENURE) variables. Compared to other settings, it seems that fewer audit clients are audited by Big 3 auditor in Germany.17 Furthermore, the average auditor tenure is shorter for the German setting in comparison to the average tenure found in studies conducted at a cross-country level (e.g. Gaver & Utke, 2019).18

The Pearson correlation coefficients are provided in table 6. As expected, the absolute value of abnormal discretionary accruals (ABSDACC) negatively correlates with the industry specialist variables19 and the control variables in most predicted directions.20 However, the unseasoned industry specialist variables are insignificant, which suggests that unseasoned industry specialists are not associated with the amount of accrual-based earnings management. Furthermore, the control variables (IFRS) and (LIT) are insignificant.

4.2 Results

The results of multivariate fixed effects regression models (5) and (6) are presented in table 7. Each of the three hypotheses introduced in section 2.3 is tested four times, following the two market share-based definitions from Reichelt & Wang (2010) used in model (5) and the two seasoning definitions from Gaver & Utke (2019) used in model (6).

16_{Please refer to table 1 for the variable definitions.}

17_{As explained in section 3.2, I exclude Deloitte because of its dissimilarities compared to other Big N auditors}

KPMG, PWC and EY. In my sample, 52,7% of the client firms are audited by a Big 3 auditor (or 59,0% when Deloitte also is included, as is the case in most other papers). Other studies report percentages of 70% or higher (e.g. Reichelt & Wang, 2010; Minutti-Meza, 2013).

18_{It should be noted that, due to the limited availability of data, I determine tenure based on the number of}

consecutive years that the audit client has hired the same auditor within the sample period of 1999 until 2011. This means that I start to count the years of tenure from 1999. However, I acknowledge that it might be possible that audit clients already hired the auditor before the start of the sample in 1999, which means that auditor tenure might actually be longer than calculated.

19_{The levels of the industry specialist variables – i.e. firm (national) level, office (city) level and joint (firm and}

office) level are abbreviated in the correlation matrix. The first letter(s) of the industry specialist variables ISP and SEASONED/UNSEASONED refer to the level of analysis: N represents the firm level, C represents the office level and NC represents the joint-level.

20_{As mentioned in section 3.1.2 and section 3.2., I measure firm-level, office-level and joint-level industry}

specialization using two market share-based definitions (Reichelt & Wang, 2010) in regression model (5). In regression model (6), these definitions are split up into two new variables: seasoned and unseasoned (Gaver & Utke, 2019). For brevity, I computed an average correlation in which the two market share-based definitions used in model (5) are combined. This method is also applied for the two seasoning definitions used in regression model (6).This is because the correlation coefficients of the two market share-based and seasoning-based definitions are similar to each other when they are separately analyzed in a Pearson correlation analysis.

(27)

27 TABLE 6

Pearson Correlation Matrix

Variables |ABSDACC| NISP CISP NCISP NSEASONED NUNSEASONED CSEASONED CUNSEASONED

|ABSDACC| 1.000 NISP -0.086*** _1.000 CISP -0.089*** _0.338*** _1.000 NCISP -0.071*** _0.791*** _0.512*** _1.000 NSEASONED -0.100*** _0.828*** _0.307*** _0.699*** _1.000 NUNSEASONED -0.003 0.545*** _0.143*** _0.366*** _-0.019 _1.000 CSEASONED -0.107*** _0.285*** _0.725*** _0.423*** _0.335*** _0.008 _1.000 CUNSEASONED -0.001 0.148*** _0.581*** _0.236*** _0.046** _0.196*** _-0.139*** _1.000 NCSEASONED -0.075*** _0.597*** _0.397*** _0.768*** _0.718*** _-0.009 _0.526*** _-0.052*** NCUNSEASONED -0.023 0.531*** _0.331*** _0.654*** _0.243*** _0.583*** _0.039* _0.430*** lnSIZE -0.213*** _0.292*** _0.374*** _0.292*** _0.266*** _0.122*** _0.311*** _0.172*** ROA -0.134*** _0.049** _0.089*** _0.055*** _0.036* _0.033* _0.072*** _0.043** CFO -0.111*** _0.032 _0.079*** _0.043** _0.018 _0.031 _0.054*** _0.050** LEV 0.105*** _0.014 _0.022 _0.010 _0.008 _0.013 _0.025 _0.002 LOSS 0.126*** _-0.098*** _-0.109*** _-0.094*** _-0.091*** _-0.040** _-0.084*** _-0.058*** MB 0.040** _-0.019 _-0.022 _-0.018 _-0.011 _-0.018 _-0.013 _-0.016 SGROWTH 0.089*** _0.007 _-0.024 _-0.013 _0.015 _-0.010 _-0.020 _-0.011 lnTENURE -0.149*** _0.096*** _0.083*** _0.093*** _0.170*** _-0.084*** _0.221*** _-0.141*** LIT -0.018 0.155*** _0.040** _0.116*** _0.169*** _0.022 _0.067*** _-0.022 IFRS 0.014 0.051** _-0.009 _0.041** _0.083*** _-0.033* _0.037* _-0.056*** BIG3 -0.057*** _0.307*** _0.301*** _0.218*** _0.244*** _0.182*** _0.257*** _0.129*** *** p<0.01, ** p<0.05, * p<0.1 (continued)

(28)

28 TABLE 6 – continued

Pearson Correlation Matrix

Variables NCSEASONED NCUNSEASONED lnSIZE ROA CFO LEV LOSS MB

NCSEASONED 1.000 NCUNSEASONED 0.019*** _1.000 lnSIZE 0.236*** _0.178*** _1.000 ROA 0.040** _0.039** _0.241*** _1.000 CFO 0.029 0.033* _0.167*** _0.638*** _1.000 LEV -0.001 0.017 0.086*** _-0.136*** _-0.179*** _1.000 LOSS -0.075*** _-0.059*** _-0.287*** _-0.593*** _-0.358*** _0.071*** _1.000 MB -0.008 -0.017 0.001 0.107*** _0.010*** _-0.066*** _0.020 _1.000 SGROWTH -0.012 -0.006 -0.025 0.034* _-0.010 _0.021 _-0.014 _0.033* lnTENURE 0.138*** _-0.019 _0.199*** _0.144*** _0.010*** _0.031 _-0.146*** _-0.033* LIT 0.120*** _0.039** _0.063*** _-0.094*** _-0.105*** _0.037* _0.053*** _0.026 IFRS 0.049** _0.007 _0.042** _-0.041** _-0.032 _-0.036* _0.025 _-0.010 BIG3 0.153*** _0.160*** _0.327*** _0.031 _0.078*** _-0.024 _-0.058*** _-0.062***

Variables SGROWTH lnTENURE LIT IFRS BIG3

SGROWTH 1.000 lnTENURE -0.032* _1.000 LIT 0.030 0.020 1.000 IFRS 0.011 0.276*** _0.088*** _1.000 BIG3 -0.006 0.112*** _0.084*** _0.003 _1.000 *** p<0.01, ** p<0.05, * p<0.1

(29)

29 4.2.1 Firm-Level Industry Specialization (H1)

To test whether industry specialized audit firms have a decreasing impact on the amount of abnormal discretionary accruals of the audit client and thus provide higher quality audits (H1), I run multivariate regressions with year fixed effects. The results are presented in table 7, panel A. If industry specialist auditors provide higher quality audits, the industry specialist coefficients will be negative and significant.

I first examine the market share-based definitions from Reichelt and Wang (2010). Although the industry specialization coefficients in model (5) are negative, they are insignificant (p > 0.1). This means that the first hypothesis is not supported when only market share-based proxies are used. These results are inconsistent with Balsam et al. (2003) and Reichelt and Wang (2010), who found a significant and negative effect. However, my findings are consistent with results found by Minutti-Meza (2013), Audousset-Coulier et al. (2016) and Gaver and Utke (2019), supporting their argument that pure market share-based proxies are a noisy measure of industry specialization and that the results are extremely sensitive to the independent variable used to measure industry specialization.21

Panel A, model (6) presents the results of the ‘seasoning process’ of auditor industry specialization at the firm level. If seasoned industry specialist audit firms provide higher audit quality than unseasoned industry specialists, the coefficient of seasoned auditors is predicted to be negative and to exceed the unseasoned auditors coefficient (Gaver & Utke, 2019). Model (6) shows that this is the case. However, only the sign for the second seasoning definition is significant (β = -0.011, p < 0.05). In line with Gaver and Utke’s (2019) findings, I find that seasoned industry specialist audit firms provide higher audit quality than unseasoned specialists when the second seasoning definition is followed. This partly supports my first hypothesis and implies that auditors might undergo a ‘seasoning process’ before they can be classified as experienced industry specialists who truly enhance audit quality. I also find that the control variables (lnSIZE), (LEV), (MB), (SGROWTH) and (lnTENURE) are significant at the (p < 0.01) level.

21_{For instance, Audousset-Coulier (2016) tested 30 abnormal accruals models, all based on different market}

share-based approaches for industry specialization. They conclude that in only five instances the expected negative and significant association between industry specialization and abnormal accruals can be found. Furthermore, they find in 21 of the instances that the pure market share based measure leads to a significant positive association between industry specialization and the amount of discretionary accruals. This contradicts the predictions from Balsam et al. (2003) and Reichelt & Wang (2010). Thus, results of industry specialist research should be interpreted with caution when only market share-based proxies are used.

(30)

30 4.2.2 Office-Level Industry Specialization (H2)

As mentioned in section 2.3, I expect that the amount of abnormal accruals will be lower for audit clients who hire industry specialist audit offices to audit their financial statements (H2). As shown in table 7, panel B, all coefficients for market share-based and seasoning-based industry specialization are negative. The market share-based definitions presented in model (5) are insignificant22, as well as the second seasoning definition in model (6) (β = -0.005, p > 0.1). However, when the first definition of seasoning is followed, the negative association between seasoned industry specialist auditors and the amount of abnormal accruals is significant (β = -0.007, p < 0.1). This means that the second hypothesis is partly supported, implying that audit offices experience a development process before they can be classified as audit quality enhancing industry specialists.

4.2.3 Joint-Level Industry Specialization (H3)

My final expectation is that joint-level industry expertise at both the firm (i.e. national) and the office level (i.e. city) will lead to a lower amount of abnormal accruals (H3). The results of this prediction are presented in table 7, panel C. Although most coefficients of market share-based industry specialization and seasoning are negative as predicted, they all are insignificant (p > 0.1). However, I find a positive and significant association between unseasoned industry specialist auditors and the amount of abnormal accruals when the first seasoning definition is followed (β = 0.020, p < 0.01). This implies that, at the joint level, unseasoned industry specialists increase the amount of abnormal accruals and thus have a decreasing impact on audit quality in their first year as industry leader. Because the market share-based coefficients in model (5) and the seasoned coefficients of industry specialization in model (6) are insignificant, however, hypothesis 3 is not supported.

22_{Again, the insignificance of the pure market share-based proxies is consistent with the arguments from}

Minutti-Meza (2013), Audousset-Coulier et al. (2016) and Gaver and Utke (2019). For a more detailed explanation of these arguments, please refer to section 3.1.2.

The impact of firm-level, office-level and joint-level industry specialization on audit quality: Evidence from the German audit market