The effect of auditor size on real earnings management in a
mandatory joint audit setting
Student name: Fatih Sarikaya Student number: 11137932 Date: 26-06-2017
Word count: 14023
Supervisor: Dr. J.J.F. van Raak
MSc Accountancy & Control, specialization Accountancy Faculty of Economics and Business, University of Amsterdam
Statement of Originality
This document is written by student Fatih Sarikaya 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
In this study, a database research is conducted to examine the effect of auditor size on earnings management in a mandatory joint audit setting. Specifically, it is hypothesized that companies audited by at least one Big 4 auditor will exhibit lower levels of accrual-based earnings management but higher levels of real earnings management. The Modified Jones (1991) model is used to measure the discretionary accruals while real earnings management is proxied by abnormal levels of cash flow from operations, abnormal levels of production costs, abnormal levels of discretionary expenses, and a combined measured of these three. Using data over the period 2012-2015, I find evidence that companies audited by two Big 4 auditors exhibit more accrual-based earnings management. Additional analysis indicates that this is because companies audited by two Big 4 auditors have more income-decreasing discretionary accruals. Although a joint audit by two Big 4 auditors does not decrease the use accrual-based earnings management itself, it does lead to more conservative reporting as demonstrated by the higher income-decreasing discretionary accruals. A possible explanation for this is that Big 4 auditors have to maintain their reputation capital and thus are more conservative in their audit. Regarding real earnings management, I find evidence that companies audited by at least one Big 4 auditor exhibit lower levels of abnormal production costs. This constraining effect is even greater when a company is audited by two Big 4 auditors. This is inconsistent with prior studies. Based on these results the hypotheses of this research are rejected.
Keywords: Joint audit, auditor size, Big 4, non-Big 4, accrual-based earnings management, real earnings management.
Data availability: Data used in this study are obtainable from sources that are mentioned in this research.
Contents
1 Introduction ... 6
2 Literature review and hypothesis development ... 9
2.1 Joint audit ... 9
2.1.1 Joint audit and the European Commission ... 9
2.1.2 The French joint-audit model ... 10
2.1.3 Joint audits and accrual-based earnings management ... 11
2.1.4 Auditor-pair choice effect in a joint audit ... 13
2.2 Real earnings management ... 14
2.2.1 Real earnings management and the role of audit ... 15
2.3 Hypotheses development ... 16
3 Research design ... 19
3.1 Measuring earnings management ... 19
3.1.1 Accrual-based earnings management ... 19
3.1.2 Real earnings management ... 20
3.2 Sample selection ... 21 3.3 Regression model ... 23 3.3.1 Control variables ... 24 4 Results... 26 4.1 Pre-tests ... 26 4.2 Descriptive statistics ... 26 4.3 Correlation coefficients ... 28 4.4 Regression analysis ... 32 4.5 Additional analysis ... 35 5 Conclusion ... 38 References ... 41
Appendix A: Variable description ... 45 Appendix B: Observations distribution ... 46
1 Introduction
The global banking crisis has led to the questioning of the quality and the independence of auditors. The collapse or bailing out of financial institutions within a short period of receiving unqualified opinions fuelled the suspicion that auditors lacked the claimed expertise to render an independent and objective account of corporate affairs (Sikka, 2009). The crisis increased the demand for legislation and governance that improves the quality of the audits and that safeguards the independence of the auditors (Eilifsen & Willekens, 2008). In 2010, the European Commission published a Green Paper called Audit Policy: Lessons from the Crisis in which they among other things proposed mandatory joint audits under the assumption that it might increase auditors’ independence, audit quality, and audit market concentration (European Commission, 2010). A joint audit could encourage the growth of small and medium-sized audit practices for the audits of large companies (European Commission, 2010). The question is how this could impact the audit of financial statements. While the European Commission decided to not make joint audits mandatory, as of June 2016 new rules on statutory audit became applicable throughout the European Union. Public-interest entities can extend the audit engagement by an additional 14 years in the case of a joint audit compared to the 10 years in the case of a regular audit (European Commission, 2016) which might popularize the joint audit throughout Europe.
This study investigates how auditor size in a mandatory joint audit setting affects the level of accrual-based earnings management and real earnings management. In order to better understand the relationship between auditing and earnings management agency theory can help. In an agency relationship, one person (the principal) engages another person (agent) to perform a service on their behalf. This involves delegating some decision-making authority to the agent (Jensen & Meckling, 1976). If both the principal and the agent want to maximize their utility, it is likely that the agent will not always act in the best interest of the principal. Managers (agents) have incentives (e.g. management compensation plans) to adjust earnings to maximize their wealth (Becker et al. 1998). The adjustment of the earnings might not give a true and fair view of the economic state of the firm. This coupled with the information asymmetry between the principal and agent caused by delegation of decision rights makes it difficult for the principal to verify if the agent has behaved appropriately. Auditing can reduce information asymmetry between the principal and the agent by verifying the validity of the financial statements (Wallace, 1980). Becker et al. (1998) mention that the effectiveness of auditing and its ability to constrain the management earnings is expected to vary with the quality of the auditor and especially with the size of the auditor.
Francis et al. (2009) examine the French mandatory joint audit setting and find that firms audited by two Big 4 auditors have smaller income-increasing abnormal working capital accruals compared to firms audited by other auditor-pair choices. André et al. (2016) find that companies audited by two non-Big 4 auditors exhibit more earnings management than those audited by at least one Big 4 auditor. However, firms can also adjust earnings by manipulating real activities. Graham, Campbell & Shiva (2005) find that managers prefer real earnings management over accrual-based earnings management because the latter is more likely to draw auditor and regulatory scrutiny. This is further strengthened by Cohen & Zarowin (2010) and Chi et al. (2011) who find that Big N auditors while constraining accrual-based earnings management they increase the probability of firms engaging in real earnings management. Zang (2012) finds that companies use these types of earnings management as substitutes. The previously mentioned statutory audit rule change of June 2016 by the European Commission alongside the preference of managers to use real earnings management provides an incentive to revisit the concept of joint audits once again. To the best of my knowledge, no prior study has examined the effect of auditor size in a joint audit setting on the level of real earnings management. The aim of this study is to research how auditor size in a joint audit setting impacts the level of accrual-based earnings management and real earnings management. This study argues that given the implied substitutive nature (Zang, 2012) between accrual-based earnings management and real earnings management and the finding of Francis et al. (2009) and André et al. that firms audited by two Big 4 auditors exhibit less accrual-based earnings management and the assumption that scrutiny is at its highest under an auditor-pair combination of two Big 4 auditors, it is expected that firms audited by two Big 4 auditors will most likely exhibit lower levels of accrual-based earnings management and higher levels of real earnings management than firms audited by other auditor-pair combinations.
Like many other joint audit studies, this research focuses on France for a couple of reasons. France has mandatory joint audits for companies preparing consolidated financial statements. It has a large capital market compared to other countries with mandatory joint audits, meaning that the number of companies subjugated to the joint audit requirement is high. The audit market concentration in France is quite low. Huber (2011) finds that the Big 4 have a market share of 58% in France which implies there is significant diversification in the auditor-pair combinations.
The findings of this study are not in line with expectations. The results show that companies audited by two Big 4 auditors exhibit more accrual-based earnings management. An additional test indicates that this is mainly due to that the companies audited by two Big 4 auditors have more income-decreasing discretionary accruals. Although a joint audit by two Big 4 auditors does not decrease the use accrual-based earnings management itself, it does lead to more
conservative reporting as demonstrated by the higher income-decreasing discretionary accruals. This result partly shows the auditors’ asymmetric monitoring of the earnings management problem. That is more leniency toward income-decreasing earnings management (Piot & Janin, 2007). Regarding real earnings management, I find a significant negative effect on abnormal production costs. A joint audit by at least one Big 4 auditor thus constrains the ability of a company to manage earnings through overproduction compared to a joint audit by two non-Big 4 auditors. This constraining effect is even greater when a company is audited by two Big 4 auditors. This finding is in contrast with Cohen & Zarowin (2010) and Chi et al. (2011).
This study contributes to the literature of joint audits by examining the auditor-pair choice effect on accrual-based earnings management and real earnings management. By examining both earnings management methods the research gives a broader view on the effect of auditor size on earnings management. It is interesting to research this relationship because the new European statutory audit rules of June 2016 may increase usage of joint audits. This study also has practical implications for audit committees of companies using joint audits when they have to decide what auditor-pair combination they want from an earnings management perspective. It also partly extends prior literature (Chi et al. 2011; Cohen & Zarowin 2010) that examines the effect an auditor can have on the use of accrual-based earnings management and real earnings management.
The remainder of this thesis is as follows. In section two the relevant literature regarding the concepts of joint audit and earnings management is described. Based on the literature the hypotheses of this study are formulated. In section three the research design is described. This section covers how earnings management is measured in this study along with the empirical models that have been used to test the hypotheses. In section four the results of the research are discussed while in chapter five the conclusion is given.
2 Literature review and hypothesis development
The first part of this section provides an overview of the practice and the research that has been undertaken regarding joint audits. The joint audits studies seem to be centered around its effect on audit quality (measured by abnormal accruals) and audit fees. The second part provides a short overview of the research that has been undertaken regarding real earnings management with an emphasis on the role of audit. By combining the literature around joint audits and real earnings management the hypotheses are presented in the third part of this section.
2.1 Joint audit
The general concept of a joint audit is an audit in which financial statements are audited by two independent audit firms with shared audit effort where there is one single auditor's report signed by both audit firms for which they are jointly liable (Audosset-Coulier et al. 2012). It is important to emphasize that a joint audit is not the same as a dual audit or a double audit. In a dual audit, each audit firm audits different financial information and provides their own audit opinion. In a double audit, the audit work is fully performed twice. This is not the case in a joint audit (Audosset-Coulier et al. 2012).
2.1.1 Joint audit and the European Commission
Although the concept of a joint audit has been around for quite some time it attracted a lot of attention in 2010 when the European Commission in their green paper Audit Policy: Lessons from the Crisis stated that they could consider making joint audits mandatory in order to encourage the emergence of other players and the growth of small and medium-sized audit practices for the audits of large companies (European Commission, 2010). In the green paper, it was mentioned that in a lot of European member states the Big 4 had a market share of more than 90% in the audit of listed firms. In France where the joint audit is mandatory, the Big 4 have a market share of around 58% (Huber, 2011, p. 6)
The European Commission received a lot of responses to their green paper. The responses also covered their joint audit proposal. The professional bodies and associations linked with the profession think that joint audits could dynamise the audit market but that further research regarding audit quality, costs, and liability issues should be done before making joint audit mandatory (European Commission, 2011). The Big Four have opposed the proposal of the European Commission as they felt that joint audits will impair audit quality and will cause co-coordination problems (European Commission, 2011). Mid-tier audit firms strongly support joint audits. Some of them believe that joint audits will increase competition and can potentially raise
quality (European Commission, 2011). Many investors feared that a joint audit could increase the costs of audits. Academics mentioned that joint audits should be a choice and not mandatory. Preparers are not opposed to joint audits if it is well balanced, well-framed and with very strict requirements (European Commission, 2011). Some companies mention that a joint audit might make collusion with the management more difficult but a majority of the companies mention for example that the costs incurred with a joint audit are 15-20% higher than a regular audit (European Commission, 2011). The European Commission decided to not make joint audits mandatory but instead encourage the application of it (European Commission, 2011). As of June 2016, Member States can allow public interest entities to extend the audit engagement by an additional 14 years in the case of a joint audit compared to the additional 10 years in the case of a regular audit (European Commission, 2016).
2.1.2 The French joint-audit model
The practice of joint audits and especially mandatory joint audits is not something that is widely applied in most countries. France is part of the few countries that have mandatory joint audits for all listed companies in France. The requirement started with the 1966 Act which made joint audits mandatory for listed firms and non-listed firms with a share capital value exceeding a certain threshold (Audosset-Coulier et al. 2012). The emergence of the joint audit requirement can be explained by two rationales (Bennecib, 2004). First, it can deal with the problem of default by one of the auditors and second, it can safeguard auditor independence by dividing managerial pressure from the client over multiple auditors.
In 1984 the scope of the joint audit requirement was modified and since then according to the Code de Commerce Art. L 823-20 all companies preparing consolidated financial statements have to be jointly audited (Azibi & Velte, 2015). The law requires a balanced division of the work of both auditors in order to ensure an efficient dual control mechanism (Gonthier-Besacier & Schatt, 2007). The work is usually allocated either by regions or by divisions of the client. After the auditors finished their part of the audit they review each other’s work. The audit firms sign the audit report on the whole financial statements and not just on their own work (Deng et al. 2014). Another requirement that sets France apart is that the statutory auditors are appointed for a six-year term (renewable) during which the auditors cannot resign and cannot be dismissed. A court decision is needed when one of the parties wishes to end the engagement (Francis et al. 2009). With the implementation of the law on financial security in 2003, there is mandatory audit partner rotation every six years for listed companies (Audousset-Coulier, 2015).
Besides France, there are a couple of other countries that either have or used to have mandatory joint audits or have voluntary joint audits. Audosset-Coulier et al. (2012) have mapped these countries. Their findings show that mandatory joint audits were primarily aimed at listed companies and/or financial institutions and it also shows that the joint audit models differ to that of France. For example, up until 2005, Denmark had mandatory joint audits for listed companies in which at least one of the audit firms had to be state authorized (Azibi & Velte, 2015). The law in Denmark did not specify how the workload had to be shared between the audit firms. Canada and Sweden abolished their joint audit requirement for banks respectively in 1991 and 2006. These countries mentioned the higher costs associated with joint audits as one of the reasons for abolishing the mandate. Algeria and Saudi-Arabia have a joint audit requirement for banks since 2003 and 1966, respectively. Kuwait and Morocco adopted mandatory joint audits for listed companies in 1994 and 1996, respectively.
2.1.3 Joint audits and accrual-based earnings management
There are two main types of earnings management; accrual-based earnings management and real earnings management. Managing accruals are achieved by changing the accounting methods or estimates when presenting a given transaction in the financial statement (Zang, 2012) while real earnings management are deviations from normal operational practices in order to meet certain financial reporting goals in the normal course of operations (Roychowdhury, 2006). Regarding accrual-based earnings management, there is a differentiation between nondiscretionary accruals and discretionary accruals (Geiger & North, 2006). Nondiscretionary accruals are the normal level of accruals for the company based on factors such as company size, revenue growth, and type of operating industry. The discretionary accruals are the unexpected component reported by the company and it is the difference between the actual levels of accruals and the expected (normal) levels of accruals for the company (Geiger & North, 2006).
Most studies regarding joint audits, as can be read below, examine whether joint audits can positively affect audit quality. Audit quality, however, is difficult to measure because the amount of assurance auditors provide is unobservable (DeFond & Zhang, 2014). DeAngelo (1981) divides audit quality into two components; auditor competence and auditor independence. DeAngelo (1981) defines auditor competence as the auditor’s ability to discover errors or breaches in the accounting system and defines auditor independence as the conditional probability that, given a breach has been discovered, the auditor will report the breach. Related to the two audit quality components of DeAngelo (1981), Piot (2007) mentions that a joint audit can thus enhance the audit quality in two ways. The first way is that a joint audit can ensure that an audit is performed
with due diligence because the co-auditors cross-review and monitor each other which is related to the component auditor competence. The different auditors in a joint audit might also have an area of internal specialization which can also enhance audit quality (Baldauf & Steckel, 2012). The second way being that the presence of auditors might protect auditor independence because the effect of managerial pressure is divided across multiple auditors. In the literature audit quality is often measured by examining the accrual quality of companies (DeFond & Zhang, 2014) and this is also the case for joint audits (André et al. 2016; Azibi & Velte, 2015, Francis et al. 2009, Lesage et al. 2017; Zerni et al. 2012). The studies examine whether joint audits constrain the level of accrual-based earnings management.
The empirical evidence regarding the effect of joint audits on audit quality as measured by accrual quality is mixed. Azibi & Velte (2015) compare the levels of abnormal working capital accruals and discretionary accruals between French and German firms during the period 2008-2012. Their analysis shows that joint audits do not have a significant effect on the abnormal working capital accruals and on the discretionary accruals. The authors could not find a clear positive link between joint audits and audit quality. The higher audit costs they find for French firms compared to German firms do not translate into higher audit quality. Lesage et al. (2017) examine the Danish setting between 2002 and 2010. In 2005 mandatory joint audits were abolished in Denmark. They also find that the association between joint audits and abnormal accruals is insignificant and that the higher audit fees they observe for joint audits do no result in a higher audit quality. André et al. (2016) confirm the findings of Azibi & Velte (2015) and Lesage et al. (2017). André et al. (2016) find based on a matched sample that a French joint audit with at least on Big 4 audit firm costs between 35% and 70% more than a British or an Italian Big 4 audit. The higher observed audit fees do not appear to be associated with higher audit quality.
On the other hand, Zerni et al. (2012) do find a positive effect regarding joint audits and audit quality. They have compared Swedish private and public firms that use voluntary joint audits and single audits. The authors find that firms using joint audits have a “higher degree of earnings conservatism, lower abnormal accruals, better credit ratings and lower perceived risk of becoming insolvent within the next year than other firms”. The authors also find that the choice of a joint audit is associated with substantial increases in the audit fees.
It thus seems that in a mandatory setting joint audits do not lead to lower abnormal accruals compared to single audits despite the higher audits costs associated with joint audits. In a voluntary setting joint audits can have a positive effect on audit quality but further research needs to be performed regarding voluntary joint audits. Furthermore, most of the studies focussed on
accrual-based earnings management while none examined the relation between joint audits and real earnings management.
2.1.4 Auditor-pair choice effect in a joint audit
A joint audit can consist of multiple audit firms varying in size. This thesis examines the effect of the auditor size i.e. auditor-pair choice in joint audits on earnings management. The specific research regarding the effects of auditor-pair choice is limited. Francis et al. (2009) examine based on a French sample size of 467 firms for the year 2003 whether a firms’ ownership structure affects its auditor-pair choice and subsequently whether this choice affects the earning quality. They find that firms with less concentrated ownership structures and with greater ownership by investors are more likely to have a Big 4 auditors as one or both auditors. Two Big 4 accounting firms are more likely to be used when firms’ ownership structure is less dominated by family holdings. Regarding the effect of earning quality, the authors find that firms using one Big 4 auditor (paired with a non-Big 4 auditor) or two non-Big 4 auditors have smaller income-increasing abnormal working capital accruals compared to firms that use no Big 4 auditors. Firms audited by two Big 4 auditors have even smaller income-increasing abnormal working capital accruals compared to firms audited by just one Big 4 auditor. The findings of André et al. (2016) are similar to those of Francis et al. (2009). While they do not find a significant difference between the levels of abnormal working capital accruals between the different auditor-pair combination they find, albeit weak, based on a matched sample that companies audited by two non-Big 4 auditors exhibit more earnings management proxied by abnormal accruals than those audited by at least one Big 4 auditor. Piot & Janin’s (2007) findings show that companies audited by Big Five do not differ from others in terms of absolute and signed abnormal accruals which is in contrast to Francis et al. (2009) and André et al. (2016).
Deng et al. (2014) examine a different type of effect regarding auditor-pair choice. They compare whether the audit evidence precision differs in an audit performed by one big firm; joint audits by two big firms; joint audits by one big firm and one small firm. Deng et al. (2014) argue that joint audits could potentially impair audit quality because it could lead to the free-riding problem. This problem may occur when one of the audit firms saves costs by investing less in its own audit work and thus taking advantage of the other audit firms hard work. Free-riding could reduce information precision and therefore audit quality. The researchers find that when a joint audit is composed of one big audit firm and one small audit firm the total precision of audit evidence is lower than that of a single audit by a big audit firm. A joint audit by two big firms leads to the same audit evidence precision as a single audit by one big firm.
2.2 Real earnings management
The previous paragraphs showed that a lot of the joint audit research focus on its effect on accrual-based earnings management and use this as a proxy for audit quality (André et al. 2016; Azibi & Velte, 2015; Francis et al. 2009; Lesage et al. 2017; Zerni et al. 2012). It is also possible to manage earnings through manipulation of real activities (Roychowdhury, 2006). This method is called real earnings management. Roychowdhury (2006) defines real activities manipulation as “departures from normal operational practices, motivated by managers’ desire to mislead at least some stakeholders into believing certain financial reporting goals have been met in the normal course of operations”. Some methods of real earnings management such as price discounts and lowering discretionary expenses can be the right thing to do in certain economic circumstances. According to Roychowdhury (2006) managers are engaging in real earnings management if they use more real activities manipulation than is normal given certain economic circumstances.
The main difference between accrual-based earnings management and real earnings management is that real earnings management have direct effects on cash flows (Cohen & Zarowin, 2010). Graham et al. (2005) find that managers prefer real activities manipulation over accruals-based activities. 80% of the survey participants mentioned they would decrease discretionary spending on R&D, advertising, and maintenance to meet an earnings target. It also seems that there is a trade-off between the two earnings management strategies. Zang (2012) analyzed the trade-off decision as a function of the costs of the two activities and provides evidence that there is a substitution effect between the two earnings management strategies. Zang (2012) finds that real earnings management is constrained by firms’ competitive status in the industry, financial health, scrutiny from institutional investors, and the immediate tax consequences of manipulation.
Cohen & Zarowin (2010) mention two reasons why there might be a greater willingness towards real earnings management than accruals-based earnings management. The first reason is that it is more likely that accrual-based earnings management will draw auditor or regulatory scrutiny than real earnings management. It is more difficult for auditors or regulators to challenge real economic actions that are taken in the operations of business than to challenge accounting decisions in the case of accrual-based earnings management. In 2002, the Sarbanes-Oxley Act (SOX) was implemented with the intention to improve accountability and transparency in companies and deter future abuses. This enhanced regulatory scrutiny might have decreased the usage of accrual-based earnings management. There are studies that provide evidence for this. Cohen et al. (2008) find that accrual-based earnings management increased steadily from 1987 until
2002, the year SOX was implemented. After SOX, there is a significant decline in the use of accrual-based earnings management. Prior to SOX, the use of real earnings management was declining while after SOX the use increased. Bartov & Cohen (2009) find that in the Post-SOX period there is a decline in upward accrual-based earnings management and an increase in upward real earnings management activities. Lobo & Zhou (2006) document an increase in conservatism in financial reporting following SOX and as a result of this firms report lower discretionary accruals than before SOX. Zang (2012) also confirms that accrual-based earnings management is constrained by SOX. Further, Cohen & Zarowin (2010) find that firms in high litigation industries are more likely to use real earnings management instead of accrual-based earnings management. Litigation is the primary penalty for earnings manipulation and since accrual-based earnings management is more likely to be detected firms in high litigation industries have fewer incentives to use accrual-based earnings management. The second reason why managers might be more willing to use real earnings management is that relying only on accrual-based earnings management is risky.
The use of real earnings management can be more harmful to a firm than accrual-based earnings management because real earnings management “can adversely affect firm value by distorting normal operating activities and thereby harming the relationship with key stakeholders such as customers, employees, and communities” (Cho & Chun, 2016). Cohen & Zarowin (2010) examine the effects of accrual-based earnings and real earnings management on a firm’ future performance by comparing the periods preceding and following a seasoned equity offering. They find that real earnings management is more likely to be associated with earnings declines than accrual-based earnings management. Bill, Iftekhar, & Lingxiang (2016) examine the impact of firms’ abnormal business operations on their future crash risk in stock prices. They find that REM firms are about 30% more likely to experience a crash in the following year than non-REM firms are.
2.2.1 Real earnings management and the role of audit
Prior research shows that Big 8 auditors constrain earnings management through abnormal accruals (Becker et al. 1998; Francis, Maydew, & Sparks, 1999). The question is whether this also applies to real earnings management. Commerford et al. (2016) interviewed 20 auditors in-depth about their perceptions of and responses to REM. The authors find that auditors are aware of real earnings management and identify real earnings management through formalized protocol that includes analytical procedures, discussions with management, or their knowledge of the business.
Commerford et al. (2016) document that real earnings management is a significant source of auditor discomfort.
Cohen & Zarowin (2010) assume that firms facing greater scrutiny prefer to use real earnings management because this type of earnings management is less likely to be scrutinized by auditors and regulators. Cohen and Zarowin (2010) assume that scrutiny increases with the presence of a Big 8 auditor, and with the auditors’ tenure. They predict that the probability for seasoned equity offering firms to use real earnings management is positively related to both the presence of a Big 8 auditor and the auditors’ tenure. These two components should increase the probability of detecting accruals-based earnings management but should not have an effect on detecting real earnings management as that typically falls outside of the auditor’s responsibility. The authors find that a Big 8 auditor and longer auditor tenure increases the probability of real earnings management by about 8% and 5.5%. In addition to Cohen & Zarowin (2010), Chi et al. (2011) research if enhanced audit quality is associated with greater real earnings management. They find for firms that meet or just beat earnings benchmarks and firms that issue seasoned equities that city-level auditor industry expertise and audit fees are associated with higher levels of real earnings management. Chi et al. (2011) also find that the presence of a Big N auditor and longer auditor tenure is associated with greater real earnings management.
Contrary to the aforementioned studies, Zangs’ (2012) results do not indicate that real earnings management increases with the presence of a Big 8 auditor and auditor tenure. A possible reason the authors adduce is that 93.7 percent of the firms in their sample have a Big 8 auditor. However, 96.8% of the sample of Chi et al. (2011) were audited by Big N audit firms. A possible explanation for the difference in auditors’ effect on real earnings management is that Zang’s (2012) sample is between 1987-2008 while the sample period of Chi et al. (2011) is between 2001-2008. In paragraph 2.2 of this thesis, it was mentioned that implementation of SOX in 2002 has lead firms using more real earnings management.
2.3 Hypotheses development
It is interesting to know if the general findings of Francis et al. (2009) and André et al. (2016) that a joint audit by two Big 4 auditors leads to lower earnings management measured by abnormal accruals than other joint audit compositions still hold 1. The findings of Francis et al. (2009) are
1 As mentioned before, Francis et al. (2009) only find a significant effect when examining the income-increasing
based on a sample from 2003. With the implementation of IFRS in 2005, there was a change in financial reporting standards for European listed firms. Research regarding the effect of IFRS on earnings management is mixed. (Ipino & Parbonetti, 2017) find that in EU countries with strong legal enforcement IFRS has led to a decrease in accrual-based earnings management but an increase in real earnings management while Doukakis (2014) in a similar research does not find a significant impact of IFRS on earnings management. The findings of André et al. (2016) are based on a sample from 2007-2011. Within this period, the financial crisis happened and Cimini (2015) find that during the financial crisis fiscal years 2008-2012 earnings management decreased in the large majority of European countries. The change in reporting standards and the financial crisis event thus provide an incentive to once again research the relation between auditor-pair choice and accrual-based earnings management.
Based on the literature which indicates that companies audited by Big N auditors exhibit lower abnormal accruals (Becker et al. 1998; Chi et al. 2011; Cohen & Zarowin, 2010; Francis et al. 1999) and that Big 4 auditors provide higher audit quality than non-Big 4 auditors because Big 4 auditors have more auditing and industrial expertise (Francis, 2004), I pose the following hypotheses:
H1-A: Companies audited by an audit-pair combination of two Big 4 auditors exhibit lower levels of accrual-based earnings management than companies audited by one Big 4 auditor and one non-Big 4 auditor.
H1-B: Companies audited by an audit-pair combination of one Big 4 auditor and one non-Big 4 auditor exhibit lower levels of accrual-based earnings management than companies audited by an audit-pair combination of two non-Big 4 auditors.
Regarding real earnings management, the literature review shows that in the last decade real activities manipulation has become more popular as a form of earnings management than manipulating accruals (Graham et al. 2005; Cohen et al. 2008). Zang’s (2012) research implies that there is a substitutive relationship between real earnings management and accrual-based earnings management. Graham et al. (2005) find that managers prefer real earnings management over accrual-based earnings management because the latter is more likely to draw out auditor and/or regulatory scrutiny. It is more difficult for auditors or regulators to challenge real economic actions (read: real earnings management) that are taken in the operations of business than to challenge accounting decisions in the case of accrual-based earnings management. More important for this research is that Cohen et al. (2010) and Chi et al. (2011) assume and find that due to the increased
scrutiny in the presence of a Big N auditor a firm’s ability to manage earnings via accruals is constrained which results in those firms to resort to more real earnings management.
The question now is if the effect of a Big N auditor on real earnings management holds when extended to a mandatory joint audit setting such as France. Francis et al. (2009) and André et al. (2016) find respectively that a firm audited by two Big 4 audit firms exhibit lower levels of abnormal working capital accruals and lower abnormal accruals than other joint audit compositions. Deng et al. (2014) find that the audit evidence provided two big audit firms is higher than that provided by one big audit firm and one small audit firm. Given the implied substitutive nature between real earnings management and accrual-based earnings management alongside the finding of Francis et al. (2009) and André et al. (2016), one could expect that firms audited by two Big 4 audit firms will exhibit higher levels of real earnings management as they are more constrained in their use of accrual-based earnings management compared to firms audited by a different auditor-pair combination. Furthermore, based on Deng et al (2014), I assume that companies audited by two Big 4 audit firms face greater scrutiny as proxied by higher audit evidence and thus resort to real earnings management as the higher audit evidence might also constrain a company’s ability to use accrual-based earnings management. This all leads to the following hypotheses:
H2-A: Companies audited by an audit-pair combination of two Big 4 auditors exhibit higher levels of real earnings management than companies audited by one Big 4 auditor and one non-Big 4 auditor.
H2-B: Companies audited by an audit-pair combination of one Big 4 auditor and one non-Big 4 auditor exhibit higher levels of real earnings management, than companies audited by an audit-pair combination of two non-Big 4 auditors.
3 Research design
This section starts with describing how the two forms of earnings management are measured in this study. Accrual-based earnings management is measured by using the Modified-Jones (1991) model. Real earnings management is measured by examining the abnormal levels of cash flow from operations, production costs, and discretionary expenses. Further, the process of the sample selection is explained and the chapter is concluded with the regression models that are used to test the hypotheses of this study.
3.1 Measuring earnings management
3.1.1 Accrual-based earnings management
Total accruals consist of discretionary accruals and discretionary accruals. The non-discretionary accruals are the expected level of accruals for a company based on its size, operating industry, and revenue growth (Jones, 1991). The discretionary accruals are the unexpected component of total accruals. It is argued that the level of these discretionary accruals is a reflection of management’s use of financial reporting discretion to either increase of decrease net income (Geiger & North, 2006). Therefore, the discretionary accruals are of interest when examining earnings management.
In order to measure accrual-based earnings management, this research uses the Modified-Jones as used by Dechow et al. (1995). The difference between this model and the original Modified-Jones (1991) model is that the change in revenues is adjusted for the change in receivables in the event period (Dechow et al. 1995). The Modified-Jones model implicitly assumes that all changes in credit sales in the event period result from earnings management. The reasoning behind this is that it is easier to manage earnings by exercising discretion over the recognition of revenue on credit sales than over the recognition of revenue on cash sales (Dechow et al. 1995).
The first step in the Modified-Jones model is determining the total accruals. The total accruals are defined as the difference between income before extraordinary items and cash flow from operations. The second step is estimating the non-discretionary part of the total accruals and the residual is the discretionary part of the accruals (Dechow et al. 1995). The non-discretionary accruals are the sum of changes in cash revenues (total revenues – credit sale revenues) and the level of property, plant & equipment. All variables are scaled by lagged total assets. The proportion of accruals that is not explained by total accruals are the discretionary accruals. The Modified-Jones model is thus as follows:
(1) TAt = β0 + β1(1/At-1) + β2(ΔREVt – ΔRECt) + β3PPEt +
ε
twhere TAt are the total accruals in year t and At-1 is the lagged total assets. ΔREVt and ΔRECt are the changes in net sales and receivables in year t. PPEt is the value of property, plant & equipment
in year t.
3.1.2 Real earnings management
To examine the level of real earnings management the empirical model developed by Dechow, Kothari, and Watts (1998) is used. Other studies have used (Cohen & Zarowin, 2010; Cohen et al. 2008; Chi et al. 2011) this empirical model with small adjustments. The empirical model estimates the normal levels of three measures in order to calculate the level of real earnings management. The three measures are cash flow from operations (CFO) as reported in the cash flow statement; discretionary expenses (the sum of advertising expenses, R&D expenses, and selling general and administrative SG&A expenses) and production costs (the sum of COGS and change in inventory during the period). The residuals from the empirical model are used as proxies for real earnings management. The residuals reflect the abnormal level of each measure.
The three measures above can capture three real activities manipulation methods (Roychowdhury, 2006). The first is by manipulating sales. Roychowdhury (2006) defines this as a manager’s attempt to temporarily increase sales during the year by offering price discounts or more lenient credit terms. The cash inflow per sale from these additional sales is lower as margins decline. As a result, the lower margins cause production costs relative to sales to be abnormally high. So, by manipulating sales it is expected that it will lead to lower current-period CFO and higher production costs than what is normal given the sales level (Roychowdhury, 2006). The second method is reducing the discretionary expenses. These expenses are generally expensed as they are incurred. By reducing discretionary expenses firms can increase earnings. Firms that use this method should exhibit unusually low discretionary expenses. The reduction in these expenses reduces cash outflows so this should have a positive effect on the abnormal CFO (Roychowdhury, 2006). The third method is overproduction. Firms can produce more goods than necessary to meet expected demand. Because of the higher production levels, the fixed overhead costs are spread over a large number of units. The total cost per unit declines as long as the reduction in fixed costs is not offset by an increase in marginal cost per unit (Roychowdhury, 2006). This method results in a higher production cost than normal given the sales levels and the cash flow from operations is lower than normal given the sales levels.
(2) CFOt/At-1 = β0 + β1(1/At-1) + β2(St/At-1) + β3(ΔSt/At-1) +
ε
twhere CFOt is the total cash flow from operations in year t and At-1 is the lagged total assets. St is the net sales in year t while ΔSt is the difference between net sales in year t and year t-1. The residuals
of the regression represent the abnormal levels of cash flow from operations (AB_CFO).
The normal level of discretionary expenses will be estimated by using the following model in which discretionary expenses are expressed as a function of lagged sales:
(3) DISXt/At-1 = β0 + β1(1/At-1) + β3(St-1/At-1) +
ε
twhere DISXt is the discretionary expenses in year t. The residuals of the regression represent the abnormal levels of discretionary expenses (AB_DISX).
The third proxy is the normal levels of production costs. The normal levels will be estimated using the following model in which expenses are seen as a linear function of contemporaneous sales:
(4) PRODt/At-1 = β0 + β1(1/At-1) + β2(St/At-1) + β3(ΔSt/At-1) + β3(ΔSt-1/At-1) +
ε
twhere PRODt is the sum of the cost of goods sold and changes in inventory during the year. The
residuals of the regression are the abnormal levels of production costs (AB_PROD).
Firms that manage earnings upwards are likely to have unusually low cash flow from operations, and/or unusually low discretionary expenses, and/or unusually high production costs (Cohen & Zarowin, 2010). The measures AB_CFO and AB_DISX are multiplied by -1 to make all measures face the same direction, so that the higher amount, the more likely it is that the firm is engaging in sales manipulation and reducing discretionary expenses to manage reported earnings upwards. These three measures are combined into a single measure (TOTAL_REM) to capture the overall effects of abnormal activities because firms that manage earnings upwards are likely to use multiple activities (Kang & Kim, 2012). The results are also reported for the three individual real earnings management measures as they may have different implications.
3.2 Sample selection
This thesis focusses on France as it is one the few countries with mandatory joint audits for listed companies. It also has a large capital market meaning that the number of companies subjugated to the joint audit requirement is high. According to Huber (2011) the audit market concentration in France is quite low as the Big 4 only has a market share of 58%. This is important considering the emphasis on auditor size in this study, therefore having a considerable amount of non-Big 4
auditors is important. The sample period is from 2012 to 2015 to ensure a IFRS and post-crisis setting. However, the proxies of earnings management require lagged variables for up to two years. This means that for the lagged variables data is also gathered for 2010 and 2011.
The sample selection process starts by using Datastream to retrieve an overview of all firms that are or were listed firms in France. This is done by filtering the stock exchange Euronext Paris with France as market and the Euro as currency. Only equities are selected for the sample. This results in a sample of 2208 firms. Only companies that have an ISIN number starting with FR are selected as these companies have to be jointly audited. This results in 1955 firms. The measures in this study require SIC codes in order to identify the industries the firms operate in. SIC codes are retrieved from Datastream and by cross-referencing the ISIN numbers to Compustat Global. For 982 companies, it was not possible to retrieve the SIC codes. Consistent with Roychowdhury (2006) and Zang (2012), companies with SIC codes ranging between 4400-4999 (regulated industries) and between 6000 – 6999 (financial institutions) and/or with General Industry Classifications ranging between 4-6 (financial institutions) are deleted from the sample as these companies may have additional reporting requirements. This results in a sample of 732 firms for which data gathering starts. In order to determine the real earnings management measures and the discretionary accruals, I require at least 10 observations per 2-digit SIC industry-year combinations2. If there are less than 10 observations for any 2-digit SIC industry-year combinations, 1-digit SIC codes are used. This is done in order to maintain a decent sample size. Appendix B shows the distribution of the observations per 2-digit SIC industry.
Audit firm name data are hand collected from the annual reports and/or registration documents. This study only uses the statutory auditors of companies. Prior studies (Cohen et al. 2008, Cohen & Zarowin, 2010; Zang, 2012) calculate the real earnings management proxy discretionary expenses as the sum of R&D, advertising, and SG&A expenditures. Datastream does not contain a separate item for advertising expenses. This study therefore only uses the item SG&A in order to estimate the normal levels of discretionary expenses. However, the item SG&A in Datastream contains advertising expenses and R&D expenses (Reuters, 2015). When SG&A is not available for a certain firm-year, Compustat Global is used to retrieve SG&A. Compustat Global does not have the item advertising expenses but the data definition guide of Compustat mentions that the item SG&A can also include advertising and R&D expenses (Compustat, 2002). However, when R&D expenses are reported separate from SG&A in Compustat, then these two variables
are combined. This is done only if the variable SG&A has a value3. The reason for using Compustat is because Datastream does not have a lot of data regarding SG&A. Table 1 shows the number of available observations for the earnings management variables for this study.
Table 1 Sample selection
3.3 Regression model
The regression models as can be seen below are used to test the hypotheses H1-A, H1-B, H2-A, and H2-B. This model tries to find whether companies audited by two Big 4 auditors exhibit lower levels of accrual-based earnings management and higher levels of real earnings management than companies audited by other auditor-pair choices. In these models, the benchmark group is NONBIG4NONBIG4 (a company audited by two non-Big 4 auditors). The models used are as follows:
DA = β0 + β1BIG4BIG4 + β2BIG4NONBIG4 + β3TOTAL_REM + β4MTB + β5SIZE +
β6LEVERAGE + β7ROA + β8LOSS + β9LITIGATION +
∑
YEAR DUMMY +∑
INDUSTRYDUMMY + ε
REM = β0 + β1BIG4BIG4 + β2BIG4NONBIG4 + β3DA β4MTB + β5SIZE + β6LEVERAGE
+ β7ROA + β8LOSS + β9LITIGATION +
∑
YEAR DUMMY +∑
INDUSTRY DUMMY + εwhere DA is the dependent variable and equals to discretionary accruals and where REM equals to the three individual real earnings management measures (AB_CFO), (AB_PROD), and (AB_DISX) and the combined measure of these three (TOTAL_REM). The higher the value of the discretionary accruals the more likely that a firm is managing earnings (Hribar & Nichols, 2007). The measures of real earnings management in this study increase when firms manage earnings through real activities. This is because in this study AB_CFO and AB_DISX are multiplied by -1.
3 I follow Cohen & Zarowin (2012) who as long as SG&A is available, set advertising expenses and R&D to zero if
Criteria Observations
Starting sample size 2.928
Less missing financial data (1.551)
Less missing auditor name or auditors not clear (157) Less to few observations per industry-year combination (18)
The coefficients of interests are the joint audit variables (β1 and β2). There are three general
options regarding the composition of a joint audit. A company can be audited by either two Big 4 auditors or by one Big 4 auditor and one Non-Big 4 auditor or by two Non-Big 4 auditors. Big 4 auditors in this study are defined as Ernst & Young, Deloitte, PwC, and KPMG. Since a higher value of discretionary accruals implies a firm is managing earnings, I expect regarding the accrual-based earnings management model that the coefficients β1 and β2 to be negative. Since the
empirical construct of the real earnings management measures in this study increases when firms use more real earnings management, I expect the coefficients β1 and β2 to be positive. The control
variables of the regression model are explained in 3.3.1.
3.3.1 Control variables
The regression models include several control variables to control for effects that might have an influence on earnings management. As mentioned before, Cohen & Zarowin (2010) and Zang (2012) find that firms engage in both accrual and real earnings management activities and that they substitute between the two methods. The regression model for accrual-based earnings management contains the variable TOTAL_REM and the real earnings management model contains discretionary accruals (DA) to control for the substitutive nature between the two earnings management methods. Roychowdhury (2006) includes the variables MTB and SIZE to control for systematic variation in abnormal CFO, production costs and discretionary expenses with growth opportunities and size. MTB or the market-to-book ratio is the ratio of market value of equity to book value of equity. In this study, SIZE is the logarithm of total assets. To control for variations in capital structure and profitability the variables LEVERAGE and ROA are added to the regression model (Healy & Wahlen, 1999). LEVERAGE is measured by dividing the total debt by the total assets. ROA, or the return on assets, is measured as the net income before extraordinary items divided by total assets. A LOSS control variable will be added as a dummy variable indicating one if net income of the current year is negative and zero otherwise.
Cohen & Zarowin (2010) model a firm’s choice to use real or accrual-based earnings management as a function of its ability to use accrual-based earnings management and the cost of doing so. They explain that the primary penalty for earnings manipulation is litigation. Because accrual-based earnings management is more likely to be detected than real earnings management, the likelihood of getting sued when using accruals to manage earnings is higher. Cohen & Zarowin (2010) find that firms in high litigation industries tend to use more real earnings management in order to avoid litigation, therefore this study controls for high litigation industries. Following Cohen & Zarowin (2010), high litigation industries are SIC codes 2833-2836, 8731-8734,
7371-7379, 3570-3577, and 3600-3674. These SIC codes are the pharmaceuticals/biotechnology, computers, and electronics industries. This study does not control for corporate governance mechanisms (e.g. outside directors and audit committee) and ownership structure (e.g. ownership concentration and investor type) effects due to limited data availability regarding these variables. I acknowledge the omitting of these variables as limitations to this study. I include year and industry as fixed effects in the regression models but these results are not reported in the regression analysis. Industry is controlled based on the 2-digit SIC codes.
4 Results
This chapter starts with describing the explanatory power of the earnings management proxies used in this study. Then, the descriptive statistics of the sample are given after which the Pearson correlation matrix is analyzed. Lastly, the main results regarding the effect of auditor size on earnings management are given in the coefficients tables and an additional test is performed. 4.1 Pre-tests
Before carrying out the regression, I examined the validity of the sample by performing several statistical tests. First, I analyzed whether the variables of the regression model are normally distributed by examining the skewness and kurtosis of the variables. In order to reduce the skewness and kurtosis, the variables have been winsorized at the top and bottom one percent. This reduced the skewness and the kurtosis. However, the majority of the variables are still not normally distributed. I have not opted for transforming the data because the values of some of the variables contain negative values. I also decided not to further winsorize the data in order to keep the originality of the data.
Second, I analyzed whether the sample data contains multicollinearity problems. Multicollinearity occurs when at least two of more independent variables strongly correlate with each other. The variance inflation factor (VIF) of all the variables in the regression models in this study show no sign of multicollinearity. In paragraph 4.3 this is explained in more detail as the VIF values are shown in Table 5.
Lastly, I examine whether heteroscedasticity is present in the regression models as this could influence the results of the regression. Heteroscedasticity is present if the standard deviation of a variable is non-constant or if the variance of errors is constant. In order to test for heteroscedasticity, I use the Breusch-Pagan test on the regression models. The null hypothesis in the Breush-Pagan test states that there is no heteroscedasticity in the data. The results of the Breusch-Pagan tests indicated that heteroscedasticity is present in the data. As a result, I have chosen to use a robust OLS-regression in which I cluster at firm-level.
4.2 Descriptive statistics
Table 2 reports the average explanatory power of the earnings management measures regressions. The residuals are estimated for each 2-digit and 1-digit SIC codes with at least 10 observations for each industry-year combination. For accrual-based earnings management, the adjusted R-squared
discretionary accruals. The rest of the total accruals (83,2%) are the discretionary accruals which are used as a dependent variable in this study. The adjusted R-squared for cash flow from operations is 23,9%. This means that 76,1% of the total cash flow from operations are abnormal cash flows. The adjusted R-squared for production costs is high with 84%. This means that 16% of the total production costs are abnormal production costs. The adjusted R-squared for the discretionary expenses is the lowest of the real earnings management measures with 18.2%. This means that 81,8% of the discretionary expenses are abnormal discretionary expenses.
The adjusted R-squared in this study partly corresponds to those of Roychowdhury (2006). He finds an adjusted R-squared of 45%, 89%, and 38% for abnormal cash flows, abnormal production costs, and abnormal discretionary expenses. The adjusted R-squared of the discretionary expenses in this study is also the lowest of the three real earnings management measures. However, the adjusted R-squared of cash flow from operations and discretionary expenses in this research is significantly lower than that of Roychowdhury (2006). This may be because residuals in this study are estimated on a 2-digit and 1-digit SIC base with at least 10 yearly observations, while Roychowdhury (2006) estimates the residuals strictly on a 2-digit SIC base with at least 15 yearly observations and this study uses European data.
Table 2 Model parameters*
* Values are rounded to three decimal places
In Table 3 the descriptive statistics are given for the variables that are used in the regression models. All continuous variables are winsorized at the top and bottom 1 percent to avoid extreme observations. As a result, the means of the earnings management dependent variables are not zero which is in line with Zang (2012). The standard deviation for real earnings management is significantly larger compared to accrual-based earnings management. For example, TOTAL_REM has a standard deviation of 0,272 compared to the 0,063 of discretionary accruals. This shows that there is more variation in real earnings management relative to discretionary accruals but this seems normal considering that real earnings management can be manipulated in three ways.
The descriptive statistics show that 18,5% of the sample firms are audited by two Big 4 audit firms while 23,4% of the sample firms are audited by two non-Big 4 audit firms. 76,6% of the sample firms are audited by at least one Big 4 audit firm. In the samples of Francis et al. (2009)
Measures Observations Adjusted R2 P25 Median P75
DA 1.202 0,168 -0,024 0,125 0,326
AB_CFO 1.202 0,239 0,059 0,176 0,394
AB_PROD 1.202 0,840 0,800 0,864 0,913
and André et al. (2016) 63,2% and 83,9% of the firms, respectively were audited by at least one Big 4 audit firm. The difference may be caused by firms switching auditors and/or smaller auditors getting taken over by larger audit firms as the time frame of the samples differs between this research and the studies of Francis et al. (2009) and André et al. (2016).
The variable LOSS indicates that 26,8% of firm-observations have a net income that is lower than zero. 22,7% of the sample firms are in high litigation industries. The average of MTB is 1,630 which means that the sample firms still have potential to grow. The mean of SIZE, which is the natural logarithm of total assets, has a value of 12,8. This is relatively similar to the value 11,39 of Francis et al. (2009). The mean of ROA is slightly negative with -0,003. This maybe because the lowest observation for ROA is -0,562 while the median and the highest observation is only 0,027 and 0,161, respectively. This shows that even after winsorizing at the top and bottom first percent, outliers still remain in the data. The mean of LEVERAGE shows that the sample firms have a debt ratio of 20,8%.
Table 3 Descriptive statistics*
* Values are rounded to three decimal places except for the earnings management proxies and the Min of LEVERAGE
4.3 Correlation coefficients
In Table 4 the correlation between the regression variables is reported. The Pearson correlation measures the statistical relationship between variables. Table 4 indicates that multicollinearity is unlikely to be a serious issue as none of the variables have a correlation higher than 0,70 or lower than -0,70 although some do come close. In order to reject multicollinearity, the variance inflation factor (VIF) is checked for the regression models. Table 5 shows the VIF values of the regression
Variable N Mean Median Std.
Deviation 25
th
Percentile
75th
Percentile Min Max
TOTAL_REM 1.202 0,0002971 -0,026 0,27 -0,188 0,167 -0,941 0,958 AB_CFO 1.202 0,0005616 -0,001 0,076 -0,042 0,041 -0,199 0,228 AB_PROD 1.202 0,0002828 0,028 0,164 -0,071 0,107 -0,619 0,321 AB_DISX 1.202 0,0005473 0,034 0,194 -0,097 0,136 -0,583 0,355 DA 1.202 -0,0000028 0,005 0,063 -0,026 0,034 -0,247 0,157 BIG4BIG4 1.202 0,185 0 0 0 0 1 BIG4NONBIG4 1.202 0,582 1 0 1 0 1 NONBIG4NONBIG4 1.202 0,234 0 0 0 0 1 LOSS 1.202 0,268 0 0 1 0 1 LITIGATION 1.202 0,227 0 0 0 0 1 MTB 1.202 1,630 1,245 1,970 0,732 2,002 -4,189 12,834 SIZE 1.202 12,800 12,426 2,247 11,004 14,331 8,189 18,396 ROA 1.202 -0,003 0,027 0,117 -0,005 0,053 -0,562 0,161 LEVERAGE 1.202 0,214 0,198 0,152 0,098 0,300 0,0004 0,793
models. The DA regression model shows a mean VIF of 1,54 while the individual VIFs of the variables are between 1,09 and 2,43. The REM regression models show a mean VIF of 1,56 while the individual VIFs of the variables are between 1,09 and 2,43. The VIF is lower than 5 which suggests that there are no multicollinearity problems.
The high correlations in the Pearson correlation matrix between the individual real earnings management and TOTAL_REM are mechanical as TOTAL_REM is the sum of the individual measures. The variable DA is positively related to all real earnings management measures suggesting that the both earnings management methods may be regarded as complements rather than substitutes. This is in contrast with Zang (2012) who finds a substitute relation in the United States. A possible explanation for the complementary relationship in France is that auditors’ legal liability is more limited in France relative to the United States (Piot & Janin, 2007). Khurana & Raman (2004) find that legal liability drives higher quality Big 4 audits more than the reputation effects, therefore auditors in the United States might hamper accrual-based earnings management more as it is more likely to be detected which results in companies resorting to real earnings management. Since auditors’ legal liability is more limited in France, companies might be given more leeway in their use of accrual-based earnings management which could result in companies using both methods in a complementary manner instead of a substitutive way.
Regarding the independent variables, the correlation matrix shows that the auditor combination of two Big 4 audit firms is negatively related to the real earnings management measures except for AB_DISX. Although, only AB_PROD (-0,1519) and AB_DISX (0,0538) are significant regarding this relationship. However, the auditor combinations of at least one Big 4 (0,0581) or two non-Big 4 (0,0525) firms are positively related to TOTAL_REM. I expected that BIG4BIG4 would also be positively correlated with real earnings management considering the hypotheses of this research. The correlations for discretionary accruals regarding auditor-pair choice are also conflicting with expectations. Table 4 shows that there is a positive relationship (0,0294) between BIG4BIG4 and DA while BIG4NONBIG4 has a negative relationship (-0,0582) with DA. I expected BIG4BIG4 to be also negative. It is interesting to see that SIZE (0,3954) has a relatively high correlation with BIG4BIG4 which indicates that larger firms tend to have two Big 4 firms as auditors.
The variable LOSS has a positive effect on real earnings management measures while the effect is negative (-0,2041) on discretionary accruals which suggest that firms with losses are more likely to use real earnings management. SIZE has a negative correlation (-0,2096) with TOTAL_REM suggesting that larger firms use less real earnings management. This is interesting
considering that SIZE, as mentioned above, is positively correlated with BIG4BIG4 while BIG4BIG4 itself is primarily negatively correlated with real earnings management. ROA has a strong negative correlation (-0,6985) with LOSS which is similar to the -0,65 of Francis et al. (2009). The negative correlation is logical because firms that report losses have a negative return on assets. Regarding LEVERAGE, it is interesting to see that its correlation with the real earnings management measures is positive while it is negative (-0,1415) with discretionary accruals.
Table 4 Correlation matrix
*, **, *** Significant at 10 percent, 5 percent and 1 percent levels, respectively
1 2 3 4 5 6 7 8 9 10 11 12 13 14 1.TOTAL_REM 1 2.AB_CFO 0,4822 1 *** 3.AB_PROD 0,6229 0,2882 1 *** *** 4.AB_DISX -0,6748 -0,0373 0,0919 1 *** *** 5.DA 0,1636 0,3667 0,1144 0,012 1 *** *** *** 6.BIG4BIG4 -0,1312 -0,0001 -0,1519 0,0538 0,0294 1 *** * 7.BIG4NONBIG4 0,0581 0,0284 0,0106 -0,0607 -0,0582 -0,5611 1 ** ** ** *** 8.NONBIG4NONBIG4 0,0525 -0,033 0,127 0,0215 0,0408 -0,2629 -0,6511 1 * *** *** *** 9.LOSS 0,1397 0,3562 0,1045 0,0327 -0,2041 -0,0991 0,1399 -0,0722 1 *** *** *** *** *** *** ** 10.LITIGATION -0,0693 -0,0296 -0,1041 -0,0032 -0,0749 -0,0175 0,0412 -0,032 0,1474 1 ** *** *** *** 11.MTB -0,0739 -0,0294 -0,1396 -0,0267 -0,0682 -0,026 0,0672 -0,0546 0,0932 0,0885 1 ** *** ** ** * *** *** 12.SIZE -0,2096 -0,1134 -0,1474 0,1224 0,019 0,3954 -0,0684 -0,2828 -0,2748 -0,2081 -0,067 1 *** *** *** *** *** ** *** *** *** *** 13.ROA -0,1713 -0,4097 -0,1106 -0,0148 0,3407 0,0861 -0,1759 0,126 -0,6985 -0,1573 -0,2521 0,2842 1 ** *** *** *** *** *** *** *** *** *** *** 14.LEVERAGE 0,0415 0,1051 0,0633 0,0368 -0,1415 0,0256 0,0298 -0,0582 0,1693 -0,1418 -0,1284 0,1731 -0,1798 1 *** ** ** ** *** *** *** *** ***
Table 5 VIF values4
4.4 Regression analysis
Table 6 reports the results of the empirical models used in this study to test the hypotheses. The dependent variables of the empirical models are DA, TOTAL_REM, AB_CFO, AB_PROD, AB_DISX. The empirical models are tested by using a robust OLS in which I cluster at firm-level. The variable NONBIG4BIG4 is the benchmark group (i.e. all coefficients of the joint auditor mix dummy variables have to be interpreted with reference to NONBIG4NONBIG4).
Table 6 column 5 shows the results for the accrual-based earnings management hypotheses. The model explains 18,2% of the variation in the dependent variable discretionary accruals. This model tries to find if hiring at least one Big 4 auditor constrains the use of accrual-based earnings management. The results show that BIG4BIG4 has a positive coefficient of 0,014 which is significant at a 5 percent level. This means that a joint audit by two Big 4 auditors leads to an increase in the use of accrual-based earnings management. The coefficient (0,004) BIG4NONBIG4 is not significant meaning there is no difference between a joint audit by one Big 4 auditor and one non-Big 4 and a joint audit by two non-Big 4 auditors in the use of accrual-based earnings management. I expected a negative coefficient for BIG4BIG4 and BIG4NONBIG4. The results show that the opposite is true.
Based on these results of the accrual-based earnings management model hypotheses H1-A and H1-B are rejected. It seems that having at least one Big 4 auditor does not have a constraining effect on accrual-based earnings management relative to having no Big 4 auditors.
4 The VIF values of the real earnings management empirical models are all the same, therefore I have only included
Variable VIF 1/VIF Variable VIF 1/VIF
TOTAL_REM 1.10 0,911 DA 1.15 0,867 MTB 1.14 0,880 MTB 1.12 0,891 SIZE 1.55 0,645 SIZE 1.53 0,655 LEVERAGE 1.17 0,857 LEVERAGE 1.17 0,852 ROA 2.28 0,439 ROA 2.43 0,411 LOSS 2.02 0,494 LOSS 2.03 0,493 LITIGATION 1.11 0,904 LITIGATION 1.09 0,915 BIG4BIG4 1.87 0,536 BIG4BIG4 1.87 0,536 BIG4NONBIG4 1.61 0,621 BIG4NONBIG4 1.61 0,621
Mean VIF 1,54 Mean VIF 1,56
