Effects of a Joint Audit on Audit Quality and Accounting
Conservatism
A comparison in Denmark
Name: Sander Boom Student number: 10220712
Thesis supervisor: Ir. Drs. A.C.M. de Bakker Date: 24-6-2018
Word count: 12.234
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam
Statement of Originality
This document is written by student Sander Boom 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
This study aims to provide an in-depth research on the relationship between a joint audit engagement on the one hand and audit quality and accounting conservatism on the other hand. The main reason to research this relationship is because the European Commission issued a green paper in 2010 with the possible implementation of a mandatory joint audit environment because it could enhance audit quality and lower the market concentration. This research is performed by using data from Danish listed companies in the period 2000-2015, which provides an unique setting in the sense that the joint audit environment was abolished in 2005. The audit quality is measured by earnings management using the discretionarily accruals and classification shifting. Accounting conservatism is measured by the book-to-market ratio compared to the dummy variable for joint audit. The results of the linear regression models have rejected all three hypotheses because no significant relation exists between joint audit and discretionary accruals, classification shifting and the book-to-market ratio, which means that, in respect to our setting, a joint audit is not associated with higher audit quality and more conservative accounting policies. The main contribution of the paper is to measure audit quality with a larger data sample size and with proxies (classification shifting and book-to-market ratio) that are not yet linked to a joint audit environment in the existing literature.
Contents
1 Introduction ... 6
2 Theoretical framework ... 9
2.1 Joint Audit ... 9
2.2 Positive and negative effects of the joint audit ... 9
2.2.1 Positive effects of a joint audit ... 9
2.2.2 Negative effects of a joint audit ... 10
2.3 Relation between audit quality and independence... 11
2.4 Accounting Conservatism ... 12
2.5 Accrual-based earnings management ... 13
2.6 Classification shifting ... 15
2.7 Measuring Accounting conservatism ... 16
2.8 Hypothesis development ... 18 2.8.1 Hypothesis 1 ... 18 2.8.2 Hypothesis 2 ... 18 2.8.3 Hypothesis 3 ... 18 3 Research Method ... 19 3.1 Conceptual framework ... 19
3.2 Regression per hypotheses ... 20
3.2.1 Hypothesis 1 ... 20
3.2.2 Hypotheses 2 ... 22
3.2.3 Hypothesis 3 ... 25
4 Data ... 26
4.2 Data for hypotheses 2: Measuring classification shifting ... 29
4.2.1 Database and help model for Unexpected Core Earnings ... 29
4.2.2 Other data sources hypothesis 2 ... 30
4.3 Descriptive statistics hypothesis 1 and 3 ... 31
4.3.1 Core statistics ... 31
4.3.2 Correlation and multicollinearity ... 32
4.4 Descriptive statistics hypothesis 2 ... 34
4.4.1 Core statistics ... 34
4.4.2 Correlation and multicollinearity ... 35
5 Regression analysis per hypothesis ... 36
5.1 Regression model hypotheses 1 ... 36
5.2 Regression model hypotheses 2 ... 38
5.3 Regression model hypotheses 3 ... 40
6 Conclusion and limitations ... 42
1 Introduction
After the financial crisis and its global impacts, many regulatory bodies issued papers containing possible measures to improve audit quality. For example, in the Netherlands the NBA published ‘In publiek belang’ to improve independency and quality. On a European level the EC (European Commission) issued ‘Audit policy: Lessons from the crisis’ (EC, 2010). They were particularly concerned with the risks of the high market concentration and the related accounting scandals (Enron). Another loss of one of the biggest accounting firms in the world would restrict competition to an unreasonable low level (Lesage et al., 2013). One of the possible solutions for this problem is, according to the European Commission (EC, 2010), to perform an audit in a joint engagement between audit firms. This (re)started the debate on whether a Joint Audit environment could improve the quality of the profession, with two major party’s responding to the green paper of the European Commission, the Big 4 firms and the so called 2nd Tier firms (Kettunen, 2012).
The Big 4 firms criticized the mandatory joint audit because it would increase the audit costs, while the 2nd Tier firms (for example BDO and Grant Thorton) argued that a joint audit
could improve audit quality (Kettunnen, 2012). This contradiction led to several studies researching the effect of the mandatory joint audit on audit fees, audit quality and market concentration. This study focuses on claim of the 2nd Tier firms of the improved audit quality
regarding joint audits.
Prior research on joint audits has indicated that a joint audit with two ‘Big 4’ auditors reported the lowest amounts of accruals and are less aggressive towards earnings management (Vanstraelen et al., 2009). They conducted a research within the scope of French companies where the joint audit is still mandatory. Although, Lesage et al. 2012, find in their study that a mandatory joint audit environment is not associated with significant less accruals in Denmark in the period 2002-2010, where the joint audit was mandatory until 2005, Zerni et al. 2012, finds that a joint audit leads to a higher degree of earnings conservatism and auditor independence in the form of perceived pressure in client-auditor negotiations on accounting choices, thus leading to higher audit quality. The research of Zerni et al., 2012, was conducted in a voluntary joint audit setting with Swedish firms. This could also contain the effect that high quality financial reporting firms are using a joint audit as a signal to the market about their reporting methods and therefore missing the causality.
characteristics are the most important determinants of audit quality and that restatements are the most useful signals of low audit quality. Also, Krishnan and Visvanathan (2008) use accounting conservatism to measure financial reporting quality as they consider accounting conservatism as the most influencing principal in valuating accounting. This means that the prior research in audit quality and joint audits can be incomplete in terms of defining audit quality as they primarily focuses on abnormal accruals. Therefore the research question is:
What is the effect of a mandatory joint audit environment on accounting quality and accounting conservatism?
This study will contribute to the existing literature in several ways. First, although other studies have already researched the effects of a joint audit on audit quality, they all used the same proxy to measure audit quality: discretionary accruals. According to Guay et al. (1996), discretionary accruals can indeed be caused by managerial opportunities, but can also be used to improve performance measurement. Also, earnings are not only managed by using accruals but can also be accomplished by using classification shifting (McVay, 2006). Therefore, other proxies of audit quality should be examined to provide an in depth insight of how a joint audit can improve audit quality. Classification shifting as a proxy for audit quality and the link with a joint audit environment has not yet been researched in the current literature.
Second, former studies combined their joint audit research on audit quality with audit fees and market concentration, provided a broad understanding of the effects of a joint audit. This study aims to provide a more in-depth paper by researching the effects of a joint audit on audit quality and accounting conservatism by using different proxies, from financial information to investors’ perceptions of audit quality. For example, accounting conservatism, in the current literature, is only researched within the scope of voluntary joint audit environments.
At last, because of the influence of time, a larger and more relevant data sample can be used to measure the period after the joint audit was mandatory in Denmark (until 2005). This is necessary because prior research all used data samples with a maximum of a 5 years of the ‘single audit period’. This contains an important limitation, it includes the ‘transition phase’ from a joint audit to either a voluntary joint audit or a single audit. This voluntary joint audit/single audit ratio is expected to be closer to 50/50 in 2005 than in 2016. This means that, with the larger data sample, the effect on audit quality in the long-run can be measured.
The study will provide positive or negative arguments on whether a country should adopt mandatory joint audit structure to improve the audit quality. Therefore, not only regulatory bodies, but also stakeholders (investors), who rely on the quality of annual reports, will benefit from the study.
The structure of the thesis is as follows. Section 2 will provide the theoretical background. Section 3 and 4 will cover the research method and the data (including descriptive statistics). Finally, section 5 and 6 contains the regression analysis, the interpretation of the results and the conclusions.
2 Theoretical framework
The theoretical background of this paper will provide an explanation of the joint audit, the pros and cons of a joint audit, the theory supporting the effect of joint audit on audit quality and the hypotheses.
2.1 Joint Audit
‘An enterprise which is under an obligation to prepare an annual report in accordance with the rules for
reporting class B, C or D must have its financial annual report audited by one or more auditors. The audit shall not comprise the supplementary reports forming part of the annual report, cf. section 2(2)’ (Danish Act no. 448 of 7 June 2001, p.34).
According to the above equation, the joint audit in Denmark was mandatory from 1930 until the abolishment in the Act of 2001 (effective since 2005). Every listed company during this period in Denmark must be audited by at least two different firms. In practice, this means that the two accounting firms divide their work, perform a second review on the work done by the other firm and both sign the annual report. However, if misstatements are found the responsibility is not necessary equal, according to the following quotation:
‘Fault, if found, may not be at the same level for each audit firm. For example, if the audited inventory is materially misstated, then the auditor who is responsible for inventory could be held more responsible than the other auditor, who simply reviewed the work’ (Deng et al., 2012, p.7).
2.2 Positive and negative effects of the joint audit
The introduction of a mandatory joint audit regime could have some positive and negative effects towards audit quality and audit costs. As stated in the introduction, the Big 4 firms argue that the audit costs will rise and the 2nd Tier firms argue that the joint audit will increase audit
quality. The effects in the existing literature will be summarized below to provide a clear overall view of the joint audit implications
2.2.1 Positive effects of a joint audit
The positive effect of a joint audit is theoretically based on two assumptions of audit quality: independence and competence (Ittonen and Tronnes, 2015).
A joint audit agreement between two accounting firms will increase the independency of the auditor in several ways. First, according to Zerni et al., 2012, two different auditors on a single assignment would result in less pressure towards managers and stakeholders and they will
report their opinions more independently This is strengthened by the European accounting environment, which is largely in accordance with the principal-based IFRS, where more judgment is required. Second, a related benefit is that, for allowing misstatement/misjudgments in the annual report, a collaboration of, not two, but three parties is necessary. The penalty for doing so would more likely exceed the potential benefit due to the shared audit fee (Zerni et al., 2012). Also, when such situations occur, the cost to ‘bribe’ the audit firms to signs the report will be higher. Third, the shared benefit of the audit fee means that the audit firm is in fact more independent from its client because it’s reliability, in terms of real numbers, decreases. Finally, Zerni et al. (2012), found that the users of financial statements perceive a higher audit quality in a joint audit compared to a single audit due to the increased independency. They measured the perceived quality by credit ratings and risk forecasts.
The joint audit will increase the competence not only because ‘two can see more than one’ but also because the auditors have more pressure to justify their assessment on reporting issues (Ittonen and Tronnes, 2015). Also, the experience and the expertise of two audit firms instead of one will be higher, which will result in improved task performance. This argument only holds when the two auditors complement each other.
2.2.2 Negative effects of a joint audit
Although related to improved audit quality by the 2nd Tier firms, other studies have found
possible downsides of a joint audit.
Deng et al. (2014), find in their study that the cooperation of two audit firms can induce two downside effects: free riding and opinion shopping. The free riding problem can occur when one firm holds on to its resources and take advantage of the work done by the other audit firm. Opinion shopping is an opportunity for the firm’s management to choose between the most favorable opinions in every accounting related issue that occurs during the audit.
However, the work done by the two audit firms must be divided equally and must be reviewed by one another by law. Therefore, if the free riding problem occurs in one year it is most unlikely that the firm with the most effort will still engage in the joint audit the next year. Furthermore, the two firms must sign the annual report but are most likely not equally responsible for misstatements. Therefore free riding can be a problem for the free riding firm when misstatements due to a lack of work are found.
Second, the increasing audit costs in a joint audit environment is researched in several studies. Zerni et al. (2012), found in their setting in Sweden that voluntary joint audits are paid
to higher costs because in Denmark, were the mandatory joint audit was abolished in 2005, companies who switched from a joint audit to a single audit received a fee discount in the next year. This is supported by Andre et al. (2012), who performed a cross country research of France, Italy and the UK and found higher audit fees in France related to the other countries.
However, Zerni et al, 2012, concluded that the premium paid by firms are related to the highest value of the audit. This means that a firm, under free market conditions, is willing to pay more for higher audit quality and the potential benefits: better credit ratings and risk forecasts and lower cost of capital.
Given the potential costs and benefits for a joint audit environment, the existing literature associates a higher independence for auditors in a joint audit compared to a single audit. Therefore the theory of independency will be the starting point to develop the hypotheses. 2.3 Relation between audit quality and independence
In order to develop an understanding of the potential increase in audit quality, several theories of audit independency must be evaluated. The starting point for the relationship is the agency theory. This theory describes the relationship between the principal and the agent. Management, who manage the company on a daily basis, can have different incentives towards providing information as do shareholders (the owners of the company) need. For example, financial statements are used by shareholders to evaluate managements performance, therefore management has an incentive to present the information in their own favorable way (Antle, 1984). The misrepresentation of information is also known as ‘earnings management’.
This information problem arises when the owners and investors cannot distinguish good from bad information. Hence, the capital market will most likely undervalue good investments and overvalue bad investment. This information asymmetry in the principal-agent structure leads to the need of information intermediaries who confirm the provided information by management (for example by signing the annual report) (Healy and Palepu, 2001). Therefore, in the basic theory, in order to provide a quality report, the auditor must be independent from the firm’s management.
The relationship between auditor independence and audit quality is examined in several studies. For example, Carey and Simnett (2006), concludes that long auditor partner tenure is associated with a reduction in audit quality. Also if auditors provides other services besides the audit, perceived audit quality decreases due to compromised objectivity and skepticism (Francis, 2004). This means that if joint audits can improve auditor independence in a different way, it has the potential to increase audit quality.
2.4 Accounting Conservatism
While audit quality related to earnings management is the focus of interest in prior literature, audit quality can also be measured using the principles of accounting conservatism. Most Generally Accepted Accounting Principles are based on conservatism, that is, to record expenses and liabilities as soon as possible and to record revenue and assets only if they will certainly occur. When the auditor is both more competent and more independent in a joint audit engagement, it is more likely the financial statements are prepared in accordance with local GAAP and therefore, contain more conservative accounting. The link between audit quality and accounting conservatism is described below.
Prior literature associated the extend of using accounting conservatism with high quality financial reporting because of the sustainability of the earnings. According to Penman & Zhang (2002), this only holds if the accounting principle ‘conservatism’ is used consistent over time. Current period earnings could be used as a predictor of the current share price because it provides information about the expected future earnings (Nichols and Wahlen, 2004). If earnings are more sustainable it will be a better predictor for the future earning and the current share price and therefore be of higher quality.
Also, conservatism is associated with audit quality because it benefits the users of the firm’s financial reports in several ways (Watts, 2003a). First, it constrains the opportunities for management to add bias and noise in the information provided to investors because it is asymmetric towards verifiability (‘anticipate no profit’). Second, according to Krishnan & Visvanathan (2008) and Watts (2003a), there is more litigation risk towards overstatement of assets than understatement and the litigation risk has increased since the recent accounting scandals worldwide.
Because a mandatory joint audit increases the competence (section 2.2.1) and the independence of the auditor it could improve audit quality and have a positive effect on accounting conservatism. While accounting conservatism is a fundamental characteristic of the financial reporting process (Krishnan & Visvanathan, 2008) and that two auditors have more expertise and more pressure to justify their accounting issues, it is more likely that a firm, when audited by two or more audit firms, reports conservatively.
2.5 Accrual-based earnings management
If audit quality is improved by the level of auditor independence and a joint audit is related to higher auditor independence, higher audit quality can be measured at firms who have their audit performed by two audit firms. Many studies have aimed at operating audit quality.
For example, the level of audit failure in a specific setting can be used as a proxy to measure audit quality (Francis, 2004). Audit failure occurs when accounting principles are not met in the annual report and when an auditor fails to issue an adverse or qualified opinion. Christensen et al. (2015), have found similar results. They stated that restatements are the most available signals of low audit quality. Francis (2004), concludes that legal procedures against auditors is the best measurement for audit failure. Due to lack of data, the amount of restatements of financial reports in Denmark as a proxy for audit quality cannot be hypothesized. However, another proxy to determine the audit quality is the ability of the auditors to detect and correct earnings management. This is in line with the positive effect of a joint audit that two auditors can perform more pressure towards management in accounting related issues because they are more competent and more independent from the client. Earnings management is most frequently tested by the discretionary accruals to measure accrual-based earnings management.
The existing literature provides different models on how to measure accrual based earnings management. For example, Dechow et al. (1995) tested five different models to measure discretionary accruals (Healy Model, DeAngelo Model, Jones Model, Industry Model and The Modified Jones Model) and concluded that the best practice on measuring earnings management comes from the Modified Jones Model. The reason for this is that the Modified Jones Model takes the changes in net receivables into account. Therefore, the modified model implicitly assumes that the delta in credit sales is associated with earnings management while the other models assume that management choices are not reflected in revenue (Dechow et al., 1995). Despite that Islam et al. (2011) tested the model in a developing economy and found that it is not effective in measuring earnings management through discretionary accrual it can still be used in a developed setting like Denmark. The researchers acknowledge that the model has explanatory power in mostly developed countries.
The model itself tests the amount of discretionary accruals in relationship to the total accruals of a company in a specific year. This distinction between discretionary and non-discretionary accruals is made because, although accruals can be used by management to manage their earnings, there are also accruals that arise during the normal course of business. For example, accrual accounting is a tool for organizations to meet the matching principle, which
means to report revenues in the same period as the related expenses. The latter form of accruals are categorized as non-discretionary. These accruals helps management to match the company’s earnings to their true economic value and therefore increase earnings quality (Krishnan, 2003). The downside is that the communication of inside information through earnings is susceptible to aggressive reporting by managers to, for example, meet analyst forecasts and/or their own compensation plans. The ability to detect and correct the discretionary accruals by the auditor is therefore associated with audit quality.
In this research the Modified Jones Model, as introduced by Dechow et al. (1995), is followed to measure the discretionary part of the total accruals and the extent of earnings management in annual reports in relation to joint audits in Denmark. The following equations are used to measure the discretionary part of the total accruals. First total accruals is measured by the first equation (for an explanation on the measurement of the variables below, refer to table 1).
(1) Second, the non-discretionary part of the total accruals are calculated by the second equation and the regression model to estimate the year specific parameters:
(2)
The above parameters , and are estimated by the Ordinary Least Squares (OLS) method in the following regression model:
(3)
From the above calculations we can then derive the discretionary accruals from and as:
(4)
In this research, the Modified Jones Model is used to determine the effect of a joint audit on the amount of discretionary accruals. This regression, as well as an explanation of the control variables and the variables used in the above equations , will be discussed in detail in section 3.
2.6 Classification shifting
As mentioned above, in the existing literature audit quality is often measured based on the degree of earnings management in accounting information because, as noted in section 2.3, management has an incentive to do so. The potential benefit of a joint audit, that is increasing auditor independence, can mitigate the information asymmetry between principal and agent. Prior literature regarding joint audit and audit quality all measured the accrual-based earnings management: the degree to which management used abnormal accruals to manage the company’s earnings. There is, however, another way for management to manipulate earnings, called classification shifting (McVay, 2006). This implies that revenue in the income statement is shifted upwards (and expenses are shifted downwards) to present higher ‘core earnings’. There are several reasons for management to prefer classification shifting above accrual-based earnings management:
First, prior research indicated that ‘Sales’ is the most permanent item on the income statement (Lipe, 1986) and that, the closer an item is to sales, the more permanent it is expected to be. This distinction in the income statement is also recognized by investors, as they value an increase in core earnings higher than an increase in special items (McVay, 2006). Therefore, management can use classification shifting as a tool to manage their perceived performance.
Next, classification shifting doesn’t change GAAP net income because it only is a shifting in the income statement itself. This limits the scrutiny of auditors (Nelson et al, 2002) and it doesn’t reduce future earnings because there is no reverse effect of an abnormal accrual.
To measure classification shifting, the research model of McVay (2006) is followed. This model calculates Core Earnings based on Sales – Cost of Goods Sold – Selling, General and Administrative Expenses with the exclusion of Depreciation and Amortization, all divided by Sales. The reason behind this model is to measure the shift in expenses in the income statement when a firm reports a special item. McVay (2006) finds that if a special item is reported, management tends to shift core expenses to special items. This leads to an increase in core earnings due to earnings management. Because of limited time and scope to conduct this research, we calculate a simplified version of McVay’s model. This means the exclusion of the delta unexpected change in core earnings (the reverse effect of the earnings management) because she intended to research if classification shifting is used by managers in the first place. In this research however, the aim is to only measure if there is any change before or after the period of a joint audit in combination with classification shifting. Therefore, the reversal in next period is excluded.
certain control variables are added to the model. To make sure the relation between joint audit and classification shifting is measured by the core earning, there must be controlled for Sales, Accruals, Change in Core Earnings (core earnings are persistent over time) and Asset Turnover Ratio (ATO) (related to profit margin) as they could all have an impact on core earnings. These are the same control variables as used in the study of McVay (2006).
The model itself tests if there is any change in the difference between the reported core earnings and the predicted core earnings in relation to the special items. Unexpected core earnings is measured as follows:
Where core earnings is calculated by: (Sales – COGS – S,G&A expenses)/Sales with the exclusion of depreciation and amortization. Expected core earnings is measured by the following regression, which means that the residual is equal to the unexpected core earnings (for a detailed explanation of the measurement of the variables below, see table 3 in section 3.2.2).
So, if managers use special items to shift their core expenses downwards and therefore increase their core earnings (which are unexpected), than the special items are positively related to the unexpected core earnings. The regression models as well as the necessary calculations to test hypothesis 2 are described further in section 3.
2.7 Measuring Accounting conservatism
Next, prior literature associated the extend of using accounting conservatism with high quality financial reporting (Krishnan & Visvanathan, 2008). For example, Penman & Zhang (2002) linked high quality earnings with accounting conservatism. Earnings in year t are considered of high quality when they can be used as an indicator for next year’s earnings (t+1). These ‘sustainable’ earnings can only be reported if it is associated with less discretionary accruals because accruals have a positive effect on volatility. An increase in volatility will lead to lower predictive value of the earnings and therefore, have a negative effect on the quality of earnings. For this reason, the same control variables as in hypothesis 1 apply on measuring accounting conservatism.
companies are more likely to report their financial statement conservatively (with less abnormal accruals, as measured in prior literature related to joint audits).
In prior literature, accounting conservatism is calculated in several ways. Many studies used the abnormal accruals, as they indicate whether or not the information provided by management is opportunistically biased. Because abnormal accruals are already tested, this study aims to calculate accounting conservatism in a different way to measure potential benefits in reporting quality and to provide a different perspective on the effect of joint audits.
According to Krishnan & Visvanathan (2008), accounting conservatism can be measured by using the book-to-market ratio. This is distracted from Basu (1997), who noticed that conservatism leads to the reducing of earnings to ‘bad news’ but not increasing earnings to ‘good news’. Therefore, if a firm is reporting conservatively, the book value of the net assets should be lower than its market value. The measurement of book-to-market ratio (BTM) is separated in a bias and lag component by Beaver & Ryan (2000). They find in their study that only the bias component (book value is persistently lower than market value) is associated with measures of accounting conservatism and that the lag component (unexpected losses are recognized in book value over time and not in the period they occur) is not associated with conservatism. Therefore, if the book to market ratio (BTM) is lower in a joint audit environment, the overall reporting tend to be more conservatively. The dependent variable is calculated as follows:
The book-to-market ratio can then be analyzed in the pre and post joint audit phases of a company.
2.8 Hypothesis development
From the theoretical framework the following hypotheses are derived.
2.8.1 Hypothesis 1
Although other studies regarding joint audit and audit quality also used accrual-based earnings management as a proxy for audit quality, this study aims to provide a more recent (and therefore more relevant) data set to test the following hypothesis:
H1: A joint audit has a negative effect on accrual-based earnings management
2.8.2 Hypothesis 2
There are several reasons for management and auditors to prefer classification shifting above accrual-based earnings management (as described in section 2.6). Therefore, in this study classification shifting is used as a proxy for audit quality, which leads to hypothesis 2:
H2: A joint audit has a negative effect on classification shifting
2.8.3 Hypothesis 3
If the audit quality in a joint audit environment is higher, the companies are more likely to report their financial statement conservatively (with less abnormal accruals, as measured in prior literature related to joint audits). Also, from section 2.4, conservative accounting is in the best interest of the users of financial reports. Therefore, the final hypotheses will be:
3 Research Method
This section contains the research method which will operate the variables to measure audit quality, classification shifting and accounting conservatism.
3.1 Conceptual framework
To provide a clear oversight of the hypotheses that will be tested, a conceptual model is provided per hypothesis:
Hypothesis 1: Joint audit negatively (-) related to accrual-based earnings management
Hypothesis 2: Joint audit negatively (-) related to classification shifting
Hypothesis 3: Joint audit positively (+) related to accounting conservatism
Independent Variable Joint Audit Dependent Variable Accrual-based Earnings Management Control Variables Audit firm Firm Size Financial crisis Leverage Growth Independent Variable Joint Audit Dependent Variable Accounting Conservastism Control Variables Audit firm Firm Size Financial crisis Leverage Growth Independent Variable Joint Audit Control Variable Financial crisis Dependent Variable Classification Shifting
3.2 Regression per hypotheses
In this section the regression models to test hypothesis 1, 2 and 3 are described in detail.
3.2.1 Hypothesis 1
Recall from paragraph 2.8 that hypothesis 1 is:
H1: A joint audit has a negative effect on accrual-based earnings management
In this research the Modified Jones Model is used to measure accrual-based earnings management by the amount of discretionary accruals. To test whether the discretionary accruals are negatively related to a joint audit the following regression models is used:
The calculations of the variables in the above regression model are described below, with first the depended variable and then the independent and control variables.
Dependent variable
The dependent variable in this regression is the amount of abnormal or discretionary accruals which functions as an indicator for whether management of an organization is practicing earnings management. Because earnings can be managed either upwards or downwards, depending on the incentives of those who are responsible for practicing earnings management, the variable is calculated as an absolute value. The discretionary accruals are calculated by the equations derived from the Modified Jones Model. For the equations used in this model, refer to section 2.5, equations (1)-(4).
Table 1 summarizes the calculations to measure the needed variables for the discretionary accruals.
Table 1: Calculating the dependent variable for hypothesis 1 Variables used to
calculate
Measurement
Revenue in year t minus revenues from year t – 1. Scaled by lagged total assets
Net receivables in year t minus net receivables in year t – 1. Scaled by lagged total assets
Gross Property, Plant and Equipment in year t. Scaled by lagged total assets
Lagged assets, calculated by total assets of the firm at t - 1
Year-specific parameters, calculated by the OLS regression model as
described in section 2.5
Total accruals in year t divided by the lagged total assets. Where total accruals are calculated by earnings before extraordinary items minus cash flows from operations
Independent test variable
The independent test variable in the regression model to analyze hypothesis 1 is the joint audit . This variable is measured as a dummy variable, whereby 1 indicates a company who is audited by at least two firms and 0 if only one firm audited the company’s annual report. For the collection and distribution of the data, see section 4.
Control Variables
The control variables are the variables which could have an effect on the dependent variable as shown in prior research. In this research discretionary accruals are controlled for whether a audit is perform by at least one ‘big4’ audit firm, the financial crisis, the firms size, the amount of leverage and the growth in sales.
According to Bengston (2011), the financial crisis may have an influence on the data set in the period of 2007-2009. Therefore the dummy is added with 1 indicating the financial crisis and 0 otherwise.
management. In their research they used the audit firm as a proxy for audit quality because they should have more competences available to audit complex annual reports. The relationship with discretionary accruals is controlled for size and leverage. They concluded that the size of the audit firm had a negative relation with earnings management. Therefore, in this research, earnings management is controlled for size of the audit firm, leverage and size.
In a more recent study (Carey & Simnett, 2006) earnings management is controlled for growth and also leverage. For this reason in this regression model we add growth as the final control variable. In table 2 the calculations for the independent variables are summarized. Table 2: Calculating the independent variables for hypothesis 1
Dependent variables Measurement
Variable (dummy) whereby 1 equals joint audit and 0 equals audit performed by only one audit firm
Variable (dummy) whereby 1 equals the audit company is a ‘Big4’ company and 0 equals otherwise
Variable (dummy) whereby 1 equals the period during the financial crisis and 0 equals the other period
Natural logarithm of assets in year t: Total liabilities divided by total assets at year t
Assets in year t minus assets in year t - 1 scaled by lagged total assets:
3.2.2 Hypotheses 2
To test earnings management, in the form of classification shifting, the following hypothesis is tested:
H2: A joint audit has a negative effect on classification shifting
regression is already controlled for economic performance by predicting unexpected core earnings in section 2.6):
Dependent variable
The dependent variable, unexpected core earnings in year t, is calculated by the following equation:
Where the predicted core earnings is measured by the regression model from section 2.6 (p.16). The unexpected core earnings equals to the standard error . In table 3 the calculations to measure the dependent variable ‘unexpected core earnings’ are summarized.
Table 3: Calculating the dependent variable for hypothesis 1 Variables used to
calculate
Measurement
Core earnings in year t measured by (Sales – COGS - S,G&A expenses)/Sales. Depreciation and amortization excluded.
Core earnings in year t – 1, calculated as above
Asset Turnover Ratio: Sales / ((NOA in year t minus NOA in year t -1)/2). With Net Operating Assets calculated as Operating assets – Operating liabilities
(Net income before extraordinary items – Cash from operations) / Sales Accruals as calculated as above at year t - 1
Change in sales measured by percentage:
Delta sales as calculated above when percentage equals 0 or lower
Independent variable
The independent variables in the regression model to test hypothesis 2 are joint audit and special items . Joint audit is the same dummy variable as used in hypothesis 1, with 1
equals a joint audit and 0 a single audit. The second independent variable, special items, is calculated by McVay (2006) as follows:
Note that special items are multiplied by -1 because only income decreasing special items are included in de model. The reason for this is that the model implicates that managers only tend to shift between core expenses and special items if the special items are income decreasing. If, for example, a special item increases income it wouldn’t provide an opportunity for managers to shift core expenses downwards. So, to include positive percentages in the regression model the items are multiplied by -1 and the positive items are set to 0 as in the model of McVay (2006). Special items are retrieved from DataStream and doesn’t include any further calculations.
The control variables are already used in the regression model to calculate the unexpected core earnings. The following quotation shows the intention of these variables:
This model attempts to control for economic performance as well as for macroeconomic and industry shocks (McVay, 2006, p.511).
Therefore, the variables, as mentioned in hypothesis 1, to control for accrual-based earnings management are not included in the regression model to test hypothesis 2, except for the dummy variable , which is added because of the selected time period. The dummy is calculated in the same way as in the regression model for hypothesis 1.
3.2.3 Hypothesis 3
Recall from section 2.7 that hypothesis 3 is:
H3: The mandatory joint audit is positively related to accounting conservatism The regression model for testing H3 is as follows:
Dependent variable
The dependent variable ‘book-to-market ratio’ is calculated by dividing the book value of a company by its market value. The book value of a company is measured as the total shareholders’ equity reported on the balance sheet whereas the market value of a company is measured by its market capitalization, which is the share price in year t ( ) * total number of shares outstanding in year t ( ).
Independent variable
The independent variable is the same as used in hypothesis 1 and 2, namely as a dummy for a joint or single audit. Because accounting conservatism is often calculated the same as accrual-based earnings management (by discretionary accruals), the same control variables used for earnings management also apply to accounting conservatism. Therefore accounting conservatism is controlled for whether the audit of the annual report is conducted by a ‘Big4’ audit firm, the financial crisis, the firms size, the amount of leverage and the growth in year t relative to year t – 1. Refer to table 2 (section 3.2.1) for a summary of the calculations used to measure the independent and control variables.
4 Data
In this section the necessary data will be collected to perform the analysis and answer the research question. The research will be performed using public available data of Denmark in the period 2000-2015 (see different periods for the hypotheses).
The mandatory joint audit was effective until 2005 and therefore the dataset will be distributed among this year. The years 2016/2017 are not included because not all information will be available due to the time period of the research. If, however, the period contains much more joint audits than single audits, the time frame needs to be adjusted.
Data will be retrieved from Thomson Reuters Worldscope, which is available through DataStream. This database contains circa 70 companies in Denmark with the code LCOSEASH This dataset contains all non-financial public listed firms on the Denmark stock market (OMX Copenhagen Stock Exchange). Financial firms are excluded because of their heavily regulated market and additional reporting requirements.
In the remaining paragraphs of section 4, the descriptive statistics for the three hypotheses are shown. Because the regression models for H1 and H3 (accrual-based earnings management and accounting conservatism) have the same independent variables, they are discussed together in 4.3. The descriptive statistics for hypothesis 2 are shown in section 4.4
4.1 Data for hypotheses 1 and 3
4.1.1 Database and Modified Jones Model
Data to provide the necessary financial information to calculate the dependent and independent variables is retrieved from DataStream in the period 2001-2014. The period is selected based upon a 5 year transition phase, starting from 2005, where companies tend to move from a joint audit to a single audit. The period before 2005 contains 4 years to provide sufficient data of joint audit companies. This includes a total of 918 Danish listed companies on the OMX. Recall from section 2.6. that multiple regressions are performed to provide the year specific parameters
for hypotheses 1. The regressions are performed per year.
The original Modified Jones Model calculates the specific parameters by year and by industry (Becker et al., 1998). The data sample of Denmark is simply too small to perform year specific and industry specific regressions. Categorizing only by industry will cause a different problem, namely that total accruals of an industry in year 2014 will effect discretionary accruals
excluded. For example, the financial industry is known for its high regulation and therefore, SIC codes 6000-6999 are omitted from the data sample (Becker et al., 1998).
After excluding certain industries and calculating the necessary variables, the data sample contains 754 samples. This means that 164 samples are deleted due to missing data. The year specific parameters are then used to calculate for the whole data sample. Total accruals and the non discretionary accruals are then used to provide the dependent variable ), see table 4 for the results.
Table 4: Year specific parameters using the linear regression model in 2.5 (equation 2)
Year 2000 -8344.253 0.075 -0.167 2001 -1659.358 0.359 -0.009 2002 14777.190 0.390 0.000 2003 -2035.704 -0.075 -0.009 2004 -3310.304 0.055 -0.041 2005 -1586.068 0.066 -0.043 2006 1409.807 0.096 -0.033 2007 -266.246 0.134 0.065 2008 3114.974 0.189 0.011 2009 433.147 0.035 -0.021 2010 -1285.755 0.058 -0.042 2011 -1277.779 0.114 -0.053 2012 2692.911 0.077 0.009 2013 -2718.382 0.097 -0.007 2014 14365.260 0.067 -0.038 2015 5891.133 0.046 -0.048
4.1.2 Other data sources hypothesis 1 and 3
Besides the collection of all the variables from the database DataStream and the regressions from the Modified Jones Model to calculate the discretionary accruals, the dummy variables need to be collected manually due to missing data. Also, the data need to be adjusted for specific outliers, who may influence the regression significantly.
Outliers
According to Kothari et al., 2005, data trimming on outliers is most common for the 1st and 99th
percentile of the whole data sample. This means to downgrade and upgrade samples sizes to the values of the next percentile (called winsorizing). In this study winsorizing is chosen to deal with specific outliers because of the limited data of listed Danish companies, instead of simply
deleting the outliers.
Because the 1st and 99th percentile still provides minimum and maximum values who
substantively differ from the total sample size, the data is winsorized at the 5th and 95th
percentile. The total data sample selection for H1 and H3 therefore contains a sample size of 754 (summarized in table 5).
Table 5: Sample size hypothesis 1 and 3
Denmark (OMX) 2001 - 2014
Total Observations 918
Less: Firms with insufficient data 164
Total data sample for H1 and H3 754
Winsorizing (percentile) 5-95
Manually collected data
Annual reports are used to subtract information about a joint audit manually. Because the mandatory joint audit environment changed to a voluntary one, there will be firms who remain audited by two companies after 2005. Therefore, the annual reports will provide the dummy variable in the regression models from section 3. The annual reports are retrieved from either the company’s website or from ‘quicktake.morningstar.com’, which contains investment and analyst information collected by Morningstar Inc.
Furthermore, missing data on the dummy variable will also be filled by these annual reports. In total, 266 annual reports have been checked manually for whether a joint audit or a single audit is performed, in which year the companies switched from a joint audit to a single audit and by which audit firm the reports where signed (for the distribution of the dummy variables, see table 6). The dummy variable regarding joint audit equals 1 if there is a joint audit, else the dummy equals 0. The dummy variable regarding ‘big4’ equals 1 if the (leading) auditor is a ‘big4’ audit firm, else the value is 0.
At last, the dummy variable is added because of the time period of the data sample. The dummy is added with 1 indicating the financial crisis (all samples in year 2007, 2008 and 2009) and 0 otherwise.
The total sample distribution for the dummy variables is displayed in table 6, see below. From this distribution we see that without collecting the dummy joint audit manually a total of 68 samples would be categorized wrong (from 2005 all companies would have been given a 0 for
65% single audits. At last, almost all (95%) listed companies in Denmark are audited by at least one ‘big4’ audit firm in the time period of 14 years.
Table 6: Distribution of manually collected data
Year 2001 43 42 0 2002 48 47 0 2003 51 50 0 2004 52 51 0 2005 15 50 0 2006 13 50 0 2007 12 52 56 2008 9 52 56 2009 5 53 57 2010 4 55 0 2011 4 58 0 2012 3 57 0 2013 2 51 0 2014 1 50 0 Total 262 718 169
4.2 Data for hypotheses 2: Measuring classification shifting
4.2.1 Database and help model for Unexpected Core Earnings
Like the data sample of H1 and H3, the sample is retrieved from DataStream. For hypothesis 2 the sample is extended to the period 2000-2015 to increase the data sample size, but not further than the year 2000 because of data irrelevancy. The sample contains a total of 1367 Danish listed companies on the OMX.
Like the study of McVay (2006) and Haw et al. (2011) the unexpected core earnings is calculated by predicting the core earnings based on the regression model in section 2.6. The regressions are performed per year, while McVay (2006) performed the regression per year and per industry. The same explanation for the Modified Jones Model, as noted in section 4.1.1, applies to this regression model, namely due to limited data and time the regression are only
performed per year.
Again the financial industry (SIC code 6000-6999) is excluded and, due to missing data to calculate the necessary variables, another 644 firms were deleted from the sample size. The regressions, as tabulated below (table 7), are measured to estimate the expected core earnings , which is then used to calculate the unexpected core earnings (for the formula see section 2.6)
Table 7: Year specific parameters using the linear regression model in 2.6 Year 2000 0.010 0.919 0.000 0.052 -0.057 0.001 0.000 2001 -0.026 1.212 0.000 -0.710 1.755 -0.338 0.072 2002 -0.004 1.025 0.000 -0.062 0.208 0.041 0.008 2003 0.035 0.860 0.000 -0.166 0.048 -0.056 -0.005 2004 -0.001 0.907 0.000 0.496 -0.427 -0.054 0.052 2005 0.065 0.804 0.000 0.226 -0.157 0.085 -0.035 2006 -0.325 1.386 0.000 0.582 1.633 0.179 0.315 2007 0.034 0.825 0.000 0.101 -0.323 0.431 -0.063 2008 0.066 0.798 0.000 0.196 0.208 0.611 -0.125 2009 -0.004 1.161 0.000 -0.980 0.787 0.251 0.003 2010 0.069 0.620 0.000 0.169 -0.377 0.222 0.000 2011 0.068 0.638 0.000 -0.033 0.718 0.162 0.013 2012 -0.007 0.933 0.000 0.083 -0.263 0.288 -0.013 2013 -0.033 1.213 0.000 -0.305 0.252 -0.118 -0.007 2014 0.160 0.669 0.000 0.196 0.037 0.711 -0.125 2015 0.010 0.937 0.000 -0.131 0.201 0.287 -0.024
4.2.2 Other data sources hypothesis 2
Besides the collection of all the variables from the database DataStream and the regressions from the model from McVay (2006), the information on joint audits need to be collected manually due to missing data. Also, the sample needs to be adjusted for specific outliers, who may influence the regression significantly. The data is winsorized at the 1ste an 99the percentile of the sample size, in according with the method used in hypotheses 1 and 3, see section 4.1.2. In total, a sample size of 723 observations remains for testing hypothesis 2.
Table 8: Sample size hypothesis 2
Denmark (OMX) 2000 - 2015
Total Observations 1367
Less: Firms with insufficient data 644
Total data sample for H2 723
Winsorizing (percentile) 1-99
Manually collected data
Like hypothesis 1 and 3, because the mandatory joint audit environment changed to a voluntary one, there will be firms who remain audited by two companies after 2005. Therefore, the annual reports will provide the dummy variable in the regression model from section 3. This
Furthermore, missing data on the cost of goods sold and selling, general and administrative expenses will also be filled by these annual reports. In addition to the 266 checked annual reports for hypothesis 1 and 3, data from another 12 financial reports is extracted to complete the dataset (for the distribution of the dummy variables see table 9). As in hypothesis 1, the dummy for financial crisis is added for the period of 2007-2009.
From the distribution we see that without collecting the dummy joint audit manually a total of 37 samples would be categorized wrong (from 2005 all companies would have been given a 0 for indicating a single audit). Also, the total dataset of N=723 contains circa 29% joint audits and 71% single audits, which is in line with the data distribution of H1 and H3.
Table 9: Distribution of joint audit
Dummy 2000-2004 2005-2012 2013-2015 %
Joint Audit 170 37 0 29%
Single Audit 0 356 160 71%
Total 170 393 160 723
4.3 Descriptive statistics hypothesis 1 and 3
4.3.1 Core statistics
From table 10 we see that the discretionary accruals have a mean of 0.073 with a minimum value of 0.008 and a maximum value of 0.223 which means that every company in the data sample have reported discretionary accruals over the selected time period and are associated with a certain degree of earnings management. The reported mean is in the middle of other study’s measuring abnormal accrual in a joint audit environment (Francis et al., 2009 and Lesage et al., 2012).
Furthermore, the book-to-market ratio has a minimum of 0.098 and a maximum of 2.170. The mean of 0.792 indicates that on average firms are reporting book value of circa 79% of the total market value. This means that the larger part of the firms in the data sample have a higher market value than reported book value and are therefore associated with accounting conservatism.
The statistics of the dummy variables show that the complete data sample consists of 35% joint audits and 65% single audits and the sample is therefore not distributed equally. The amount of ‘Big 4’ audit firms involved in the Danish listed companies throughout 2001-2014 is circa 95%. In total only 36 firms (not tabulated) are audited by a non ‘Big4’ audit company in the past 14 years. The crisis variable indicates that 22% of the sample size is reported during the financial crisis. This is consistent with the 3 years of the crisis (2007-2009) out of 14 years of the
total sample size.
Finally, the control variable size reports a mean of 14.605 with a relative low standard deviation and no special values. Leverage has a mean of 0.513 which indicated that the Danish listed firms consist of circa 51% liabilities and 48% own equity. The maximum shows that there are no firms in the sample size with more liabilities than total assets. Growth has a mean of 0.059 which means that, on average, the Danish listed market (with respect to our data sample) reports a growth of 6%.
Table 10: Summary of statistics H1 and H3 (N=754)
Variable Mean Std. Dev. Min Max
DA 0.073 0.059 0.008 0.223 BTM 0.792 0.592 0.099 2.170 d_Joint 0.347 0.476 0 1 d_Big4 0.952 0.213 0 1 d_Crisis 0.224 0.417 0 1 Size 14.605 1.659 11.707 17.748 Lev 0.513 0.182 0.179 0.812 Growth 0.059 0.208 -0.305 0.607
4.3.2 Correlation and multicollinearity
The Pearson Correlation Matrix, see table 11, is displayed to see which correlations between the regression variables exists. The dependent variable discretionary accruals is positively related with the dummy representing the financial crisis and negatively correlated with size and the dummy for audit firm. The book-to-market ratio is negatively related to d_Big4, Size and Growth.
The positive correlation between DA and d_Crisis means that during the financial crisis management tends to participate more in earnings management. This could be caused by the impact of the crisis where it is harder for a company’s management to meet certain benchmarks. For example, according to Burgstahler & Eames (2006), managers tend to avoid negative surprises regarding the company’s earnings and doesn’t want to report lower than analyst forecast, which ultimately results in more earnings management. Therefore the financial crisis is an important control variable to the regression model of H1.
The negative correlation with d_Big4 is explained by the fact that the discretionary accruals are decreasing over time and that 95% of the data sample is audited by ad ‘Big4’ audit firm (almost all values are 1). The correlation with Size means that larger companies tend to report less discretionary accruals and therefore participate relatively less in earnings management.
management and therefore controlled for the same variable.
The book-to-market ratio is also negatively related to d_Big4 and Size . This is explained by the fact that accounting conservatism in the current literature is often measured by either the book-to-market ratio or, like earnings management, by the discretionary accruals. Furthermore, BTM is negatively correlated with Growth which means that the market reacts faster than the accounting policies if companies report an increase in specific assets.
Another mentionable correlation exist between d_Joint and DA/BTM. The variable d_Joint is, although not significantly, positively correlated to the both dependent variables for hypothesis 1 and 3. This means that over time companies tend to report less discretionary accruals, although the theory predicts a negative correlation due to the change from a joint audit environment to a single audit environment.
The positive direction in the book-to-market ratio however suggests that companies with a joint audit report more conservatively than companies with a single audit. The implications of both relations are further discussed in the analysis and conclusions section (see section 5 and 6). Table 11: Pearson Correlation Matrix H1and H3
Variable DA BTM d_Joint d_Big4 d_Crisis Size Lev Growth
DA 1 BTM -0.0399 1 d_Joint 0.0338 0.0312 1 d_Big4 -0.2126*** -0.0891** 0.0981*** 1 d_Crisis 0.1517*** 0.0120 -0.2186*** -0.0586 1 Size -0.1400*** -0.1255*** -0.0489 0.2595*** 0.0164 1 Lev -0.0042 0.0564 -0.0155 -0.0172 -0.0142 0.0811** 1 Growth 0.0330 -0.1852**** -0.0417 -0.0037 0.0487 0.1026*** -0.1067*** 1 ***Significance 0,01, **Significance 0,05
At last, the variance inflation factor (VIF) is calculated to test the regression model for multicollinearity. If two or more independent variables tend to correlate to much it could influence the reliability of the regression model because it would impact the beta’s from the regression. According to Mansfiels & Helms (1982), the variance inflation factor is the best predictor for multicollinearity. The VIF is not a problem if the coefficients are not unusually larger than 5. In our analysis, the highest multicollinearity exist for Size with a variance inflation factor of 1.10, see table 12. This means that the calculated factors are far below 5 and no multicollinearity problem in the regression models for H1 and H3 exists.
Table 12: VIF calculation H1 and H3 Variable VIF d_Joint 1.06 d_Big4 1.09 d_Crisis 1.05 Size 1.10 Lev 1.02 Growth 1.03 Mean VIF 1.06
4.4 Descriptive statistics hypothesis 2
4.4.1 Core statistics
In table 13 the summary of statistics are show for the regression model H2 as well as the variables to measure the dependent variable: unexpected core earnings. The variables and
are omitted because they are directly subtracted from one other variable
included in the table, which means that correlation is not an issue.
The dependent variable has a mean of 0.002 with a standard deviation of 0.09. These values are in line with the research of McVay (2006), who reported a mean for unexpected core earnings of 0.001 with a standard deviation of 0.148. If perfectly predicted by the estimated model (and without data trimming), the mean should be equal to zero. The minimum and maximum value of -0.285 and 0.310 is a smaller part of the minimum and maximum value of the reported core earnings. This is in line with the purpose of the model: to measure a shift towards core earnings year t when a special item is reported.
The distribution of the independent dummy variable d_Joint is already discussed in section 4.2.2 where it was shown that the ratio was 29% joint audit to 71% single audit. The mean of the dummy variable d_Crisis is 0.145 which means that circa 15% of the data sample is measured during the financial crisis. The independent variable %SI has a minimum value of zero because all income increasing special items are set to this value. Again, the reported mean of 0.012 is in line with the 0.027 reported mean by McVay (2006). The value indicates that the income decreasing special items on the income statement are, on average, as high as 1.2% of the reported sales.
Table 13: Summary of statistics H2 (N=723)
Variable Mean Std. Dev. Min Max
UE 0.002 0.090 -0.285 0.310
%SI 0.012 0.089 0.000 0.77
d_Joint 0.286 0.452 0 1
d_Crisis 0.145 0.353 0 1
4.4.2 Correlation and multicollinearity
The Pearson Correlation Matrix for hypothesis 2, see table 14, is displayed to see which correlations between the regression variables exists.
There exists no significant correlation between unexpected core earnings, special items and the dummy representing the joint audit.
Also, the dummy variable d_Crisis is positively correlated with d_Joint because the crisis is only reported during 2007-2009 and the joint audits are declining from year 2004.
Table 14: Pearson Correlation Matrix H2
Variable UE_CE %SI d_Joint d_Crisis
UE 1
%SI 0.0101 1
d_Joint -0.0130 0.0138 1
d_Crisis 0.0431 0.0303 -0.2003*** 1
***Significance 0,01
Table 15 represent the VIF analysis for the regression model of hypothesis 2. Because the variance inflation factor is not unusual larger than 5 (recall the detailed explanation of the VIF analyses in section 4.3.2) no problem regarding multicollinearity for hypothesis 2 exists.
Table 15: VIF calculation H2
Variable VIF
%SI 1
d_Joint 1.04
d_Crisis 1.04
5 Regression analysis per hypothesis
5.1 Regression model hypotheses 1Recall from section 3 that the regression model, to test whether joint audit has a positive relation with audit quality (which means less abnormal accruals if the audit is performed by two or more auditor firms), was:
The linear regression is performed by using statistical software program Stata and the results are tabulated below (table 16).
Table 16: Regression analysis Hypothesis 1 Number of observations 754
F(6, 682) 10.99
Prob > F 0.000***
R-squared 0.081
Adjusted R-squared 0.074
β Std. Err. T-value p-value
0.161 0.020 8.21 0.000*** 0.010 0.004 2.32 0.021** -0.052 0.010 -5.09 0.000*** 0.022 0.005 4.41 0.000*** -0.003 0.001 -2.57 0.010*** 0.003 0.011 0.22 0.822 0.011 0.010 1.09 0.277 ***Significance 0,01, **Significance 0,05
The above results of the regression formula shows that the model has a value for R-squared of 0.081, which means that the explanatory power of the model is 8,1% (percentage of the variation in the outcome of the model that is explained by the model, Field (2013)). In other words, 8.1% of the variance in the absolute value of discretionary accruals is explained by the independent variable of the joint audit together with the control variables.
The F value equals 10.99 and has a significant value of 0.000, which means that the null hypothesis (all beta coefficients equals zero) is rejected. According to Field (2013), this means
joint audit and audit quality. For example, Francis et al. (2009), reported an R-squared in their models for abnormal accruals between 4 and 6 percent and Lesage et al. (2017) reported an average R-squared of circa 13%.
The signs of the betas for the explaining variables are positively significant for and which means that the dependent variable increases 0.010 point if the audit is performed by at least two audit firms and increases 0.022 point during the financial crisis.
This means that the outcomes of the model suggest that a joint audit not only doesn’t improve audit quality, but it also leads to more discretionary accruals and therefore, less audit quality. The sign of the variable is opposite of what was expected and hypothesized based on current literature. This could be due to the effect of the mandatory implementation of IFRS in 2005 for all listed companies in Europe (Soderstrom & Sun, 2007) and/or the increasing corporate governance regulations over the past time. Corporate governance, if independent from management, is negatively associated with earnings management and should therefore be controlled for (Lin & Hwang, 2010). However, due to limited time and missing data, this control variables couldn’t be adopted to the model.
The positive sign of the financial crisis related to de discretionary accruals was as expected because management has an incentive to avoid reporting negative results and therefore, to participate in earnings management when economic performance is low (see also section 4.3.2 - correlation).
Furthermore, the discretionary accruals are significantly negative related to and , which is in line with the reported correlation between the variables in section 4.3.2. The dummy for the audit firm is negatively related because 95% of the data sample is audited by a ‘Big 4’ audit firm and the discretionary accruals are decreasing over time. The negative correlation of the size of the company is in line with current literature (Becker et al., 1998) and the main reason why it was added as a control variable to test hypothesis 1.
The sign of growth and leverage is, although not significant, positive as expected. When a company increases its assets by practicing earnings management (for example to meat analysts forecast) both growth, leverage (total liabilities held constant) and abnormal accruals increases. Concluding, the regression model shows that a joint audit is not negatively related to the discretionary accruals in the period 2001-2014 for Danish listed companies. This means that a mandatory joint audit environment is not associated with less accrual-based earnings management and therefore, hypothesis 1 is rejected.
