The Link between CSR and Earnings Management: A Quantitative Study in the Manufacturing Sector of the EU

FACULTY OF EDUCATION AND BUSINESS STUDIES

Department of Business and Economics Studies

The Link between CSR and Earnings Management

A Quantitative Study in the Manufacturing Sector of the EU

Bereket Gebregiorgis Josefine Dijkstra

2020

Student thesis, Bachelor degree, 15 HE Business Administration

Supervisor: Jan Svanberg

Examiner: Arne Fagerström

i

Abstract

Purpose: The aim of this thesis is to examine the link between CSR and EM in the EU manufacturing sector. The former is measured by ESG scores whereas the latter is approximated by discretionary accruals. Moreover, the corporate governance theories of stakeholder and agency are used as starting points.

Method: Accruals are calculated using the so-called Modified Jones model. A linear regression equation with five variables is used to calculate the effect of CSR scores on the levels EM. In order to capture the difference, if any, regarding the link between CSR and EM between companies with high CSR scores and those with lower scores, three models are developed – the top 30 percent (model 1), the bottom 70 percent (model 2) and all companies (model 3).

Findings: The results indicate CSR scores negatively affect the levels of EM. The causation is statistically significant.

Implication: The findings of this study imply CSR practices are not used to hide weaker earnings in the manufacturing sector of the EU.

Value: The study contributes to the literature on the link between CSR and EM by focusing on a specific sector that is often scrutinized for its negative environmental footprint. The choice of using three models is also seen to be of additional value.

Keywords: Corporate social responsibility, earnings management, stakeholder theory,

agency theory, accruals

ii

Contents

1 Introduction ... 1

1.1 Background ... 1

1.2 Problem statement ... 2

1.3 Aim ... 3

1.4 Additional research topics ... 3

1.5 Scope ... 4

1.6 Outline ... 4

2 Theory ... 5

2.1 Corporate governance theories ... 5

2.1.1 Agency theory ... 5

2.1.2 Stakeholder theory ... 6

2.2 CSR ... 6

2.2.1 Historical and current trends... 6

2.2.2 Components of CSR ... 8

2.2.3 CSR in practice ... 9

2.3 Earnings management and its determinants ... 10

2.3.1 Earnings management... 10

2.3.2 Factors that determine EM ... 11

2.4 Empirical studies on the link between CSR and EM... 13

2.5 Hypothesis ... 16

3 Method ... 17

3.1 Methodological framework ... 17

3.2 Data collection ... 18

3.2.1 Publications ... 18

3.2.2 Raw data and sampling ... 18

3.3 Measuring CSR ... 21

3.4 Estimating EM ... 22

3.5 Regression model ... 24

3.6 Statistical tests ... 26

4 Result ... 27

4.1 Descriptive statistics ... 27

4.2 Bivariate analysis ... 29

4.3 Multivariate analysis ... 32

5 Discussion ... 36

5.1 The variables... 36

5.1.1 The dependent variable AEM ... 36

5.1.2 The independent variable CSR ... 36

5.1.3 The control variables ... 37

5.2 Correlation ... 37

5.2.1 The interaction between CSR and the control variables... 37

5.2.2 The interaction among the control variables ... 39

5.3 Causation ... 40

5.3.1 The effect of CSR on AEM ... 40

5.3.2 The control variables’ effect on AEM ... 43

iii

6 Conclusion ... 45

6.1 Recap ... 45

6.2 Findings ... 46

6.3 Reflection ... 47

6.4 Tips for future studies ... 49

7 References ... 51

8 Annex ... 57

Figures Figure 1. The pyramid of CSR ... 9

Figure 2. Types of financial report manipulation ... 11

Figure 3. The composition of the sample of manufacturing companies in EU ... 20

Figure 4. Pillars and subcategories of the ESG score ... 21

Tables Table 1. Descriptive statistics for CSR reporting firms... 27

Table 2. Correlation analysis for all firms ... 30

Table 3. Correlation analysis for top 30% firms ... 30

Table 4. Correlation analysis for bottom 70% firms ... 31

Table 5. The link between AEM and the independent variables ... 32

Table 6. The link between AEM and the independent variables - stepwise method ... 34

iv

Abbreviations

AEM - accrual based earnings management CSP - corporate social performance

CSR - corporate social responsibility EM - earnings management

ESG - environmental, social and corporate governance LEV - leverage

MB - market to book value ROA - return on assets

2SLS - two-stage least squares

1

1 Introduction 1.1 Background

The importance of corporate social responsibility (hereafter CSR) reporting to

corporations and their stakeholders is growing. Abdelhalim and Eldin (2019) attribute this to the rise of globalization and international trade. In an ever interconnected world, the environmental and social footprint of transnational corporations is going beyond employees to influence the livelihood of consumers and communities. Policy makers are as a result enforcing a more responsible and transparent corporate behavior that considers the environmental, social and economic concerns of their constituencies. In addition to countries such as France that have done so for some time, several other major EU countries, among them, Germany and Italy, have increasingly engaged in mandating CSR reporting in recent years (Mion and Adaui, 2019).

Even though CSR reporting involves complex and hard to quantify issues such as the environment, human resources and local community relations, it is generally seen as a positive endeavor for companies themselves. With greater CSR reporting, studies show, a firm can expect to achieve lower cost of capital, increased market share, good

reputation with stakeholders and so forth (Buertey et al., 2019; Suyono and Farooque, 2018). According to Suyono and Farooque (2018), it is because of these benefits a considerable amount of companies make yearly disclosures voluntarily.

Despite the benefit of CSR to both corporations and their stakeholders, there is a

concern that they are sometimes being used for more sinister goals, hiding earnings

management being one. Earnings management is the practice of manipulating financial

reports by managers to fit a specific and predetermined goal. It is generally seen as

harmful since the resulting financial information about the company fails to capture the

company's true worth (Al-Haddad and Whittington, 2019). The concern is that some

studies are finding more earnings management practices to be positively linked to CSR

reporting. This suggests managers are using the good reputation they get from being

involved in CSR reporting to hide their less-than-acceptable financial reporting

practices (Liu and Lee, 2019). As shown below, CSR and earnings management are

however not always positively related.

2 1.2 Problem statement

In regard to its link to financial reporting, accounting literature provides us with two broad reasons why firms are involved in CSR-related practices (Gerged et al., 2020).

Firstly, firms may disclose CSR reports to signal that they are a responsible member of the society. In those firms, CSR reporting is expected to be negatively related to manipulative financial reporting (Suyono and Farooque, 2018). This stems from the argument that having good relationships with various stakeholders is important to socially responsible firms, and as a result those firms are less likely to engage in aggressive earnings management that threatens their long-term reputation. This

argument is supported by the stakeholder theory (Mohmed et al., 2020). And, empirical studies by Kim et al. (2012), and Pyo and Lee (2013) support it.

CSR activities may not however be wholly motivated by a firm’s social standing – managers may report CSR in pursuit of their self-interest (Mohmed et al. 2020).

Research in two dozen countries by Prior et al. (2008) for instance suggests managers use CSR as a tool to protect their own careers. One way of achieving this is to use CSR reporting as a shield to hide earnings management (Mohmed et al., 2020). This tactical move by managers which results in a positive link between CSR reporting and the practice of earnings management is usually explained by the so-called agency theory.

This theory postulates the divergence of interest and asymmetric information between owners and managers creates a condition where managers extract private benefits at the expense of the owners (Prior et al., 2008). Among others Hemingway and Maclagan (2004); and more recently, Martínez-Ferrero et al. (2016) show CSR reporting to be an act of self-interest by managers and find a positive relation between CSR disclosures and earnings management indicating that CSR reporting is indeed used to shield discretionary accounting practices. In this case, being just a managerial tool of self- preservation, CSR reporting ends up hurting stakeholders.

In addition to those mentioned above, studies on the relation between CSR reporting and the occurrence of earnings management are plentiful, and they have been carried out for decades. This is especially true in regard to how accounting decisions are affected by managers’ self-interest, which is to say along the lines of agency theory.

Healy (1985) for example looks at the effects of bonus schemes have on the accounting method managers end up choosing (no surprise for cynics here, managers select

accounting procedures that maximize their bonus awards). There are also a couple of

3 studies that support the stakeholder theory by empirically finding a negative link

between CSR reporting and earnings management. There are however far fewer studies that consider both the agent and stakeholders theories simultaneously. This is despite some recent studies, Mohmed et al. (2020) being one, that suggest companies with high CSR performance and those with low performance may have different rationale for being involved in CSR reporting, and one theory is not sufficient to capture that.

Moreover, Suyono and Farooque (2018) argue that managing earnings is more pronounced in manufacturing firms which, when compared to other similarly-sized firms in other sectors, tend to have more complex financial transactions and high volatility in cash flows. The manufacturing sector in the EU is declining (Marschinski and Martinez, 2019). To exacerbate the situation, the sector happens to be constantly under pressure in the union to improve its environmental-friendly credentials (Masud et al., 2019). However, most studies in this subject matter tend to focus on the major listed corporations in a specific national state without considering the sector they operate in.

There is thus an opening for a study such as this that sets sail to examine the link between CSR and earning management in the manufacturing sector. To that end, both theories, i.e. agency theory and stakeholder theory, are used as starting points. Starting a study by being based on established theories is not new. When discussing the

relationship between theory and research, Hartwig (2018) states theory is either a starting point of a study or its output. Hence, it can be said this thesis takes the former route in order to investigate the link between the two phenomena in one of the most important economic and political blocks in the world, the EU.

1.3 Aim

The aim of this thesis is to examine the link between Corporate Social Responsibility (CSR) and levels of Earnings Management (EM) in the manufacturing sector of the EU.

1.4 Additional research topics

On the way to examining the link between CSR and EM, the thesis deals with

 Calculating levels of EM and measuring CSR,

 Identifying some relevant factors that influence EM, and

 Analyzing the interplay between those factors, CSR and EM

4 1.5 Scope

The scope of this thesis is limited to publicly traded manufacturing companies in the EU. At the time of publication, the EU includes 27 countries. Because of longer experience with CSR reporting, the sample is however dominated by manufacturing companies in the northern and western parts of the union. In this thesis, companies that make physical products are considered manufacturing and range from paper factories to heavy equipment assemblies. Companies that process minerals are also included.

The period of analysis is from 2010 to 2019. The data used for all calculations is obtained from the online database DataStream compiled by Thomas Reuters. All financial data is calculated using euros, whereas CSR is measured by a score out of 100 maximum possible points.

1.6 Outline

So far in the introduction chapter, the background for this thesis, why the study area is of interest and, most importantly, the specific aim of the thesis are provided. In the next chapter, theory, terms and concepts that are introduced in the background and in the problem formulation sections, such as CSR and earnings management, are explained more in depth. Previous studies that deal with the same area of research are also summarized. At the very end of this chapter, two hypotheses are made to be tested later on.

Method deals with providing the steps and calculations that are made in this thesis in the quest to answer the research questions put forward in chapter one. The result of those calculations is presented in chapter four which is appropriately called result.

The aim of this chapter is to present the findings of this thesis without any analysis.

In chapter five, analysis, the result of the study which is presented in the preceding chapter is discussed in detail, and furthermore, comparisons are made to previous work. In conclusion, all the steps taken in all the preceding chapters are briefly summarized and the questions raised in chapter one are answered. There, tips for further research are also to be found.

Finally, reference and annex are available following chapter six.

5

2 Theory

2.1 Corporate governance theories

The decision making of managers, in regard to CSR or any other matter, can be

explained by two broad groups of theories – economic and contextual (Hartwig, 2018).

The former group of theories emphasizes the economic aspect of managerial decision making, and is based on the assumptions of profit maximization and interest conflict between the management of a corporation and its stakeholders especially shareholders.

It is noted, this category, among others, includes the theories of signaling and proprietary cost, but the most well-known is probably the agency theory.

2.1.1 Agency theory

In big companies, the owner and the leader of the organization are almost always not the same person. The owners, i.e. shareholders, entrust the manager with the responsibility of running the day-to-day affairs of the company, and to look after their long term benefits. However, being close to the organization, the manager has more information on the current state of the company than the owners which results in what is widely known as information asymmetry (Hartwig, 2018). Because of the information asymmetry, even though the manager is delegated to do what is best for the owners, he/she is able to use the resources of the company to maximize his/her benefits instead.

Agency theory is thus a theory developed by Jensen and Meckling (1976 cited in

Hartwig, 2018) to explain and resolve this conflict of interest. The resulting inefficiency due to the misalignment of the interests of the owners and the manager is called agency cost (Deegan and Unerman, 2011).

Shareholders can take some steps in order to minimize agency cost. The first is to put resources into monitoring the manager. According to Hartwig (2018), auditors are the main part of monitoring. The other option shareholders have to minimize the agency cost is called bonding. In this case the shareholders take steps that align their interests with that of the manager’s. Basically using a carrot instead of a stick. One way of doing so, according to the author, is to use stocks as a means of manager compensation (together with salaries). In this case, the manager has interest to take actions that maximize his/her share which then end up benefiting the shareholders as well.

However, the interests of the manager and the shareholders cannot be perfectly aligned,

i.e. monitory and bonding actions shareholder undertake can only reduce the agency

6 problem, not completely eliminate it (Hartwig, 2018). It is thus quite common in

research in the field of business administration to use the agency theory as an explanation whenever managers are found to be engaged in activities that benefit themselves while harming the company they head or the wellbeing of the society at large (some details in 2.4).

2.1.2 Stakeholder theory

The so-called contextual group of theories tries to explain how norms, standards, laws, moral, ethics, institutions and the like influence managers’ decision making, and overall the conduct of companies (Hartwig, 2018). Stakeholder theory, a part of this group, reasons a manager’s actions to the most part are determined by the wish to do what is best for the company’s most important stakeholders which includes the well-being of the society (Deegan and Unerman, 2011). In this theory the company is expected to revise and adapt its priorities and conduct every time the interests of the most important stakeholders or, the stakeholders themselves, change.

Hartwig (2018) states that some theories in the contextual group tend to overlap one another. The so-called legitimacy theory for example states that companies conduct their business in line with the society’s norms in order to be considered legitimate entities (Deegan and Uneman, 2011). This seems to be very similar to that of the stakeholder theory except it explicitly puts forward a motive to why corporations confine themselves to the society’s norms (to be considered legitimate). In both

theories, by stakeholders, it is meant customers, suppliers, the capital market, the state, the employees, the media and so forth.

As opposed to the agency theory, whenever a company is found to be acting in the best interest of its stakeholders especially those that are external to the company like the society at large, the stakeholder theory and the legitimacy theory are put forth as possible explanations (again, some details in regard to these theories and their implication to CSR and EM are presented in 2.4).

2.2 CSR

2.2.1 Historical and current trends

According to Abdelhalim and Eldin (2019), sustainability has become a global trend in recent years exerting an increasing amount of pressure on corporations. Corporations’

choices of suppliers, products and strategic decisions are now more than ever critically

7 followed by consumers and other stakeholders. The authors note the increasing pressure can be attributed to some social and environmental problems which have raised an awareness among consumers that are nowadays holding corporations accountable. CSR is however not new, it has been a discussion topic since the first half of the 20 ^th century (Gavin and Maynard, 1975). Sial et al. (2019) even trace back CSR to the 19 ^th century.

It was not however widely accepted immediately after it emerged, and as late as the 1960s, many managers did not see the need for a change in how they conduct their business (Sethi, 1975). The research available at that time could not prove those

managers wrong. But following the 1960s, CSR started entering the mainstream (Gavin and Maynard, 1975). And, as stated earlier, it gained prominence following the

expansion of transnational corporations (Abdelhalim and Eldin, 2019).

Earlier studies on CSR focus on why it arises. Most often than not, they base their reasoning on the theories of legitimacy and the social contract (Fitch, 1976; Sethi, 1975). Society needs corporations for employment. Whereas corporations need the society as a consumer – if there is no or low consumption, corporations will not see any growth or can even cease to exist. As being solely dependent on the society, it is hence important for corporations to be aware of the expectations of the society and fulfil them.

In short, corporations need to follow the social contract if they are interested in gaining legitimacy, and as a consequence, continue to operate profitably (Sethi, 1975). CSR is thus seen as a tool for doing just that. Fitch (1976) warns if the social contract is not observed voluntarily, it will instead be imposed by governments. Additionally, both Sethi (1975) and Fitch (1976) emphasize that CSR is not only about solving current problems instead the main focus and goal of CSR is to look forward and to prevent or, in case they occur, solve future difficulties which in the long term would lower costs for the corporations themselves.

Currently, there are strong laws in the EU to protect the environment, think of air and

water regulations for instance (Dijken, 2007; Sial, et al., 2019). CSR with its various

dimensions is however mostly encouraged by guidelines and principles issued in the

union. Over the years, the European Commission (2020) has introduced a mix of

voluntary and mandatory actions plans to promote CSR. The commission believes

promoting CSR would benefit companies by widening their access to capital and

improving their consumer relationships whereas the whole economy is expected to gain

by being more innovative. In Europe where CSR is prevalent, the society is also

8 expected to be more cohesive. To that end, CSR reporting of large companies has

passed from being voluntary to mandatory in the past few years (Arraiano and Hategan, 2019). Arraiano and Hategan (2019) note that reporting practices are improving for all member countries but there is still some gap of quality for instance between the eastern and western states. Furthermore, CSR reporting practices seem to still vary among different industries and companies of different sizes (Vukic, 2015). In most of the developing countries of the world, clear CSR regulations and guidelines do not exist (Sial, et al., 2019). However, there are signs even in these countries the issue is getting prominence. China is a good example of that (Liu and Lee, 2019).

According to Grimmer and Bingham (2013), customers are more likely to buy a product if it is manufactured by a company with good CSR reputation. Mohr and Webb (2005) go even further by concluding customers are willing to pay an extra premium for products from companies that engage in CSR practices. This suggests, besides the increasing push by regulators, there are factors that pull companies towards engaging in CSR practices and reporting their achievements. The pull factors to CSR reporting can contemporary be seen as stronger in manufacturing firms since, according to Torugsa et al. (2012), those firms have closer relationship with their stakeholders than their peers in other industries.

2.2.2 Components of CSR

Gavin and Maynard (1975) introduce to CSR problems on the corporate level such as the overall well-being of the employees, but also on a worldwide scale like poverty, civil rights and ecology. It is, though, Carroll (1979 cited in Sheehy, 2014) who anchors CSR around four clear components that are the responsibilities of a company –

economic, legal, ethical and philanthropic. These four responsibilities are later described as forming a pyramid. As seen in figure 1, the base is the economic responsibilities which include producing goods and services, and to make profit.

According to Carroll (1991), the base is needed for all the other layers to exist. The second layer is legal responsibilities. This means to follow the law, and the right and wrong set by the society. The next component is ethical responsibilities which include to be just, fair and to avoid harm. On the top of the pyramid are philanthropic

responsibilities. Philanthropic responsibilities entail to be a good citizen and contribute

in improving the quality of life of the society the corporation operates in.

9

Figure 1. The pyramid of CSR (Carroll, 1991)

Furthermore, the UN has erected ten CSR principles for corporations to follow, the initiative being called the United Nations Global Compact. Corporations can use these guidelines as a framework when developing their own practices. The principles

constitute four areas: human rights, labor, the environment and anti-corruption (United Nations Global Compact, 2014). In 2014, the initiative had more than 12 thousand participants in more than 160 countries that, in total, had close to 60 million employees, making it the biggest initiative in the world. In that year, most of the participating corporations were located in the EU, with Spain and France holding the top positions.

2.2.3 CSR in practice

Empirically measuring CSR is difficult due to the many dimensions of CSR and the difficulty in quantifying those dimensions (Ehsan et al., 2018). Which dimensions and aspects that should be taken into account is thus widely discussed among researchers and regulators. It is therefore not surprising that there is no globally uniform

measurement of CSR and, as a consequence, researchers can be found choosing from the smorgasbord of methods already available, or customizing their own measurement which in their assessment fits their specific area of research.

According to Ehsan et al. (2018), the most commonly used method to measure CSR is the use of social ratings or reputation indices. Some of the widely used indices include the KLD index, Moskowitz’s tripartite ratings and Domini Social Index (DSI). The agencies that make the indices use publicly available information about corporations, and collect some more by conducting surveys and interviewing employees of various corporations. Based on the information collected, corporations are given CSR ratings.

In addition to a reputation index, it is possible to use the so-called content analysis

method to determine the level of CSR performance in corporations (Cochran and Wood,

1984). Content analysis uses public information about corporations, mainly their

10 financial reports and other publications such as their website, letters to shareholders and media which then get quantified, i.e. turned into a number, and used as a measure of CSR (Cochran and Wood, 1984; Ehsan et al., 2018). One of the advantages of using content analysis is that it is easy to calculate for a large number of companies as the process is very mechanical (Cochran and Wood, 1984). The so-called ESG score database, more on it later on, seems to combine this approach and the social ratings approach which is mentioned above.

Another common measurement for CSR is Corporate Social Performance (CSP).

According to Orlitzky et al. (2003), CSP measures CSR disclosures, CSR reputation ratings, principles and values held by the managers, and audits and processes

concerning CSR. In this approach, primarily content analysis is used when measuring CSR disclosure but letters to stakeholders, annual reports and miscellaneous corporate disclosures can also be looked at.

2.3 Earnings management and its determinants 2.3.1 Earnings management

There are two broad ways in which managers can influence the content of their financial reports that result in unrealistic numbers (Hartwig, 2018). The first is through lobbying – managers hire lobbyists to change the guidelines of accounting in order to end up with a favorable result in their balance sheet. This is usually known as macro manipulation.

On the other hand, when managers directly influence their financial reports, it is called micro accounting manipulation. In this case, the manipulation takes place without any systematic help from the outside.

According to Hartwig (2018), earnings management (hereafter EM) is the most

common form of micro accounting manipulation. EM is said to occur “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 numbers” (Healy and Wahlen, 1999). More contemporarily, Al- Haddad and Whittington (2019) emphasize that the alteration of financial reports can take place without violating general accounting standards.

Hartwig (2018) identifies two types of EM – accrual EM where income statements are

influenced by changing accounting methods, and real EM where actual business

11 operations are influenced in order to keep costs down in the short run, for instance in the current year. The common ways of undertaking accrual EM include reporting a cost as an asset, and using longer expected asset useful life in order to decrease reported costs (Hartwig, 2018). There are also other judgments managers make such as salvage value of long-term assets, obligations for pension, deferred taxes, losses from bad debts and so on that may be used as tools for accrual EM (Healy and Wahlen, 1999). Real EM, in contrast, includes decreasing investment in research and development, advertising, and employee training (Kothari et al., 2016). According to Kothari et al. (2016), detecting real EM is more challenging since it, going beyond the choice of accounting methods, involves the managers’ investment and operational decisions.

Figure 2 . Types of financial report manipulation

One motive for the existence of EM is to elevate the performance of the management in hopes of getting benefits such as job security and good reputation. This is especially true in corporations that have management compensation contracts that are aligned with the corporations’ yearly performance (Healy and Wahlen, 1999). EM can also be undertaken in the hopes of skewing a firm’s stock price upwards (Kothari et al., 2016).

To comply with regulations such as capital adequacy requirements is another important motive behind EM (Healy and Wahlen, 1999). But whatever the motive behind it is, widespread EM threatens the credibility of financial reports and may lead to

misallocation of business resources conclude Healy and Wahlen (1999).

2.3.2 Factors that determine EM

Variables that have to do with the CEO are some of the factors that are often found to

indicate the existence of EM. This is hardly surprising since the CEO’s fate is directly

connected to the performance of the company. Jouber and Fakhfakh (2012) for instance

reach the conclusion that EM is more likely in corporations where the CEO owns stock

– a CEO that personally gains when the corporation does well has higher motivation to

engage in practices that inflate corporate performance. Another study on the role of the

12 CEO by Hu et al. (2015) establishes a link between the likelihood of EM and CEO tenure. In this study, CEOs are found to engage in EM aggressively in the middle of their tenure, usually at about the fifth year of holding office. The reason why this is the case is not explicitly stated but to presume that it takes a few years for a new CEO to start influencing the financial reporting of a company is not unreasonable. Hu et al.

(2015) also find that CEOs engage in less EM practices close to the end of their tenure.

Now that retirement is on the horizon, this perhaps can be explained by the need to protect their life-long reputation.

The form of company ownership is another well researched factor. For instance, Saleem and Alzoubi (2016), find less EM in family-owned firms where the owner and manager of the firm are the same person. Similarly, corporations that have few but large

shareholders are also found to have lower levels EM. The explanation given for the latter is that, unlike many small shareholders, large institutional shareholders have the resources needed to closely monitor managers.

Epps and Ismail (2009) show that companies with small, annually elected board of directors have smaller instances of EM. Smaller boards are efficient. Yearly elections bring about more accountability. The combination of the two is seen as an explanation for the lower levels of EM in such companies. In addition to the size of the board of directors, its composition is also of interest. There is evidence that more gender

diversity in the board of directors leads to less EM practices (Saona et al., 2019). This is attributed to the risk averseness of women. The negative relationship between having a gender diverse board of directors and EM is also supported by Damak (2018).

The likelihood of EM can also be estimated by taking a look at external controls exerted on the management, the auditing firm being front and center. The biggest auditing firms in the world and industry-specialist auditors have the resources and the experience to mitigate EM effectively, and hence companies that hire them tend to have lower levels of EM (Anissa, 2019; Krishnan, 2003). Furthermore, there is some empirical evidence that high auditing fees in contrast to comparably sized companies can indicate higher levels of EM since auditing firms adjust their fees in line with the estimated risk of litigation that comes with the work (Marinakis, 2011).

Not surprisingly, the financial status of companies also influences their quality of

financial reporting. To begin with, Buertely et al. (2019) find EM to be high in highly

13 leveraged companies. The explanation given is that companies resort to inflating their income statement because of the fear of violating their debt covenant – the agreement with their creditors to fulfill certain financial goals. Moreover, return on assets (ROA), income divided by assets, is often found to be negatively related to EM (Buertely et al, 2019). After all, when a company actually has high income, it wouldn’t make sense to artificially inflate it in order to present outwards a more positive picture of the company.

Going well beyond the company itself; institutional factors such as judicial independence, political stability, type of market and the media are some of the

additional determinants of EM mentioned in past literature (Chang et al., 2019; Kouki, 2018; Wu et al., 2016). This shows the sheer amount of factors that can influence the financial reporting of companies. To make things even more complex, the same

institutional factors can influence EM in opposite directions. In the case of the media for instance, it can aggravate EM by the pressure it puts on managers to excel while

simultaneously intense media scrutiny can discourage managers from EM practices by acting as a watchdog (Wu et al. 2016). It is therefore very important to choose control variables very carefully when examining the link between EM and any other variable, be it internal or external. Various publications, for instance the Economist (2020), prepare annual indexes that rank countries based on the strength of their institutions which, as stated a moment ago, can affect the prevalence of EM.

2.4 Empirical studies on the link between CSR and EM

There are many studies in regard to the link between CSR and EM. It is worth mentioning that due caution must be taken when comparing research across borders.

Differences in CSR regulations between countries exist. In some countries it is mandatory while in others it is not. Even how well the same regulations are implemented vary greatly between countries (Brown et al., 2014). Difference in

legislation can result in distinct accounting traditions developing, for instance Muresan and Pop Silaghi (2014) suggest that legal differences between countries itself can lead to varying levels of EM practices. By going through various studies, it is however possible to get a general view of the relation between the two variables on the ground.

A quantitative research in Indonesia by Suyono and Farooque (2018) is one recent study

that examines the relationship between CSR and EM. In addition, the authors aim to

determine if other factors change the relation between CSR and EM. Seeing the relation

14 from the point of view of agency theory, the study hypothesizes that managers use CSR practices in their advantage, such as hiding EM. After using ordinary least squares regression analysis to test their hypothesis, the authors indeed find a positive

relationship between CSR and EM, thus confirming their starting point, i.e. the agency theory perspective. Some other factors that are taken into account in the study when calculating the link between CSR and EM include corporate governance variables such as managerial ownership of equity and the size of the board of directors. In this regard, the authors find that when managers own a larger portion of the company’s equity, they engage less in EM. Perhaps because then, the quality of the company’s financial reports serves their interest too.

Another recent study, this time by Buertley et al. (2019), focuses on non-financial companies – consumer goods, consumer services, basic materials, industries, and technology/telecoms – that are listed on Johannesburg stock exchange. The study includes corporate governance factors, but the aim is to look at the relationship between CSR and EM. A regression model is used to determine the relationship which is found to be positive, the same as Suyono and Al Farooque (2018). In addition, board size and block ownership are found to moderate the relation between CSR and EM. A larger board leads towards a more responsible managerial behavior thus lowering EM. This is in direct contrast to Epps and Ismail’s (2009) finding pointed out in section 2.3.2. Block ownership, when a large portion of the company’s shares are in the hands of one person, makes managers act more responsible. This finding is supported by Saleem and Azoubi (2016) (also in section 2.3.2.)

Prior et al. (2008) too relies on the agency theory to examine the relation between CSR

and EM. In this case, data from 2002 to 2004 on 593 companies across 26 countries is

used. The theory tested in the study is that when managers practice EM they will also

practice CSR, as predicted by the agency theory. They find EM to positively impact

CSR, and going even further, they conclude this link ends up hurting the financial

performance of corporations. On the other hand, Pyo et al. (2013) look at donations and

voluntary CSR reporting from companies on the Korean stock exchange to determine

the level of CSR and whether high CSR companies have higher earnings quality (high

earnings quality means low EM). The study uses regression models which also happens

to be the choice of the above mentioned studies but this time the starting point is the

stakeholder theory. The authors reach the conclusion that companies involved in CSR

15 activities and CSR reporting have a higher earnings quality, i.e. low EM, resulting in a negative relationship between CSR and EM. Variables such as high leverage and high return on assets are found to be linked to higher EM. A negative relationship between CSR and EM is also found by Kim et al. (2012). In this case, high CSR scoring

companies from a database of more than 18 thousand companies are analyzed in order to see if they act more responsible in their financial reporting than other firms. Here too the starting point is the stakeholder theory. In this study, it is also empirically proved that socially responsible companies are less likely to be subject to investigations by the authorities.

There are some studies that use more than one theory in order to establish the link between CSR and EM one of which is on Kuwaiti listed companies by Gerged et al.

(2020). In this study, the environmental aspect is seen as a part of CSR. The theories used in the study mainly include the theories of stakeholder and legitimacy but, to lesser extent, the agency theory as well. After running a 2SLS model, the researchers arrive at a negative link suggesting that the environmentally ethical managers in Kuwait do not engage in EM practices. Though, the authors do find that corporate environmental disclosures are used by companies to fend off both political and social pressures from the public. Martinez-Ferrero et al. (2016) examine how CSR as a strategy can be used by companies to hide or even-out EM’s negative side effects by looking at the practices just less than two thousand publicly listed non-financial companies from 26 different countries in the years ranging from 2006 to 2010. Again, agency theory and positive accounting theory are used. The findings suggest that companies do use CSR to hide EM as well as to even-out the harm caused by EM.

Unlike most studies reviewed, Mohmed et al. (2020) examine the link between CSR and

EM based on both agency theory and stakeholder theory. The subjects of the study are

100 Egyptian companies that have CSR scores from 2007 to 2015. The results of the

analysis are twofold. With the top 30 CSR scoring companies, a negative relationship is

found – the companies seem to be acting ethical and using CSR for the right reasons. In

the remaining 70 companies, a positive relationship is established implying that those

companies use CSR reporting to greenwash their financial reporting and to mislead

stakeholders. The first result seems to adhere to the stakeholder theory whereas the later

conforms to the agency theory. Moreover, the size of the company, high leverage and

high ROA are found to be positively related to EM in the top CSR scoring companies.

16 2.5 Hypothesis

As seen in the section above, most of the literature on examining the link between CSR and EM is often based on a single theory, usually either agency or stakeholder theory, and then focus to determine if the relationship between the two variables is positive or negative. Just as their starting point, the results of those studies do also vary – some find the relationship to be positive and others negative.

It is thus interesting that, in the same study, Mohmed et al. (2020) manage to arrive at the conclusion that the link between CSR and EM can be positive or negative depending on the level of CSR performance. That is to say, there is no definitive link between CSR reporting and EM but there is a link between CSR and EM in firms with high CSR ratings on one hand, and CSR and EM in firms with low CSR ratings on the other. And the link tends to be negative and positive respectively, the first being explained by the stakeholder theory and the second by the agency theory.

Mirroring this, there are two hypothesis made in this study:

H1. There is a significant negative association between CSR and EM in manufacturing companies with high CSR ratings.

H2. There is a significant positive association between CSR and EM in manufacturing companies with low CSR ratings.

To sum up, those two hypotheses are made based on the results of Mohmed et al.

(2020) that seem to support the argument that firms with high CSR ratings are engaged in CSR reporting because they are interested in pleasing their stakeholders. Because they are dedicated in that pursuit, they end up getting higher CSR ratings. The

companies with lower ratings seem on the other hand to be engaged in CSR reporting in

order to be perceived as CSR conscious and they use that to hide EM. Since they are

not truly dedicated to CSR practices, their scores tend to be lower.

17

3 Method

3.1 Methodological framework

Now that the aim of the study is stated and the literature deemed relevant to the subject matter is presented, the next step is to choose an appropriate method to carry out the study. According to Bryman and Bell (2017), it is possible to identify two broad methods of conducting research – quantitative and qualitative. The frequently put forward difference between those categories is that the former is often interested in quantifying when it comes to data collection and analyses whereas the latter is more interested in understanding and interpreting behavior. Bryman and Bell (2017) further emphasize that the focus in a quantitative method lies in empirically testing theories, in other words it tends to be deductive.

This thesis clearly falls in the category of quantitative method since it is interested in measuring variables and, by using two theories as a starting point, figuring out the link between the two variables which in this case happen to be CSR scores and EM. This is not to say that the only relevant method in research that has to do with CSR and EM is quantitative. To the contrary, if the aim of this study was understanding CSR reporting and EM practices, not establishing a link between the variables, a qualitative method that uses interviews with managers of companies as data source would have made more sense. This is a long way of saying that the choice of method is largely determined by aim and problem formulation of studies (Bryman and Bell, 2017).

Estimating the link between variables is an important factor in quantitative research.

This often involves determining dependent variables and an independent variable in the

research area, and figuring out if changes in the former lead to changes in the latter

(Bryman and Bell, 2017). To this end, specific quantitative tools such as correlation

analysis – the link between two variables – and multivariate analysis – the analysis of

three or more variables can be used. Univariate analysis is on the other hand used to get

descriptive statistics of variables individually such as frequency tables, histograms,

mean value and standard deviation which help in knowing the variables in a given study

in detail (Djurfeldt et al., 2010). With this in mind, some of these forms of analysis are

used in this study.

18 Finally, quantitative research is closely connected to the philosophical school of thought known as positivism which encourages the use of natural science techniques while conducting research in the social sciences, hence the prioritization of quantifying and measuring variables, hypothesis testing and replicability in quantitative research (Bryman and Bell, 2017). However, because of these characteristics, quantitative research is criticized for being too obsessed with providing precise numbers, and ignoring the complexities of human behavior which is after all the subject of the social sciences (Bryman and Bell, 2017). To combat that, continuous effort is made in this thesis to put numbers in context.

3.2 Data collection 3.2.1 Publications

The empiric knowledge collected for this study comes from publications of different sorts, most of which are peer-reviewed articles from reputable academic journals. To ensure the quality of the academic journals, and thereby the articles, the Academic Journal Guide published by the Association of Business Schools (2015) is used. Hence, the majority of journals used here meet the preferred ranking in line with the guide, and it is those articles that are raised repeatedly throughout the thesis.

Furthermore, to guarantee current and up to date facts, newly published articles are included. To strengthen the cumulative aspect of the thesis on the other hand, the need for including older and widely-used articles has also played a role in deciding which articles to refer to. The publication date of the articles ranges from the 1970’s to 2020, with the most being published in the last decade.

In addition to articles, some course literature are seldom used. This is especially true in regard to choosing and explaining different methods. As rare is the use of

websites and other theses.

3.2.2 Raw data and sampling

Thomson Reuters DataStream is the source to all the secondary data used in this study.

The data provided by DataStream is comprehensive including 175 countries, and 35 million individual instruments and indicators (Refinitiv, 2020). Despite the

thoroughness of this particular database, the quality of secondary data can be difficult to

guarantee – simply because the researcher is not personally involved (Bryman and Bell,

2014). However, Thomson Reuters (2017) states that all necessary steps are taken to

19 make the whole data collection process as objective and transparent as possible. This assertion by the database compiler seems to be accepted in the research community as it is widely used (see for example Dhaliwal et al. 2012; Eccles et al. 2011). And this study follows that lead. Also, it is worth noting that the emphasis put on objectivity makes the database especially appealing to quantitative studies, which, as discussed in 3.1, tend to be influenced by positivism.

As this study focuses on the manufacturing sector of the EU, the first step to undertake in the practical raw data collection process is to identify companies that fulfil these two criteria. In DataStream, companies are classified into various economic sectors that, unfortunately, may include both manufacturing and services. If one goes through each category and compiles those that have to do with manufacturing, just over 2000 EU listed companies can be found. The process can however be subjective since identifying where one sector ends and the other starts is not always straightforward (see figure 3 for DataStream sectors and subsectors considered as manufacturing in this study).

Out of the about 2000 companies, 127 of them have CSR data for 2010-2019, the study period. Since these 127 companies are both in the manufacturing sector and publish CSR performance reports, they are considered as the population of the study. In order to first estimate EM and then to develop a model linking it to CSR, a dozen variables are needed (details are available in the coming two sections). 91 of the companies have complete data for nine variables and, as a result, these companies are taken as samples resulting in 910 firm-year observations (annex 8.2). Using the rather simplified formula of calculating sample size by Yamane (1967) which is seldom used by quantitative studies, the minimum sample size for a study with a population of 127 is 96 (annex 8.1).

This would make the sample size of this study so slightly lower than that.

The missing entries for the remaining three variables out of the twelve are calculated by

taking the average of the given entries. For instance, if the entry of variable_1 for

company A for the year 2019 is missing, it would be calculated by adding the entries

from 2010 to 2018 and dividing the result by nine (the number of entries). Increasing

the sample size is considered to lead to less quality in the data as it would require more

of such calculations. Moreover, the similar study carried out by Mohmed et al. (2020) is

based on firm-year observations of around 800. Decreasing the number of samples until

all data is found in DataStream would on the other hand lead to a very small sample size

20 and thus is not seen as a wise choice. Beside, out of the three incomplete variables, two (depreciation and amortization, and property, plant and equipment) are not expected to significantly fluctuate year to year. The third variable whose some of its entries are estimated by averaging is change in cash.

Figure 3. The composition of the sample of manufacturing companies in EU

Perhaps not surprisingly, the sample composition shown in figure (3) reflects that most CSR reporting companies are from larger economies such as Germany and France. That can be explained by the simple fact that large advanced countries tend to be home to many companies. Because of mandatory CSR reporting practices, even though having a considerably smaller economy, Sweden is also well represented. Not represented at all are companies exchanged within the east European countries that seem to have weak CSR reporting traditions, just like what Arraiano and Hategan (2019) suggest.

Moreover, when conducting study in Business administration, some researchers resort to excluding the UK. The reason often given is that the kingdom has an accounting system that is vastly different from the continental countries (Hartwig, 2018). Since the UK has left the union on its own will in addition to its distinct accounting system, this study does not have that dilemma. All the remaining EU-27 countries are included.

Finally, the time range between 2010 and 2019 is chosen because it is the time period

following the great recession, the data is thus expected to be less volatile making

comparisons more reliable. When calculating year-on-year change in some variables

such as liabilities and revenue, the data from 2009 is included.

21 3.3 Measuring CSR

In this study, the ESG score in Thomson Reuters DataStream is used as a measurement for CSR. ESG is an abbreviation for environmental, social and corporate governance aspects of companies which are the broad pillars included in the measure. The total ESG score for a company is given based on its performance in ten subcategories shown in figure (4) that fall under the pillars. Each subcategory can have more than 30 indicators that are graded based on data updated weekly by hundreds of analysts from various sources such as annual reports, company websites, and even news sources. Based on these updates, ESG scores are continuously recalculated (Refinitiv, 2019). The ESG score is given in percentages. According to Thomson Reuters (2017), this makes comparison between companies easier. But, the score is also available in letter grades.

One choice that must be made while using ESG score as measurement for CSR is whether to take the combined score, or whether to give more weight to one or two of the pillars that are deemed to be more relevant to the specific study. In this study, the first alternative is chosen, i.e. to use the combined scores. The reason is that CSR is defined more broadly here in line with Carrol’s model (fig. 1) where companies have

responsibilities that include being ethical and good corporate citizens. It can be argued that environmental, social and corporate governance aspects of the ESG are all relevant in identifying ethical and good corporate citizens. In studies that define CSR narrowly, weighing the social pillar more would make sense.

Figure 4. Pillars and subcategories of the ESG score (Thomson Reuters, 2017)

22 3.4 Estimating EM

First of all, in line with Mohmed et al. (2020), accrual based EM (hereafter AEM) is used in this thesis as a proxy of EM (see fig. 2 on the categories of EM). Data for the levels of AEM in companies is however not directly available – it wouldn’t make sense for a rational manager to disclose how, by using different accounting methods, the financial reports in the company are tinkered with. Instead, indirect ways of estimating AEM must be undertaken.

In previous research including but not limited to Kothari et al. (2005), the so-called discretionary accruals is used as an approximation for AEM. Remember that accruals are revenues earned or expenses incurred in a given year which impact a company’s income statement but the cash related to the revenue or expense has not yet changed hands (Tuovila, 2019). Discretionary accruals in return include the portion of accruals that are made by the discretion of management i.e. intentionally. Since the delay is made intentionally, it is seen as a way of moving revenue or expense from a given accounting period to another in order to arrive at a predetermined result, in other words they are tools of EM and, consequently, are used to estimate EM.

Throughout the years, there have been various models developed in order to measure the discretionary portion of accruals and most models build on the ones that came before. The earliest include Healy (1985) and DeAngelo (1986). The logic behind those models is the same – companies in the same sector have similar business models, account turnover ratios and depreciation schedules that should result in similar accruals.

Any deviation from what is considered “normal” levels of accruals in a given sector is thus considered to arise because of management discretion (Ivanov and Maximova, 2011).

The starting point for estimating discretionary accruals is that the sum of discretionary accruals and nondiscretionary accruals gives total accruals (equation 1). This implies that total accruals minus nondiscretionary accruals gives discretionary accruals (equation 2). Furthermore, recognized total accruals in equation (2) can be calculated from the balance sheet by the so called cash flow approach shown in equation (3), a formula used by Jones (1991).

𝑇𝐴 _𝑖,𝑡 = 𝐷𝐴 _𝑖,𝑡 + 𝑁𝐷𝐴 _𝑖,𝑡 (1)

23 𝐷𝐴 _𝑖,𝑡 = 𝑇𝐴 _𝑖,𝑡 − 𝑁𝐷𝐴 _𝑖,𝑡 (2) Where

𝑇𝐴 _𝑖,𝑡 is total accruals for company i during year t, 𝐷𝐴 _𝑖,𝑡 is discretionary accruals, and

𝑁𝐷𝐴 _𝑖,𝑡 is nondiscretionary/normal component of accruals

𝑇𝐴 _𝑖,𝑡 = ∆𝐶𝐴 _𝑖,𝑡 − ∆𝐶𝐿 _𝑖,𝑡 − ∆𝐶𝑎𝑠ℎ _𝑖,𝑡 − 𝐷𝑒𝑝𝐸𝑥𝑝 _𝑖,𝑡 (3) Where

𝑇𝐴 _𝑖,𝑡 is total accruals recognized for company i during year t,

∆𝐶𝐴 _𝑖,𝑡 is change in current assets from year t-1 to t,

∆𝐶𝐿 _𝑖,𝑡 is change in current liabilities from year t-1 to t,

∆𝐶𝑎𝑠ℎ _𝑖,𝑡 is change in cash from year t-1 to t,

𝐷𝑒𝑝𝐸𝑥𝑝 _𝑖,𝑡 is depreciation and amortization expense in year t

It is in the calculation of the nondiscretionary part of equation (2) where quantitative models are needed of which the most widely used is perhaps the so called Modified Jones model developed by Dechow et al. (1995) (equation 4). In this model, total accruals is the sum of expected normal accruals (accruals to be expected in a firm under normal circumstances given its industry type), and the error term 𝜀 _𝑡 that denotes

unexpected accruals, in other words, discretionary accruals that are based on managerial decisions. Based on this, nondiscretionary accruals can be rewritten as equation (5).

Now that the two parts of equation (2) – total accruals and nondiscretionary accruals – are known, discretionary accruals are estimated by using equation (6) which is basically an expanded version of equation (2). The coefficients in equation (6) are unique for every firm and are, according to Dechow et al. (1995), calculated by using the original Jones model shown in equation (7). The role of the coefficients is to determine expected accruals under normal circumstances given a firm’s financial variables such as revenue.

𝑇𝐴 _𝑖,𝑡

𝐴 _{𝑖,𝑡−1} = 𝛼 _0,𝑖 ∗ ¹

𝐴 _{𝑖,𝑡−1} + 𝛼 _1,𝑖 ∗ ^{∆𝑅𝐸𝑉 𝑖,𝑡 −∆𝐴𝑅 𝑖,𝑡}

𝐴 _{𝑖,𝑡−1} + 𝛼 _2,𝑖 ∗ ^{𝑃𝑃𝐸 𝑖,𝑡}

𝐴 _{𝑖,𝑡−1} + 𝜀 _𝑡 (4) 𝑁𝐷𝐴 _𝑖,𝑡 = 𝛼 _0,𝑖 ∗ ¹

𝐴 _{𝑖,𝑡−1} + 𝛼 _1,𝑖 ∗ ^{∆𝑅𝐸𝑉 𝑖,𝑡 −∆𝐴𝑅 𝑖,𝑡}

𝐴 _{𝑖,𝑡−1} + 𝛼 _2,𝑖 ∗ ^{𝑃𝑃𝐸 𝑖,𝑡}

𝐴 _{𝑖,𝑡−1} (5) 𝐷𝐴 _𝑖,𝑡 = ^{𝑇𝐴 𝑖,𝑡}

𝐴 𝑖,𝑡−1 - [𝛼 _0,𝑖 ∗ ¹

𝐴 𝑖,𝑡−1 + 𝛼 _1,𝑖 ∗ ^{∆𝑅𝐸𝑉 𝑖,𝑡 −∆𝐴𝑅 𝑖,𝑡}

𝐴 𝑖,𝑡−1 + 𝛼 _2,𝑖 ∗ ^{𝑃𝑃𝐸 𝑖,𝑡}

𝐴 𝑖,𝑡−1 ] (6) Where

𝐷𝐴 _𝑖,𝑡 is discretionary accruals in year t,

𝑇𝐴 _𝑖,𝑡 is total accruals recognized in year t calculated using equation (2),

𝐴 _{𝑖,𝑡−1} is lagged total assets,

24

∆𝑅𝐸𝑉 _𝑖,𝑡 is change in revenue from year t-1 to year t,

∆𝐴𝑅 _𝑖,𝑡 is change in accounts receivable from year t-1 to year t, 𝑃𝑃𝐸 _𝑖,𝑡 is gross property plant and equipment for year t, and 𝛼 _𝑗 are firm specific coefficients generated using equation (7)

𝑇𝐴 _𝑖,𝑡

𝐴 _{𝑖,𝑡−1} = 𝛼 _0,𝑖 ( ¹

𝐴 _{𝑖,𝑡−1} ) + 𝛼 _1,𝑖 ( ^{∆𝑅𝐸𝑉 𝑖,𝑡}

𝐴 _{𝑖,𝑡−1} ) + 𝛼 _2,𝑖 ( ^{𝑃𝑃𝐸 𝑖,𝑡}

𝐴 _{𝑖,𝑡−1} ) + 𝜀 _𝑡 (7) To make the process even clearer, the concrete steps taken to estimate discretionary accruals (proxy to AEM) are: 1 – finding total accruals using equation (3), 2 – inserting the resulting total accruals into equation (7) in order to calculate the coefficients for the sector, and finally 3 – inserting the total accruals and the coefficients into equation (6) to come up with estimates of discretionary accruals. In future calculations, the absolute value of the discretionary accruals is taken since the interest in this thesis lies in the existence of AEM and its link to CSR, whether AEM is used to inflate or decrease results is not important.

3.5 Regression model

To capture the link between AEM and CSR, a regression model with five variables is created (equation 8). In line with previous research such as Kim et al. (2012), the absolute value of AEM is taken as a dependent variable. This implies the decision of managers in regard to EM is determined by the levels of CSR. A positive 𝛽 ₁ in equation (8) would imply the agency theory, whereas a negative value where AEM decreases when CSR rises would suggest the stakeholder perspective.

But it is not only CSR that influences AEM. As a result, studies that look into the link

between AEM and CSR include control variables such as firm size and leverage (LEV)

in their regression models. This is because previous research finds those variables to

have statistically significant impact on the levels of AEM (Mohmed et al., 2020). For

instance, Pincus and Rajgopal (2002) show managers in larger firms engage in EM to

show less erratic earnings. This would indicate a positive 𝛽 ₂ . In addition to leverage,

other financial indicators like market to book value (MB) and ROA are often included

as control variables in regression models (Kim et al. 2012; Mohmed et al., 2020). It is

the case here too. It is also fairly common to include corporate governance variables,

especially board size and auditing firms in regression models because those factors also

determine AEM (discussed in 2.3.2). Since collecting data for corporate governance

25 variables involves the time-consuming process of going through the annual reports of companies one by one, the practical decision to focus on just firm size and financial indicators is thus made. It is worth mentioning that there are no specific variables that must always be included in EM and CSR regression models. There are certain

similarities but also considerable differences in the variables included by Mohmed et al.

(2020) and Prior et al. (2006) for instance. Equation (8) includes however the variables that tend to be the most common in the literature.

𝐴𝐸𝑀 _𝑖,𝑡 = 𝛽 ₀ + 𝛽 ₁ 𝐶𝑆𝑅 _𝑖,𝑡 + 𝛽 ₂ 𝑆𝐼𝑍𝐸 _𝑖,𝑡 + 𝛽 ₃ 𝐿𝐸𝑉 _𝑖,𝑡 + 𝛽 ₄ 𝑀𝐵 _𝑖,𝑡 + 𝛽 ₅ 𝑅𝑂𝐴 _𝑖,𝑡 (8) Where

𝐴𝐸𝑀 _𝑖,𝑡 is the absolute value of discretionary accruals for firm i in the year t, 𝐶𝑆𝑅 _𝑖,𝑡 is the percentage ESG score for firm i divided by 100,

𝑆𝐼𝑍𝐸 _𝑖,𝑡 is the firm size measured by the natural logarithm of total assets for firm i in the year t,

𝐿𝐸𝑉 _𝑖,𝑡 is total liabilities scaled by total assets for the firm i in year t

𝑀𝐵 _𝑖,𝑡 is the market to book ratio per share measured by the market value of equity to book value of equity for firm i in the year t, and

𝑅𝑂𝐴 _𝑖,𝑡 the return on assets measured by net income divided by the total assets for firm i in the year t

To capture more clearly the differences in AEM, if any, between firms with high CSR ratings and those with lower CSR ratings, the firm-year data is divided into two categories – those in the top 30 percent (model 1) and the remaining 70 percent (model 2). The two categories are then regressed separately using the statistical software SPSS to come up with the statistical description of all the variables in equation (8). The correlation between those variables, and the causality between the five independent variables and the dependent variable AEM is calculated using bivariate and multivariate analysis respectively. This, it is hoped, would help in seeing the two theories more distinctly in the data set in addition to making

comparison to Mohmed et al. (2020) easier. Note that Mohmed et al. (2020) divide the firms in a similar way. And finally, the entire firm-year data (model 3) is inserted into the regression model to see if a negative or positive link is stronger overall. The results of the three models are presented in the upcoming chapter.

Going forward, annex 8.3 can be used as a systematic summary of the variables

involved in the regression and how they are measured.

26 3.6 Statistical tests

To begin with, in order to test how strong the relation between each pair of variables in equation (8) is, the so-called Pearson’s coefficient r is used. According to Bryman and Bell (2017), the value of r lies on or between -1 and 1. As the value of r gets closer to -1 or 1, the relation between two variables becomes stronger. The negative or positive sign of the correlation coefficient indicates whether the variables under consideration are related negatively or positively.

When deciding to accept or reject coefficients – be it correlation coefficients or the betas in equation (8), the statistical significance level is set to p < 0.05. According to Bryman and Bell (2017), this is a common practice in research within the social sciences. Note that when the value of p is large, any relation established between variables is more likely to be caused by chance. When p gets closer and closer to zero however, chance plays a lesser role and coefficients become statistically sound.

In multivariate equations, the coefficient of determination R squared is used to show the proportion of the variance in the dependent variable that can be attributed to the

independent variables (Bryman and Bell, 2017). Close to 1 R squared suggests that the most of the factors that explain the changes in the dependent variable are included as independent variables in the equation.

Histogram representation of the data is used to determine if it resembles normal

distribution. The data for CSR, SIZE, LEV and ROA, more or less, approximate normal distribution while MB seems to slightly have a positive skew.

And finally, the independent variables in multivariate equations should be distinct. If they are too similar, the equation is said to suffer from multicollinearity, and keeping just one variable from those that are found to be too similar can lead to a better model (Djurfeldt et al., 2010). Multicollinearity is the last test undertaken, and the dependent variables are found to be distinct enough. There are of course other statistical tests that can be used. Mirroring similar studies out of which Mohmed et al. (2020) is one, only those tests that are believed to be necessary for the aim, scope and level of this

particular study are used.