Validity of audit quality proxies: The
impact of audit firm tenure on audit
quality in the US
Name: Veerman, Timothy Benjamin Student number: 11089881
Thesis supervisor: Dr. S.W. Bissessur Date: 10 January 2018
Word count: 16704
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam
Statement of Originality
This document is written by student Tim Veerman who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
Existing literature employs a diversity of audit quality proxies. However availability of research on whether these proxies show similarities and construct validity is scarce. Using a US dataset for the period 2000-2014 the construct validity of four audit quality proxies is tested, discretionary accruals, accrual quality, just meet or beat and auditor size respectively. A construct validity model based on the research by Audousset-Coulier et al. (2016) is used to perform a series of test. By analyzing the correlation between each proxy the internal association is determined. The relationship between auditor tenure and audit quality is used to perform a regression model for each proxy, the tenure coefficient is used to determine whether the different proxies provide coefficient which are similar in direction. In order to determine the strength and significance of the similarities between the audit quality proxies the results are testing using a Chi-squared test.
The findings show a lack of construct validity between the different audit quality proxies. The Chi-squared test shows that for four out of six pairs that the results are significantly different from each other. A series of robustness test prove that the results are robust for bias and model design. Interpretation of the results point to the fact that, as the proxies as significantly different, the choice of audit quality proxy has a great impact on the results of research. The results provide ground for future research due to the fact that on one hand not all existing proxies have been tested and on the other hand due to the fact that new models might be needed to provide more accurate results.
Contents
1 Introduction... 6
2 Literature review and hypothesis development... 9
2.1 Auditor tenure and auditor independence... 9
2.2 Audit quality... 10
2.3 Proxies for audit quality ... 12
2.3.1 Output measures... 13
2.3.2 Input measures ... 14
2.4 Auditor tenure versus audit quality, evidence from existing literature... 15
2.4.1 Results from existing literature using output measures... 15
2.4.2 Results from existing literature using input measures ... 18
2.5 Hypothesis development... 19
3 Research design... 23
3.1 Methodology ... 23
3.2 Computing and designation of audit quality proxies ... 24
3.3 Computing of internal association... 25
3.3.1 Discretionary accruals proxy (DA)... 25
3.3.2 Accrual quality proxy (AccQ)... 26
3.3.3 Earnings surprises or market reaction proxy (MorB)... 27
3.3.4 Auditor size or Big-N proxy (BigN) ... 27
3.4 Computing of external association – The impact of auditor tenure on audit quality .... 28
3.4.1 Regression model... 28
3.4.2 Chi-squared test... 29
4 Empirical results... 31
4.1 Sample development... 31
4.1.1 Compustat sample selection procedures ... 31
4.2 Descriptive statistics ... 34
4.3 Internal association for audit quality construct validity... 35
4.4 External association for audit quality construct validity... 38
4.4.1 Regression model for audit quality and auditor tenure ... 38
4.4.2 Test for significant differences between proxies using Chi-squared tests... 40
4.4.3 Robustness tests ... 41
5 Conclusion ... 45
References... 47
Appendices... 51
Appendix A – Compustat and IBES output screens... 52
1
Introduction
The relationship between auditor tenure and audit quality has been a field of interest of, amongst others, state legislators, cross-border legislators, researchers and regulators for many years1. During the last decades the field has been evolving from mandatory audit partner rotation to mandatory audit firm rotation whereas multiple countries have incorporated audit partner rotation in to their laws but only a few have incorporated audit firm rotation (EY, 2015). Events such as the recent financial crisis but also well-known audit failures (i.e. Enron) triggered
governments and regulators to, once again, start the debate on whether auditor tenure impacts audit quality.
Based on the research from Myers et al. (2003) auditor tenure can be described as the length of the relationship between the auditor and its client. In general, the interest in whether auditor tenure has an effect on audit quality can be derived from the assumption that a more lengthy relationship might negatively impact auditor independence (Read & Yezegel, 2016). Independence is expected to be a fundamental principle as it relates to the objectivity which might be impaired by a lack of independence. This is adhered by the ACIPA as auditors should not only “commit themselves to honor the public trust” (AICPA, 2016) but a member should also be “independent in fact and appearance”2.
The most commonly used description of audit quality can be found in the research as performed by DeAngelo (1981) who states that audit quality can be defined as the combined probability that the audit both detects and reports a breach in the client’s accounting system. In practice measuring audit quality based on said description seems infeasable, DeFond and Zhang (2014) state that this is due to the fact that the amount of assurance is not observable. By using proxies3to measure audit quality this problem can be solved. In addition the review by DeFond and Zhang (2014) shows that prior studies use a variaty of proxies, ranging from output based proxies to material misstatement proxies and auditor communications proxies. Furthermore DeFond and Zhang (2014) show that each proxy has its own strenghts and weaknesses, infact the
1A distinction is made between audit partner tenure and audit firm tenure, audit partner tenure is focused on the length of partner involvement while audit firm tenure is focused on actual audit firm tenure. In this study the term “auditor tenure” refers to both tenure types.
2This implies that there is a difference between being independent and being perceived as independent. A review study performed by Ewalt-Knauer et al. (2013) finds that prior research is mainly focussed on independence in appearance and is not able to generalize their findings into independence in fact which is related to the fact that factual independence can only be measured by a lack of actual independence (and thus not the factual independence on its own).
most strong proxies might also be the proxies with the worst weaknesses, thus implying that no single proxy is perfect in measuring audit quality.
Results on the effects of auditor tenure onaudit quality are inconsistent. Results are ranging from finding no significant impact of auditor tenure on audit quality (Myers et al., 2003) to a lower audit quality during the first years of an audit (Geiger & Raghunandan, 2002) and to finding that audit quality is not only lower during the first years of an audit but again after a certain period of time (Nashwa, 2004). Aforementioned studies use respectively discretionary accruals, going-concern opinions and audit failures as proxies to measure audit quality. Prior studies show, using a variaty of proxies to measure audit quality, that the results on the impact of auditor tenure on audit quality are inconsistent. Hence it is reasonable to assume that the choice of proxy has an impact on the results.
In addition to existing literature which shows inconsistent results across different studies, DeFond and Zhang (2014) note that some proxies are rarely used due to, for instance, the fact that they prove to be unfit for large-sample studies. The finding by DeFond and Zhang (2014) provides reasonable doubt on whether the choice of proxy is dependent of the type of research conducted. Or, that it might be possible that, eventhough all proxies to measure audit quality have the same objective4, not all proxies provide similar results when used on a singular dataset. This implies that existing proxies might not all be valid and provide similar, hence reliable, results. This study will therefore use the following research question:
RQ1: How valid are existing audit quality proxies when measuring the impact of auditor tenure in the US?
Using a set of different proxies used to measure audit qualitythis study will present empirical evidence on whether existing audit quality proxies show construct validity. Research will be conducted using data from US companies from 2000 up untill 2014 (15 years). Four different audit quality proxies will be computed in order to be able to test their internal and external construct validity. This study will follow a study performed by Audousset-Coulier et al. (2016) on whether Industry Specialist Proxy (hereafter 'ISP’) measures are reliable and valid. The Audousset-Coulier et al. (2016) study functions as a foundation for this research and will be reperformed with regard to the field of audit quality proxies.
Deviations from the study by Audousset-Coulier et al. (2016) are, first of all, that this study focusses on audit quality proxies instead of ISP measures, secondly this study will not use an
assigment approach to determine firms which delivery high quality audits in a certain year5. And third, this study will use the relationship between auditor tenure and audit quality to demonstrate whether audit quality proxies are valid instead of using audit fees and earnings quality.
Relevance can be obtained through the fact that prior studies might be using several combined audit quality proxies to test for robustness, but no prior study is focused on the reliability and validity of different audit quality proxies. Furthermore, prior literature such as the review performed by DeFond and Zhang (2014), provide evidence showing that a variety of proxies is used in existing literature. In addition existing literature shows that each proxy has strengths and weaknesses and that researchers should be (very) careful in their decision on what audit quality proxies to use but existing literature lacks to show whether existing proxies are valid. The fact that this study is focused on the construct validity of audit quality proxies makes it unique and relevant. In addition relevance can be obtained through the fact that a variety of proxies will be used on a singular data sample of US companies for a period of 15 years, prior studies do use multiple proxies but solely to test for robustness of results instead of actually comparing the results.
This study is divided into five sections, the literature review and hypothesis development will be presented in the second section followed by the research design description in the third section. The fourth and fifth section will respectively present the empirical results of the research and the conclusion.
5The study by Audousset-Coulier et al. (2016) used different assignment approaches to determine whether, per year, whether an auditor is an ISP. With regard to this study the assignment of the level of quality to an audit firm in a certain year is deemed irrelevant as the objective is to determine audit quality per observation by using different proxies.
2 Literature review and hypothesis development
2.1 Auditor tenure and auditor independenceInterest in the field of auditor tenure and its impact on audit quality dates back to as early as the 1920’s where Canada implemented mandatory firm rotation6up to the 1970’s where Italy implemented mandatory firm rotation7 and the U.S. implemented mandatory audit partner rotation8(EY, 2015). More recent events, such as the financial crisis, put the interest in auditor tenure back on the political agenda once again. Both the PCAOB (2011) and the European Commission (2010; 2013) showed interest in the field as they believe it might be used to restore auditor independence.
In the period before the recent financial crisis a series of audit failures9from the late 90’s and onwards lead the US Congress to implement a set of rules by accepting the Sarbanes-Oxley act on July 30 2002. Amongst others both section 201 and 203 are related to auditor independence. Section 201 contains a list of services prohibited to be performed by auditors (in combination with other services) whereas section 203 provices rules related to mandatory audit partner rotation. In order to enhance independence the US Congress, once again, implemented a rule which states that audit partners need to rotate every five years and should adhere a five year cooling down period.
The most commonly used definition of auditor tenure can be obtained from a study performed by Meyers et al. (2003). Their study states that auditor tenure can be best described as the length of the relationship between the audit firm and its client. In their review DeFond and Zhang (2014) write that auditor independence might be impaired by the length of auditor tenure as it points toward familiarity. Auditor independence is recognized by the AICPA (2016) in their Code of Conduct by stating that the auditor should be independent both in fact and in appearance. In addition the Code of Conduct clearly states the familiarity threat (the threat that the auditor becomes too familiar with its client and thus impairing indepence/objectivity).The familiarity threat is clarified by the AICPA by stating that a long association of senior personnel is part of that specific threat. The relevance of auditor independence is endorsed by the SEC (2000) which emphasizes that the role of auditors is of great public importance as investors should be able to rely on audited financial statements.
6Canada implemented mandatory audit firm rotation in 1923 and kept it in place up until 1991. 7Italy implemented mandatory audit firm rotation in 1975.
8The U.S. implemented mandatory audit partner rotation in 1978. 9Such as Enron, WorldCom, Tyco and Ahold.
Elaborating on the aforementioned review of DeFond and Zhang (2014), the risk of (excessive) familiarity might impact independence as it might lead to a situation where the auditor might place the clients interest above the public importance and by using a lower level of details (Arel, Brody, & Pany, 2005). In addition, the paper by Arel, Brody and Pany (2005) shows a situation of staleness as, within auditor tenure, the risk of entering a state of repetition is present. On recurring engagements the auditor uses prior year schedules and experiences and might even be relying on those especially when it relates to a self-review. Auditor tenure might also provide benefits, existing literature shows that audit quality is lower during the first years of an audit engagement (Nashwa, 2004; Geiger & Raghunandan, 2002), which links to another finding in existing literature which shows that during the first years of an audit engagement the auditor needs to obtain client-specific knowledge (GAO, 2003) a statement which is endorsed by Big-4 audit firms (PwC, 2013).
2.2 Audit quality
A comprehensive definition most commonly found in existing literature is that audit quality is the combined probability that the auditor will both find and report a breach in the accounting system of its client (DeAngelo, 1981). Referring to the SEC (2000), which states that “investors must be able to rely on issuers’ financial statements”, audit quality can also be described as the level in which investors are able to rely on financial statements. Combining the two definitions the level of reliance investors can obtain from financial statement is dependent of the ability of the auditor to both discover and report breaches in the reporting system. Citing the
Government Accountability Office (GAO, 2003) audit quality can be defined as being performed ‘‘in accordance with Generally Accepted Auditing Standards (GAAS) to provide reasonable assurance that the audited financial statements and related disclosures are (1) presented in accordance with Generally Accepted Accounting Principles (GAAP), and (2) are not materially misstated whether due to errors or fraud.’’, which adds to the definition that audit quality is also related to whether an audit is performed in accordance with relevant laws and regulations. The definitions from both the SEC (2000) and GAO (2003) are, in general, more related to financial reporting quality and thus the output whereas the definition used by DeAngelo (1981) is more related to the quality of the actual audit and thus the process.
A review performed by Francis (2004) on audit quality summarizes existing literature in the field of audit quality. Francis (2004) shows that there are different variables driving or at least influencing audit quality. Earlier research tends to be more focussed on auditor differentiation variables and uses, in example, firm size as a proxy for audit quality. Support can be found in the fact that a large firm does not rely on a single (or a few) client(s) for its existence and that the reputation and lititation risks are therefore higher for large firms (DeAngelo, 1981). An paper by Francis and Wilson (1988) provides evidence for the DeAngelo (1981) statement by showing that large audit firms have established brand names and thus have incentives to protect their reputation. Furthermore earlier research shows that larger audit firms have a fee premium when compared to smaller audit firms (Simunic, 1980). In turn, a more recent study by Francis (2004) concludes that higher audit fees equal higher audit quality by either implying more effort (more audit hours) or higher billing rates.
Higher billing rates are also related to industry specialization, another variable which might impact audit quality, evidence shows a fee premium for the industry specialist in the US (Francis et al., 2005). Existing literature shows evidence that audit quality is higher for audits performed by industry specialists (Balsam et al., 2003). The earlier discussed auditor tenure, described as the length of the relationship between the auditor (both firm and partner) and it’s client (Myers et al., 2003), is another variable which might influence audit quality. As discussed in section 2.1 auditor tenure research is focussed on whether tenure affects auditor independence and thereby audit quality. Other variables commonly used are, amongst others, non-audit fees and audit committees. The audit fee variable investigates whether the amount of non-audit fees, within a specific engagement, has an impact on audit quality. The review on existing literature by Francis (2004) shows that results are inconclusive but that non-audit fee might impair audit quality. The Sarbanes-Oxley act of 2002, specifically section 201, provides a list of non-audit services prohibited to be performed by auditors in combination with other services, therefore this variable is now less relevant in research related to the US market (i.e. services which are deemed to impair independence/objectivity are prohibites and thus reducing the relationship of non-audit fees and audit quality). Audit committee research is based on whether the audit committee is independent, evidence from existing literature shows that auditors are more likely to issue going-concern opinions in the presence of independent audit committees (Carcello & Neal, 2003) and that auditors are more likely to detect breaches in financial reporting when the audit committees are more independent (Dechow et al., 1996).
DeFond and Zhang (2014) describe the relationship between audit quality and financial reporting quality. First of all they describe that within financial reporting quality the quality of the
audit acts as a component due to the fact that a high quality audit implies higher quality financial reporting. Secondly, audit quality is, according to DeFond and Zhang (2014) dependent of the quality of unaudited financial statements which are derived from the clients’ financial reporting system. In turn, the quality of unaudited financial statement is dependent of the quality of the financial reporting system. In addition audit quality is dependent of other variables such the representatives of the client, who are, according to DeFond and Zhang (2014) more likely to choose the quality of the financial reporting system which matches the expected quality delivered by the auditor.
2.3 Proxies for audit quality
In addition to variables which influence audit quality there are also indicators or measures (hereafter ‘proxies’) to determine the level of audit quality, these proxies are used to measure the audit quality for a given firm in a given year. Papers by both DeFond and Zhang (2014) and Knechel et al. (2013) have respectively reviewed existing literature using archival auditing research and academic literature. DeFond and Zhang (2014) use two main categories to range existing proxies, input measures and output measures, similar categories are used by Knechel et al. (2013) but two additional categories are added, processes and context, which are, in some cases, used as subcategories of the general categories as introduced by DeFond and Zhang (2014).
Referring to the introduction, audit quality is rather difficult (if not impossible) to measure as the actual assurance delivered10is unobservable (DeFond and Zhang, 2014), therefore proxies for audit outcomes are used by researcher as indirect measurement for audit quality (Knechel et al., 2013). The review by Knechel et al. (2013) shows that most proxies are related to identifying what is not a high quality audit rather than identifying what is a high quality audit. In general output measures are used in research conducted on effect of the supply-side whereas input measures are, in general, used in research on effect of the demand-side of the spectrum. Variables which might influence audit quality on the supply-side are auditor tenure, firm size and industry specialization whereas audit fees and audit committee independence are variables which might influence audit quality on the demand-side. Output measures are focused on the outputs of the audit process such as going-concern opinions and financial reporting quality whereas input measures are focused on the inputs such as firm size and audit fees.
10The lack of audit quality can be measured by assessing audit failures, the presence of audit quality is another story and is rather difficult to capture.
2.3.1 Output measures
Within the output measures DeFond and Zhang (2014) determine four subcategories, material misstatement, auditor communications, financial reporting quality and perception based measures.
2.3.1.1 Material misstatements
According to DeFond and Zhang (2014) existing literature shows two most commonly used measures for material misstatements, restatements and Accounting and Auditing Enforcement Releases (also known as ‘AAER’s’ which are released by the SEC when violations occur). The frequency of the usage of these measures is relatively low, mostly due to the simple fact that restatements, and AAER’s in particular, are rare. The measures are very direct, have a lower measurement error and are very strong in measuring audit failure and thus low audit quality. Main concern with material misstatement proxies is that the absence of restatements of AAER’s does not imply a high quality audit (DeFond and Zhang, 2014). Using restatements Stanley and DeZoort (2007) find that the presence of restatement is negatively related with auditor tenure and audit quality. Nashwa (2004) uses audit failures and includes AAER’s in his sample, he finds that failures are more likely to occur during the first years of an audit engagement and after 7 years.
2.3.1.2 Modified opinions
The most common auditor communications used as a proxy are going-concern opinions. According to DeFond and Zhang (2014) going-concern opinions show whether the auditor has any doubt on the ability that client will continue as a going-concern, managers will have incentives to pressure the auditor in not issueing a going-concern opinion. Using the going-concern opinion as an indicator of audit quality is a direct method with small measurement errors, the fact that an auditor did not issue a going-concern modified opinion when one was due is clearly evident of a low quality audit11. However, using the modified going-concern opinion has a clear disadvantage, it is not very sensible; It’s either a high or a low quality audit. Furthermore, auditors have an incentive to inappropriately issue a going-concern opinion to mitigate the risk of litigation (Kaplan & Williams, 2013). Between 40 and 50 percent of bankrupt companies in the US did not receive a modified going-concern opinion (defined as a Type II error) whereas non-bankrupt companies received a modified going-concern opinion in 80 to 90 percent of all cases as shown in the paper by Knechel et al. (2013).
11These type of proxies show similarities with the material misstatement type proxies, using a set of companies which went bankrupt the proxy implies that a lack of a modified going-concern opinion in the year(s) before the bankruptcy equals a low quality audit.
2.3.1.3 Financial reporting quality
The basis of financial reporting quality lies in the fact that a high quality audit should be able to prevent or discourage earning management (DeFond and Zhang, 2014). Financial reporting quality measures are less direct when compared to restatements, AAER’s and going-concern opinions. According to DeFond and Zhang (2014) advantages can be found in the fact that, as referred to in section 2.2, audit quality is a component of financial reporting quality and thus makes financial reporting quality measures well suited to measure audit quality. In addition these measures have the ability to detect earnings management which is within the allowed GAAP. The most obvious disadvantage is the fact that measurement errors are high12, therefore DeFond and Zhang (2014) point to the fact that caution should be applied when using financial reporting quality measures. According to Knechel et al. (2013) the evidence, in general, shows a negative relationship between these measures and auditor tenure while research performed by Myers et al. (2003) shows no significant impact of auditor tenure on audit quality.
2.3.1.4 Perception-based
Perception-based measures use the perceptions of investors, stock market reactions and the cost of capital to measure audit quality. Using perceptions makes these measures relatively indirect (DeFond and Zhang, 2014). Advantages can be found in the fact that measurement of audit quality is comprehensive and continuous as it uses additional dimensions of audit quality rather than actual outputs. In addition, the usage of the client market share measures provides a unique measure on the perception of audit committees. It shows the dismissal of auditors after, for example, an audit failure. The biggest disadvantage of these measures can not only be found in the fact that firm value is, not in the first place, determined by financial reporting quality but also due to the fact that the variation amongst these proxies and what they measure is large, therefore the disadvantages and advantages vary across the measures in this subcategory.
2.3.2 Input measures
Input measure can be categorized in two subcategories, auditor characteristics and auditor-client contracting features.
12Financial reporting quality measures are often using (complex) calculation methods, using these methods increases the risk of a measurement error.
2.3.2.1 Auditor characteristics
Auditor characteristics are mainly based on certain characteristics of the auditor13such as whether the firm is a big N firm or an industry specialist. According to DeFond and Zhang (2014) the most unique feature of these measures is the fact that they are not engagement specific14. The biggest advantages are the fact that auditor characteristic measures are linked to most of the other proxies and that the industry specialization measure provides a variation within the big N measure. The biggest disadvantage lies in the fact that there is no distinction made within audit firms and/or industry specialization.
2.3.2.2 Auditor-client contracting features
Measures using auditor-client contracting features include audit fees. Audit fees as an proxy are based on the fact that the audit fee measures the effort put into an audit. A distinguishing factor of audit fees is that they can be used to measure both supply and demand factors. The biggest advantage is that the measure is continuous and thus able to capture subtle changes according to DeFond and Zhang (2014). The biggest limitation is the fact that audit fees also include risk premiums, audit efficiency and commercial aspects and are therefore more influenced by other variables.
2.4 Auditor tenure versus audit quality, evidence from existing literature
Existing literature on the relationship between auditor tenure and audit quality is comprehensive. Over time the amount of research performed in the field is extensive, using different proxies to measure audit quality and different data samples from countries across the world. Using the categories as introduced by DeFond and Zhang (2014) results from existing literature will be reviewed. Table 1 shows a summary of the results per proxy and includes a short summary of advantages and disadvantages per proxy.
2.4.1 Results from existing literature using output measures
2.4.1.1 Material misstatements
In general two types of proxies are used, restatements and AAER’s, where restatements are commonly used in existing literature. Restatements are used in a study by Stanley and DeZoort (2007), using a sample of 191 companies in the US with restatements in the period 2000 up until
13The characteristics are mainly based on the audit firm instead of the auditor himself. 14And thus not measuring the quality of an single engagement but of an audit firm as a whole.
2004 they use auditor tenure, industry specialization and audit firm independence variables. Using a series of logistic regression models Stanley and DeZoort (2007) calculate whether the variables prove to be significant on whether restatatement occur. They find an inverse relationship between auditor tenure and restatement and thus no evidence that auditor tenure has a negative effect on audit quality. In constrast, a study performed by Nashwa (2004) using 90 companies in the US finds other evidence. Using three types of audit failure (litigation, AAER’s and bankruptcy) he finds that the likelyhood of audit failures is higher during the first three years of an audit engagement and from the seventh year onwards. An older research performed by Anderson and St. Pierre (1984) shows that auditors are more likely to make mistakes during the first years of an audit engagement.
2.4.1.2 Modified opinions
The modified opinion proxy uses the outcome of the audit, the auditors opinion, as a proxy to measure audit quality. The most commonly used proxy is that of the modified going-concern opinion. This proxy either verifies, for a sample of bankrupt companies, whether during the year(s) prior to bankruptcy the auditor issued a going-concern opinion or by verifying whether the auditor issued a modified going-concern opinion for financially distressed companies (not necessarily leading to bankruptcy).
Geiger and Raghunandan (2002) use the going-concern proxy in their study. Using a sample for the period 1996-1998 containing companies which entered into bankruptcy, Geiger and Raghunandan (2002) find that there is a positive relationship between auditor tenure and the likelihood of the auditor issuing a going concern opinion. Audit failures (i.e. the lack of the issuance of a going concern opinion where one would be warranted) are more likely to occur during the first years of auditor tenure. Furthermore, Geiger and Raghunandan (2002) find that auditors might be more influenced by their clients during the first years of the auditor tenure. Geiger and Raghunandan (2002) suggest that this might be due to the fact that the auditor has a learning curve which leads to more specific client knowledge which enables the auditor to better assess whether the entity is able to continue as a going concern.
A study performed on the Belgium market by Knechel and Vanstraelen (2007) finds a different result than the study as performed by Geiger and Raghunandan (2002). Knechel and Vanstraelen (2007) find no evidence for a decrease of auditor independence when the auditor tenure becomes more lengthy. Furthermore Knechel and Vanstraelen (2007) find no significant evidence that the auditor becomes more able in predicting bankruptcy when tenure increases.
A sample similar to the one used by Geiger and Raghunandan (2002) but more recent, is used by Read and Yezegel (2016). Using a sample of companies which entered into bankruptcy in the period 2002-2008 Read and Yezegel (2016) find a non-significant relationship between auditor tenure and the issuance of a going concern opinion for Big 4 firms. For non-Big 4 firms the results are slightly different, a positive significant relationship was found between auditor tenure and the likelihood of the issuance of a going concern opinion. In addition, Read and Yezegel (2016) find that the likelihood of the issuance of such an opinion is lower during the first years of auditor tenure for non-Big 4 firms. Furthermore Read and Yezegel (2016) find that the general likelihood increases up until the fifth year of the relationship between the auditor and its client, after the fifth year the likelihood remains stable and no evidence was found for a decrease of the likelihood during a more lengthy auditor tenure.
2.4.1.3 Financial reporting quality
Of all financial reporting quality proxies the (modified) Jones model using discretionary accruals is by far the most popular. Myers et al (2003) use discretionary accruals and the Jones model in their research, using sample of companies for the period 1988-2000 they find no evidence for a significant negative relationship between auditor tenure and audit quality and thus no evidence that a longer tenure negatively impact audit quality.
Closely related to the discretionary accrual proxy is the conservatism proxy, this proxy also uses earnings management and accruals to measure audit quality. Jenkins and Velury (2008) use conservatism in their study, using a sample of US companies for the period 1980-2004 they find that conservatism is lower during the intitial years of an engagement (short tenure), between short and medium tenure (four to eight years) conservatism will increase and during long tenure (nine or more years) they note no detoriation of the amount of conservatism.
The paper by DeFond and Zhang (2014) also shows that meet or beat earning targets are also popular proxies used within the financial reporting quality proxies. A study performed by Davis et al. (2009), who use this proxy, uses a sample of US companies for the period 1988-2006 and pays specific attention to the period before and after the implementation of the Sarbanes-Oxley act. Davis et al. (2009) find that for the pre-SOX era audit quality is lower during both short and long periods of auditor tenure, for the post-SOX era they find no evidence for lower audit quality for short auditor tenure but find that the auditor is more tolerant after a 15-years.
2.4.1.4 Perception-based
Most commonly used proxies are the costs of capital and changes in auditor market share (DeFond and Zhang, 2014). The cost of capital proxy is used by Azizkhani et al. (2013), using a sample of Australian companies for the period 1995-2005 they find a nonlinear relationship between auditor tenure and the cost of capital and thus they find no significant relation between auditor tenure and the cost of capital (and thus audit quality) before the implementation of audit partner rotation. In addition the research by Azizkhani et al. (2013) finds that audit partner rotation is associated with increased ex-ante costs of capital as they find a significant negative relationship between partner tenure and the ex-ante costs of capital.
Another perception based proxy is used by Ghosh & Moon (2005), using earning response coefficients as a proxy for investor perceptions of earning quality evidence is found for a postive relation between auditor tenure and investor perception of earning quality. Using a sample of US companies during the period 1990-2002 a basic regression framework is used to measure the effects of auditor tenure on the perception of investors on earning quality.
2.4.2 Results from existing literature using input measures
2.4.2.1 Auditor characteristics
A paper by DeAngelo (1981), who uses auditor size as an proxy for audit quality, finds that audit quality is dependent of audit firm size. DeAngelo (1981) does however not pay attention to the effect of auditor tenure. A study performed by Bryan and Reynolds (2016) uses the period 1988-2010 for US companies, using auditor size as a proxy they find evidence that auditor tenure has a negative relationship for small audit firms but no significant relationship for large (Big-N) audit firms. In addition Bryan and Reynold (2016) also find that industry specialization may act as a substitute for firm size, for industry specialists there is no significant negative relationship between auditor tenure and audit quality.
2.4.2.2 Auditor-client contracting features
Existing literature on the effect of auditor tenure on audit quality uses many proxies, however auditor-client contracting features are not commonly used as a proxy. Research performed by Simon and Francis (1988) shows evidence that during the initial years of an engagement audit fees are significantly lower, only after year three will the audit fees have been increased to normal levels. This might provide some insights in whether tenure affect audit quality, Francis (2004) shows that higher audit fees are related to higher quality audits as higher audit fees imply either more audit hours or higher billing rates (and thus a fee premium indicating a higher quality audit).
A meta-analysis has been performed by Hay (2013), this more recent research also shows that longer tenure is related to higher audit fees. In addition the meta-analysis also shows that existing literature finds that audit fees are positively related to internal control and goverance but also that fee premiums exist for industry specialist and Big-N auditors.
Summarizing, this proxy is rarely used in research on the effect of auditor tenure on audit quality, however, combining the existing literature on this relationship it can be stated that in general a more lengthy tenure leads to increased audit fees which, in turn, point to higher audit quality due to more audit hours or higher billing rates (Francis, 2004), industry specialization or auditor size (Hay, 2013).
2.5 Hypothesis development
Review papers on existing literature such as that of DeFond and Zhang (2014) but also that of Knechel et al. (2013) show that, within the research field of audit quality, a variety of proxies is used. As tabulated in table 1, existing literature shows varying results using different proxies to measure audit quality. Multiple studies use a combination of proxies, in general to test for robustness. Examples can be found in the research performed by Myers et al. (2003) who use discretionary accruals as a proxy for audit quality, during their sensitivity analysis they also use auditor specialization to determine robustness. Kaplan and Williams (2013) use both going concern and litigation risks as measure for audit quality and another sample can be found in the research of DeZoort and Stanley (2007) who use restatements as an audit proxy and test for sensitivity using, amongst others, non-audit fees and auditor specialization.
The main complication of the the forementioned research is that proxies and models are used to measure audit quality and that assumtions are made with regard to the reliability and especially validity of those proxies. As stated by Defond and Zhang (2014), all proxies have advantages and disadvantages but also that the strongest proxies might also witness the greatest weaknesses. This leads to the fact that researchers should pay great attention and care when making choices on which proxy to use but also that further research is needed on the reliability and validity of the existing proxies to measure audit quality.
Reliability can be described as the extent to which a measurement procedure provides consistent results on large scale test with repeated trials (Audousset-Coulier et al. , 2016). Conducting this research, audited indicators are used which are obtained directly from digitial
sources such as Computstat and IBES which ensure that the indicators used are reliable15. The concept of validity is described in detail by Carmines and Zeller (1979) who describe the concept of validity as the fact that measures (i.e. proxies) are measuring what they are supposed to measure. Audousset-Collier et al. (2016) state that the main element in verifying validity is construct validity, in addition a refferal is made to a study by Campbell and Fiske (1959) who describe two types of validity, convergent and discriminant validity. A situation of convergent validity exists when multiple measurement methods are used to measure the same trait (i.e. audit quality) those measures should converge on the same result which means that the results should be consistent across the methods. Discriminant validity is based on the fact that the same method of measurement is used to measure different traits leads to different results. Construct validity can be measured by first testing internal association and second by testing external association (Carmines and Zeller, 1979). Using internal association the correlation between a set of proxies will be tested, in case of validity it is expected that correlation exists between the measures. The concept of external association means that different proxies should behave similar in terms of direction, strenght and consistency (Audousset-Collier et al., 2016) with regard to relevant external variables. The relationship between auditor tenure and audit quality is used to measure external assocation. Based on the results from existing literature it is expected that different proxies lead to significantly different results, thus the following hypothesis is presented:
H1: The use of different audit quality proxies leads to significantly different results regarding the impact of auditor tenure on audit quality.
The following section will further develop the research method used to test the proposed hypothesis. The concept of construct validity, convergent validity and both internal and external association will also be explained in more detail.
15Compustat and IBES are commonly known databases provided by Wharton Research Data Services and are deemed reliable sources.
Table 1
Audit quality proxy comparatives and results from existing literature
Proxy category Advantages Disadvantages Existing literature Results
Output measures
Material misstatements - Very direct - Strong evidence of low audit quality Absence of restatements or AAER's does not imply high quality
Stanley & DeZoort (2007) + Inverse relationship between auditor tenure andrestatements
Nashwa (2004) - Higher audit failures in first 3-years and after 7 years
St. Pierre and Anderson
(1984) - More likely to make mistakes during initial years ofengagement
Modified opinions - Very direct
- Strong evidence of low audit quality - Small measurement errors - Not very sensible - Limited use is possible
Geiger and Raghunandan
(2002) + Positive relationship, likelihood of issuance is lowerduring initial years of engagement Knechel and Vanstraelen
(2007) +/- No significant evidence for decrease of independence
Read and Yezegel (2016) +/- Not significant for Big 4 but significant for non-Big 4. No evidence for decrease in likelihood for longer tenure (but lower likelihood during initial years of engagement) Financial reporting
quality - Well suited- Capable of showing subtle variations - Detect EM within GAAP allowance - High measurement error - Limited consensus on measurement
Myers et al. (2003) +/- No significant evidence for impaired audit quality
during longer tenure
Jenkins and Velury (2008) + Evidence shows that audit quality is lower during the
initial years, increases over time and does not detoriate
Davis et al. (2009) - Evidence shows lower audit quality during short and
long tenure (pre-SOX) and lower quality after 15-years (post-SOX)
Perception-based Comprehensive
and continuous No strongrelationship with audit quality
Azizkhani et al. (2013) - Negative nonlinear relationship between auditor tenure
and cost of capital
Ghosh and Moon (2005) + Longer tenure is associated with better earnings quality
Table 1 (continued)
Audit quality proxy comparatives and results from existing literature
Input measures
Auditor characteristics Linked to most
other proxies No distinctionwithin audit firms/ISP
Bryan and Reynolds (2016) - Negative relationship between auditor tenure and audit
quality for small audit firms
DeAngelo (1981) N/a Audit quality is dependent of firm size (but tenure is not
included in this paper)
Contracting features Can be used
for both supply and demand
Might be impacted by other variables
Hay (2013) + Increased tenure is related to higher audit fees which, in
turn, are related to higher audit quality
Simon and Francis (1988) N/a Initial years of engagement show lower audit fees which,
according to existing literature, points to lower audit quality
3 Research design
3.1 MethodologyThe research is conducted using archival research as it enables the usage of a large data sample covering multiple firms, years and variables (i.e. Big Data). Empirical research is performed using the concept of construct validity, in particular by conducting research on the validity of proxies used in existing literature to measure audit quality. This study will, for most parts, follow the research conducted by Audousset-Coulier et al. (2016). Following the principles of construct validity, Audousset-Coulier et al. (2016) use the different measures available to determine the ISP and conduct research on whether these measures provide significantly different results. Using correlation between the different ISP measures the coefficients are compared on, mostly, direction. In addition two regression models are used, an audit fee and earning quality regression model. These models are used to compute whether the ISP variable has a significant relationship with both audit fees and earning quality, this is performed for each ISP measure. By testing comparability of the regression coefficient for ISP by using a Chi-squared test to determine the significance.
This study will deviate from the method used by Audousset-Coulier et al. (2016), in particular due to fact that this study will not designate a level of audit quality to certain audit firms in certain years. The designation of a level of audit quality is not deemed relevant as the objective is not to determine which auditor delivers high audit quality in a given year but to determine whether existing proxies to measure audit quality are valid. Furthermore this study will use the relationship between auditor tenure and audit quality to measure external association and thus validity, whereas Audousset-Coulier et al. (2016) use audit fees and earning quality as models to measure the effect of different measures.
A two-step research design is used to investigate the research question and to verify the hypothesis’. In line with Audousset-Coulier et al. (2016) the construct validity will be tested using internal and external association (Carmines & Zeller, 1979). First audit quality is calculated for each sample (i.e. for each year per firm16), the results are compared using internal association to determine correlation. Secondly this study will investigate external association by investigating whether the proxies are convergent. Using the relationship between auditor tenure and audit quality this study will examine whether the results, using different audit quality proxies, converge
and are consistent. Using an audit quality regression model it will first be investigated whether auditor tenure is a significant variable and secondly whether results are consistent (on direction and strength) using a Chi-squared test.
The following sections will provide a detailed analysis of the models used to investigate the research question and to verify the hypothesis.
3.2 Computing and designation of audit quality proxies
The review by DeFond and Zhang (2014) is very extensive on summarizing existing proxies used to measure audit quality. By comparing the summarized audit quality measures which the review paper by Knechel et al. (2013) reliability and appropriateness of the most commonly used proxies is verified. Based on the studies by DeFond and Zhang (2014) and Knechel et al. (2013) this study will adopt four proxies as input to test construct validity.
The theory of construct validity and specifically convergent validity (Campbell and Fiske, 1959) states that different methods are used in order to measure the same same trait. The review paper by DeFond and Zhang (2014) shows two main categories of proxies, output- and input measures. Each category consists of subcategories, respectively four and two subsections. Within each subsection a selection of proxies is explained, which each have their own characteristics but are determined similar on the fact that their angle of measuring audit quality is the same. As stated above, the basic principle of convergent validity is that using different methods on the same trait shouls provide similar results. Due to this basic principle it is of particular importance that the adopted proxies to measure audit quality are indeed different measures. Due to this importance the following four proxies are adopted: discretionary accruals (output, financial reporting quality measure), accrual quality (output, financial reporting quality measure), just meet or beat or market reaction (output, perception-based measure) and auditor size or Big-N (input, auditor characteristics)17. The adopted proxies and their designation through-out this paper are tabulated in table 2.
Table 2
Adopted audit quality proxies and their designation
Proxy # Audit quality proxy name Designation
1 Discretionary accruals DA
2 Accrual quality AccQ
3 Just meet or Beat MorB
4 BigN auditor BigN
3.3 Computing of internal association
As explained, the research model used consist of two steps; Internal association and external association. First the selected proxies to measure audit quality are used to determine the audit quality per firm-year, the combined results enables this study to calculate the internal association. By calculating pairwise correlations between the proxies this study will first analyze whether there is any correlation between the audit quality proxies, using a p-value of 0,05 it will be determined whether the correlation is significant. Below the models used to calculate audit quality are explained per proxy, the designation (i.e. abbreviation) per proxy is shown for convenience.
3.3.1 Discretionary accruals proxy (DA)
The first proxy uses discretionary accruals as a proxy to determine audit quality. Existing literature shows the usage of multiple models however the most commonly used model is that of the modified Jones model as by Dechow et al. (1995). The modified Jones model is similar to the original Jones model but the firm specific parameter for change in revenues is complemented (or as Dechow et al. (1995) calls it, “adjusted”) by the change in accounts receivables. The modified Jones model is considered to be an appropriate model to determine discretionary accruals. Using the modified Jones model both the total accruals and discretionary accruals will be calculated, data will be obtained using the Compustat database from the Wharton Research Data Services (hereafter ‘WRDS’), more details on variables will be provided in section 4.1 where the dataset is described.
The basis of the modified Jones model is an regression analysis, the regression analysis will be used to determine the specific coefficient from the modified Jones model which can be found in the following formula:
= 1 + ∆ − ∆ + + Where: = = − 1 ∆ = ℎ ∆ = ℎ = =
A cross-sectional analysis will be performed to calculate the coefficients, these coefficients will be used per firm to calculate the firm specific discretionary accruals. The is done by using the following formula: = 1 + ∆ − ∆ + Where: = − = − 1 ∆ = ℎ ∆ = ℎ =
Next the discretionary accruals are calculated, this is performed by calculating the residue value by using the following formula:
= −
For this study all statistical tests will be performed using Stata. The above formulas show the background of the modified Jones model and how it is calculated. The usage of a tool such as Stata leads to a clear advantage as Stata is familiar with the modified Jones model, using codes the modified Jones model can be calculated. Stata will be used to create the modified Jones model variables per observation, to create a two-digit SIC number for each year in order to be able to perform the cross-sectional analysis, to create numeric variables for the regression residuals and for the coefficient estimated from the regression and to perform the actual modified Jones model for all combination of industry code and year.
3.3.2 Accrual quality proxy (AccQ)
The second proxy is accrual quality. The regression used in this study, is the model as implemented by McNichols (2002). The model is similar to the model by Dechow and Dichev (2002) but includes two control variables for revenue and property, plant and equipment. This proxy consists of a two-step approach, the first step is the following equation from McNichols (2002):
= + + + + ∆ + +∈
Where:
= Working capital accruals
= ℎ − 1
= ℎ + 1
∆ = ℎ
= ∈ =
The second part of the statistical model is estimating the variability of the residuals over time. The model used by Dechow and Dichev (2002) estimates the standard deviation of 5 years, this study will use the 3-5 year period as used by Ashbaugh-Skaife et al. (2008), this method will only eliminate observaton with less than 3 years instead of 5 years and therefore will ensure a larger data sample (and thus might increase reliability).
3.3.3 Earnings surprises or market reaction proxy (MorB)
The third proxy is based on the model as presented by Davis et al. (2009). The model presented is based on the fact that companies might use discretionary accruals (such as calculated using the modified Jones model in section 3.1.1) to influence earning targets (i.e. earnings management). Management has control over the accruals, using this power leads to discretionary accruals which might be used to meet or beat earnings forecasts. The model consists of three steps; The first step is to calculate the discretionary accruals, in this study the results from the first proxy will be used (refer to section 3.1.1). The second step is to calculate the non-discretionary earnings per share (hereafter ‘EPS’), this is done using the following model:
= Actual EPS − Discretionary accruals per share
The third step is to calculate the adjusted forecast error, this is done using the following model:
= −
Using this model it can be determined for each firm year whether ADJFE < 0, this means that the initial non-discretionary EPS doesn’t meet the earnings forecast but that the adjusted (discretionary) EPS does lead to the company meeting or beating the forecast18. The dummy equals 1 for just meeting or beating the earnings forecasts and 0 otherwise.
3.3.4 Auditor size or Big-N proxy (BigN)
The fourth and final proxy is that of auditor size or the Big-N proxy, this proxy is based on the fact that auditor size is related to audit quality. The proxy assumes that Big-N auditors deliver higher quality audits as they have stronger incentives and better competencies to deliver a high
quality audit (DeAngelo, 1981). Similar results are presented by Francis and Wilson (1988), their research shows that Big-N auditors have established brand names and thus have incentives to deliver high quality audits to avoid litigation and damage to their brand names.
Based on existing literature it can be concluded that an audit of a specific firm in a certain year is deemed a high quality audit if the firm is audited by a Big-N firm. Using a dummy variable this study will determine whether an audit is conducted by a Big-4 audit firm, PwC, KPMG, EY, Deloitte (dummy equals 1) or by a non-Big-4 audit firm (dummy equals 0).
3.4 Computing of external association – The impact of auditor tenure on audit quality The second step in the research model is calculating external association. As presented earlier, this study will use the relationship between audit quality and auditor tenure to determine whether construct validity of different audit quality proxies. The second step consist of two sub-steps; The first sub-step will use a logistic regression model to determine whether different proxies have different effects on the relationship between auditor tenure and audit quality. The second sub-step will use a Chi-squared test to compare the audit quality coefficients in pairs to verify whether they significantly differ from each other.
3.4.1 Regression model
As presented earlier in this study the relationship between auditor tenure and audit quality is used as to determine whether the usage of different proxies to measure audit quality have different effects. The relationship between auditor tenure and audit quality is based on the consensus that increased auditor tenure impacts auditor independence and thus audit quality. Using a regression model statistical evidence can be obtained on whether auditor tenure is of significant influence on audit quality. Performing the regression for each individual proxy will generate results which can be compared, first by using comparative descriptions and secondly by performing a Chi-squared to determine whether the results are significantly different from each other.
Within existing literature a large variety of regression models can be found, each with different dependent and independent variables. For this study a general regression model, such as used by Myers et al. (2003), will be used. The dependent variable will be audit quality (AQ), tenure is used as an independent variable along with several control variables. The control variables used are obtained from Myers et al. (2003) except for age, the control variable age shows the number of years for which total assets are reported in Compustat for this study we will use the total assets within the sample, therefore age is not considered to be relevant. Furthermore, please also note that the auditor size control variable is excluded from the initial regression model. This is due to
the fact that the BigN proxy is based on the assumption that an audit performed by a BigN auditor equals a high quality audit.
= + + + + + ℎ + ℎ + Where: = ℎ ( 3.3) = ℎ ℎ ℎ ℎ = ℎ log ℎ ( ℎ ℎ ) = ℎ − (65 ) = ℎ ℎ = ℎ ℎ ℎ = ∑ ∑ , − ∑ , , −
The regression model will be performed for each individual proxy, variable is of particular interest as this shows whether auditor tenure is significant with regard to audit quality. Based on existing literature it is expected that the results might be different depending on the proxy choice, in general table 1 shows that for the financial reporting quality proxies (DA and AccQ) increased tenure will have a either no or a positive effect on audit quality (non-significant or positive relationship) whereas the expected results for the other proxies are either that there is a negative effect on audit quality (negative relationship) or inconclusive. The other variables act as control variables, each single variable is believed to have an potential impact on the audit quality and are therefore added as control variables. The combined results will be summarized and tabulated, furthermore the results will be analyzed to determine the direction of the coefficient of interest (tenure).
3.4.2 Chi-squared test
The results which are derived from the regression model will be analyzed using descriptive statistics providing evidence whether tenure is significant using different proxies to measure audit quality. However, actual evidence on whether the results using different proxies are indeed significantly different is lacking when using descriptive statistics. Using Chi-squared tests for all pairs statistical evidence can be obtained on whether the results using different proxies are
significantly different from each other. For this study a p-value < 0,05 is accepted (generally accepted p-value within existing literature).
The regression model will provide coefficients for each independent variable, the variable of interest is , therefore using the Chi-Squared test the tenure coefficient amongst all proxies will be compared to determine whether they are significantly different. As explained earlier, Stata will be used to perform all statistical calculations. The results of each pair will be tabulated as evidence for the hypothesis.
4 Empirical results
4.1 Sample developmentThe section will summarize the procedures performed with regard to the sample selection for this study. The sample selection procedures can be divided into two sub procedures; Compustat sample selection and IBES sample selection. Proxy 1, 2 and 4 (discretionary accruals, accrual quality and BigN auditor respectively) are based on Compustat data whereas proxy 3 (just meet or beat) is based on IBES forecasts and actuals.
4.1.1 Compustat sample selection procedures
As presented earlier, this study will focus on the period 2000-2014, however due to the usage of lag variables19the obtained sample will contain the period 1999-2014. The initial data is obtained using Compustat, the Compustat database contains all relevant variables to calculate audit quality using three different proxies, discretionary accruals, accrual quality and BigN auditor. The relevant variables and their corresponding descriptions have been tabulated in table 3, in addition table 3 also shows which variables are relevant for which proxy and in which dataset the variables are gathered (i.e. dataset 1 for the Compustat based proxies and dataset 2 for the IBES based proxy). A singular dataset has been obtained which forms the basis for calculating proxy 1, 2 and 4 but also for calculating auditor tenure and the relevant regression model variables.
The initial sample consist of all firm years from 1999-2000, as explained above the firm years for 1999 are used to calculate several lagged variables and are later dropped (as they fall outside the selected period 2000-2014). The Compustat “North America - Annual Updates” and in particular the “Fundamentals Annual” database is used to obtain all relevant variables for the specified period. The initial sample contains all companies, specified by gvkey (i.e. no companies and/or SIC’s are excluded at this point). The standardized screening variables are used except for the fact that Financial Services is excluded as an industry (following prior audit tenure research), only domestic (i.e. US) data is obtained and only observations in USD are used. The output screen of the Compustat data extraction can be found in appendix A.
The initial dataset obtained from Compustat is cleaned for irrelevant variables (no impact on firm years). As stated above, the initial dataset will be used not only to calculate audit quality using different proxies, but also to calculate auditor tenure and the control variables for the regression
model. The sample selection procedures have been tabulated in table 4, panel A, showing the development of the number of firm years during the procedures. Initially the sample is cleaned to generate a main database including auditor tenure and the calculated value for each control variable from the regression model per firm year. The initial sample consists of 154.709 firm years, after the procedures the main database contains 41.550 firm years, these firm years will be the starting point to calculate the different proxies. As tabulated in table 4, the number of firm years are 36.167, 32.948 and 41.550 for respectively the discretionary accrual, accrual quality and BigN auditor proxies.
Table 3
Obtaining data and the variables used per proxy
Included in dataset # Description Variable Relevant
for proxy*
1 2 3 4 Compustat variables
1 Company identifier Compustat gvkey x x x
1 Company identifier cusip x x x x
1 Company name conm x x x x
1 Fiscal year end datadate x x x x
1 Fiscal year fyear x x x x
1 Auditor AU x x x x
1 Total assets at x x x x
1 Income before extraordinary items ibc x x
1 Cash flow from operations oancf x x x x
1 Gross property plant and equipment ppegt x x
1 Sales revenue sale x x x x
1 Receivables rect x x
1 Depreciation dpc x
1 Industry code sich x x
I/B/E/S variables
2 Company identifier I/B/E/S ticker x
2 Company identifier cusip x
2 Earnings per share value x
2 Forecast Period Indicator fpi x
2 Company name Company name x
2 Forecast period end date fpedats x
2 Announce date of actual anndats_act x
2 Actual value actual x
* Variables relevant for all proxies are either used as part of the regression variables or to merge Compustat and I/B/E/S datasets
4.1.2 IBES sample selection procedures
For the third proxy, the market reaction or just meet/beat proxy relevant variables are obtained from IBES, the IBES database contains analysts’ forecast and actuals. As this study is
focused on the period 2000-2014 data is obtained for the forecast period end date ranging from 2000-2014, data is obtained using the IBES ticker and no companies and/or industries are excluded. The Earning Per Share (hereafter ‘EPS’) is obtained for the ‘current’ fiscal period (using Fiscal Period Indicator 1), the sample does also contain the actuals for the selected period as this variable is specifically selected. The selected variables from the IBES database have been tabulated in table 3, which shows the variables of interest per proxy. All IBES data is obtained from the US File as this study is focused on the US. The IBES output screen can be found in appendix A.
The initial sample from the IBES database contains all analysts’ forecast per fiscal year, for the proxy calculations only the last forecast prior to the actual is relevant, therefore all prior forecasts’ are dropped. Furthermore all observations with missing data or forecasts after the actual are dropped, the results are tabulated in table 4, panel B. The initial sample contained 2.522.838 observations, after cleaning the data 16.416 firm years are remained.
Table 4
Sample selection procedures
Panel A: Compustat sample selection procedure
Compustat sample 1999-2014* 154.709
Less:
Observations with missing data and to
little observations 98.401
Observations with missing auditor data 2.190
Observations with less than 10
companies per two-digit SIC 1.533
Observations before 2000, after 2014 and
to little observations 11.035
Sample for proxy level analysis 41.550
Proxy level analysis Proxy 1 Proxy 2 Proxy 4
Discretionary accruals Accrual quality BigN auditor
Full sample 41.550 41.550 41.550
Less:
Observations with missing data 5.268 13.769 0
Observations with less than 10
companies per two-digit SIC 115 103 0
Table 4 (continued)
Panel B: I/B/E/S sample selection procedure
I/B/E/S sample 2000-2014 2.522.838
Less:
Observations with missing cusip 8.554
Observations with missing actuals 67.587
Observations with forecast after actual 4.141
Observations prior to last forecast before
actual 2.375.647
Observations with less than 15
observations per cusip 50.493
Sample for proxy level analysis 16.416
* The original dataset included samples ranging from 1999-2014, the year 1999 is used to generate lagged variables and dropped afterwards.
4.2 Descriptive statistics
From section 4.1 it can be derived that in total two datasets are obtained, one from Compustat and one from IBES. In this section the descriptive statistics of all relevant variables will be tabulated. The dataset obtained from Compustat is used as the general dataset where both auditor tenure and the control variables for the regression model are calculated. Next to that, this dataset is also used as the basis for the DA and AccQ proxy, separate descriptive statistics will be generated for each of these subsets as the number of firm years differs. The BigN proxy dataset is unchanged compared to the main database except for the fact that only the gvkey, fyear, au and BigN variable are of interest, the descriptive statistics are therefore included in the main database descriptive statistics. The dataset as obtained from IBES is used solely for the usage of the MorB proxy and therefore tabulated once. The results are presented in table 5, panel A shows the main dataset and the BigN proxy, the DA and AccQ proxies are tabulated in respectively panel B and C and the IBES descriptive statistics are presented in panel D. All variables obtained from either Compustat or IBES are explained in table 3, in addition all variables (both obtained and calculated) are explained and described in appendix B.
Table 5
Descriptive statistics per dataset
Panel A: Main dataset and BigN proxy
Variable N Mean Std. Dev. Min Max
at 41.550 13.676,5700 85.964,8500 0,0010 3.270.108,0000
oancf 41.302 772,6191 3.499,2880 -85.822,0000 121.664,9000
sale 41.526 5.679,1310 21.135,9000 -15.009,3300 483.521,0000
au 41.550 6,8931 3,7586 0,0000 26,0000
