Audit fees and audit quality in family firms

(1)

Audit Fees and Audit Quality in Family Firms

Name: Bas Bleeker

Student number: 10574840 Date: June 19, 2018 Word count: 13,277

Thesis supervisor: Dr. Alexandros Sikalidis

MSc Accountancy & Control, specialization Accountancy Amsterdam Business School

(2)

Statement of Originality

This document is written by student Bas Bleeker 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: In this study I examine the differences in the issue of auditor-client dependency

for family firms and non-family firms. Specifically, I investigate the impact of family control (proxy: family ownership) on the relation between positive abnormal audit fees and auditor audit quality. Positive abnormal audit fees impair the independence of the auditor. The relationship between the client and the auditor is different for family firms than for non-family firms, because of the advisory role of the auditor in family firms. I empirically demonstrate that the independence of the auditor (proxy: audit quality) in family firms is more impaired when abnormally high audit fees are paid by the client. I argue that this is because of the trust and the long-term commitment between the auditor and family firms, which increases the willingness to cooperate. The too high willingness to cooperate in combination with an economic bond (created by positive abnormal audit fees) in family firms create a dependency for the auditor on the audit evidence provided by the client. The results are robust to a variety of robustness tests and to different audit quality proxies such as abnormal discretionary accruals and various forms of real activities manipulation.

Keywords: auditor-client, family firm, audit fees, audit quality, economic bonding,

(4)

1 Introduction

In this paper the relationship between audit fees and audit quality in family firms is examined. To see if there is a different relation between non family firms and family firms, specifically the positive abnormal audit fees are used to see if they influence the allowance for both real activities manipulation and accrual-based earnings management.

For a long time there has been a debate over the role of ownership structures on firm performance. In this case the debate is specifically about the effect of family ownership on the audit quality. For instance combined ownership and control can create a situation where such owner can expropriate wealth form other shareholders for their own gain (Fama and Jensen, 1983; Shleifer and Vishny, 1997). On the other hand there are arguments that the family ownership decreases the misalignment of interest, so that the total costs associated with this misalignment is lower (Ho & Kang, 2013). Furthermore, Anderson and Reeb (2003) suggest that the family understands the business and that involved family members view themselves as the stewards of the firm.

Homayoun and Hakimzadeh (2017) did a similar research. They show that there is a negative relation between family ownership and audit fees. Moreover, they found that there is no significant relation between family ownership and demand for audit quality (auditor size and expertise as proxies). Nevertheless, they did not examine if the relation between the audit fees and the audit quality is different for family and non-family firms and this is what will be addressed in this research.

Prior research states that “founding-families represent a unique class of shareholders that hold poorly diversified portfolios, are long-term investors (multiple generations), and often control senior management positions” (Anderson and Reeb, 2003, p. 1304). So founding family members regularly play an influencing role within an organization. In these family firms there is less separation between ownership and control, because of controlling family members that hold shares, but more separation between controlling family members and outside investors (Ho & Kang, 2013).

Two competing theories suggest opposite effects of family ownership on the earnings quality or reporting quality. The entrenchment effect suggests that managers from the founding family have incentives to expropriate wealth from other shareholders for personal gain (Wang, 2006; Bhaumik & Gregoriou, 2010). On the other hand there is the alignment effect which suggests that the interests of founding families and other shareholders are better aligned

(6)

because of the large blocks of stock owned by family members and their long-term presence (Wang, 2006; Bhaumik & Gregoriou, 2010). Family firms choose the auditor as the preferred external advisors (Jaffe, Lane, Dashew & Bork, 1997; Chrisman, Chua, Sharma & Yoder, 2009) and they more often have a long-term relationship with their auditor (Khalil, Cohen and Trompeter, 2011). Therefore it is interesting to study the impact of family ownership on the independence of an auditor and the corresponding allowance for earnings management.

The focus of this paper is on family firms in the United States for a couple of reasons. Firstly, because family firms are considered an important part of modern economies, because the value of listed family firms accounts for almost 10 percent of the total market capitalization of U.S.-listed stocks (Srinidhi, He and Firth, 2014). Secondly, the concept of a family firm leaves some room for interpretation. In recent literature it is mostly described as a firm where the founding family members remain a significant influence in the firm, where voting rights (ownership) and positions within the firm play a role (Wang, 2006; Ghosh & Tang, 2015; Khalil & Mazboudi, 2016). According to this approach approximately 45-50% of the S&P1500 and 30-40% of the S&P500 are family firms (Ghosh & Tang, 2015; Khalil & Mazboudi, 2016). This means that family firms play a significant economic role in the US. Secondly, family firms in the United States are the subject of recent research, and more specifically the auditor-client relationship in family firms (Khalil & Mazboudi, 2016). And thirdly, prior research states that founding-families represent a unique class of shareholders tied by blood and kinship (Khalil & Mazboudi, 2016), that hold poorly diversified portfolios, are long-term investors (multiple generations), and often control senior management positions (Anderson and Reeb, 2003). So the relation between the auditor and these kind of owners/managers can be significantly different for family firms, leading to different independence and audit quality.

In this research the focus will be on positive abnormal audit fees instead of negative abnormal audit fees and therefore the economic bonding story, because I try to investigate if this economic bonding influences auditor independence and ultimately the audit quality more in family firms than in non-family firms. I do this by employing a fee prediction model to measure abnormal audit fees and several models to estimate the use of accrual-based earnings management and real activities manipulation, including one based on the modified Jones model. Sequentially, I run multiple regressions with these measures of earnings management as dependent variable and family firms and abnormal audit fees as the independent variables.

To add to prior literature I extend the research by investigating whether the positive abnormal audit fees for family firms are more (or less) associated with audit quality than the

(7)

positive abnormal audit fees for non-family firms. I find evidence that there is a positive relation between abnormal audit fees in family firms and the audit quality. This suggests a difference in independence of auditors in family firms, because of the different role the auditor has within family firms and the accompanying bond between the client and auditor.

(8)

2 Literature review

2.1 Auditor client relationship

The auditor client relation is different than most buyer-seller relationships, because there is a third party who benefits from the work that is done by the auditor, other than the client. This construction has influence on this relationship between the auditor and the client. This influence is the subject of a lot of scientific research, where this thesis focusses on the relationship between family firms and their auditor.

The IESBA handbook (2013, p. 47) states: “In the case of audit engagements, it is in the public interest and, therefore, required by this code, that members of audit teams, firms and network firms shall be independent of audit clients”. Where independence (in mind) is: “The state of mind that permits the expression of a conclusion without being affected by influences that compromise professional judgement, thereby allowing an individual to act with integrity and exercise objectivity and professional skepticism”. These definitions are assumed to be widely accepted and are used throughout this thesis.

On the other hand, the auditor is employed by the client and needs cooperation from the client to perform a good audit. This is because the client needs to provide useful information to the auditor, so that a good audit can be carried out. There is an information asymmetry problem between the client and the auditor who needs to be solved, where the client holds important information that the auditor needs (Ruyter and Wetzels, 1999).

Audit quality is defined as the probability that the auditor will both discover and report a breach in the client's accounting system (DeAngelo 1981a). An explanation for variations in audit quality involve auditor reputation and power conflicts (managerial pressure) (Deis & Giroux, 1992). The commitment-trust model of Morgan and Hunt’s (1994) provides arguments that power was used in ‘sick’ relationships and that the restrained use of power by the parties involved leads to long-term commitment and trust (Fontaine & Pilotti, 2012). Furthermore, previous literature suggests the positive relationship between interdependence (sharing of power) and relationship commitment (Ruyter and Wetzels, 1999).

To cooperate, the client needs to have benefits from the arrangement. Fontaine & Pilotti (2012) argue, based upon the power model of Goldman and Barlev (1974), that the majority of audit services are routine jobs with little value added. Because of this the client can easily change auditors, which is their main source of power. On the other hand they argue, based on

(9)

Beattie, Fearnley and Brandt (2000), for the added-value audit. This concept means that the findings of the auditor are communicated to the client, so that the client can use this information to run its organization. Because of this added value, the client is willing to cooperate with the auditor, because it is not a routine and some of the power is shifted towards the auditor. In this case there can be a ‘healthy’ relationship with a balanced and restrained use of power (Fontaine & Pilotti, 2012). The findings of Fontaine and Pilotti (2012) are in line with the theory of added-value.

2.2 Audit fees and audit quality

There are important factors that influence auditors to compromise on audit quality according to scholars, investors and regulators. One of these factors is the independence of an auditor (Asthana & Boone, 2012). The payment of abnormal audit fees can influence the independence of an auditor, because of conflicts of interest.

The expected audit fee charged by an auditor is a function of units of audit resources expended, the cost per unit of these resources and the auditor’s expected losses/costs from the engagement (litigation, penalties, etc.) (Simunic, 1980). In extant research models the expected audit fee is estimated as a function of observable factors that are proxies of the auditor’s cost in performing the audit, including auditor effort, expected future litigation losses, and normal profit. If the model is well defined and specified, the residual audit fee reflects the abnormal audit fees from the audit engagement (Asthana & Boone, 2012).

Audit fees can influence auditor behavior in various ways. Large fees paid to auditors may increase the effort of auditors, increasing audit quality (Hoitash, Markelevich & Barragato, 2007). On the other hand, unexpected large fees paid to auditors, make auditors more economically dependent on their clients. These positive abnormal audit fees can indicate economic bonding between the client and auditor (Hoitash et al., 2007; Asthana & Boone, 2012; Krauß, Pronobis & Zülch, 2015). This economic bonding occurs when a client pays in excess of the ‘normal’ audit fee. This creates a situation where there is a client specific financial bond between the client and the auditor, leading to lower independence. Because of financial dependency, the auditor is willing to compromise the audit work done, to prevent the loss of the highly profitable fee (Hoitash et al., 2007).

Economic bonding between the auditor and the client can be seen as a power conflict, because the auditor is dependent on the fees the client pays, so that the power over the audit lies within the client. It is assumed in prior literature that auditors who receive abnormally high

(10)

audit fees have an incentive to allow clients to engage in opportunistic earnings management. This relation holds to the extent that the perceived net benefits of the audit engagement are greater than the associated costs (Asthana & Boone, 2012; Krauß et al., 2015). If this allowance for earnings management is high, the audit is of poorer quality (Krauß et al., 2015).

The measure of unexpected auditor profitability is believed to measure the relation between audit quality and auditor independence, because it is likely that auditor independence is influenced by the amount of the total fees relative to their expected amounts, rather than the level of fees received from clients (Hoitash et al., 2007). The use of unexpected audit fees addresses concerns about the use of levels of total audit fees, without controlling for their source (Larcker & Richardson, 2004).

2.3 Agency theory in family firms

In publicly traded companies there is separation between ownership and control and this can create conflicts of interest between managers and outside investors. This is because managers have incentives to report financial information that does not represent the underlying economic transactions to maximize their private benefits at the cost of shareholders or creditors. On the other hand, earnings are used as an incentive for managers (in the form of bonus contracts) to mitigate agency conflicts by aligning the interests with outside shareholders or creditors (Wang, 2006; Bhaumik & Gregoriou, 2010). Wang (2006) therefore concludes that ownership structures affect the supply of quality financial reporting and that a demand for this high quality reporting is created by shareholders, creditors, and other users of financial statements for efficient contracting and monitoring. This demand creates the incentives to provide high-quality financial statements to obtain better contracting terms (Wang, 2006; Bhaumik & Gregoriou, 2010).

There is no consensus about when an organization is a family firm. In recent literature it is mostly described as a firm where the founding family members maintain a significant influence in the firm. Voting rights (ownership) and positions within the firm play a role (Wang, 2006; Ghosh & Tang, 2015; Khalil & Mazboudi, 2016). According to this approach approximately 45-50% of the S&P1500 and 30-40% of the S&P500 is a family firms (Ghosh & Tang, 2015; Khalil & Mazboudi, 2016). For instance Srinidhi et al. (2014, p. 2298) say: “We define family firms as firms in which the members of one family have 20 percent or more of the voting rights and at least one family member serves as a director on the board”. So they use 20 percent of voting power, but Anderson, Duru, and Reeb (2009) and Anderson, Reeb, and Zhao (2012) use

(11)

a threshold of 5% ownership of voting rights plus a second variable where dual class plays a role. They say that it is not necessary for a family member to serve as CEO to have significant influence and that founders and heirs can gain additional influence through dual-class share structures.

Family firms are differently governed than non-family firms (Ho & Kang, 2013). Because often one or multiple family members run the business, they have less segregation between ownership and control (type I agency costs). On the other hand there will be more conflict of interest between insiders (managerial owners) and outside investors (type II agency costs). Ho and Kang (2013) found evidence that the total agency cost will decline, so the decline in type I dominates the increase in Type II agency costs. Ho and Kang (2013) and Ghosh and Tang (2015) both found that audit fees are lower for family firms, due to firms’ lower demand for external auditing services, lower auditors’ perceived audit risk and they found that auditors work less to provide assurance in family firms.

Wang (2006) gives two competing theories regarding the effect of the family ownership structure on the quality of earnings, namely the entrenchment effect and the alignment effect. The entrenchment effect is based upon the argument that concentrated ownership, and the separation between non-family ownership and control which comes from it, creates incentives for controlling (family) shareholders to expropriate wealth from other shareholders, because of greater information asymmetries between founding families and other shareholders.

The alignment effect is based upon the notion that the interests of founding families and other shareholders are better aligned because of the large blocks of stock owned by family members and their long-term presence, leading to better monitoring mechanisms (Wang, 2006). Wang (2006) finds evidence that supports the alignment effect of family ownership on the supply of earnings quality, or alternatively, the entrenchment effect on the demand for earnings quality.

The two theories Wang (2006) introduced could also be used to explain the influence of audit fees on audit quality. This is because the alignment effect in family firms can mean that managers are willing to cooperate more with the external auditor to benefit all stakeholders and not only themselves. This willingness to cooperate with the auditor has in this way a positive effect on audit quality. On the other hand the entrenchment effect can mean that, because controlling (family) shareholders are incentivized to expropriate wealth from other shareholders, they will not cooperate closely with the external auditor. This means that the auditor is not able to perform the best audit, so the audit quality is negatively affected.

(12)

The inclusion of emotions in the research in management of family firms is increasingly called for by scholars (Gomez-Mejia, Cruz, Berrone & De Castro, 2011). This theory for family firm management is also referred to as the socio emotional wealth theory. This is because family members who have influence over the firm have non-economic goals (Chrisman, Chua, Pearson & Barnett, 2012) as well. These non-economic goals include ‘perpetuation of family values through the business’, ‘preservation of family dynasty’, ‘conservation of the family’s social capital’ and ‘family’s strong desire to infuse its values into the business as the spring well of organizational culture’ (Gomez-Mejia et al., 2011). In addition, in family firms the members within a family are generally ‘stuck’ in the business and cannot walk away when they have emotional issues. These emotions are likely to play a more important role in family firms than in non-family firms, because this exit is not a viable option (Gomez-Mejia et al., 2011). Concluding, because family members’ long-term and sustainable presence in the firm and their intention to preserve the family name, founding families have a greater stake in the firm than non-family executives (Wang, 2006).

2.4 Willingness to cooperate in family firms

The willingness to cooperate with the external auditor can be a mitigating factor to the effect of economic bonding on audit quality, because of the entrenchment or alignment effect introduced earlier. A higher willingness to cooperate with the auditor influences auditor independence, because the auditor can rely more on the evidence he can obtain, although a too high willingness can decrease independence, because the auditor more dependent on the evidence the client provides (Fontaine & Pilotti, 2012). According to Fontaine and Pilotti (2012) two influencing factors on the willingness to cooperate are trust and (long-term) commitment. They based this on the commitment-trust model of Morgan and Hunt (1994). It is argued that commitment is present in every successful relationship (Fontaine & Pilotti, 2012).

Prior literature has shown some differences on the relationship between auditor and client for family firms. In family firms the auditor can be more of an advisor and auditors are among the most preferred external advisors of family firms (Jaffe et al., 1997; Chrisman et al., 2009). Dobler (2014) documents that family firms rely on a broad range of auditor-provided services beyond the scope of financial and tax accounting. Also Khalil et al. (2011) find that auditors and family firms more often have a long-term relationship. These factors are indicators for trust and commitment and therefore the willingness to cooperate may be different for family firms.

(13)

According to the model of Morgan and Hunt (1994), this can be a result of the restraint use of power in family firms.

Assuming there is a gap in knowledge and capabilities in family firms, auditors can act as advisors to fill that gap with their knowledge resource and expertise (Strike, 2012). This statement is closely aligned with the value-added theory. By the sharing of knowledge, advisors may help the firm in developing knowledge resources that lead to fruitful outcomes (Strike, 2012). If clients believe in the added value of the audit and therefore are more cooperative, they could be willing to accept price increases and be less willing to leave the auditor in the case of less than satisfactory service quality (Fontaine & Pilotti, 2012)

2.5 Earnings management

Following prior literature, ‘‘Earnings management occurs when managers use judgment in financial reporting and in structuring transactions to alter financial reports to either mislead some stakeholders about the underlying economic performance of the company or to influence contractual outcomes that depend on reported accounting practices’’ (Healy and Wahlen, 1999, p. 368). Roychowdhury (2006) explained that earnings managerial intervention can also occur through operational decisions. The definitions of Scott (2011) and Cohen and Zarowin (2010) will be used as explanation for both forms of manipulation. According to Scott (2011), accrual-based earnings management is: “a situation where managers are incentivized to manipulate accounting data willfully for their own interests. On the other hand, Cohen and Zarowin (2010, p. 5) defined real activities manipulation as: “management actions that deviate from normal business practices, undertaken with the primary objective to mislead certain stakeholders into believing that earnings benchmarks have been met in the normal course of operations”.

It is believed that real activities manipulation has a greater long-term effect on the performance and on the shareholders than accrual-based earnings management, because it will influence future cash flows and can potentially hurt the total firm value. Accrual-based earnings management, however, does not have an impact on actual cash flows, and therefore there is no impact on the total profit of the firm, but only on the reported annual profits (Roychowdhury 2006; Cohen, Dey & Lys, 2008; Cohen et al., 2010).

To perform earnings management a couple of criteria have to be met, according to prior literature (Dechow, Sloan & Sweeney, 1995). First of all there has to be a form of agency problem as described earlier. Because of this agency problem there is a situation where the agent (management) has an information and knowledge advantage over the principal

(14)

(stakeholders), so that the agent can pursue self-interests without the knowledge of the principal (information asymmetry). This agency problem is arguably less in family firms, as stated before. On the other hand, there can be an increased information asymmetry between family members and other stakeholders. Secondly, there needs to be a certain management bonus contracts in place so that there is also the incentive to manipulate earnings in order to receive the bonus. Lastly, there needs to be discretion in the decisions the management can make. This discretion creates a situation where the manipulation of earnings is possible.

According to Bhaumik and Gregoriou (2010) there are some additional motives which drive management to apply earnings management. Firstly, there is the expectation of the market and the accompanying market price. It is argued that high reported earnings result in higher stock values, which will create more job security and wealth, if the manager is also a stockholder. Secondly, managers do not only manage their earnings now to gain current bonuses, but also maximize the expected future bonuses.

2.5.1 Earnings management as a proxy for audit quality

High quality audits lower the degree of earnings management, through the lower allowance of the auditor for earnings management, and increases the informativeness of the financial statements. Balsam, Krishnan and Yang (2003) find that there is positive relation between audit quality and the quality of financial reporting. The proxy they used for the quality of financial statements is earnings quality (earnings management measured by accruals). Lennox, Wu and Zhang (2015) found that audit adjustments cause earnings to become smoother and more persistent and that the adjustments result in higher accrual quality. This suggests that audit quality (more audit adjustments) has a strong positive relation with earnings quality. Recent literature also argues that audited earnings is a good proxy for audit quality (Francis, Michas & Seavey, 2013). It can be seen in prior literature that the earnings management proxy is widely used (Asthana & Boone, 2012; Srinidhi et al., 2014; Krauß et al., 2015).

According to the same reasoning as for accrual-based earnings management it can be argued that real activities manipulation is a good proxy for audit quality. This is because auditors can recognize real activities manipulation. On the other hand, the auditor cannot request a correction for the published financial statements. The only known response to real activities management is the resignation of the auditor (Kim and Park, 2014). It can be argued that auditors in family firms are more able to prevent real activities management. This is both because of the willingness to cooperate, so that auditors have more information to build their

(15)

advice on during the audited period and due to the fact that they have more of an advisory role and are among the most trusted advisors within family firms. Chi, Lisic and Pevzner (2011) find that longer auditor tenure is associated with greater real earnings management. This last finding suggests that if family firms really have longer auditor tenure, than there can be a risk of more real earnings management.

In the case of the family firms it can be argued that both measures are suited as proxy for audit quality, because this thesis is about the impact of family ownership structure (willingness to cooperate) on the relation between positive abnormal audit fees on audit quality (allowance for earnings management). But caution has to be taken, because of the implication that a lower allowance for accrual-based earnings management could result in an incentive for higher real activities manipulation (Chi et al., 2011).

2.6 Hypothesis development

There is an information asymmetry between the auditor and the client, where the client has information the auditor needs to perform a good audit. This information asymmetry exists, because of audits being routine jobs, with no benefits for the client. On the other hand, there is the theory of added value, where the auditor shares the findings of the audit with the client. The client therefore gets more value out of the arrangement the better the audit is performed and is therefore willing to cooperate with the auditor. Assuming there is a gap in knowledge and capabilities in family firms, auditors can act as advisors to fill that gap with their knowledge resource and expertise (Strike, 2012).

Prior literature states that positive abnormal audit fees result in a lower independence of the auditor, because of economic bonding. This financial dependency can result in the auditor allowing more earnings management. Prior literature has shown that lower audit risk and audit effort (as proxy for reporting quality) result in lower audit fees in family firms, possibly because of lower type I agency costs leading to higher reporting quality and lower audit risk (Ho & Kang, 2013). The results of Hoitash et al. (2007) are consistent with economic bonding rather than auditor reputational concerns being a determinant of auditor behavior.

The agency costs arise from the separation of ownership and control, which is different for family firms. In family firms there is less separation between ownership and control (type I), which is called the alignment effect, because owners are more aligned with the management. On the other hand, there is more separation between controlling owners and other shareholders (type II), which effect can be called the entrenchment effect, where controlling owners have

(16)

the incentive to expropriate wealth from other shareholders. The inclusion of socioemotional wealth theory is popular, because it is understood that managers also have non-economic goals. In family firms these non-economic goals are perceived to be higher, because of family values and because family members are usually ‘stuck’ in the business. Concluding, because family members’ long-term and sustainable presence in the firm and their intention to preserve the family name, founding families have a greater stake in the firm than non-family executives (Wang, 2006).

All these differences between family firms and non-family firms can create different incentives for management to cooperate with the auditor. A higher willingness to cooperate can lead to auditor independence, because the auditor can obtain more appropriate evidence on which he can rely, but a too high willingness to cooperate can cause the auditor to be less independent, because of the high dependency on the evidence the client provides. According to Fontaine & Pilotti (2012) the two factors that are an indication of willingness to cooperate are long-term commitment and trust. Prior literature found evidence for differences in these factors between family and non-family firms. If clients are more cooperative they could be willing to accept price increases and be less willing to leave the auditor in the case of less than satisfactory service quality.

From the perspective of the client the differences in trust and commitment in the auditor client relationship for family firms can mean that the client is more willing to cooperate with the auditor to find and correct misstatements in the financial statements. Also it is possible that they are more willing to provide the information needed for the audit. So this willingness to cooperate can have a positive effect on the audit quality by mitigating the risk of non-independence following positive abnormal audit fees.

Nevertheless, from the perspective of the auditor, a higher willingness to cooperate with the auditor influences auditor independence, because the auditor can more rely on the evidence he is able to obtain, although a too high willingness can decrease independence, because the auditor can be too dependent on the evidence the client provides (Fontaine & Pilotti, 2012). This too high willingness to cooperate creates a situation where the auditor is too dependent and overly trusting of the information provided by the client. The higher willingness to cooperate can therefore also enhance the negative effect of positive abnormal audit fees on audit quality. Because there is not an explicit expected direction of the results, the hypotheses will be of a neutral kind. The main null hypothesis is:

(17)

The sub hypotheses will be more specific and are as follows:

H0.1: Family ownership has no influence on the relation between positive abnormal audit fees and accrual-based earnings management

H0.2: Family ownership has no influence on the relation between positive abnormal audit fees and real activities manipulation

(18)

3 Data and method

3.1 Sample selection

This study investigates the difference in the influence of family and non-family firms on the influence of abnormal audit fees on the audit quality. The data used on family ownership is a combined sample from the studies of Anderson, Duru, and Reeb (2009) and Anderson, Reeb, and Zhao (2012). This is a database I downloaded from the website of Ron Anderson. They use a threshold of 5% ownership of voting rights for family control. They also check if the firm has dual class shares. The combined data on family ownership from these two studies can be found on the website of Ron Anderson: http://www.ronandersonprofessionalpage.net/data-sets.html.

The process of the previous mentioned authors for collecting data on family ownership as follows. First they pulled all firms from CompuStat (North America) for date-year 2001 with all data available for total assets. They excluded foreign firms, regulated public utilities (SIC codes 4812, 4813, 4911 through 4991), and financial firms (SIC codes 6020 through 6799) because government regulation potentially affects firm equity ownership structure, corporate opacity, and performance. Furthermore, master limited partnerships are excluded, because this is an ownership structure that does not fit the research of family ownership structures. Next the firms with a share price less than 0,25 dollar are excluded and ultimately the 2000 largest firms based on total assets are selected as the sample. Corporate histories (family lineage) are taken from ReferenceforBusiness.com, FundingUniverse.com, and individual company websites. The data on family ownership is then collected, by the authors, from corporate proxy statements and 10-k’s for the years 2001 through 2010. To control for survivorship bias, they allow firms to exit and re-enter the sample. The final sample consists of 2,000 largest firms for 2001 and spans from 2001 through 2010, providing 16,230 firm-year observations.

The data found in the downloaded file is a list of 2000 largest firms for 2001, with data for these firms from 2001 through 2010. Two dummy variables are included. One equals 1 for family control, where the threshold is 5% (ownership or votes) and a dummy variable that equals 1 if the firm has a dual-class share structure. The rest of the data collection will be explained after the empirical models description.

(19)

3.2 Research design

3.2.1 Proxy family control

As mentioned before the proxy for family control is a dummy variable which is 1 if founders or descendants own or control more than 5% of the shares or the voting rights. According to Anderson, Duru, and Reeb (2009), Anderson, Reeb, and Zhao (2012) and Ghosh and Tang (2015), it is not necessary for a family member to serve as CEO to have significant influence and that founders and heirs can gain additional influence through dual-class share structures, because non-family members more likely have shares with less/no voting power. So according to this reasoning in this study the data previously described will be used, where a family firm will be denoted by a dummy variable equaling 1 if it is a family firm and 0 otherwise.

3.2.2 Proxy positive abnormal audit fees and fee prediction model

To get the proxy for the abnormal audit fees, I use a fee prediction model drawn from prior literature (Hoitash et al., 2007). I perform a cross-sectional regression by firm year observations to capture changes in fees over time, due to expected changes in expected auditor effort and audit risk. The error terms of this model are the abnormal audit fees, which will represent either a premium paid or a discount provided. The emphasis of this research will be on the premiums, because these relate to the economic bonding theory and are an impairment for auditor independence. The effect of negative abnormal audit fees on audit quality seems to be ambiguous and inconsistent (Krauß et al., 2015). I also add the dummy variable of family firms to the prediction model. This is because previous literature suggests that the audit fees for family firms are structurally lower and therefore I want to control for that.

The other control variables used for the fee prediction model will be the same as in the research of Hoitash et al. (2007), because these variables are most of the other factors that are correlated with audit fees. So based on prior literature the control variables are included are explained. The first one is client size (proxy: natural logarithm of total assets (LNTA), and the next one is audit complexity/audit risk (proxies: liquidity (current ratio), return on assets and a dummy variable which is 1 if the company had a loss in either of the two previous years). Also additional auditor work (proxy: (accounts receivable + inventory)/total assets and new equity issues are included (FINANCE = 1 if equity issue > 5% of lagged total assets). The last two variables are engagement tenure, as number of years under engagement over the last ten years (TENURE) and a dummy which is 1 if there occurred a merger or acquisition in the previous year and 0 otherwise (MERAC). See appendix A for a list with clear variable descriptions. The

(20)

model (1) will look like this, where LNTFEE is the natural logarithm of total audit fees and the residuals from this regression represent the abnormal audit fees denoted as ABNTFEE:

𝐿𝑁𝑇𝐹𝐸𝐸 = 𝛼 + 𝛽₁𝐿𝑁𝑇𝐴 + 𝛽₂𝐿𝑂𝑆𝑆 + 𝛽₃𝐿𝐼𝑄 + 𝛽₄𝑅𝑂𝐴 + 𝛽₅𝐼𝑁𝑉𝑅𝐸𝐶 + 𝛽₆𝐹𝐼𝑁𝐴𝑁𝐶𝐸 + 𝛽₇𝐵𝐼𝐺4 + 𝛽₈𝑇𝐸𝑁𝑈𝑅𝐸 + 𝛽₉𝐹𝐴𝑀 + 𝛽₁₀𝑀𝐸𝑅𝐴𝐶 + 𝜀 (1)

The descriptive statistics and results from model (1) are untabulated for brevity reasons. The average R2 is 0.6340, which is in line with prior literature (Hoitash et al., 2007; Asthana & Boone, 2012; Krauß et al., 2015). Most of the control variables were significant in all years, except for TERURE. TENURE was only significant in two of the ten years, indicating that there may be a better way to estimate experience with a client or that experience does not influence the total audit fees.

3.2.3 Proxies for audit quality

For audit quality the first proxy will be absolute discretionary current accruals as proxy for accrual-based earnings management. I perform an annual cross sectional regression per industry-year groups on the data (by year and 2 digit sic code). The model used follows from Asthana and Boone (2012) and Kim, Park and Wier (2012), which is based on the modified Jones model, where Asthana & Boone (2012) also control for firm performance as return on assets and Kim et al. (2012) use no constant, because 1 divided over lagged assets represents the constant. The model looks like this:

𝑇𝐴 𝐿𝐴 = 𝛼 ( 1 𝐿𝐴) + 𝛽 ( 𝛥𝑅𝐸𝑉𝑡− 𝛥𝑅𝐸𝐶𝑡 𝐿𝐴 ) + 𝛾 ( 𝑃𝑃𝐸 𝐿𝐴 ) + δ(𝑅𝑂𝐴𝑡) + 𝜀𝑡 (2)

Total accruals (𝑇𝐴) is net income minus operating cash flows. 𝐿𝐴 is total assets at the beginning of the fiscal year, 𝛥𝑅𝐸𝑉_𝑡 is equal to net sales in year 𝑡 less net sales in year 𝑡 − 1 and, 𝛥𝑅𝐸𝐶_𝑡 is equal to accounts receivables in year 𝑡 less accounts receivables in year 𝑡 − 1. I first estimate equation (2) and then use the parameters 𝛼, 𝛽, 𝛾 and δ from this equation to estimate the expected accruals. Then I subtract this expected accrual measure from the dependent variable of equation (2). This value represents the discretionary accruals. This method gives the same values as the residuals from the model (2). Because I do not look at the difference between income increasing and income decreasing discretionary accruals, I use the absolute value of discretionary accruals as evidence of accrual-based earnings management. This variable is denoted as ABSDISACC.

The second measure for audit quality will be based on three methods of real activities manipulation, namely the abnormal cash flows from operation (ABCFO), abnormal production

(21)

(ABPROD) and discretionary expenditures (ABEXP). For all three methods I also perform an annual cross sectional regression per industry-year groups on the data (by year and 2 digit sic code). To measure the real activities manipulations, the models employed by Kim et al. (2012) are used, which are derived from Roychowdhury (2006). The first of the three estimations is abnormal cash flows from operations:

𝐶𝐹𝑂 𝑇𝐴𝑡−1 = 𝛽0+ 𝛽1( 1 𝑇𝐴𝑡−1 ) + 𝛽2( 𝑅𝐸𝑉_𝑡 𝑇𝐴𝑡−1 ) + 𝛽3( 𝛥𝑅𝐸𝑉_𝑡 𝑇𝐴𝑡−1 ) + 𝜀𝑡 (3)

Here 𝐶𝐹𝑂 stands for cash flows from operations in year 𝑡, 𝑇𝐴_𝑡−1 for total assets in year 𝑡 − 1, 𝑅𝐸𝑉_𝑡 stands for net sales in year 𝑡 and 𝛥𝑅𝐸𝑉_𝑡 for the change in net sales of year 𝑡 to 𝑡 − 1. The residuals from this regression (3) is the abnormal cash flows from operations used as a proxy for real activities manipulation.

The second estimation is the abnormal production: 𝑃𝑅𝑂𝐷 𝑇𝐴𝑡−1 = 𝛽0+ 𝛽1( 1 𝑇𝐴𝑡−1 ) + 𝛽2( 𝑅𝐸𝑉𝑡 𝑇𝐴𝑡−1 ) + 𝛽3( 𝛥𝑅𝐸𝑉𝑡 𝑇𝐴𝑡−1 ) + 𝛽4( 𝛥𝑅𝐸𝑉𝑡−1 𝑇𝐴𝑡−1 ) + 𝜀𝑡 (4) Where 𝑃𝑅𝑂𝐷 stands for the sum of cost of sales in year 𝑡 plus the mutation in inventory in year 𝑡 relative to 𝑡 − 1. The rest of the variables are the same as before, except for 𝛥𝑅𝐸𝑉𝑡−1 which stands for the change of net sales of year 𝑡 − 1 to 𝑡 − 2. The abnormal production costs are the residuals from this equation.

The third and last estimation is the abnormal discretionary expenses: 𝐸𝑋𝑃 𝑇𝐴_𝑡−1= 𝛽0+ 𝛽1( 1 𝑇𝐴_𝑡−1) + 𝛽2( 𝑅𝐸𝑉𝑡−1 𝑇𝐴_𝑡−1) + 𝜀𝑡 (5)

𝐸𝑋𝑃 stands for the sum of research and development, selling, general and administrative expenses. The other variables are again explained above. The abnormal discretionary expenditures are the residuals from this equation.

The descriptive statistics and results from the models (2), (3), (4) and (5) are untabulated for brevity reasons.

3.2.4 Regression models

In the last two regressions I test if the influence of positive abnormal audit fees for family firms is more or less associated with audit quality than the positive abnormal audit fees for non-family firms, due to the possible different independence for auditors in non-family firms. I add the control variable LNTFEE, because total audit fee is a proxy of auditor effort, which can

(22)

increase the audit quality (Hoitash et al., 2007). Furthermore, Ghosh and Tang (2015) point out that audit fees are expected to increase with higher expected audit effort and having more specialized personnel on the audit team.

Then I split the abnormal total fees (ABNTFEE), where I only use the positive abnormal audit fees. This variable is denoted as HIABNTFEE and where the abnormal audit fee was negative this variable gets a value of zero. I can then see if the positive abnormal audit fees for family firms, given that the optimal fees are lower than for non-family firms, are associated more with audit quality through economic bonding. This is because the dummy variable for family control is also added in these models (6) and (7).

𝐴𝐵𝑆𝐷𝐼𝑆𝐴𝐶𝐶 = 𝛼 + 𝛽1𝐿𝑁𝑇𝐹𝐸𝐸 + 𝛽2𝐻𝐼𝐴𝐵𝑁𝑇𝐹𝐸𝐸 + 𝛽3𝐹𝐴𝑀 + 𝛽4(𝐻𝐼𝐴𝐵𝑁𝑇𝐹𝐸𝐸 ∗ 𝐹𝐴𝑀) + 𝛽₅𝐿𝑁𝑀𝑉𝐸 + 𝛽₆𝐶𝐴 + 𝛽₇𝐹𝐼𝑁𝐴𝑁𝐶𝐸 + 𝛽₈𝐿𝐸𝑉𝐸𝑅𝐴𝐺𝐸 + 𝛽₉𝑀𝐵 + 𝛽₁₀𝐿𝐼𝑇𝐼𝐺𝐴𝑇𝐼𝑂𝑁 + 𝛽₁₁𝐿𝑂𝑆𝑆 + 𝛽₁₂𝐶𝐹𝑂 + 𝛽₁₃𝑀𝐸𝑅𝐴𝐶 + 𝛽₁₄𝑅𝑂𝐴 + 𝜀 (6) 𝑅𝐴𝑀 = 𝛼 + 𝛽₁𝐿𝑁𝑇𝐹𝐸𝐸 + 𝛽₂𝐻𝐼𝐴𝐵𝑁𝑇𝐹𝐸𝐸 + 𝛽₃𝐹𝐴𝑀 + 4(𝐻𝐼𝐴𝐵𝑁𝑇𝐹𝐸𝐸 ∗ 𝐹𝐴𝑀) + 𝛽₅𝐿𝑁𝑀𝑉𝐸 + 𝛽₆𝐶𝐴 + 𝛽₇𝐹𝐼𝑁𝐴𝑁𝐶𝐸 + 𝛽₈𝐿𝐸𝑉𝐸𝑅𝐴𝐺𝐸 + 𝛽₉𝑀𝐵 + 𝛽₁₀𝐿𝐼𝑇𝐼𝐺𝐴𝑇𝐼𝑂𝑁 + 𝛽₁₁𝐿𝑂𝑆𝑆 + 𝛽₁₂𝐶𝐹𝑂 + 𝛽₁₃𝑀𝐸𝑅𝐴𝐶 + 𝛽₁₄𝑅𝑂𝐴 + 𝜀 (7)

Where RAM is ABCFO, ABPROD, ABEXP or COMBINEDRAM. COMBINEDRAM is calculated as a combined (aggregate) measure of all real activities manipulation activities, which is ABCFO – ABPROD + ABEXP. The direction of abnormal production costs is the opposite of the other two measures. This is because it is expected that a higher abnormal audit fee leads to higher abnormal cash flows from operations and abnormal discretionary expenditure, but leads to lower abnormal production costs.

The control variables used in models (6) and (7) with the audit quality as dependent variable and abnormal audit fees as independent variable are derived from the method of Hoitash et al. (2007). These variables are market value of equity (proxy: the natural log of the firm’s price per share at fiscal year-end multiplied by the number of shares outstanding (LNMVE)), current accruals, which is again net income before extraordinary items plus depreciation and amortization minus operating cash flows, scaled by beginning of year total assets (CA), the firm’s total assets less its book value divided by total assets (LEVERAGE), the market-to-book ratio (proxy: market value of equity divided by book value of equity (MB)), litigation risk (proxy: dummy variable with value 1 if risky industry and 0 otherwise, where high litigation

(23)

industries are firms where the SIC-code is 2833- 2836, 3570-3577, 3600-3674, 5200-5961, or 7370), LOSS is a dummy variable set as 1 if the firm reported a net loss in the reporting year or the two preceding years and 0 otherwise, CFO are the cash flows from operations, scaled by the beginning of year assets and MERAC is a dummy which is 1 if there occurred a merger or acquisition in the previous year and 0 otherwise. All variables used in the models (6) and (7) can also be found in appendix A, where a clear description of the variables is given.

Table 1 gives a short overview of the samples used and the corresponding observations per sample. The first sample is the sample used in model (6) and the second sample is used in model (7). The start is the sample provided by Anderson et al. (2009) and Anderson et al. (2012). I downloaded the data on audit fees and auditor information from AuditAnalytics and the firms’ financial information from CompuStat, where the firm observations with complete information for all dependent and independent variables are the samples 1 and 2. Because the information for the real activities manipulation models was less available or empty, I deleted a lot of observations and therefore the sample for model (7) is smaller than that for model (6).

Table 1. Sample selection

Procedure Observations: Sample

Data available from Anderson et al. (2009) and Anderson et al. (2012) 16,230 Data on firm observations also available on Compustat file 14,326 Data on firm observations also available on AuditAnalytics file 13,400

Data for AEM on firm observations available 12,911 1

(24)

4 Results

4.1 Descriptive statistics and diagnostic procedures

Tables 2 and 3 provide the descriptive statistics for respectively models (6) and (7). I identified specific outlier observations using visual methods and deleted these observations. Additionally, I winsorized most of the continuous variables at the 1 and 99 percent level if these outer values seem extreme. As can be seen in the first sample 33.71 percent is a family firm and 27.28 percent is a family firm in the second sample. The means and medians for most variables are close to each other, indicating an appropriate distribution.

For completeness, the correlation tables are added in the appendix B, because of readability reasons. Most of the bivariate correlations are low. The highest significant correlation is between ROA and CFO (0.6295 for sample 1 and 0.6543 for sample 2), which is logical and expected, because these are both measures of firm performance. The correlation between LNMVE and LNTFEE is also positive and significant and respectively 0.5864 and 0.5985 for samples 1 and 2. This is also expected because larger firms require more audit work and are expected to pay higher audit fees. These correlations don’t seem to generate a problem in interpreting the regression results, because all the variance inflation factors (VIF) for the explanatory variables are below 3.10, as can be seen in the regression results (Tables 4 and 5).

Belsley, Kuh and Welsch (1980) suggest that a condition index of 15 (or 30) or higher can be considered too weak for further consideration. The condition index for the independent variables in the samples 1 and 2 are 3.8149 and 4.0770, respectively (untabulated). Overall, correlation and multicollinearity does not seem to generate a problem for the multivariate analyses. The descriptive statistics are in line with prior research on audit fees and audit quality (Hoitash et al., 2007; Asthana & Boone, 2012; Krauß et al., 2015).

(25)

Table 3. Descriptive statistics sample 2

Variable Mean Min P25 Median P75 Max SD

ABCFO 0.0587 0.0009 0.0207 0.0441 0.0804 0.2661 0.0526 ABPROD 0.1062 0.0016 0.0371 0.0793 0.1468 0.4761 0.0951 ABEXP 0.1175 0.0019 0.0435 0.0911 0.1637 0.5142 0.1009 COMBINEDRAM 0.0695 -0.1295 0.0089 0.0533 0.1122 0.4414 0.0961 LNTFEE 14.1223 10.9966 13.3212 14.1032 14.8908 18.3175 1.1372 HIABNTFEE 0.2825 0.0000 0.0000 0.1374 0.4821 3.0034 0.3607 FAM 0.2728 0.0000 0.0000 0.0000 1.0000 1.0000 0.4454 FAM*HIABNTFEE 0.0828 0.0000 0.0000 0.0000 0.0000 2.7359 0.2353 LNMVE 7.1934 0.6549 6.0931 7.0794 8.2116 12.8814 1.7065 FINANCE 0.1187 0.0000 0.0000 0.0000 0.0000 1.0000 0.3235 LEVERAGE 0.4938 0.0689 0.3078 0.4868 0.6369 1.5650 0.2502 MB 2.7431 -11.0855 1.3990 2.1758 3.4747 18.8197 3.0426 LITIGATION 0.3954 0.0000 0.0000 0.0000 1.0000 1.0000 0.4889 LOSS 0.3682 0.0000 0.0000 0.0000 1.0000 1.0000 0.4823 CFO 0.0826 -1.3641 0.0456 0.0918 0.1376 0.5626 0.1084 MERAC 0.2153 0.0000 0.0000 0.0000 0.0000 1.0000 0.4110 ROA 0.0071 -0.8569 -0.0072 0.0432 0.0813 0.2369 0.1578 PPE 0.4862 0.0000 0.2301 0.4064 0.6772 2.1839 0.3287

Notes: The table provides summary statistics for the main variables in sample 2 (n=7,147) for firm-year observations from fiscal years 2001–2010. All numbers are rounded up to fourth decimal place. Variable definitions are shown in the Appendix A.

Table 2. Descriptive statistics sample 1

Variable Mean Min P25 Median P75 Max SD

ABSDISACC 0.0453 0.0006 0.0147 0.0325 0.0597 0.2538 0.0453 LNTFEE 14.0084 10.9647 13.2036 13.9940 14.7580 18.3175 1.1377 HIABNTFEE 0.2403 0.0000 0.0000 0.0223 0.4083 3.0034 0.3490 FAM 0.3371 0.0000 0.0000 0.0000 1.0000 1.0000 0.4727 FAM*HIABNTFEE 0.0801 0.0000 0.0000 0.0000 0.0000 2.7359 0.2269 LNMVE 7.0132 -0.6533 5.9415 6.9453 8.0405 12.9922 1.7332 FINANCE 0.1112 0.0000 0.0000 0.0000 0.0000 1.0000 0.3144 LEVERAGE 0.5346 0.0689 0.3600 0.5213 0.6662 1.5650 0.2581 MB 2.5125 -11.0855 1.2570 1.9896 3.1968 18.8197 3.0586 LITIGATION 0.2851 0.0000 0.0000 0.0000 1.0000 1.0000 0.4515 LOSS 0.3739 0.0000 0.0000 0.0000 1.0000 1.0000 0.4838 CFO 0.0826 -1.4095 0.0451 0.0912 0.1381 1.4835 0.1134 MERAC 0.2135 0.0000 0.0000 0.0000 0.0000 1.0000 0.4098 ROA 0.0060 -0.76991 -0.0110 0.0392 0.0774 0.2502 0.1493 PPE 0.5534 0.0000 0.2363 0.4470 0.7658 6.2728 0.4248

Notes: The table provides summary statistics for the main variables in sample 1 (n=12,911) for firm-year observations from fiscal years 2001–2010. All numbers are rounded up to fourth decimal place. Variable definitions are shown in the Appendix A.

(26)

4.2 Empirical results

The fee prediction model, as discussed before, has an average R2 _{is 0.6340, which is in line} with prior literature (Hoitash et al., 2007; Asthana & Boone, 2012; Krauß et al., 2015). The coefficients of the control variables were mostly significant in al years. The variable TENURE was ambiguous, because the coefficient was only significant in two of the ten observed years at the 10 percent level. I found that family firms pay structurally less audit fees, because in all years the coefficient of FAM was negative and significant at the 1 percent level, which is also in line with prior literature (Hoitash et al., 2007; Asthana & Boone, 2012; Krauß et al., 2015). The results from the regression models (2) through (5), to estimate various forms of abnormal earnings, are untabulated, because I regress these estimations per 2 digit industry group and year. For sample 1, this means that I run 350 regressions and for sample 2 these are 195 times 4, or 780 regressions. But the average R2 and regression estimates are in line with prior literature on earnings management (Kim et al., 2012; Chi et al., 2011; Hoitash et al., 2007).

In table 4 the results from the models (6) and (7) are tabulated. LNTFEE captures the effect of higher total audit fees on earnings management. As can be seen, all coefficients of LNTFEE are negative and significant at the 1 percent level. This indicates that a higher audit fee is associated with a higher auditor effort, which decreases the allowance for earnings management. HIABNTFEE captures the relation between positive abnormal audit fees and earnings management. All the coefficients are positive and significant at the 1 percent level except for the relation with abnormal production, which is significant at the 5 percent level. This indicates, in line with prior research, that positive abnormal audit fees create an economic bond between the auditor and the client. This leads to a higher allowance for earnings management.

The coefficients for FAM are negative and significant for accrual-based earnings management, abnormal cash flows from operations and combined real activities manipulation. This indicates that family firms employ less earnings management. The coefficient for FAM*HIABNTFEE are positive and significant for the model with the dependent accrual-based earnings management (at 10 percent level), abnormal expenses (at 1 percent level) and combined real activities manipulation (at 5 percent level). The other coefficients are positive, but not significant. This provides some evidence that auditors have a different role in family firms and that there can be an increased economic bond between the auditor and family firms when the client pays a positive abnormal audit fee.

(27)

In model (6) all control variables are significant. Where abnormal production is the dependent variable the smallest number of control variables is significant, indicating a possible misspecification of the model. The R2 for this model is also the lowest, but this is in line with other research on real activities manipulation (Kim et al., 2012).

Most of the client size related control variables are negatively associated with earnings management, which can indicate that larger firms generally have better governance structures. These structures prevent managers from engaging in managing of the earnings. The coefficients for MB are positive and significant for all models. This indicates that if firms are overvalued or have ‘hidden value’ off balance sheet (intangibles), they have a higher incentive to manage earnings. Additionally, in all models the coefficient for FINANCE is positive and significant at the 1 percent level. Additionally, the coefficient for MERAC is positive and significant at the 1 percent level in three out of the five models. This indicates that firms that issued equity or were involved in a merger or an acquisition have more abnormal earnings, which is expected.

The coefficients for LITIGATION are positive and significant at the 1 percent level for four out of the five regressions. This indicates that managers in high risk industries perform more earnings management.

Concluding, because the results of the main variables (LNTFEE, HIABNTFEE and FAM) are in line with prior research, there is evidence found that supports the individual sub hypotheses and therefore the main hypothesis. FAM*HIABNTFEE is positive in all models and significant in most, which supports the theory of economic bonding between the auditor and the client. Additionally, the results support the supply side of the alignment effect or alternatively the demand side of the entrenchment effect, where the family firm provides or the users demand higher quality financial statements, respectively. On the other hand, the results can indicate that the willingness to cooperate in family firms is so high that it decreases auditor independence. This, in turn, increases the allowance for earnings management.

(28)

Table 4. Family control on the relation between audit fees and audit quality

Model: (6) (7) (7) (7) (7)

Dependent: ABSDISACC ABSABNCFO ABSABNPROD ABSABNEXP COMBINEDRAM

LNTFEE -0.0073*** (0.000) -0.0099*** (0.000) -0.0046*** (0.007) -0.0076*** (0.000) -0.0114*** (0.000) HIABNTFEE 0.0122*** (0.000) 0.0121*** (0.000) 0.0105** (0.017) 0.0125*** (0.007) 0.0130*** (0.003) FAM -0.0053*** (0.000) -0.0063*** (0.000) 0.0056* (0.092) -0.0006 (0.867) -0.0120*** (0.000) FAM*HIABNTFEE 0.0042* (0.071) 0.0045 (0.215) 0.0081 (0.238) 0.0188*** (0.009) 0.0143** (0.035) LNMVE 0.0021*** (0.000) 0.0031*** (0.000) 0.0002 (0.848) -0.0017 (0.157) 0.0003 (0.789) FINANCE 0.0124*** (0.000) 0.0184*** (0.000) 0.0287*** (0.000) 0.0312*** (0.000) 0.0211*** (0.000) LEVERAGE 0.0054*** (0.002) 0.0060** (0.022) 0.0150*** (0.002) 0.0238*** (0.000) 0.0102** (0.034) MB 0.0004*** (0.003) 0.0019*** (0.000) 0.0034*** (0.000) 0.0034*** (0.000) 0.0020*** (0.000) LITIGATION 0.0061*** (0.000) 0.0118*** (0.000) 0.0016 (0.502) 0.0172*** (0.000) 0.0258*** (0.000) LOSS 0.0070*** (0.000) 0.0101*** (0.000) -0.0009 (0.751) 0.0085*** (0.003) 0.0184*** (0.000) CFO -0.0118*** (0.009) -0.0646*** (0.000) 0.0057 (0.688) 0.0091 (0.545) -0.0500*** (0.000) MERAC 0.0036*** (0.000) 0.0005 (0.729) -0.0027 (0.334) 0.0089*** (0.002) 0.0116*** (0.000) ROA -0.0464*** (0.000) 0.0052 (0.370) 0.0249** (0.010) -0.0027 (0.790) -0.0230** (0.016) PPE -0.0033*** (0.001) -0.0019*** (0.001) 0.0088* (0.017) -0.0059*** (0.125) -0.0203*** (0.000) Constant 0.1246*** (0.000) 0.1646*** (0.000) 0.1397*** (0.000) 0.1970 (0.000) 0.2089*** (0.000) R2 _0.0853 _0.0994 _0.0323 _0.0414 _0.0798 N 12,911 7,147 7,147 7,147 7,147 Prob > F 0.0000 0.0000 0.0000 0.0000 0.0000 Highest VIF 2.85 3.10 3.10 3.10 3.10

Notes: This table reports the results of the OLS regressions using a proxy for accrual-based earnings management (for model (6)) and four proxies for real activities manipulation (for model (7)) as the dependent variables, denoted as ABSDISACC, and ABSABNCFO, ABSABNPROD, ABSABNEXP and COMBINEDRAM respectively. This is done for firm-year observations

from fiscal years 2001–2010. The samples for model (6) and (7) consist of 12,911 and 7,147 respectively. The models that are estimated are as follows: for model (6): ABSDISACC=⨍(x), and for model (7): RAM=⨍(x), where x is the group of variables related to each specification. The p-values are displayed in parentheses. All numbers are rounded up to fourth decimal place. Variable definitions are shown in the Appendix A.

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

(29)

4.3 Additional analyses and robustness tests

I reperform the regressions of models (6) and (7) without the control variables for the sake of comparison. The results can be seen in appendix C. The direction of LNTFEE, HIABNTFEE, FAM, and FAM*HIABNTFEE do not change, but the coefficient for FAM*HIABNTFEE is now significant at the 5 percent level. Overall, this does not change the results found in the complete models.

Arthur Andersen stopped their audit activities in 2002 and my sample begins in 2001. This can impair the results of audit fees charged in the years following the change from Arthur Andersen to another auditor. Therefore, I reperform the regressions without the companies where the auditor was Arthur Andersen in 2001. The results and my main conclusion remains the same.

The quality of financial reporting can also be affected by the corporate governance, as previously suggested. Therefore, I downloaded the governance index (Gompers, Ishii and Metrick, 2003) from the Institutional Shareholder Services database as an extra control variable for the years 2002, 2004 and 2006. The untabulated results remain the same for both models (6) and (7), with comparable, but slightly higher R2 scores and slightly lower P-values for LNTFEE, HIABNTFEE, FAM and FAM*HIABNTFEE. The governance index is negatively related to ABSDISACC and to COMBINEDRAM and is in both cases significant at 10 percent level. Because the observations for model (6) are reduced 3.106 observations and the observations for model (7) are reduced to 1.830 observations I do not report these findings as my main results.

Additionally, I use different cut-off points for TENURE at 5 and 3 years. This is done to see if my models and results are sensitive to alternative definitions of auditor tenure. For both alternative cutoff points, the results remain the same and my main conclusion is not altered.

Another important robustness test I performed is the deletion of the observations in 2001. This is because the Sarbanes‐Oxley Act (SOX) was introduced in 2002, which mandates management evaluation and independent audits of internal control effectiveness. This can effect the relations found in my regression analyses. When all 2001 observations are excluded, the coefficient for FAM*HIABNTFEE is still positive and now significant at the 5 percent level for model (6). For model (7) the coefficient is only significant for the models with ABSABNEXP at the 5 percent level and at the 10 percent level for the model with COMBINEDRAM as the dependent variable. These last two are slightly less significant, but the overall results remain the same.

(30)

Lastly, I run the regressions without LNTFEE, because of the high correlation with HIABNTFEE. All key coefficients remain the same and have the same significance level. The R2 is slightly lower. In my opinion the model is better with the natural logarithm of total audit fees included, because these total fees have a positive relation with audit quality, possibly through higher audit effort. On the other hand, the abnormal positive fees have a negative relation with audit quality, explained by the economic bonding theory. If LNTFEE was not included, the relation between audit fees and audit quality will be captured incorrectly in my opinion.

(31)

5 Discussion and conclusion

Prior literature suggests that family firms have less total agency costs, that they provide higher quality financial reporting and that auditors therefore have less audit risk and charge lower audit fees to family firms. However, prior literature also suggests that the relationship between family firms and the auditor is different and that auditors play more of an advisory role, are among the most preferred advisors and generally have a longer tenure with family firms. Additionally, according to prior literature, positive abnormal audit fees seem to have a negative effect on auditor independence and therefore on the allowance for earnings management. Because of the different relation between family firms and their auditor, I investigated if higher abnormal audit fees in family firms have a different impact on the relation with the auditor than in non-family firms.

First, I found evidence that family firms pay structurally less audit fees than non-family firms from the fee prediction model that I employed, in line with prior literature. I sequentially estimated various unexpected earnings as a proxy of earnings management. With the (unexpected) residuals from the fee prediction model, I regressed the final two models (6) and (7) with the various measures of earnings management as the dependent variable. I found evidence for an increased effect of positive abnormal audit fees in family firms on multiple measures of earnings management.

I explain the findings using multiple theories. The main point of focus is the relationship between the auditor and family firms. First of all, the lower information asymmetry in family firms, between ownership and control (type I) can explain the overall positive relation between family firms and reporting quality. Therefore, auditors may need less effort in the audit of family firms, which explains the negative relation found in the fee prediction model between family firms and total audit fees.

Positive abnormal audit fees have a positive relation with all proxies of earnings management. This can be explained by the economic bonding theory. Because auditors are more dependent on the client, prior literature states that auditor can compromise audit quality, by allowing more earnings management. Prior literature states that this economic bonding theory is a problem, because auditors are more influenced by this bond than by reputational concerns.

My findings suggest, because of the different relation between family firms and their auditor, that the auditor is more inclined to compromise audit quality. In family firms the

(32)

controlling family members have more non-economic goals than regular managers or shareholders. This also explains why family firms report higher quality financial statements. These non-economic goals include ‘perpetuation of family values through the business’, ‘preservation of family dynasty’, ‘conservation of the family’s social capital’ and ‘family’s strong desire to infuse its values into the business as the spring well of organizational culture’. This explains why the family members have a more forward looking vision and want sustainability on the long-term.

According to prior literature the willingness of the client to cooperate with the auditor is determined by two important factors, namely long-term commitment and trust. Prior literature found that these two factors exist more in family firms, through longer auditor tenure and the advisory role of the auditors. The increase in willingness to cooperate can have a positive effect, because the auditor can more depend on the evidence that can be gathered. On the other hand, it can have negative effects, because if the client has a too high willingness to cooperate, the auditor can be overly trusting and too dependent on the provided information.

According to this prior literature and according to my findings, the auditors in family firms are compromised in their independence if the family firm pays positive abnormal audit fees. They have a higher allowance for earnings management in this case. I find evidence that fits these findings in prior literature and I argue that auditors in family firms are subject to a greater threat to compliance with ethical behavior. This is because there is more long-term commitment and trust between auditors and family firms, so that the independence is more easily compromised when higher abnormal audit fees are paid.

As every research, this paper also has limitations. In the two regression models (6) and (7) the R2 measures were slightly lower than in prior literature on earnings management. This indicates that the inclusion of more or better control variables could lead to a better specification of the model. These could include currently unobservable effects on audit quality, such as audit team composition, and the relative quality of internal controls and financial reporting systems.

Additionally, I use four measures of earnings management as proxies for audit quality. These measures might be a noisy proxy for management’s discretion over earnings. In addition to these four measures, the meeting or beating the market or accruals quality could also be used to increase the validity of the findings.

Another limitation is the fact that there have been a lot of regulatory changes within the audit profession in the period that I investigated, such as the introduction of SOX, changes in

Audit fees and audit quality in family firms