The relationship between audit opinion and information risk : evidence from the stock market

(1)

The relationship between audit opinion and

information risk

Evidence from the stock market

by

Zeeshan Malik Mohammad

Student number: 10814132

Master’s thesis submitted in support of the degree of

Master of Science in Accountancy and Control

Track: Accountancy

University Of Amsterdam

Faculty of Economics and Business

Thesis supervisor: dr. A. Sikalidis

Word count: 12,575

(2)

Statement of Originality

This document is written by student Zeeshan Malik Mohammad who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

This paper examines the association between the level of information asymmetry and the content of the audit opinions. We predict and empirically demonstrate that the level of information asymmetry is perceived as higher by market participants when a modified audit opinion is given. We further show that when the audit opinion is quantified that there is a non-significant positive effect on the level of information asymmetry. Furthermore, we empirically demonstrate that the level of information asymmetry is perceived as higher when the manager has the ability to hide private information through research and development costs.

(4)

Table of Content

Abstract ... 3

Table of Content ... 4

1 Introduction ... 5

2 Literature review and hypothesis development ... 7

2.1 Role of an auditor ... 7

2.2 Information asymmetry and audit opinion ... 8

2.3 Hypothesis development ... 13

3 Research design ... 16

3.1 Sample selection and data ... 16

3.2 Measuring information asymmetry ... 17

3.2.1 Proxies for information asymmetry ... 18

3.2.2 Proxies for audit opinion classification... 20

4 Empirical findings ... 22 4.1 Descriptive statistics ... 22 4.2 Univariate analysis ... 24 4.3 Multivariate analysis ... 25 4.4 Robustness analysis ... 28 5 Conclusion ... 30 References ... 31 Appendix ... 35

(5)

1 Introduction

Over the last few decades, the accounting world has had many scandals. Year after year the accounting boards release reports with the message that audit quality is poor and should improve. Recently, in August 2017, KPMG received a fine of more than 6.2 million dollars from the US Securities and Exchanges Commission (SEC). The reason for this fine is that they gave a qualified opinion while the firm they were doing the audit on had overvalued assets by more than hundred times (Murphy, 2017). The year before 2017, we have had another audit scandal. While employees of Wells Fargo ordered credit cards for customers without consenting them, created fraudulent checking and savings accounts and many other frauds, the auditors did not find all these misstatements. According to Mckenna & Riquier (2017), KPMG wrote a letter to the senators stating that they were aware of the illegal activities for a while, but they did not assess the unlawful activities as material for the users of financial statements. To solve this audit quality issue, many researchers have researched how to improve the quality of the audit.

However, while doing this, they do not take the main subjective into account, namely the decision makers. According to Arens et al. (2014), an audit report serves as an assurance for the decision makers. The audit report has the function to assure that the assertions of the management on the financial statements are fairly stated in accordance with applicable standards like the General Accepted Accounting Standards (GAAP) or the International Financial Reporting Standards (IFRS). However, how effective are the auditors in providing assurance for the decision makers? Do they ensure that the auditor’s opinion of the financial statements reduces information asymmetry?

Hence, this paper investigates the relationship between the auditor’s audit opinion and information asymmetry. This relationship will be assessed through empirical research.

There are various reasons why providing an answer to this research question is important. One of the reasons for this is that having high-quality audit information benefits the efficient allocation of resources in capital markets. Having high-quality audit information also is also known as a mechanism that reduces information asymmetry and agency conflicts. In the past, most of the researchers focused on aspects of the audit tenure, the audit firm size, the independence of the auditor, the gender of the auditor, the audit fee etcetera. However, there is not a lot literature available discussing the effectiveness of reducing information asymmetry of the audit report. In our study, we employ the relative effective spread, the price impact, the probability of information-based trading and a composite of these three models to measure the level of information asymmetry for different types of audit opinions. Resulting in a sample of 7747 firm-year

(6)

observations of American non-financial, NYSE listed companies from 2002 to 2016. The reason why we have selected American companies is that according to Van der Meulen et al. (2007) the U.S. GAAP outperforms the IFRS in providing accounting information. This increased account information leads to higher audit quality and probably more effective in reducing the level of information asymmetry. The period after 2002 is selected because in the year 2002 the government of the United States adopted the Sarbanes-Oxley act resulting in a radical change in the accounting world (Defond & Francis, 2005). Moreover, the reason why we have selected NYSE and not the NASDAQ is that under NASDAQ stocks are sold through dealers, while market participants can directly buy stocks from the NYSE market. So, under the NYSE market, we can directly measure the level of information asymmetry as perceived by the market participants.

Analogous to our expectations, we demonstrate that firms with modified audit opinions show a higher level of information asymmetry in the market. However, we do not find credible evidence for that when a modified audit opinion is quantified that it reduces the level of information asymmetry. Moreover, we demonstrate that when a company can hide private information through research & development costs that the level of information asymmetry is significantly higher in the eyes of market participants. The results are robust to controlling for the financial situation of the firm.

Finally, this paper contributes in multiple ways. Firstly, this paper contributes to the literature on the microstructure and the economic consequences of financial reporting by

showing that different types of audit opinions have different effects on the information effect of the audit opinion on the financial statements of a firm in the eyes of market participants.

Secondly, this paper demonstrates how the type of audit opinion can deliver economic value to the market participants. Lastly, this paper contributes by showing that even when several stakeholders believe that the actual audit quality is poor, the market participants perceive the audit opinion as credible.

The rest of the paper proceeds as follows. Section 2 provides a summary of related literature on the association between the level of information asymmetry and several audit opinions. Also, we discuss our hypothesis in Section 2. In Section 3, we describe how we select our sample and how we will use our data to measure the level of information asymmetry. In Section 4, we discuss our results, and we present some additional tests. In Section 5, we present our conclusions.

(7)

2 Literature review and hypothesis development

In the first subsection of this chapter, we discuss the role and the importance of an auditor in solving agency conflicts and the role of an auditor in reducing the level of information asymmetry. In the second subsection of this chapter, we analyse prior literature related to the effectiveness of an audit opinion and in the third subsection we discuss our hypotheses.

2.1 Role of an auditor

According to Arens et al. (2014, p. 24) auditing is defined as the ‘accumulation and evaluation of evidence about information to determine and report on the degree of correspondence between the information and established criteria. Auditing should be done by a competent, independent person’. The purpose of auditing by an auditor is to prove reliable information for decision-makers. Examples of decision makers are the managers, shareholders, prospective investors, lenders, suppliers, employees, customers and the government. As we can see, there are various types of decision makers, which may lead to a conflict of interest between the decision makers and the one who provides information. This conflict of interest is also classified as agency conflict.

According to Eisenhardt (1989), agency conflicts arise when the agent tries to maximize his utility at the cost of the principals. In the context of auditing, it could mean that the manager maximizes his utility by changing information in the financial statements knowingly that the utility of the principal, the stakeholders, will be impacted negatively by it. This manipulation of the manager to change information in financial statements does not only increase the number of agency conflicts but also increases information risk for decision maker (Arens et al., 2014). The information risk rises because when the managers hide information, the principals will have trouble finding the hidden information. To put it differently, the principals lack information to monitor the agents, which leads to these agency conflicts. Lack of information becomes a problem when agents commit moral hazard behavior. To solve or reduce this problem of information risk problem, which is also known as information asymmetry problem, and the agency conflicts the role of the auditor is established.

As has been noted by Arens et al. (2014), the role of the auditor is to determine and report on the degree of correspondence between the information and established criteria. To put it another way, the role of the auditor is to reduce / to tackle the agency conflicts and information asymmetry problem by determining and reporting on the degree of correspondence between the information and established criteria. To explain further, under the term ‘information’ we classify the firm's financial statements, the individuals’ federal income tax returns, other quantifiable

(8)

information and also subjective information like the effectiveness of computer systems and the efficiency of manufacturing operations (Arens et al., 2014). With this in mind, it is easier to understand the corresponding criteria. When auditing financial statements, the criteria could be the U.S. Generally Accepted Accounting Principles (GAAP) or the International Financial Reporting Standards (IFRS). This means that the role of the auditor used in our paper, using data from companies in the Netherlands, is to determine if the information from the financial statements has been prepared in accordance with the International Financial Reporting Standards. As a result of the several audit scandals before 2002, the government of the United States adopted the Sarbanes-Oxley act. This act improved the ability of auditors to fulfill their role by requiring public companies to strengthen audit committees and increasing the level of disclosures. Moreover, this act established the Public Company Accounting Oversight Board that has the ability to create rules like the audit partner rotation every five years to reduce the level of conflict of interests. This all had a positive effect on the work of the auditors (e.g., Banerjee et al., 2015).

2.2 Information asymmetry and audit opinion

In the past various researchers have investigated the topic of information asymmetry in different fields. From their investigations, we learn that companies providing disclosures have an impact on the level of information asymmetry that the users of the financial statements and traders have in the stock market. When the quality of their disclosures is higher, the information asymmetry is lower. They explain this with that the users of the financial statements will receive more information, so there are fewer traders who can trade with an information advantage, and the traders will have fewer incentives to search for private information Arens et al. (2014). More specifically, Leuz and Verrecchia (2000) examined the effect of information asymmetry by discussing the existence of adverse selection in the transactions of traders and investors. Due to traders and investors having more knowledge about some information about the firm they decreased the level of liquidity for the stocks. Reducing information asymmetry would lead to a lower level of risk of adverse selection and also would improve the level of liquidity of shares. Moreover, Lafond and Watts (2008) have analyzed the stock market demand for conservatism. In their paper, they argue that information asymmetries between traders and investors engender accounting conservatism.

In contrast to prior literature, this paper is not looking into characteristics of auditors and things like the independence of the auditor. This paper looks into the role of the audit report. How does it impact the quality of financial statements reported by firms? How does it reduce information risk for the users of financial statements? According to Butler et al. (2014), the audit

(9)

report is an essential element for users of financial statements when using the statements for making decisions. This is because the audit report increases the reliability of the financial statements. According to Khasharmeh (2003), an opinion of an auditor is seen as an appropriate information tool for users to make financial investment decisions. As mentioned in the previous subsection, the function of an audit is to assure the users of financial statements so that the users can make investment decisions with reliable information. The opinion of an auditor can help users of financial statements in making business investment decisions by providing the user's information over the financial statements that are prepared by the manager of the company that is audited. This information is needed because users of the financial statement are unable to gather the information & assurance is necessary for their investment decisions (Leuz, 2000).

Moreover, the information provided by an auditor, known as audit opinion, can be categorized in four categories (Arens et al., 2014). The four types for audit opinions are (1) unqualified audit opinion (with or without emphasis of matter paragraph / explanatory paragraph), (2) qualified audit opinion, (3) adverse audit opinion, (4) disclaimer of opinion. An unqualified audit opinion is an opinion given when the auditor believes that the financial statements of the entity inspected is stated in a true and fair manner. An emphasis of matter paragraph is added to the unqualified audit opinion when the auditor believes that there is a significant uncertainty or other matter. This matter is disclosed appropriately in the disclosures of the financial statement, but the auditor concludes that this matter is essential at a level that he or she should warn the users of financial statements separately.

Different from the unqualified audit opinion, the qualified audit opinion is given when the auditor believes that the financial statements of the entity inspected is stated in a true and fair manner, except for that there are uncertainties that might be material but not persuasive for the users of financial statements. In contrast to the previous two mentioned audit opinions, the adverse audit opinion does not believe that the financial statements of an entity are stated in a true and fair manner. When the auditor gives this adverse audit opinion, he or she believes that there are material misstatements in the financial statements that are persuasive in a way that it has a significant impact on the investment decisions of the users. However, when the auditor is not able to collect enough evidence or is not satisfied with the evidence he/she has collected he can give a disclaimer of opinion.

According to Abad et al. (2017), Melumad and Ziv (1997) and Martinez et al. (2004) a qualified audit opinion has two types. The first type is the quantified audit opinion. A quantified opinion is issued when the auditor believes that the financial statements contain material misstatements that are not persuasive and in how far the misstatements are material /

(10)

non-compliance with accounting standards or from significant changes in accounting standards and principles applied by the entity in the financial statements. The second type is non-quantified. From the same papers, we read that the definition of a non-quantified audit opinion is that the auditor is not able to determine the effect of giving a qualification on the financial statement. With qualification is meant going concern, scope limitations and other uncertainties.

According to Pucheta-Martinez et al. (2014), qualified audit opinions do not increase the information that users of financial statements have. They explain this with that one of the reasons for this is that the result of the audit opinion can be anticipated. However, this is in contrast with what Soltani (2000) has found in his research. According to his analysis, the price of the stocks decreases significantly when a company receives a qualified audit report. This event of a decrease in stock prices indicates that the qualified audit report has value and reduces adverse information to the market.

In the paper of Christensen et al. (2014) examined if critical audit matter paragraphs in the audit report change nonprofessional investors' decision to invest. They do this by performing an experiment with nonprofessional investors who are business school graduates, have invested in individual stocks and do company financial data analysis. From their research, they have concluded that investors who receive a critical audit matter paragraph are more likely to change their investment decision than investors who did not receive an audit report with a critical audit matter / a standard audit report. This effect on the investors is called the information effect. Christensen et al. (2014) also found that investors who have received a critical audit matter paragraph are more likely to change their investment decision than investors who received the same information as mentioned in the critical audit matters paragraph through the management’s footnotes. That investors trust the critical audit matter of the auditor more is called the credibility effect. The so-called information and credibility effect in the paper indicate that investors perceive less information asymmetry when receiving a critical audit matter paragraph in the audit report and also it indicates that there is less information asymmetry when the auditor reports it than when the management reports it. Hence, according to the paper of Christensen et al. (2014), the opinion of the auditor is perceived as necessary in the battle against information asymmetry by the users of the financial statements.

In addition to the paper of Christensen et al. (2014) and Rena et al. (2016) have investigated the importance of the opinion of an independent auditor as perceived by individuals and institutional investors with whom the banks are in a trading relationship. They both believe that the opinion of the independent auditor should have vital importance in decreasing the information risk and increasing the reliability of the information that is submitted in financial tables and form

(11)

the basis of the decisions made by the users of financial statements. To examine the relationship between the opinion of the independent auditor and the investments made to banks they looked at two different points. Firstly, they examined whether the trading relations of the banks with third parties, mainly their customers, are affected by the audit opinion. They do this by looking at the deposit reimbursed by the entities and institutions to the 10 banks, the credits they received and the total assets that might be affected, depending on the net interest income, the trading profit and losses, the net profit and losses that have been used in the processes that reflect the trading relationship of the banks and the entities and institutions. Secondly, they looked at those who make investments in the banks and if they are affected by the opinion of the independent auditor. They do this, by looking at the equity shares. A limitation of this study of Rena et al. (2016) is that they ignored the effect of adverse audit opinions and the disclaimer of audit opinions. They only used unqualified with or without emphasis of matter paragraph and qualified audit opinions. Nevertheless, in their study, they found that the opinions of independent auditors given to the ten banks as unqualified have an improving impact on the variables, mentioned earlier, when the variables are compared to the variables of the banks that are given a qualified audit opinion by the independent auditors. Assuming ceteris paribus for the other variables, Rena et al. (2016) has found that the opinions of the independent auditors are one of the factors that change the decisions of investors in the same direction as seen with the variables as deposits, credits, total assets, net interest incomes, net trading profits and losses, net profits and losses and the price of equity shares obtained from the financial tables.

Likewise, this relationship between the opinion of an audit has a strong effect on the investment decisions of the users of financial statements (Wisdom et al., 2017). Wisdom et al. (2017) assume that it is widely believed that auditors are responsible for business failures that affect the going concern of the companies. So, they evaluate the relationship between the report of an auditor and the investment choices of users of the audit report in Nigeria. They did this by doing a survey and asking academics in universities within Ado-Odo, Ota and Ogun state. On the data, they applied the Pearson Product moment Correlation technique. The results of this research reveal a strong and positive relationship between the report of an auditor and the investment decisions of users of audit reports in Nigeria. However, the interviewed academics do acknowledge that the audit report should be improved upon. The reason for this is that the audit report should guarantee that the communication of information to stakeholders is useful and helps with decision-making.

Different from the studies discussed earlier; there were also studies who do not find that the opinion of an independent auditor has an influence on the decision making of users, which

(12)

indicates that an audit opinion does not reduce information asymmetry in the eyes of the users. In the study of Tahinakis et al. (2010) they investigated whether the appraisal of the audit qualification has an impact on the firm's stock exchange price fluctuations. To do this, they looked at 39 companies with shares in the Athens Stock Exchange (ASE) during 2005, 2006 and 2007. From these 39 companies, they collected 40 unqualified opinions and 16 qualified opinions. The 39 companies selected did not receive an adverse audit opinion or disclaimer of opinion report. In their study they have found that audit reports, be it unqualified-, qualified-, adverse or disclaimer of opinion, have limited information content for investors. Also, they concluded that the audit reports do not influence the decision-making process of users of financial statements. According to Tahinakis et al. (2010), the reason for this is that the users of audit reports may lack the knowledge to understand the contents, importance, and value of these audit reports. Moreover, in the study of Duflo et al. (2013), they looked at third-party entities reports to firms on the environment of internal plants. Their results show that the resultant conflict of interests between the management of the entity and stakeholders could corrupt information provision, which leads to higher information asymmetry, and also it undermines regulation. Backcheck data has shown that when auditors are directly paid by the management of the entity, they are significantly more likely to report falsely so that the entity can meet regulatory limits. This falsely reporting leads to higher information asymmetry, because the users of financial statements believe that the auditor has checked the facts. While, when the auditors are paid from a central pool, they report more accurately. Hence, auditors are strongly influenced by incentives. This also shows us that whenever the auditor gives his opinion, it does not mean that the opinion reduces information asymmetry. Under certain circumstances, it is possible that the opinion of the auditor increases the actual information asymmetry. However, it is not always the case that when the actual information asymmetry increases that the users of financial statement perceive the opinion of the auditor as an increase in information asymmetry. Karjalainen (2011) has looked into the value relevance of the perceived quality of the audit. In their study, the researcher notes that primarily the quality of the audit as perceived by the users of financial statements has an impact on the information asymmetry and not the intended quality of the audit. The perceived quality of the audit opinion is higher when the firm is audited by a Big 4 audit firm and audits with more than one responsible auditor. This results in less information asymmetry. This also goes for modified audit opinions, opinions like the qualified opinion, adverse opinion and disclaimer of opinion. Furthermore, in the paper of Jones and Soloman (2010), they look at the credibility of the opinion of the external auditor by interviewing managers of firms. For their research, they interviewed managers from 20 UK listed companies and asked them if they believed that the

(13)

opinion of an external auditor increases the credibility. Half of the respondents are negative about it. They believe that the external auditor does not increase the credibility / reduces the information asymmetry.

Last, in the paper of Chen et al. (2016), they investigated if clients of audit firms successfully engage in audit opinion shopping. With audit opinion shopping is meant that when the incumbent audit firm is likely to give a modified audit opinion, which is a qualified/adverse or disclaimer of opinion, the auditee looks for an alternative audit firm that is prepared to give an unqualified audit opinion without emphasis of matter or explanatory paragraph. This paper shows that companies successfully engage in audit opinion shopping. Also, they have found that the effectiveness of audit opinion shopping depends on two factors namely, the client importance and the organizational form of the audit firms. Client importance refers to the pressure of the auditee on the audit firm to yield to a request for a partner switch. The higher the independence of the audit firm, the less likely it becomes that the auditee does audit opinion shopping. The organizational form refers to the company as being a partnership or a corporation. Furthermore, they compared incoming partners with outgoing partners. The incoming partners have a significantly higher propensity to issue an unqualified audit opinion without emphasis of matter paragraph than the outgoing partners. This audit opinion shopping might lead to a decrease in the trust of audit opinion and effect of the audit opinion on prices.

The crux of the matter becomes more clear when we look at the report ‘In het Publieke Belang’ of the NBA (2014). This article states that a requirement to provide assurance to the users of financial statements is that the audit profession should be trusted by the users of financial statements. However, this trust has been abused several times in the past by audit firms. As we can see in the paper of Chen et al. (2016), audit firms are willing to abuse this trust by having a significantly higher propensity to issue an unqualified audit opinion without an emphasis of matter paragraph to get the audit job from the company, even when the financial statements of the auditee company has some material misstatements in it. This behavior leads to a decreased trust in the auditing profession.

2.3 Hypothesis development

Since an audit report has as function to assure that a company's financial statements give a true and fair view, it is expected that an audit opinion increases the relevance and reliability of financial statements and therefore reduces information asymmetry Arens et al. (2014). This enhancement of the credibility of financial information may be seen by the investors as evidence that the firm has less information risk. However, this is only true if the investors if the audit opinion is seen as

(14)

credible (Hope et al., 2011). In addition to this, Limperg's theory of inspired confidence (1932) states that when the society has no more confidence in the effectiveness of the auditor, than the social usefulness of the audit opinion is lost. From the reports on audit quality of the Dutch AFM (e.g., AFM, 2017) in the last decade, we read that the quality of audit opinion is inadequate. More specific, on the effects of information asymmetry on the stock market, the information asymmetry problem leads to adverse selection and moral hazard. According to Klein et al. (2016), the adverse selection and moral hazard problem lead to an increase in transaction costs between the buyers and sellers. Their research suggest that when information asymmetry is reduced, then the market transparency is increased and therefore the market becomes more efficient.

Undeterred by the effects of information asymmetry on the stock market and the importance of the audit opinion to reduce these effects, we find little evidence about the relevance of an audit opinion in the stock market. Most of the prior literature focuses on the reaction of the announcement of the audit opinion and not on the sustainable effects of the audit opinion in the year after. Pucheta-Martinez et al. (2004) suggest in their research that the negative impact of not receiving a qualified opinion can be perceived before the release of the audit opinion by the market participants. So, on the day of the announcement the market participants already know about the opinion, and therefore there is no reaction. On the other hand, Soltani (2000) suggests that an audit opinion provides new adverse information to the market participants and therefore it has an effect on the stock prices.

To examine the level of information asymmetry among market participants, we examine the relationship in an association study context. To do this, we follow study designs like the effect of the audit opinion on contract in the private debt market (Chen et al., 2014) and the level of information asymmetry in Spain (Abad et al., 2017). We expect that when an unqualified audit opinion is given the investor's perception of the quality of the financial report is better and that there is more certainty of earnings. Hence, our first hypothesis is as follows:

Hypothesis 1: A qualified audit opinion is positively associated with the level of information asymmetry.

Especially relevant, is that the hypothesis stated above is in an alternative form. The opposite of it, the null hypothesis, is that firms with qualified audit reports do not have higher information asymmetry.

Following Pucheta-Martinez et al. (2004), we distinct audit opinions between opinions that prove extra quantitative information like GAAP compliance and audit opinions that cannot be

(15)

quantified. We expect that opinions that provide quantified information reduce information asymmetry significantly stronger than those that do not provide quantified information. Hence, our second hypothesis is as follows:

Hypothesis 2: a quantified audit qualification is positively associated with the level of information asymmetry.

Especially relevant, is that the hypothesis stated above is in an alternative form. The opposite of it, the null hypothesis, is that firms with non-quantified audit qualification do not have higher information asymmetry.

From papers like Iqbal & Santhakumar (2018) and Jin and Myers (2006), we learn that firms, whereby the manager has much private information, can commit more earnings management. They can hide private negative information and share private positive information with users of financial statements to make the price go up. This is confirmed by Aboody and Lev (2000). Furthermore, they mentioned in their paper that R&D investments are a significant source of private information from the perspective of the investor. They explain this with that most of the R&D projects, like new drugs/software programs are unique to the company. This makes it harder for investors to determine the actual value of these R&D assets. As mentioned before, one of the functions of an audit report is to reduce this level of private information. So, when an unqualified audit opinion is given, the users could assume that the private information that is not provided is not material for the making of decisions based on the financial statements.

Hence, these theoretical arguments combined lead to the following third & fourth hypothesis: Hypothesis 3: Higher information asymmetry will be detected for companies with unqualified audit reports and high level of private information than companies that have received unqualified audit opinions and low level of private information.

Especially relevant, is that the hypothesis stated above is in an alternative form. The opposite of it, the null hypothesis, is that firms with high information risk and an unqualified audit report do not have higher information asymmetry.

(16)

3 Research design

To provide an empirically contrast for the proposed research hypotheses, we have developed a quantitative study aimed to analyze the relationship between the level of information asymmetry amongst users of financial statements, in our case market participants, and the audit report itself. In the next few subsections, the data that we are going to use, e.g., the sample selected, and we will explain the models that we will use in our research.

3.1 Sample selection and data

To analyze the relationship between the level of information asymmetry amongst users of financial statements and the audit report itself we will look at companies based in the United States of America. One of the reasons why we have chosen to look at companies based in the United States of America is that these companies make use of the U.S. General Accepted Accounting Standards, while European companies use the International Financial Reporting Standards. According to Van der Meulen et al. (2007), the U.S. GAAP accounting information outperforms the accounting information under IFRS, even after controlling for differences in firm characteristics such as size, leverage, and the audit firm. The increased accounting information under U.S. GAAP results in a higher quality of the audit and thus in a higher quality of the audit opinion.

As for the sample, we investigate the impact of an audit opinion and the level of information asymmetry for firms listed on the NYSE during 2002-2016. The reason why we have selected this timeframe is that in 2002 the government of the United States has adopted the Sarbanes-Oxley act in that year. As mentioned in the literature framework, this law has a significant impact on the quality of audit opinions. The reason why 2016 is the cut-off point is that 2016 is the last year where we could find all the data needed for our analysis.

For this paper, we select a sample consisting of 7747 non-financial firms. The main reason why we exclude financial firms from our data is that financial firms have high leverage and having high leverage is typical for firms, while high leverages are not normal for non-financial firms. For non-financial firms high leverage indicates distress. This also results in different outcomes in models between financial firms and non-financial firms as empirically confirmed by Foerster and Sapp (2005). The sample is reported in Panel A of Table 1 and the construction of the sample is as follows. First, we collect all firm-year observations of the NYSE market with an unqualified or modified audit opinion from the database Audit Analytics for the period 2002 - 2016, representing 32,016 non-financial firm-year observations. The data from Audit Analytics covers all SEC registrants filing on Edgar. The reason why we have selected the NYSE and not the National

(17)

Association of Securities Dealers Automated Quotations (NASDAQ) is that the NASDAQ is a market whereby stocks are sold through dealers, while market participants can directly buy from the NYSE market. So, when using the NYSE market, we can directly measure the level of information asymmetry from the perception of the market participants. This allows us to do our research with the most comprehensive sample of audit opinions possible. Next, we eliminate 6,443 firm-year observations because we cannot match the data with the CRSP database and Compustat database. Likewise, we drop 17,826 firm-year observations due to lack of data or missing data from the Compustat's Fundamentals Annual Table and the CSRP database, resulting in a sample of 7,747 firm-year observations from 709 different firms. Last, we collect data manually. For the second hypothesis, we need to know if an audit opinion is quantified or not. To determine this, we manually look at the opinions that are not unqualified in the database of Audit Analytics.

Table 1

Sample Selection

Panel A: Sample Selection

Description Observations Firms on Audit Analytics for years 2002-2016 32,016 Less: no match with CSRP / Compustat 17,826 Less: missing data items CSRP / Compustat 6,443

Final sample 7,747

Notes: Panel A reports the sample. The sample consists of firms listed on the NYSE in the United States of America whereby data is available on Audit Analytics, Compustat, and CRSP. The observations from Audit Analytics are first matched with Compustat based on the Central Index Key (CIK) identifier. The results from this are matched with CRSP using the eight-digit Committee on Uniform Security Identification Procedures (CUSIP) identifier. Firms whereby the CUSIP could not be matched or whereby data was missing are dropped.

3.2 Measuring information asymmetry

As stated in the previous chapter, our paper aims to examine the association between several audit reports and the level of information asymmetry as perceived by market participants by following the regression model used in Abad et al. (2017) in our model specification:

𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖,𝑡𝑡 = 𝛽𝛽0+ 𝛽𝛽1𝑄𝑄𝐴𝐴𝑄𝑄𝑖𝑖,𝑡𝑡−1+ 𝛽𝛽2𝐴𝐴𝑆𝑆𝑆𝑆𝑆𝑆𝑖𝑖,𝑡𝑡+ 𝛽𝛽3𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑄𝑄𝑇𝑇𝑖𝑖,𝑡𝑡+ 𝛽𝛽4𝑇𝑇𝑄𝑄𝑉𝑉𝐴𝐴𝑇𝑇𝑖𝑖,𝑡𝑡+ 𝛽𝛽5𝐵𝐵𝑆𝑆𝐵𝐵4𝑖𝑖,𝑡𝑡+

𝛽𝛽6𝑆𝑆𝐴𝐴𝑍𝑍𝑄𝑄𝑇𝑇𝑆𝑆𝑖𝑖,𝑡𝑡+ ∑ 𝛽𝛽𝑗𝑗 𝑗𝑗 𝑆𝑆𝑇𝑇𝐼𝐼+ 𝜀𝜀𝑖𝑖,𝑡𝑡 (1)

In the equation mentioned above the i reflects the firm and the t reflects the firm-year observation. The variable ASY reflects the proxy for information asymmetry. In our paper, we are going to use four regression models, whereby the difference between every regression model is the measurement for information asymmetry. The four measures for the level of information asymmetry are the Relative Effective Spread (RES), the Price Impact (PI), the Probability of Information-based Trading (PIN) and a mixed proxy of these three proxies. The variable QAO

(18)

stands for Qualified Audit Opinion. As there are several types of audit opinion, this variable could differ per hypothesis. We use a one-year lagged QAO because the opinion of the auditor reflects the opinion of the financial statements of the previous year. If there are effects created by the opinion of the auditor, it is logical that these effects arise in the years after the release of the audit opinion. In our analysis, we control for factors associated with the level of information asymmetry. We use factors like the firm size regarding assets value (Size), the average daily trading volume (Turnover), the stock return volatility (Volatility), the audit quality (BigN) and the credit risk score (Zscore). From the papers of Bartov and Bodnar (1996) and Chae (2005) we learn that the size of the firm regarding asset value, the trading volume and the volatility of stocks play a significant role in determining the stock liquidity and information asymmetry.

More specifically, we measure the control variable Size as the logarithm of the total assets at the end of the year. The control variable Turnover is measured as the logarithm of the average daily trading volume in dollars scaled by the market value of the firm's equity at the end of the year. The volatility variable is a proxy for stock return calculated as the daily squared close-to-open mid-quote return. The last variable BigN is a proxy for audit quality. In the paper of Clinch et al. (2012) is stated that companies of the Big 4 provide higher audit quality and thus are associated with a lower information asymmetry between market participants. Last, we control for the credit risk score measured with the Altman Z-score model (Altman, 1968). The Zscore shows the financial strength of the company. From the paper of Raghunandan & Rama (1995), we learn that when a firm experiences financial distress, it also experiences a different type of audit qualification. The Zscore model is calculated as follow:

𝑆𝑆𝐴𝐴𝑍𝑍𝑄𝑄𝑇𝑇𝑆𝑆 = 0.106𝑆𝑆𝐵𝐵𝑆𝑆𝑇𝑇 + 0.169𝐴𝐴𝐴𝐴𝑉𝑉𝑆𝑆𝐴𝐴 + 1.01𝑇𝑇𝑆𝑆 + 0.104𝑊𝑊𝑍𝑍_{𝑇𝑇𝐴𝐴} + 0.003 𝑀𝑀𝑇𝑇𝑆𝑆_{𝐵𝐵𝑇𝑇𝐼𝐼} (2) Whereby, EBIT stands for earnings before interest and taxes, SALES for the total sales, RE for retained earnings, WC for working capital, TA for total assets, MVE for market value of equity and BVD for the book value of the debts.

Moreover, we also control for temporal and industrial effects by adding the dummy variables year and the industry of the company. As for the betas in the regression model we expect that β2 < 0,

β 3 < 0, β 4 > 0, β 5 < 0.

3.2.1 Proxies for information asymmetry

In this subsection, we will explain which and why we have selected the three proxy measures for the variable ASY. Following Abad et al. (2017), the three proxy measures used in our study for

(19)

information asymmetry are the Relative Effective Spread (RES), the Price Impact (PI) and the Probability of Information-based Trading (PIN).

Firstly, we will take a look at the relative effective spread (RES). The RES is also known as the effective bid-ask spread. The reason for this is that this proxy measures the difference between the highest price for which a market participant wants to buy the stock and the lowest price for which a market participant intend to sell the stock. More directly, it reflects the difference between the highest bid and lowest ask. Hence, that it is called the effective bid-ask spread.

Copeland and Galai (1983) state that an individual that decides to act as market-maker is assumed to optimize his position by setting an RES which maximizes the gap between expected revenues received from liquidity-motivated traders and expected losses to information-motivated traders. This indicates that different market participant has different information at its disposal. This difference reflects the information asymmetry between market participants. If this gap increases, there is more information asymmetry, while if this gap becomes smaller, there is less information asymmetry. The model used to measure the relative effective spread comes from the paper of

Easley & O’Hara (1992) and is calculated by doubling the absolute value of the difference between the actual trade price and the midpoint of the market quote divided by the midpoint of the market quote. More specific, the relative effective spread is calculated on a daily basis by averaging all the observations within a day. After doing this, these observations are recorded in trade time and the average is calculated on the basis of the volume. Finally, the yearly relative effective spread observed is averaged by the days within the year of measuring.

𝑇𝑇𝑆𝑆𝐴𝐴𝑡𝑡= 2 ∗ | 𝑃𝑃_𝑄𝑄𝑡𝑡− 𝑄𝑄𝑡𝑡|

𝑡𝑡 (3)

Where P reflects the price paid for the stock in trade t and Q reflects the midquote. The midquote is the average of the bid and ask quotes.

Secondly, we will discuss the Price Impact model of Huang and Stoll (1996) which we use as a proxy measure to determine the level of information asymmetry. This proxy measure looks at the spread that happens due to the price impact that lasts a period after the trade. With this, they can extract the presence of new information. To calculate the Price Impact, we deduct the current midquote from the midquote in the future and multiply it with a 1 or -1. The average is determined just like how we did with the relative effective spread. First, we compute the trade time by calculating the average based on the volume of all the trades within the day. And then averaging the results with the days in the year. The model to determine the Price Impact is defined as follows:

(20)

Where Q is already defined in the first model. The X reflects if the trade is a sell or a buy. If it is a sell X is -1 and if it is initiated on the buy side X is 1.

Thirdly, we’ll discuss the Probability of Information-based trading (PIN) model of Easley et al. (1996). That we use as a proxy measure to determine the level of information asymmetry. This probability of information-based trading model examines the unconditional probability that a randomly selected trade is done by an informed trader. This proxy measure is not easy to observe, to do this we need to look at parameters of a microstructure model that are estimated by numerical maximization of a likelihood function. As for these estimates, we follow Abad et al. (2017). In their study, they note that 30 trading-day windows are sufficient for the estimations. Thus, using data of one month is enough to produce reliable estimates for the PIN model. After calculating the monthly PIN, we compute the yearly by averaging the monthly PIN values. After estimating these parameters, the PIN variable is determined as follows:

𝑃𝑃𝑆𝑆𝑇𝑇 = _{𝛼𝛼𝛼𝛼 + 𝜀𝜀}𝛼𝛼𝛼𝛼

𝑏𝑏+ 𝜀𝜀𝑠𝑠 (5)

Where α reflects the probability that an information event occurs, µ reflects the arrival rate of orders from the informed trader, 𝜀𝜀𝑏𝑏 reflects the arrival rate of the buy orders and 𝜀𝜀𝑠𝑠 reflects the

arrival rate of the sell orders. To estimate if a particular trade is buyer ora seller initiated, we perform the Lee and Ready (1991) tests consisting of a quote test, tick test & a randomized test.

Lastly, to remove the underlying adverse selection of the three proxies for the level of information asymmetry, we combine them into one variable that measures information asymmetry. To do this, the principal components analysis is used by Abad et al. (2017). This lead to the new information asymmetry proxy determined as follow:

𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖𝑡𝑡 = 0.405𝑇𝑇𝑆𝑆𝐴𝐴𝑖𝑖𝑡𝑡+ 0.396𝑃𝑃𝑆𝑆𝑖𝑖𝑡𝑡 + 0.320𝑃𝑃𝑆𝑆𝑇𝑇𝑖𝑖𝑡𝑡 (6)

3.2.2 Proxies for audit opinion classification

As mentioned before QAO is a variable that reflects the relationship between audit opinions and information asymmetry. As for the first hypothesis, the variable QAO reflects the opinion of the auditor as required by the GAAP standards. When QAO has the value 1, it means that a modified audit opinion is given, be it an opinion with emphasis of matter paragraph, be it a qualified opinion, be it adverse or be it a disclaimer of opinion. On the contrary, when QAO has the value 0, it means that an unqualified audit opinion has been given. As for the extra variable for the second hypothesis, the extra dummy variable Quantified has the value 1 if we find that the audit opinion is quantified. If it is non-quantified this dummy variable will return 0.

(21)

As for the third and fourth hypothesis, we distinct companies into two categories, namely research & development companies and non-research and development companies. Following,

Kim & Zhang (2016) and Aboody & Lev (2000), a company is recognized as an R&D company when the database COMPUSTAT reports R&D expenditures for the company. This leads to a new dummy variable named RDCAT in our regression model. RDCAT has the value 1 when the company is seen as an R&D company and returns zero when the company is not seen as an R&D company.

(22)

4 Empirical findings

To provide an empirically contrast for the proposed research hypotheses, we provide summary statistics of our data. Followed by a Pearson and Spearman Correlation matrix. After that, we perform our regression analyses and in addition to this we perform an additional robustness test. 4.1 Descriptive statistics

Table 2 summarizes the statistics the determinants of the level of information asymmetry of our initial sample. Panel A of Table 2 reports the number of observations, the mean, the standard deviation, the minimum, the 25th_{percentile, the median, the 75}th_{percentile and the maximum.}

Consistent with prior literature, we see that the mean (median) of the proxies for information asymmetry are: PIN 0.2010 (0.1840), RES 0.0010 (0.0009), PI 0.0095 (0.0040) and ASY 0.0685 (0.0624), suggesting that three of the four proxies of the firm-year observations is symmetric (e.g.,

Brown and Hillegeist, 2007; Abad et al., 2017). Moreover, we see that the PI has a significantly lower value than the RES and PIN variables because the PI is as a less noisy measure of the level of information asymmetry than RES and PIN. As for the market control variables (SIZE, TURNOVER, VOLAT, and ZSCORE), we see from the median and mean that they are symmetric distributed and that these variables show a significant level of dispersion, suggesting heterogeneity in our initial firm-year observations. Panel B reports the dummy variables for audit opinions. We see that 99.2% of the observations are unqualified audit opinions, 92.38% of the audit opinions are audited by the Big 4 and that 49.05% of the firms have research & development costs. Moreover, we see that 66.13% of the modified opinions are not quantified.

In Table 3, we report the correlations between Probability of informed-trading (PIN), Relative effectiveness spread (RES), Price Impact, a mix of these proxies ASY and other market control variables related to the level of information asymmetry. In the upper-right corner of the Table, we present the Pearson product-moment correlation, while in the lower-left portion of the Table we present the Spearman rank order correlations. As can be read from Table 3, the correlation between the proxies is significant at the 10% level. Likewise, the vast majority of the market control variables are significantly correlated with the proxies; this suggests that multicollinearity is not a severe problem.

(23)

Table 2

Summary statistics for the level of information asymmetry and control variables

Panel A: Level of information asymmetry determinants (continuous variables)

No. Obs Mean St. dev. Min. P 25 Median P 75 Max.

PIN 7747 0.2010 0.1305 0.0080 0.1140 0.1840 0.2580 0.8230 RES 7747 0.0010 0.0005 0.0002 0.0005 0.0009 0.0014 0.0020 ASY 7747 0.0685 0.0449 0.0027 0.0391 0.0624 0.0862 0.3074 PI 7747 0.0095 0.0138 0.0000 0.0012 0.0040 0.0127 0.1095 SIZE 7747 8.1483 1.3821 4.4700 7.2061 8.0299 9.0177 11.7255 TURNOVER 7747 13.4615 1.3149 9.3840 12.5231 13.4221 14.3182 16.5339 VOLAT 7747 0.1773 0.2488 0.0019 0.0442 0.0980 0.2011 1.6053 ZSCORE 7747 0.5195 0.4014 -1.7247 0.3073 0.5355 0.7528 1.4616

Panel B: Level of information asymmetry determinants (dummy variables)

No. Obs 0 (%) 1 (%)

QAO 7747 7685 99.20% 62 0.80%

BIGN 7747 590 7.62% 7157 92.38%

RD 7747 3947 50.95% 3800 49.05%

No. Obs Quantified (%) Non-quantified (%)

Opinions 62 21 33.87% 41 66.13%

Notes: The variables used in this table are defined in the Appendix. This table provides the summary statistics for the variables used in the regression model for firm-year observations from the year 2002 to 2016. Panel A reports summary statistics for the continuous variables used to determine the level of information asymmetry for the sample. All continuous variables mentioned in Panel A are winsorized to the 1st and 99th percentiles to improve the normal distribution of the variables. Panel B reports summary statistics for the dummy variables used to determine the level of information asymmetry for the sample. The numbers in panel A and panel B are rounded up to fourth decimal place.

Table 3

Pairwise correlation matrix

1 2 3 4 5 6 7 8 1 RES 0.44* 0.28* 0.33* -0.36* -0.25* -0.27* -0.03*** 2 PI 0.41* 0.41* 0.48* -0.17* -0.36* 0.41* 0.24* 3 PIN 0.29* 0.52* 0.99* -0.19* -0.22* -0.02*** 0.01 4 ASY 0.32* 0.61* 0.99* -0.19* -0.23* 0.02*** 0.03*** 5 SIZE -0.34* -0.10** -0.16* -0.16* 0.78* 0.17* -0.05*** 6 TURNOVER -0.24* -0.24* -0.20* -0.21* 0.79* 0.03*** -0.06** 7 VOLAT -0.15* 0.32* -0.02*** 0.02*** 0.09** 0.01 0.17* 8 ZSCORE -0.04*** 0.16* 0.02*** 0.04*** 0.03*** -0.06** 0.10*

Notes: We define the variables used in this table in the Appendix. The table reports the correlation between the continuous variables used in the regression model for firm-year observations from 2002-2016. All mentioned variables are winsorized to the 1st and 99th percentiles of their distributions. Top right shows the correlation based on the Spearman method, while the bottom left shows the correlation based on the Pearson method.

∗∗∗ indicate significance at the 1% level. ∗∗ indicate significance at the 5% level. ∗ indicate significance at the 10% level.

(24)

4.2 Univariate analysis

Table 4 reports the results from the univariate analysis. In Panel A of Table 4, we determine if there is a significant difference in the level of information asymmetry between firms with a clean unqualified audit opinion and those who have a modified audit opinion by using the t-test and Mann-Whitney U-test. Our results suggest that when the auditor does not give a clean unqualified audit opinion the level of information asymmetry is significantly higher.

Similarly, we determine in Panel B of Table 4 if there is a significant difference in the level of information asymmetry between firms audited by PwC, KPMG, Deloitte or EY and firms audited by other audit firms. Our results suggest that when a firm is audited by one of the Big 4 firms than the level of perceived information asymmetry is lower in the eyes of market participants.

Likewise, we examine in Panel C of Table 4 if there is a significant difference in the level of information asymmetry between firms that have research and development costs and thus opportunity to hide information and firm who do not have this opportunity. Our results suggest that when a firm has the opportunity to hide information through research and development costs the level of information asymmetry as perceived by market participants is significantly higher.

Table 4

Univariate analysis: mean differences of audit opinion, audit quality and R&D costs.

Panel A: mean differences by the opinion of the auditor

QAO No. Obs Mean SD t-test z-statistic

RES 0 7685 0.0023 0.0032 -11.940*** -6.394*** 1 62 0.0072 0.0074 PI 0 7685 0.0145 0.0189 -3.481*** -2.546*** 1 62 0.0178 0.0185 PIN 0 7685 0.2301 0.1493 -1.431 -2.252*** 1 62 0.2573 0.1193 ASYf 0 7685 0.0803 0.0525 -1.799* -2.888*** 1 62 0.0923 0.0408

Panel B: mean differences by the quality of the audit

BigN No. Obs Mean SD t-test z-statistic

RES 0 590 0.0042 0.0048 14.821*** 15.505*** 1 7157 0.0022 0.0030 PI 0 590 0.0003 0.0032 2.303** 2.359*** 1 7157 0.0001 0.0027 PIN 0 590 0.2789 0.1863 8.271*** 6.157*** 1 7157 0.2263 0.1449 ASYf 0 590 0.0985 0.0658 8.755*** 6.739*** 1 7157 0.0789 0.0509

(25)

Table 4

Continued.

Panel C: mean differences by the opportunity to hide information through R&D

R&D No. Obs Mean SD t-test z-statistic

RES 0 3947 0.0025 0.0035 4.409*** 5.500*** 1 3800 0.0022 0.0030 PI 0 3947 0.0145 0.0190 2.001** 1.449* 1 3800 0.0145 0.0189 PIN 0 3947 0.2345 0.1504 2.511** 2.634*** 1 3800 0.2260 0.1477 ASYf 0 3947 0.0818 0.0528 2.398 2.554*** 1 3800 0.0789 0.0520

Notes: the variables used in this table are defined in the Appendix. This table reports the level of information asymmetry by audit opinion (Panel A), audit quality (Panel B) and research and development firm (Panel C). To determine significant differences for each proxy for the level of information asymmetry by the group we use the t-test and Mann-Whitney U-test (z-statistic).

4.3 Multivariate analysis

To examine the relationship between the qualification of an audit opinion and the level of information asymmetry, we estimate the OLS regression described in the model (1). The only difference between the four models is the proxy variable for the level of information asymmetry ASY. Table 5 shows the results of the RES model, PI model, PIN model and ASYf model.

As shown in Table 5, the level of information asymmetry perceived by market participants is significantly higher, when the auditor does not give a clean unqualified audit opinion, under the RES and PI model. However, under the PIN model and the ASYf model we see that even when the level of information asymmetry as perceived by market participants, when the auditor does not give a clean unqualified audit opinion, is higher these two models do not provide credible evidence in favor of this statement. These results are in accordance with our expectations that an unqualified audit opinion leads to a lower level of information asymmetry. The results mentioned in Table 5

are in accordance with our expectations. Even, while we do not find credible evidence for every model that we use to determine the level of information asymmetry. The four models suggest that an unqualified audit opinion leads to a lower level of information asymmetry. The results and adjusted R2_{are also comparable to prior research (e.g.,}_{Brown and Hillegeist, 2007}_;_{Clinch et al.,}

2012; Abad et al., 2017).

Moreover, to examine the relationship between a quantified audit opinion and the level of information asymmetry we estimate the OLS regression by using model (1) with as

(26)

2017), we do not find credible evidence to suggest that a quantified audit opinion is associated with a lower or higher level of information asymmetry. In three of our four regression models we find a statistically insignificant positive relationship between a quantified opinion and the level of information asymmetry. The results are shown in Table 6. The results are not in

accordance with our expectations. Our expectations were that when information is quantified in an audit opinion that the level of information asymmetry would be lower. Our results suggest that in three out of our four models the information asymmetry is higher. A possible explanation for this could be that, even when the audit opinions are quantified and that more information is provided, the report does no longer maintain user readability (Ascioglu et al., 2015). Therefore, we need to examine the readability of quantified audit opinions to conclude if this explanation is true.

Table 5

Level of information asymmetry and the audit qualification

Dependent variable RES model PI model PIN model ASYf model

ASY Coeff. t-stat Coeff. t-stat Coeff. t-stat Coeff. t-stat

Test variable

QAO 0.0040 4.12*** 0.0011 2.17** 0.0093 0.55 0.0056 1.02

Market control variables

SIZE -0.0003 -3.7*** 0.0001 1.11 0.0047 2.11** 0.0023 2.86*** TURNOVER -0.0006 -6.53*** -0.0001 -2.59*** -0.0281 -12.08*** -0.0117 -14.6*** VOLAT -0.0006 -4.25*** 0.0006 1.92* -0.0299 -3.62*** -0.0002 -0.07 BIGN -0.0007 -2.73*** -0.0002 -1.24 -0.0176 -1.59 -0.0062 -1.57 ZSCORE -0.0014 -7.24*** 0.0001 0.32 0.0001 0.02 0.0010 0.65 INTERCEPT 0.0136 13.6*** 0.0017 3.22*** 0.5693 26.26*** 0.0075 29.1***

Industry dummies Included Included Included Included

No. Observations 7,747 7,747 7,747 7,747

Adj. R2 _23.72% _9.00% _7.61% _9.60%

Notes: In this table, we examine the association between the level of information asymmetry and qualification of the audit report in the previous year of the relevant OLS regression. The selected sample consists of 7,747 firm-year observations from the years 2002 to 2016, whereby ASY is the dependent variable. How the sample is selected is mentioned in Table 1 and the definitions of all variables used are presented in the Appendix. The continuous variables are winsorized to the 1st and 99th percentiles of their distributions. Robust standard errors clustered at the firm level in parentheses. The dummy for the industry is based on 2-digit SIC industry classification from CRSP.

(27)

Table 6 Level of information asymmetry and the quantified audit opinion

ASY RES model PI model PIN model ASYf model

Coeff. t-stat Coeff. t-stat Coeff. t-stat Coeff. t-stat

Test variable

QUANTIFIED 0.0004 0.30 -0.0002 -0.37 0.0168 0.82 0.0067 0.98

Market control variables

SIZE -0.0022 -2.83*** 0.0008 1.43 0.0517 2.49** 0.0170 2.26** TURNOVER -0.0006 -0.89 -0.0009 -2.13** -0.0485 -3.04*** -0.0187 -3.39*** VOLAT -0.0013 -0.43 0.0033 1.47 -0.0884 -0.99 -0.0136 -0.48 BIGN 0.0038 1.57 -0.0013 -0.63 -0.0220 -0.34 -0.0072 -0.30 ZSCORE -0.0052 -2.78*** 0.0014 1.80 0.0258 0.72 0.0074 0.60 INTERCEPT 0.0243 3.63*** 0.0084 1.91* 0.6723 4.67*** 0.2613 5.18***

No. Observations 7,747 7,747 7,747 7,747

Adj. R2 _42.54% _26.14% _31.25% _33.30%

Notes: In this table we examine the level of information asymmetry and firms that received a quantified audit opinion in the previous year of the relevant OLS regression. The selected sample consists of 7,747 firm-year observations from the years 2002 to 2016, whereby ASY is the dependent variable. How the sample is selected is mentioned in Table 1 and the definitions of all variables used are presented in the Appendix. The continuous variables are winsorized to the 1st and 99th percentiles of their distributions.Robust standard errors clustered at the firm level in parentheses. The dummy for the industry is based on 2-digit SIC industry classification from CRSP.

∗∗∗ indicate significance at the 1% level. ∗∗ indicate significance at the 5% level. ∗ indicate significance at the 10% level.

Furthermore, to look into the relationship between the level of information asymmetry and the level of private information that a company has measured with research and development costs as proxy, we estimate the OLS regression using model (1) with an addition that all the firms should have research and development costs in their income statement.

As shown in Table 7, the level of information asymmetry perceived by market participants is significantly higher when the auditee has research and development costs. More specifically, we see that RDQAO is significantly associated with a higher level of information asymmetry under the RES and PI model. The PIN and ASYf model do signal that RDQAO is associated with a higher level of information asymmetry but do not find the evidence significant. The results are in accordance with our expectations. Furthermore, these results are also in accordance with research in related segments. For example, Kim and Zhang (2016) found that research and development costs increased the risk of a stock price crash.

(28)

Table 7

Level of information asymmetry for R&D firms

Test Variable RDQAO 0.0025 2.88*** 0.0011 1.84* 0.0131 0.65 0.0071 1.13 Control variables SIZE -0.0003 -3.75*** 0.0001 1.08* 0.0047 2.1** 0.0022 2.86*** TURNOVER -0.0006 -6.56*** -0.0001 -2.61*** -0.0281 -12.08*** -0.0117 -14.6*** VOLAT -0.0006 -4.27*** 0.0006 1.91* -0.0299 -3.62*** -0.0002 -0.07 BIGN -0.0007 -2.76*** -0.0002 -1.23 -0.0176 -1.59 -0.0062 -1.57 ZSCORE -0.0014 -7.29*** 0.0000 0.28 0.0001 0.03 0.0010 0.65 INTERCEPT 0.0137 13.68*** 0.0017 3.27*** 0.5694 26.3*** 0.2168 29.12***

No. Observations 7,747 7,747 7,747 7,747

Adj. R2 22.62% 0.83% 7.61% 9.60%

Notes: In this table we examine the level of information asymmetry for firms that have research and development costs and also received an unqualified audit opinion from the auditor in the previous year of the relevant OLS regression. The selected sample consists of 7,747 firm-year observations from the years 2002 to 2016, whereby ASY is the dependent variable. How the sample is selected is mentioned in Table 1 and the definitions of all variables used are presented in the Appendix. The continuous variables are winsorized to the 1st and 99th percentiles of their distributions.Robust standard errors clustered at the firm level in parentheses. The dummy for the industry is based on 2-digit SIC industry classification from CRSP. ∗∗∗ indicate significance at the 1% level.

∗∗ indicate significance at the 5% level. ∗ indicate significance at the 10% level.

4.4 Robustness analysis

In addition to the multivariate analyses from the previous subsection, we try to control for endogeneity in our findings by doing a robustness analysis. We repeat our multivariate analysis and control for the impact on the results when the auditee is audited by a Big Four firm.

Firstly, we examine the association between the level of information asymmetry as perceived by market participants and the qualification of the audit opinion when the firm that does the audit is a Big Four firm. As shown in Table 8, the coefficient that measures the level of information asymmetry is above the zero. This suggests that the level of information asymmetry increases when the auditor does not give a clean unqualified audit opinion. However, the t-statistics of these coefficients suggest that two out of four of our models for the level of information asymmetry does not find credible evidence to suggest that these coefficient differ significantly.

Secondly, we examine the association between the level of information asymmetry as perceived by market participants and an audit opinion that is quantified. The coefficients in Table 8, suggest that a quantified audit opinion is associated with a lower level of information asymmetry. However, the t-statistics in the four models suggest that there is no credible evidence to suggest

(29)

that a quantified audit opinion is associated with a significantly lower level of information asymmetry.

Thirdly, we look at the relationship between a firm that has received a clean unqualified audit opinion whereby the management can hide information through research and development costs and the level of information asymmetry. As shown in Table 8, the coefficient and t-statistics of the variable RDQAO suggest that there is credible evidence to say that when the management has the ability to hide information through research and development costs and the firms has received a clean audit opinion from one of the Big Four firms, then the level of information asymmetry is significantly higher.

Table 8

Robustness analysis - the impact of high audit quality on previous results

Test variables

QAO 0.0043 4.15 0.0008 1.98 0.0114 0.66 0.0062 1.08

QUANTIFIED -0.0003 -0.23 -0.0001 -0.14 -0.0001 0.04 0.0012 0.18

RDQAO 0.0029 2.95*** 0.0010 1.65*** 0.0242 1.35*** 0.0109 1.90***

No. Observations 7,157 7,157 7,157 7,157

Notes: In this table, we examine the level of information asymmetry for firms that are audited by one of the Big 4 audit firms. We repeat the OLS regressions mentioned in table 5, 6 and 7 with as only difference that we exclude non-Big Four audit firms. The selected sample consists of 7,157 firm-year observations from the years 2002 to 2016, whereby ASY is the dependent variable. How the sample is selected is mentioned in Table 1 and the definitions of all variables used are presented in the Appendix. The continuous variables are winsorized to the 1st and 99th percentiles of their distributions.Robust standard errors clustered at the firm level in parentheses.