Does it financially pay off to act responsible? A comparison between two capitalist economies in the Western world.

(1)

Does it financially pay off to act responsible? A comparison

between two capitalist economies in the Western world.

Ruud Kok (S3901645), supervised by dr. N. Selmane MSc Finance Thesis

Date: 04-06-2020

Faculty of Economics & Business, University of Groningen

Article info Abstract Keywords: CSR ESG ROA Tobin’s Q Capitalism Economy

Drawn upon the bulk of corporate social responsibility - corporate financial performance (CSR-CFP) literature, this study extends this CSR-CFP relationship by providing an analysis and comparison between coordinated market economy (CME) countries and liberal market economies (LME). First, a general CSR-CFP regression is run. Secondly, by using the CME-LME taxonomy created by Hall and Soskice (2001), this paper tries to get a better understanding of the CSR-CFP relationship in the Western capitalist world. This is done by regressing ESG on two dependent variables, return on assets (ROA) and Tobin's Q, by controlling for the two economies. Results for the general CSR-CFP relationship show a minor

(2)

countries in the liberal market economies (LMEs) and coordinated market economies (CMEs) for this relationship is very limited in literature though. This is interesting because the relation is likely to be different for these two types of economies, given the institutional differences. Law systems, type of firm financing, protection of shareholders and stakeholders are just some fields where these two types of capitalist economies differ from each other. Because of these

differences it is likely that shareholders, investors, customers etc. will react differently to CSR in these two types of economies. Therefore, significant differences are expected between these two set of countries. Testing the CSR-CFP relationship at the CME-LME taxonomy adds value since it enables firms and managers to get more insights into this relatively new relationship; it gives them a perception whether and how much should be spent on CSR to reach certain financial goals. The concept of CSR is operationalized by means of ESG ratings, because there is a high availability of ESG ratings for many companies and years. CFP is operationalized by two common measurements: return on assets (ROA) and Tobin’s Q. The former indicates the firm’s profitability and the latter shows the firm’s value. This study holds two research questions:

- Are corporate social responsibility and corporate financial performance positively related? - Does this CSR-CFP relationship hold more for coordinated market economy countries than fore liberal market economy countries?

The remainder of this paper is structured as follows: the second chapter will highlight the importance of ESG ratings as a proxy for CSR, followed by explaining the CSR-CFP

(5)

2. Literature review

2.1 Using ESG to determine CSR of companies

Due to increase in social and especially environmental awareness during the last years, ESG is becoming more and more important. This is straight-forward since ESG is in fact an

operationalization or a proxy of the concept CSR. According to Shackleton et al (2019), “an ESG rating is an assessment of the positive and negative environmental, social and governance

performance of a corporation.” The first indicates the firm’s effort to reduce its emissions and resource consumption. The second shows how much the firm respects human rights, product responsibility, its workforce, and the community. The third indicates how well the firm’s governance mechanisms ensure that its executives and board members act in the best interest of its long-term shareholders (Velte, 2017). To get a better understanding what the E, S, G pillars and its controversies contain, the categories and themes of the three pillars and controversies are displayed at Appendix F. Due to pressures from outside, managers are increasingly concerned to improve the image of companies, and CSR is a major tool to accomplish this. The increase of socially responsible investments (SRI) is also reflected by the numbers: according to the Forum for Sustainable and Responsible Investment, the SRI market has risen to $12 trillion which is a 20-fold increase since 1995.

In the literature, there is no precise definition of CSR; therefore, many different definitions exist in the literature. For example, the European Commission (2001), defines it as “a concept whereby companies integrate social and environmental concerns in their business operations and in their interaction with their stakeholders on a voluntary basis.” According to Heal (2004), “CSR is defined as a program of actions taken to reduce externalized costs or to avoid distributional conflicts”. These negative externalities stem from the consumption or production of a product or service which results in a cost for an unrelated third party. McWilliams and Siegel (2001) refer in their paper that CSR can be considered as “actions that appear to further some social good, beyond the interests of the firm and that which is required by law”. Since there is no precise consensus of CSR in the literature, a standardization and operationalization of the concept is fruitful for the adaption of CSR in companies. This is where ESG comes in.

(6)

importance of the concept (Dyck et al, 2019). According to Shackleton et al (2019), even most of the top 50 institutional investors integrated ESG in their equity screening, portfolio construction and quantitative analysis. Institutional investors and regulators use ESG amongst others for evaluating the risks and opportunities of a company. Another tool that is used by e.g. pension funds and endowments, is the exclusion of so called “sin firms”, those involved in alcohol, tobacco, weaponry, and gambling business. Given the increasing global socially responsible awareness, it is likely that the number of institutional investors using ESG will increase in the future.

During recent years, the utilization of ESG information by stakeholders and mainly investors has expanded (Tarmuji et al (2016). Before the emergence of quantitative measure indicators such as ESG, stakeholders and investors had to rely on qualitative information which are bothersome when one e.g. wants to compare companies based on CSR. In the late 1990s, 35% of the world 250 largest companies disclosed their ESG activities; in 2017 this number has increased to 93%.1

ESG data is also becoming more available for researchers because the number of companies reporting their ESG scores are increasing for years already. Given the increasing use of ESG, more and updated research in this field adds value. Please keep in mind that ESG and CSR are used interchangeably in this research; again, ESG is a proxy for CSR after all.

2.2 CSR – CFP relationship

As previously mentioned, CSR received increasing attention during the last years. As a result, much has been written about the relation between CSR and financial performance. The list of studies covering this relation is big. Not only has the CSR-CFP been tested on a worldwide basis, it is also applied on for example a country, industry, and law system basis. Given the wide variety of institutional environments, it implies that this relation hardly can be used in a vacuum but is dependent on its context. Nevertheless, Cavaco and Crifo (2014) for example states that no consensus has emerged so far on whether the relation between CSR and financial performance is significant. However, Friede et al (2015) performed a metastudy on the CFR-CFP relation by combining the findings of 2200 individual studies. About 90% of the studies show a nonnegative

1_See

(7)

CSR-CFP relation, so an academic consensus that a high CSR performance leads to a better financial performance, is more or less confirmed by this study.

Resource-based view (a theory that states when companies posit so called VRIR resources, they have a competitive advantage (Barney, 1991)) scholars advocate that CSR can improve firm-stakeholder relationship and enhance the firm’s reputation, which could point that there exists a positive relation between CSR and financial performance (Tang et al, 2012). This is consistent with the good governance or stakeholder view which “argues that socially responsible firms can and often do adhere to value-maximizing corporate governance practices” (Ferrell et al, 2016). This stakeholder theory argues that firm value maximization should incorporate the value of stakeholders, not merely shareholder value. In contrast, some neoclassical economists such as Milton Friedman (1970) argue that CSR does not improve financial performance. These economists look through an agency view lens: the only responsibility for corporations is to make profit and CSR is often seen as an agency problem because it is not in the interest of the shareholders. According to this view, managers that engage in “time-consuming CSR activities lose focus on their core managerial responsibilities” (Ferrell et al, 2016). This view is akin to the shareholder theory. In this agency view, the state must correct the market failures which occur due to the externalities (Bénabou & Tirole, 2010). One would expect a weaker CSR-CFP relationship in environments where the shareholder theory is prevalent, compared to environments following the stakeholder theory.

CSR has also emerged as a differential strategy rather than a minimum responsibility: companies increasingly try to use CSR for their strategy to create a brand insurance against adverse events (Werther and Chandler, 2005). In this way, chances on conflicts with stakeholders are lowered. Moreover, these CSR activities positively impact the corporate reputation which provides a mean to create competitive advantage (Gupta 2002; Rhou (2016)). Companies also benefit by acting transparently and disclose information, since rewards from CSR can only be achieved if the stakeholders are aware of the CSR activities (McWilliams and Siegel, 2001). When the company is not reporting these CSR activities, the stakeholders cannot be kept informed after all.

(8)

CSR-CFP studies applied to different environments, countries, regions etc. Recently, Walker et al (2019) performed a study very similar to this study: they compared liberal and coordinated market economy companies in a CSR-CFP relation. As expected, the results demonstrate the influence of the institutional environment, which suggests that corporations behave contingent on their external environment. That is, the CSR-CFP relationship holds for coordinated market economies but not for liberal market economies. From a managerial point of view, this is very important because he/she can estimate whether efforts spent in CSR activities do pay-off from a financial perspective. Section 2.3 will go in depth about the comparison between LMEs and CMEs. This study tries to extend on this same CSR-CFP relationship and comparison of the same set of economies by adding the return on assets (ROA) as an independent variable, extending the timespan, and altering the control variables slightly. On top of that, the causality of the CSR-CFP relationship will be tested by means of a Granger causality test. ROA is added as a dependent variable since it is an accounting based indicator; next to the Tobin’s Q, which is a hybrid between accounting and market based indicators, this gives a more comprehensive picture of CFP .Moreover, integrating both accounting and market-based measures is in line with the literature and in this way, a useful contribution is made. These measurements will be thoroughly explained in section 3.1. Extending the timespan gives us a better picture of the CSR-CFP relationship throughout the years. The addition of the reverse causality test provides us with the direction of the CSR-CFP relationship, which is very meaningful for firms since it indicates whether and how much should be allocated to CSR spending.

2.3 Coordinated and liberal market economies

There is still debate in the literature whether the CSR-CFP relationship holds. Most studies point towards a positive relationship, nevertheless there are studies that claim the opposite (e.g. Vance, 1975; Wright and Ferris, 1997). Therefore, Walker et al (2019) suggested that this inconsistency might has something to do with the institutional differences across countries. More specific, he hypothesized that CSR-CFP relationship is only existent for companies coming from the CME world and not from the LME world. Why this is so, will be explained in the hypotheses overview. The two sets of institutional environments will be analyzed now.

To operationalize exactly which countries belong to which region, the taxonomy created by Hall and Soskice (2001) will be used. They suggested that there are roughly two varieties of

(9)

economies (CME). Hall and Soskice (2001) refer to the following countries regarding the liberal market economy countries: Australia, Canada, Ireland, New Zealand, United Kingdom and United States. These countries are known to follow the Anglo-Saxon model of capitalism. For the coordinated market economy countries, they refer to Austria, Belgium, Denmark, Finland,

Iceland, Germany, Japan, Netherlands, Norway, Sweden and Switzerland. They are known to have a social market economy. Besides these countries in Europe, Hall & Soskice claim that another “world” exists, consisting of France, Italy, Spain, Portugal, Greece and Turkey, also called the “Mediterranean”. The focus of this study is on the other two models; however, it is important to mention these countries because these Mediterranean countries have many

characteristics which are similar to especially coordinated market economy countries such as an equal law system, common currency, geographical proximity etc. An update of the classification made by Soskice and Hall cannot harm since it dates from 2001, therefore a short recent literature revise is done. Hall and Soskice (2001) received criticisms due to its simplistic dichotomy

(10)

Companies in the liberal market follow the shareholder theory whereas the stakeholder theory is directive in the coordinated market world (Ferrell et al, 2016). The former asserts that the sole social responsibility of a company is to use its resources and create profits, bounded to the law and regulation. The latter advocated that managers have duties to all the stakeholders of the organization. Passive monitoring is prevalent in the liberal market world whereas in the

coordinated market world, active monitoring is used most frequently. One can also consider these two models as approximately the Rhineland and Anglo-Saxon world.2 The Rhineland world name is confusing though, since for example two prosperous common law country where the Rhine (and its river branch) also goes through are not in the Rhineland world included (France and Luxembourg), whereas for example Scandinavian countries and Japan are included in that same model while they are not even close to the Rhine. In the Rhineland model, there is for example more emphasis on the collective, a big government and the role of the labor unions is big. In the Anglo-Saxon world, individual responsibility is more accentuated, the government has less influence and the labor market relies more on the invisible hand. A full comparison between the two models is displayed in the table in the Appendix B. Another analogy that can be made partially with the CME-LME taxonomy, is the common vs civil law system, since it shows many similar characteristics with the liberal vs coordinated market economy taxonomy. So, when analyzing the CSR-CFP relationship in this setting might give us an indication that could be similar to the CME-LME dichotomy. An example of a study that compares civil and common law countries in a CSR-CFP relationship, is the study done by Azar and Zhou (2017). In this study, CSR has a negative effect on financial firm performance for common law countries

whereas the relation is positive for civil law countries. Still, the two taxonomies show similarities but are still different and therefore, research between CME and LME countries add value.

3. Hypotheses

3.1 CSR and firm performance 3.1.1 Hypotheses 1a-1b

CSR is expected to have a positive relation with CFP. Companies with stronger CSR

performances, face less lawsuits, environmental scandals, and consumer boycotts (Raza, 2012). Besides that, CSR improve employees ‘productivity and the relation between humans (Raza,

(11)

2012). The positive relation between CSR and firms ‘financial performance is also supported by disclosure literature. According to several studies, greater disclosure of CSR activities leads to an increased investors’ awareness of a firm and an improved risk-sharing. Moreover, Dimson and Karakas (2015) found that ESG activities give positive abnormal returns followed by improved accounting performance and governance and increased institutional ownership. Besides that, information asymmetry between corporate managers and investors is reduced whereas the cost of equity capital decreases (Merton, 1987; Verrecchia, 2001; Dhaliwal 2013 et al., 2011; Kim et al (2013)). Moreover, it is found that CSR has modest positive effects on the acquirement of private debt by banks: companies with CSR concerns pay between 7 and 18 basis points more than responsible firms for a loan (Goss and Roberts, 2011). This is because “banks register CSR concerns as risks and respond with less attractive loan contract terms”. However, keep in mind that overinvestments in CSR that are unlikely to add value, are punished as well by lenders, which suggests that there might be an optimal point of CSR dedication for firms in terms of debt financing. Lastly, using CSR as a differential strategy as explained in section 2.2 can bring many financial benefits on the long term such as risk mitigation and competitive advantage. Given all these arguments, the following is hypothesized:

Hypothesis 1a: Combined ESG score and return on assets are positively related.

Hypothesis 1b: Combined ESG score and firm value are positively related.

An accounting-based and a hybrid, i.e. a mix between market-based and accounting-based, measurement item is used to measure CFP in this study. The study of Galant and Cadez (2017) gives a short overview of the advantages and disadvantages of these measurement items. The first sub hypothesis will deal with a typical accounting-based measurement: the return on assets (ROA). An advantage of accounting-based measures is that they are widely available. A major disadvantage is that it exists of historical data. The dependent variable of the second hypothesis will be measured with a mix of market- and accounting- based measurement item: Tobin’s Q. An advantage of a market-based measurement item is the directness: changes in CSR will be

reflected faster with market-based measurement items than with accounting-based items. A major drawback is the fact that there are many corporations which only have private equity. Tobin’s Q has become a standard in the literature for the measurement of CFP, which allows for

(12)

In the three upcoming hypotheses, the ESG Scores are separated by three different coefficients, because it enables us to differentiate between the three ESG factors.

3.1.2 Hypotheses 2a-2b

Firm value measured by Tobin`s Q, one of the two proxies used in this study for CFP, is found to be significantly lower for so-called “toxic stocks” compared to “neutral stocks” (Fernando et al, 2017). This implies that mitigating environmental risk exposure contributes to stock valuations. Also, firms from “toxic” industries have a lower number of institutional investors ((Fernando et al, 2017). However, “green firms” have a lower Tobin’s Q than “neutral firms” which suggests that there might exists some kind of optimal environmental focus when one wants to optimize its firm value. Regarding the financing of the firm itself, CSR seems to have positive effects as well: firms scoring high on corporate environmental performance have a lower cost of equity (El Ghoul et al, 2011). According to the International Federation of Accountants, operational activities that are guided by good environmental practices can generate cost savings and keeps away the business effect of contamination issues as well (International Federation of

Accountants, 2005). Besides above given arguments, many arguments given for hypothesis 1 are also applicable to this hypothesis. A positive relation is also in line with the literature. Therefore, the following is hypothesized:

Hypothesis 2a: Environmental scores (E) and return on assets are positively related

Hypothesis 2b: Environmental scores (E) and firm value are positively related

(13)

CSR-CFP literature and previously given arguments in hypothesis 1 stating that social

performance leads to better financial performances. It would make sense after all, that companies doing well in terms of for example data privacy, human rights and health & safety (some

indicators for the S score), would yield better financial results than peer companies that do not so. Given all above arguments, the following is hypothesized:

Hypothesis 3a: Social scores (S) and return on assets are positively related

Hypothesis 3b: Social scores (S) and firm value, are positively related

In line with other hypotheses, it comes as no surprise that it is expected that a strong governance performance leads to better financial performance. There is a vast number of articles dealing with this relationship in the literature. For example, a study done by the Credit Lyonnais Securities Asia indicated an existence of a positive relation between governance and financial performance on almost 500 developing economy firms (Gürbüz et al, 2010; CLSA, 2002). A more recent study, performed by Association of British Insurers in 2008 showed that there exists a positive link between governance systems and its financial performance for UK firms (Selvaggi and Upton, 2008). However, results from a less recent metastudy indicated there exists for example no relationship between board composition and financial performance (Dalton et al, 1998) An important remark here, is that the study might be outdated.

Notwithstanding the contradictions in the literature, a positive relation between governance and financial performance is expected. After all, it would make sense if shareholder rights are

preserved and takeover defenses are sound (which are two indicators for the G score), a company shares are more attractive for investors, leading to a higher firm value. A good governance system is in the best interests of the shareholders and limits agency costs (Fama and Jensen, 1983). Besides above given arguments, many given for hypothesis 1 are also applicable to this hypothesis. A positive relation is also in line with many similar other CSR-CFP studies such as Tarmuji et al (2016) and Velte (2017). Therefore, the following is hypothesized:

Hypothesis 4a: Governance scores (G) and return on assets are positively related

(14)

3.2 Coordinated and liberal market economies

The two hypotheses in this section deal with the comparison of the set of capitalism economies: the CMEs and LMEs. As previously stated, companies in coordinated market economies mainly follow the shareholder theory whereas companies from liberal market economies primarily pursue their operations and strategies in accordance with the shareholder theory. Evidently, a higher CSR performance is expected from companies which adhere to its stakeholders instead to the shareholders only, but the main question here is whether this CSR performance has a stronger (weaker) effect on CFP in coordinated (liberal) market economy countries? If true, this implies that for CME companies, CSR performance is rewarded more by financial performance.

According to Hall and Soskice (2001), in CMEs there is a strong preference for consensus decision-making which “encourages the sharing of information and the development of

reputations for providing reliable information, thereby facilitating network monitoring.” Given this, one would expect a strong CSR-CFP link since e.g. reputations are important in CMEs. Also, in CMEs there is more focus on the collectivism whereas the individualism is more emphasized in LMEs. Moreover, since the stakeholder theory is prevalent in CMEs, it is likely that CSR would be more appreciated by investors. This would lead to a bigger influx of

investments. On top of that, CSR is lower in common law countries than in civil law countries (Liang and Renneboog, 2017). Given this, it is not strange to think that the CSR-CFP relationship will hold much more for the coordinated market economies: after all, all liberal market

economies have a civil law system whereas all the coordinated market economies have a common law system. Please beware that the opposite does not hold; definitely not all common (civil) law countries are coordinated (liberal) market economy countries. Finally, firms in CMEs are more long term oriented which is endorsed by investors and investment companies that screen on ESG and make use of SRI.

(15)

easier so that they can take (financial) advantage of these flexible opportunities. Lastly, LMEs are characterized by a stronger competition policy, which weakens the CSR-CFP link.

The hypotheses based on all given arguments are given below:

Hypothesis 5a: The positive relationship between ESG combined score and return on assets, will be significantly stronger for coordinated market economy companies than for liberal market economy companies

Hypothesis 5b: The positive relationship between ESG combined score and firm value, will be significantly stronger for coordinated market economy companies than for liberal market economy companies

4 Methodology & data 4.1 Measurement

CSR is measured by ESG ratings, which are already given in the database so there is no need to calculate them. The ASSET4 Thomson Reuters database is used as data input for this study, which is the second most popular ESG database used in ESG literature (Shackleton et al, 2019). ESG scores are retrieved from 2002 until 2018; 2019 is excludes due to many missing

observations. The database has several ESG scores, this study will use the combined ESG score and the individual E, S and G pillar scores. The combined ESG score consists of a combined score of the three E S G factors, complemented with the ESG controversies score. The latter is added because it adds value to account for possible controversies which are not explained by the ESC scores themselves. In this way, a more comprehensive evaluation of the firm’s sustainable behavior and impact is gauged (Shackleton et al, 2019). For a visual explanation, please see Appendix C.

(16)

policies and activities.3 However, this does not necessarily mean that these larger firms actually take better care of the environment and society. Thirdly, correlation between ESG ratings of different ESG data providers seems to be low, in some cases only 0.3, which is in high contrast with e.g. credit ratings where correlation between rating companies of 0.99 exists.4

As previously mentioned, ROA and Tobin’s Q are used to measure CFP. In section 3.1, the main differences between the two methods are already described.

ROA is measured using different formulas in Datastream, contingent on the nature of the company. Roughly, most companies are measured by following formula:

𝑅𝑒𝑡𝑢𝑟𝑛 𝑜𝑛 𝑎𝑠𝑠𝑒𝑡 (𝑅𝑂𝐴) = 𝑁𝑒𝑡 𝑖𝑛𝑐𝑜𝑚𝑒

𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑙𝑎𝑠𝑡 𝑦𝑒𝑎𝑟′_{𝑠 𝑎𝑛𝑑 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 𝑦𝑒𝑎𝑟}′_{𝑠 𝑡𝑜𝑡𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠}∗ 100

To measure the firm value, Tobin’s Q will be used. Since Thomson Reuters does not provide data on this variable, it should be calculated. There is no consensus in the literature how to calculate the Tobin’s Q. In this study, a simplified version of Tobin’s Q will be used which equals the market value of a company divided by its assets' replacement cost (Investopedia). As is explained in Guenster et al (2010), the book value of assets is approximately equal to replacement costs. Therefore, the following formula arises:

𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄 =𝐸𝑞𝑢𝑖𝑡𝑦 𝑚𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒 + 𝐿𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠 𝑚𝑎𝑟𝑘𝑒𝑡 𝑣𝑎𝑙𝑢𝑒 𝐸𝑞𝑢𝑖𝑡𝑦 𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 + 𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠 𝑏𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒

4.2 Data collection / sample

Tobin’s Q is calculated according to the formula given in section 4.1. There is no need to calculate ROA since it is already given by Datastream. Please see Appendix D for Datastream codes.

For the financial data, the same database is used. Data will be denoted in US Dollar currency. When necessary, non-US currencies are converted to US dollar. Since ESG ratings are a relatively new phenomenon, there is a great deal of missing observations, implying an unbalanced data panel. In table X, the countries with their indices used in this study, are displayed. Important to mention is that Iceland is omitted since Datastream produces errors for

3_{See https://www.nnip.com/en-INT/professional/insights/esg-ratings-friend-or-foe}

(17)

companies from this country. Moreover, the indices originally contain more companies, but some observations produced errors. Moreover, some companies have secondary listings on other exchange markets or split shares. Companies are assigned to the country where their headquarter is based. For stocks with double listings by means of class shares, the alternative class is omitted. RDS Shell, Alphabet, Fox are examples of such companies. Furthermore, some companies do not report their ESG ratings, resulting in a missing observation.

The sample size for LME companies is much bigger than for the CME countries. Firstly, this is due to the fact that the former has a significantly bigger population (487 vs 283 million

inhabitants respectively). More importantly, LME companies finance their company by means of the mechanisms of the financial markets more often whereas the CME companies finance

themselves more frequently via financial institutions such as banks.

4.3 Empirical model / equations

To test for significant relations in panel data, different methods can be exploited. The most convenient way is running a pooled OLS. Because of its simplicity, it has a major drawback: pooled OLS assumes homogeneity among all firms in the dataset. Therefore, a redundant fixed test is applied to check whether it is appropriate to use a pooled OLS. With this F-test, one can see whether there is heterogeneity among firms in the dataset. The null hypothesis states that fixed effects are redundant, whereas the alternative hypothesis states that they are not redundant. Homogeneity among all firms in the dataset is far from realistic and this thought is also

statistically substantiated: in all FE regressions in this study, the F-test is highly significant at the 1% level. So, there is evidence of heterogeneity among firms in our dataset. This holds for both dependent variables. Therefore, FE is preferred above pooled OLS for hypotheses 1 to 4. Another way of testing the hypotheses, is a random effects (RE) model. This model assumes that the correlation between the fixed effects and other explanatory variables is equal to zero. In all FE regressions in this study, this correlation is existing, i.e. not 0. Therefore, the RE is not

(18)

FE is highly preferred to test hypotheses 1-4. An advantage of FE is that it automatically controls for omitted variable bias5.

For hypotheses 5a and 5b, a pooled OLS is used, because the variable economy is an invariant variable, i.e. a variable that does not change over time. A FE model would again be preferable; however, a FE model does not allow independent dummy variables that change over time in its regression. This is because of the time demeaning process of the data: the 0s and 1s drop out, thus the whole dummy variable will be omitted.

Hypotheses 1 to 4 will be tested by means of a two-way FE OLS regression. For the ROA as an independent variable (i.e. hypotheses 1a, 2a, 3a and 4a), the equation is as follows:

𝑅𝑂𝐴_𝑖,𝑡 = 𝛼 + 𝛽₁𝐸𝑆𝐺_{𝑖,𝑡−1}+ ∑𝛽_𝑚𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠_{𝑖,𝑡−1}+ 𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑖𝑒𝑠 + 𝑇𝑖𝑚𝑒 + 𝜖𝑖,𝑡

(1)

Where α stands for the intercept, ε for the error term and FE for industry and time fixed effects. The same regression is also run with E, S and G as separate regressors. Control variables consist of the following:

- Firm size. It is common to include this control variable since it affects CFP. Larger firms for example, can perform more efficiently since they take advantage of

economies of scale which lead to stronger purchasing power and reduced costs (Rhou 2016; Riordan and Williamson, 1985). Also, big firms want to avoid becoming a target for NGO campaigns and government regulations or want to lead the industry (Tang et al,2012). Therefore, for bigger firms there exists a stronger motive to engage in CSR activities. Besides that, larger firms are better able to apply CSR engagement strategies since they are more familiar with diversified operations (Tang et al,2012). So, indirect via ESG ratings can this have its effect on CFP. Firm size is measured by the natural logarithm of the total amount of assets.

- Leverage. This also can be considered as a firm risk that could affect the financial performance in the future (Atan et al, 2017; Prior, Surroca, and Tribo, 2008).The capital structure of a company affects the spending and increases or decreases costs.

(19)

The amount of leverage used and the resulting tax shield effects, also called the capital structure, has its influence on the costs of a company and has therefore effect on CFP (Rhou, 2016; Miller and Modigliani, 1961).

Moreover, this control variable is included because managers tend to disclose more ESG / CSR information as leverage increases, because a higher leverage leads to additional scrutiny from financial institutions (Lanis and Richardson, 2013) and a lower cost of capital (Atan et al (2017); Francis, Nanda, and Olssson, 2008; Healy and Palepu, 2001). Since ESG is expected to have an influence on CFP, this leverage can have an indirect effect on CFP in this way. Leverage is measured by the ratio of total liabilities to equity.

- Research and development intensity measured by the R&D expenditures divided by sales. In recent literature, R&D intensity has been verified as an important factor that affects the relationship between CSR and financial performance in firms. (Tang et al, 2012; Hull and Rothenberg, 2008; McWilliams and Siegel, 2000). The reason for this is, is that R&D and CSR are considered as competing fund allocations. They are also highly correlated (Tang et al, 2012). Therefore, it is hard to measure the effects of CSR in e.g. highly innovative companies (Tang et al, 2012). This variable is

troublesome since not all companies report R&D expenditures. Therefore, it is better to delete observations with missing R&D values since filling in a mean or 0 will distort the data.

Hypotheses 1b, 2b, 3b, 4b deal with the firm value, where Tobin’s Q is used as the dependent variable:

𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄𝑖,𝑡 = 𝛼 + 𝛽1𝐸𝑆𝐺𝑖,𝑡−1+ ∑𝛽𝑚𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡−1

+ 𝐼𝑛𝑑𝑢𝑠𝑡𝑟𝑖𝑒𝑠 + 𝑇𝑖𝑚𝑒 + 𝜖𝑖,𝑡

(2)

With again the same abbreviations as in the previous equations. The same regression is also run with E, S and G as separate regressors. Control variables consist of the following:

(20)

and analyst coverage, reducing information asymmetries and therefore improving firm value (Aoudi and Marsat (2016); El Ghoul et al (2011). Another side of the coin is that small firms are better able to avoid public scrutiny (Aouadi and Marsat (2016); Salancik and Pfeffer 1978; Meznar and Nigh 1995).

- Leverage (see explanation below previous equation). - R&D intensity (see explanation below previous equation).

- ROA. According to Lee (2013), short term profitability of a firm generally has positive results for value performance. Please beware that this control variable is not selected for the equation with ROA as a dependent variable, but only with firm value as a dependent variable.

In all equations, the dependent variables are lagged. In this way, the effect of ESG and the control variables on financial performance can be gauged. Rather than pooling all industries together in an economy wide model, industry FEs are preferred above company FEs given the differences across industries. Also, time FE are added since it is possible that the CSR-CFP relationship has altered over time. Moreover, this is in line with similar studies (e.g. Ioannou and Serafeim (2010); Fatemi et al (2017); Aoudi and Marsat (2018)).

Besides using control variables, the dataset allows us to group samples. This is interesting because the effects can be analyzed per industry, country, type of economy and during different times. The latter is interesting since it makes sense that ESG has more effect on financial

(21)

industry, where this study is set, one would expect a different CSR-CFP relation before and after the crisis, since the crisis did have a very big impact on the banking industry, also in terms of CSR.

Hypotheses 5a and 5b are concerned with the comparison between CMEs and LMEs: 𝐶𝐹𝑃(𝑅𝑂𝐴𝑖,𝑡, 𝑇𝑜𝑏𝑖𝑛𝑠𝑄𝑖,𝑡)

= 𝛼 + 𝛽1𝐸𝑆𝐺𝑖,𝑡−1+ 𝛽2𝑋𝑖,𝑡 + 𝛽3𝑋𝑖∗ 𝐸𝑆𝐺𝑖,𝑡−1

+ ∑𝛽𝑚𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡−1(+𝑅𝑂𝐴𝑖,𝑡−1 ) + 𝜖𝑖,𝑡

(3)

Where X stands for the type of economy. All other (control) variables are equal to that of previous equations. ROA is used as an independent variable when Tobin’s Q is a dependent variable. Hypothesis 5 is not tested by means of a FE, but with a pooled OLS, since type of industry is a time-invariant variable.

As previously mentioned, ordinary least regression (OLS) will be used to analyze the panel data. OLS is a statistical procedure that is known among a wide audience (Pohlmann and Leitner, 2003). To correctly apply it, the data and equation model should suffice roughly 5 assumptions, which will be shortly discussed now.

- Assumptions 1: the errors have a zero mean. (𝐸(𝑢_𝑡 ) = 0)

This assumption is very easy to solve because when a regression has a constant, this assumption will not be violated.

- Assumptions 2: homoskedasticity (𝑣𝑎𝑟(𝑢_𝑡 ) = 𝜎2₎

This assumption is solved by using White errors. In Stata this is also known as robust standard errors, which diminishes the heteroscedasticity of the variance. In layman terms, it makes sure that the variance is constant.

- Assumptions 3: no autocorrelation (𝑐𝑜𝑣(𝑢_𝑡, 𝑢_𝑠) = 0)

When this assumption holds, there is no pattern in the errors. By lagging the independent variables and using clustered standard errors in the FE equations, it is assumed that the presence of autocorrelation is sufficiently minimized.

- Assumptions 4: no endogeneity (𝑐𝑜𝑣(𝑢_𝑡, 𝑥_𝑡) = 0)

(22)

cases in which this assumption is not met. One of them, reverse causality, will be thoroughly explained and tested. Although the main relation in this paper focuses on the direction from CSR to CFP, one could also argue that the relation holds for the opposite direction. For example, companies with strong financial performance, could allocate more funds to improve their CSR performance. The question about this causality remains unanswered in the literature; many studies do not even mention the possible reverse causality. It is also possible that there exists a simultaneity: the relation exists in both directions. Reverse causality is tested in this paper with the Granger causality test of Dumitrescu & Hurlin (2012). On the next page, this Granger causality test is described more extensively. Another pitfall that could lead to

endogeneity, are omitted variables. Although there are many more variables having effect on CFP, this study stays roughly in line with the literature concerning the control variables. Therefore, we do not expect an omitted variable bias.

- Assumptions 5: errors are normally distributed (𝑢_𝑡 ~𝑁(0, 𝜎2₎₎

This assumption can be considered the less stringent of all and is not a requirement to get your estimates “blue”. In line with the central limit theorem, it is expected that the errors are normally distributed because of its large sample size.

To address the direction of the CSR-CFP relationship, a reverse causality test will be performed. The Dumitrescu & Hurlin (2012) test will be used to test for Granger causality, since it can deal with heterogenous panel datasets. The test looks as follow:

𝑦𝑖,𝑡 = 𝛼𝑖 + ∑ 𝛶𝑖𝑘𝑦𝑖,𝑡−𝑘 𝐾 𝑘=1 + ∑ 𝛽𝑖𝑘 𝐾 𝑘=1 𝑥𝑖,𝑡−𝑘+ 𝜖𝑖,𝑡 (4)

(23)

1% level, implying that the null hypothesis is rejected. Therefore, at least one panel is stationary. It is assumed that is fair enough to presume with the Granger causality test. The test will be performed in both directions: CSR-CFP and CFP-CSR.

4.4 Descriptive statistics / sample characteristics

In table 1, the descriptive statistics per country is displayed. As expected, the ESG combined scores of CME companies are on average higher than those of their LME peers. So, the

difference between a shareholder and stakeholder view becomes clear directly. Focusing on CME countries, it comes as no surprise that Nordic countries such as Sweden, Finland and Denmark score high to very high. This is consistent with the conventional wisdom and previous studies (Dahlberg and Wiklund, 2018; Afrooz and Kruusman, 2019; Lämsä and Viljanen, 2014). At first glance, it is strange that another Nordic country, Norway, has the one of the lowest ESG rating of all CME countries. However, taking into account the presence of several big oil and gas

companies in the country, it makes sense that they drag down the combined ESG score

significantly. The USA and Canada have the lowest score of all LMEs. The USA is known as the ultimate example where companies focus substantially more on shareholders instead of

stakeholders. To statistically test the differences of ESG scores between CMEs and LMEs, a simple two sample t-test is performed. The output is shown in table 2.

In Appendix E and figure 1, the development of average individual E, S, G pillar and combined scores throughout the years is displayed. As expected, the trend is sloping upwards for all dimensions. This is in accordance with the conventional wisdom too since more emphasis is put on non-financial performance over the years. One can think of e.g. environmental durability and increasing social stigmas about racism and child labor. The trend is less steep for the governance factor; this might be due to the reason that a strong governance is in the interest of shareholders since quite some years already.

(24)

CME Share N%

ESG mean

ESG

median ESG σ Indices

Austria 32 3% 52,72 51,69 14,41 Wiener Boerse Index Belgium 45 5% 48,68 52,79 19,21 Bel 20, -Mid, -Small

Denmark 32 3% 55,04 50,61 15,51 OMX Copenhagen 20, - Mid Finland 30 3% 63,22 62,365 12,57 OMX Helsinki 25, -Mid

Germany 176 19% 49,33 49,32 19,75 CDAX

Japan 403 44% 47,35 48,92 20,67 TOPIX 500

Netherlands 61 7% 54,36 53,98 19,06 All share Index

Norway 25 3% 48,81 55,48 20,94 OBX

Sweden 51 6% 52,99 53,24 17,53 OMX Stockholm 25, -Mid Switzerland 67 7% 51,22 49,275 20,11 SMI, SPI Extra

922

LME Share N%

ESG mean

ESG

median ESG σ Indices

Australia 187 8% 46,64 46,09 18,23 ASX 200

Canada 225 10% 41,00 39,26 17,65 TSX Composite

Ireland 18 1% 45,41 43,74 21,98 ISEQ 38

New Zealand 20 1% 42,64 38,955 17,50 NZX 20 United

Kingdom 329 15% 46,23 47,62 20,28 FTSE 100, FTSE 250

United States 1477 65% 41,11 38,06 17,78 S&P 500, S&P 400, S&P 600

2256

Table 1: descriptive statistics (N, mean, median, standard deviations) per country including indices used in this study.

Variable Economy N ESG Mean ESG Std error ESG σ Significance ESG CME 10.219,00 43,99 0,20 20,47 LME 19.186,00 40,27 0,13 18,52 Combined 29.405,00 41,56 0,11 19,30 P(T>t) = 0.0000 E CME 10.219,00 44,02 0,29 29,39 LME 19.191,00 29,62 0,21 28,39 Combined 29.410,00 34,63 0,17 29,55 P(T>t) = 0.0000 S CME 10.219,00 41,88 0,26 26,33 LME 19.191,00 42,78 0,16 22,27 Combined 29.410,00 42,47 0,14 23,77 P(T>t) = 0.0009 G CME 10.219,00 50,56 0,23 23,12 LME 19.204,00 52,31 0,16 22,49 Combined 29.423,00 51,70 0,13 22,72 P(T>t) = 0.0000

(25)

Industry N ESG mean ESG median ESG σ

Automobiles & Parts 66 46,13 47,70 14,8

Banks 173 38,45 36,71 16,36

Basic Resources 109 48,92 48,86 20,47

Chemicals 92 53,34 54,87 18,74

Construction & Materials 106 45,67 41,24 21,5 Consumer Prod & Services 163 42,34 40,96 19,11 Drug & Grocery Stores 57 53,23 54,11 19,99

Energy 122 45,14 45,22 19,31

Financial Services 195 35,27 34,12 18,12

Food, Beverages and Tobacco 116 47,73 47,80 20,66

Health Care 255 45,22 41,70 19,42

Industrial Goods & Services 465 45,78 46,15 19,00

Insurance 100 48,44 47,49 16,29 Media 47 41,60 37,24 21,04 Real Estate 190 46,57 45,16 20,71 Retailers 114 46,29 44,34 17,38 Technology 246 44,47 43,10 17,79 Telecommunications 75 44,52 45,96 19,64

Travel & Leisure 110 42,86 43,62 17,07

Utilities 93 50,41 48,93 15,98

(26)

Figure 1: individual E, S, G and combined ESG score over the years of all CME and LME companies. CME companies score on average better than LME companies. Moreover,

environmental, social, and combined ESG score show a strongly increasing trend during the last 16 year. 0,00 10,00 20,00 30,00 40,00 50,00 60,00 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

ESG score averages of CME companies

E S G ESG 0,00 10,00 20,00 30,00 40,00 50,00 60,00 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

ESG score averages of LME companies

(27)

4.5 Correlation matrix

Although the assumptions for OLS are met, additional issues can arise during data analysis. One of these is the presence of multicollinearity. As one can see, is the correlation between the environmental, social, governance and combined ESG score high to very high, which is reasonable since companies often see CSR as an overarching concept which is covered by the three pillars. Thus, to perform well, firms have to score strong on all three pillars. Moreover, the ESG combined score is comprised of the three pillars. The three separate pillars and the

combined ESG score are only used in separate models, so multicollinearity is impossible between these variables. Since multicollinearity often makes an appearance when correlation is 0,9 or higher, no problems with multicollinearity is expected in this study (Franke, 2010).

Variables ESG E S G ROA Tobin’s

Q Firm size R&D Int. Lever. ESG 1 E 0.79*** 1 S 0.84*** 0.69*** 1 G 0.68*** 0.43*** 0.47*** 1 ROA -0.02*** -0.03*** 0.01 -0.04*** 1 Tobin’s Q -0.05*** -0.08*** -0.02*** -0.05*** -0.12*** 1 Firm size 0.40*** 0.42*** 0.40*** 0.36*** 0.05*** -0.05*** 1 R&D Int. -0.02** -0.01 -0.01 -0.02** -0.03*** 0.01 -0.01 1 Lever. 0.08*** 0.08*** 0.12*** 0.07*** -0.03*** -0.11*** 0.09*** -0.01 1 *** p<0.01, ** p<0.05, * p<0.1

Table 4: correlation matrix with Pearson r correlations between all continuous variables. Only the E, S, G, and ESG scores are highly correlated with each other. This is however not a threat for the model since they are used in separate models.

5 Results

5.1 Hypotheses 1-4

(28)

see whether the alternative hypotheses are accepted or rejected. Please keep in mind that the hypotheses will be accepted purely based on the whole time period (i.e. 2002-2018); the three smaller time periods are there to differentiate between different times and to dive deeper in the CSR-CFP relation.

5.1.1 Hypotheses 1a and 1b

For the whole period, the results are displayed at table 5. Starting with ROA as an independent variable, combined ESG has a highly significant positive effect, thus hypothesis 1a is accepted. The effect of combined ESG on Tobin’s Q is significant at the 10% level. Its coefficient is 0,00, implying that it has no effect on Tobin’s Q over the whole period of 2002 and 2018. So,

hypothesis 1b is not accepted. For the pre-crisis period, ESG has a positive and significant effect on ROA (table 6, panel 1), The same holds for Tobin’s Q as a dependent variable, however its effect is less strong and less significant (table 6, equation 4). During the financial crisis, the effect of combined ESG on ROA is mostly unchanged with only a 0,01-coefficient score difference (table 6, equation 2). For Tobin’s Q, the effect is not significant anymore. For the post financial crisis period, ESG has still a positive and highly significant effect on ROA. The relation between ESG and Tobin’s Q after the financial crisis is the same as during the crisis. To demonstrate the interpretation of the coefficients, the estimate at equation 1 in table 5 is taken, which is as follow: for a given firm, on average the ROA increases with 0.05 when the combined ESG score

increases with one unit. Other estimates can be interpreted in the same manner. Regarding the model fit, the model with Tobin’s Q as a dependent variable has a greater adjusted R-square, so this model explains more than the model with ROA as a dependent variable. Although it is about adjusted r-square, which does not automatically increase when any independent variable is added, it must be taken into account that the former model has one control variable more. Further, the negative coefficient estimate for firm size seems odd at first hand, but this effect has been witnessed more often in research (Hall and Weiss, 1967).

(29)

Q is significant during the financial crisis while it insignificant in the two other periods. This does not makes sense since many companies are in a “survival mode”, so less funds are allocated to non-core activities such as CSR and R&D. Surprisingly, the CSR-CFP relationship is the strongest before the crisis for ROA as an independent variable. This goes against the trend of increasing social and especially environmental awareness.

Again, the results of the environmental relation on CFP during 2002-2018 are displayed at table 5. During the whole period, the environmental score has a small significant effect on ROA. The relation between E and Tobin’s Q is however not existent. Thus, hypothesis 2a is accepted whereas hypothesis 2b is not accepted. Before the financial crisis, the relation between E and both dependent variables was non-significant (table 7, equation 1 and 4). During the financial crisis, the relation between E and Tobin’s Q is unaltered. For ROA, the opposite holds: during this period there is a positive relation between environmental scores and ROA that is significant at the 1% level (table 7, equation 2). In the post crisis years, both CSR-CFP relationships are non-significant again (table 7, panel 3 and 6).

With environmental score as a sole explanatory variable (second table), about the same pattern as with ESG exists. This is not in line with expectations since especially environment awareness has increased much in last years. It is very remarkable to observe that the ROA equation is only significant during the financial crisis. It is not a strange to observe that the effect of

environmental scores on financial performance is lower than for the other two pillars, given that environmental awareness and activism was not so much present for some years ago.

(30)

the crisis, the coefficients have decreased a bit, however its significance has remained (table 8, equation 3). For Tobin’s Q, all coefficients are the same during the three periods; only the last period has a smaller significance (table 8, equation 4,5 and 6).

Again, for S about the same patterns appear (third table). Social seems to have a bigger impact in recent years compared to during the crisis. Unexpectedly, the effect of the social pillar score on ROA during the first years is again the strongest of the three periods. S is the only score pillar that supports its hypotheses during all time periods, signaling that social corporate performance might be a correct predictor of financial corporate performance.

Again, the results of the governance score on CFP during 2002-2018 is displayed at table 5. Governance performance has positive and significant relation with ROA, but the relation with Tobin’s Q is non-existent. Therefore, only hypothesis 4a is accepted. The effect of governance on ROA throughout the three periods is stable. All relationships are positive and significant. Again, the effect is strongest before the financial crisis, although the difference between this and the other periods is minimal (see table 9, equation 1,2 and 3). The relation between G and Tobin’s Q is only significant during the financial crisis, however its coefficient is zero through all three periods, so the significant relation is not worth mentioning. (see table 9, equation 4,5 and 6) Notably, the effect of governance on ROA has not increased during the years (table 9). For firm value as an independent variable the relations remains mainly the same across the years.

(31)

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 5: FE results of the total sample (2002-2018) with no differentiation between pre, post or financial crisis. All independent variables are lagged one year. E, S, G and ESG are proxies for CFP; ROA and Tobin’s Q are proxies for CFP. Hypotheses 1 to 4 can be answered by means of this table; a further analysis of different time periods allows for a more in-depth discussion and analysis of these hypotheses (these can be seen at table 6 to 10). Control variables are displayed in italic. Vce(cluster) is used to address

heteroskedasticity and autocorrelation.

(1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES ROA ROA ROA ROA Tobin’s Q Tobin’s Q Tobin’s Q Tobin’s Q

ESG 0.05*** 0.00* (0.01) (0.00) E 0.02* 0.00 (0.01) (0.00) S 0.06*** 0.01*** (0.01) (0.00) G 0.03*** 0.00 (0.01) (0.00) Leverage -0.00 -0.00 -0.01 -0.00 0.00*** 0.00*** 0.00*** 0.00*** (0.01) (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) (0.00) Firm size -1.17*** -1.13*** -1.34*** -1.05*** -0.30*** -0.29*** -0.33*** -0.29*** (0.17) (0.18) (0.17) (0.19) (0.07) (0.07) (0.07) (0.06) R&D intensity 0.00*** 0.00*** 0.00*** 0.00*** -0.00*** -0.00*** -0.00* -0.00*** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) ROA 0.08*** 0.08*** 0.08*** 0.08*** (0.02) (0.02) (0.02) (0.02) Constant 23.03*** 23.35*** 25.71*** 21.30*** 6.33*** 6.30*** 6.78*** 6.17*** (2.65) (2.75) (2.68) (2.81) (1.08) (1.14) (1.19) (1.04)

Industry FE Yes Yes Yes Yes Yes Yes Yes Yes

Year FE Yes Yes Yes Yes Yes Yes Yes Yes

Observations 12,776 12,770 12,770 12,776 12,752 12,746 12,746 12,752

(32)

Table 6: FE results of the total sample (2002-2018) with differentiation between pre, post or financial crises. All independent

variables are lagged one year. Hypotheses 1 to 4 can be answered by means of this table. E, S, G and ESG are proxies for CFP; ROA and Tobin’s Q are proxies for CFP. Control variables are displayed in italic. Vce(cluster) is used to address heteroskedasticity and autocorrelation.

(1) (2) (3) (4) (5) (6)

VARIABLES ROA (pre-crisis) ROA (crisis) ROA (post-crisis) Tobin’s Q (pre-crisis) Tobin’s Q (crisis) Tobin’s Q (post-crisis)

ESG 0.06*** 0.05*** 0.04*** 0.01** 0.00 0.00 (0.02) (0.01) (0.01) (0.00) (0.00) (0.00) Leverage -0.02* -0.01 -0.00 -0.00 0.00*** 0.00*** (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) Firm size -1.28*** -1.41*** -1.13*** -0.30*** -0.22*** -0.31*** (0.25) (0.25) (0.17) (0.06) (0.05) (0.07) R&D intensity -0.21*** -0.15*** 0.00*** 0.04*** 0.02*** -0.00*** (0.05) (0.01) (0.00) (0.00) (0.00) (0.00) ROA 0.07*** 0.05*** 0.10*** (0.01) (0.01) (0.03) Constant 25.91*** 27.53*** 23.48*** 6.31*** 4.22*** 5.65*** (3.68) (3.72) (2.63) (1.02) (0.69) (1.10)

Industry FE Yes Yes Yes Yes Yes Yes

Year FE Yes Yes Yes Yes Yes Yes

Observations 2,622 2,469 7,685 2,610 2,464 7,678

(33)

variables are lagged one year. E, S, G and ESG are proxies for CFP; ROA and Tobin’s Q are proxies for CFP. Control variables are displayed in italic. Vce(cluster) is used to address heteroskedasticity and autocorrelation

(1) (2) (3) (4) (5) (6)

E 0.02 0.04*** 0.02 0.00 0.00 0.00 (0.02) (0.01) (0.01) (0.00) (0.00) (0.00) Leverage -0.02* -0.00 0.00 -0.00 0.00*** 0.00*** (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) Firm size -1.08*** -1.55*** -1.11*** -0.27*** -0.22*** -0.31*** (0.23) (0.32) (0.19) (0.06) (0.05) (0.08) R&D intensity -0.21*** -0.15*** 0.00*** 0.04*** 0.02*** -0.00** (0.05) (0.01) (0.00) (0.00) (0.00) (0.00) ROA 0.07*** 0.05*** 0.10*** (0.01) (0.01) (0.03) Constant 24.15*** 30.20*** 24.05*** 5.95*** 4.20*** 5.76*** (3.61) (4.38) (2.67) (0.99) (0.73) (1.17)

Observations 2,619 2,466 7,685 2,607 2,461 7,678

Adjusted R-squared

(34)

(1) (2) (3) (4) (5) (6)

S 0.06*** 0.07*** 0.05*** 0.01*** 0.01*** 0.01* (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) Leverage -0.02 -0.01 -0.00 -0.00 0.00** 0.00*** (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) Firm size -1.36*** -1.73*** -1.29*** -0.32*** -0.25*** -0.34*** (0.25) (0.27) (0.18) (0.07) (0.05) (0.08) R&D intensity -0.22*** -0.15*** 0.00*** 0.04*** 0.01*** -0.00 (0.05) (0.01) (0.00) (0.00) (0.00) (0.00) ROA 0.07*** 0.05*** 0.10*** (0.01) (0.01) (0.03) Constant 27.25*** 31.87*** 25.59*** 6.57*** 4.64*** 6.05*** (3.69) (3.83) (2.67) (1.09) (0.74) (1.23)

Observations 2,619 2,466 7,685 2,607 2,461 7,678

Adjusted R-squared 0.14 0.17 0.06 0.35 0.42 0.34

(35)

(1) (2) (3) (4) (5) (6)

G 0.03*** 0.02** 0.02** 0.00 0.00** -0.00 (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) Leverage -0.02* -0.00 -0.00 -0.00 0.00*** 0.00*** (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) Firm size -1.08*** -1.27*** -1.03*** -0.27*** -0.22*** -0.29*** (0.22) (0.23) (0.20) (0.05) (0.05) (0.07) R&D intensity -0.21*** -0.15*** 0.00*** 0.04*** 0.02*** -0.00*** (0.05) (0.01) (0.00) (0.00) (0.00) (0.00) ROA 0.07*** 0.05*** 0.10*** (0.01) (0.01) (0.03) Constant 23.17*** 25.95*** 22.45*** 5.97*** 4.22*** 5.53*** (3.26) (3.64) (2.91) (0.91) (0.69) (1.04)

Observations 2,622 2,469 7,685 2,610 2,464 7,678

Adjusted R-squared 0.12 0.14 0.05 0.35 0.41 0.34

(36)

5.2 Hypotheses 5a and 5b results and discussion

In this section, the results of the comparison between the two capitalist economies are made. This is done by adding an interaction variable. Table 10 shows the output of the tests

performed. The results are contradicting. First, ESG has a stronger impact on ROA in LME countries compared to CME countries, implying that the CSR-CFP relationship is stronger in the former set of countries. Based on this, hypothesis 5a is not accepted. On the contrary, the relation between ESG and Tobin’s Q is stronger for CMEs than for LMEs. This indicates that performing strongly in the CSR fields pays off more in CMEs than in LMEs. Therefore, hypothesis 5b is accepted. Since type of economy is a time-invariant variable, it is not possible to differentiate between time periods. Just as in the main FE model, the Tobin’s Q model has a bigger adjusted r-square than one with ROA as a dependent variable. Given the contradictions of hypotheses 5a and 5b, it is hard to make strong conclusions.

(1) (2)

VARIABLES ROA Tobin's Q

ESG 0.03*** 0.01*** (0.01) (0.00) Economy 1.86*** 0.78*** (0.36) (0.06) Economy * ESG 0.02** -0.01*** (0.01) (0.00) Leverage -0.02*** 0.00 (0.01) (0.00) Firm size -0.96*** -0.30*** (0.10) (0.01) R&D intensity 0.00*** -0.00 (0.00) (0.00) ROA 0.08*** (0.01) Constant 19.90*** 5.62*** (1.43) (0.19) Observations 12,776 12,752 Adjusted R-squared 0.06 0.28

(37)

5.3 Reverse causality test

The test output with a lag of 1 for both the ESG-ROA and ESG-Tobin’s Q is displayed below. To get a full picture, tests are run with all possible lags ranging from 1 to 3. The number of time periods T (17) must be larger than 5+3K, where K is the lag order, so a maximum of lag order 3 can be chosen (Dumitrescu & Hurlin, 2012). According to the AIC and BIC, the optimal number of lags is 1, which indicate that this lag might be the best option to base our conclusions on. The z-bar tilde is preferred above the z-bar since the latter is not suitable for unbalanced panels.6 One relation is significant at the 10% level, two relations are significant at the 5% level and all others are significant at the 1% level. In both directions, the null hypotheses are rejected. Therefore, ROA / Tobin’s Q (ESG) does Granger cause ESG (ROA / Tobin’s Q).

Lags(1) Lags(2) Lags(3)

Z-bar

tilde P-value Z-bar tilde P-value Z-bar tilde P-value ESG -> ROA 1,840 0,0658* 2,453 _0,0142** 2,394 0,0167** ESG -> Tobin’s Q 3,448 _0,0006*** 5,008 _0,000*** 4,737 0,000*** ROA -> ESG 7,021 _0,000*** 5,242 _0,000*** 3,780 0,000*** Tobin’s Q -> ESG 6,138 0,000*** 6,361 0,000*** 7,965 0,000*** *** p<0.01, ** p<0.05, * p<0.1

Table 11: results from the Dumitrescu & Hurlin (2012) Granger causality test. The z-bar tilde is preferred above the z-bar since the latter is not suitable for unbalanced panels. The results show a simultaneous relationship between ESG and CSP (both ROA and Tobin’s Q).

Given the fact that both relationship directions are significant, there is an evidence of a simultaneous relationship: CSR and CFP work both ways in our dataset. On top of that, the CFP-CSR is slightly stronger than the CSR-CFP relationship which can be considered a surprise, given the bulk of the CSR-CFP articles in literature. It is also contradicting with for example the metastudy done by Roman et al (1999) where a majority of the CSR-CFP relationships is found to be positive. Notwithstanding, some studies, claim the opposite: e.g. the results of Scholtens (2008) indicated that “the direction of the ‘causation’ predominantly runs from financial to social performance.”

(38)

5.4 Robustness checks

By changing the dependent variable, one can check how robust a certain model is. Since this study uses two CFP proxies, the robustness check is already integrated into the core study results. Since results from ROA and Tobin’s Q as a dependent variable are different, hypothesis sets a and b are contradicting (see Appendix G). Based on this, the model seems not quite robust. On the other side, one has to remind that both indicators indicate financial performance, but they measure two different but also related concepts: profitability and firm value.

(39)

(1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES ROA ROA ROA ROA Tobin’s Q Tobin’s Q Tobin’s Q Tobin’s Q

ESG 0.03*** 0.00* (0.01) (0.00) E 0.01 0.00 (0.01) (0.00) S 0.04*** 0.01*** (0.01) (0.00) G 0.02*** 0.00 (0.01) (0.00) Leverage -0.01 -0.00 -0.01 -0.00 0.00*** 0.00*** 0.00*** 0.00*** (0.01) (0.01) (0.01) (0.01) (0.00) (0.00) (0.00) (0.00) Firm size -1.08*** -1.03*** -1.20*** -1.03*** -0.32*** -0.31*** -0.34*** -0.30*** (0.18) (0.18) (0.19) (0.20) (0.07) (0.07) (0.08) (0.06) R&D intensity 0.00*** 0.00*** 0.00*** 0.00*** -0.00 -0.00 -0.00 -0.00* (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) ROA 0.08*** 0.08*** 0.08*** 0.08*** (0.02) (0.02) (0.02) (0.02) Constant 22.53*** 22.62*** 24.13*** 22.04*** 6.15*** 6.11*** 6.46*** 6.03*** (2.93) (2.72) (2.94) (3.07) (1.05) (1.09) (1.13) (1.00)

Industry FE Yes Yes Yes Yes Yes Yes Yes Yes

Year FE No No No No No No No No

Observations 12,776 12,770 12,770 12,776 12,752 12,746 12,746 12,752

Adjusted R-squared 0.05 0.05 0.06 0.05 0.31 0.31 0.32 0.31

(40)

6 Conclusion

6.1 Discussions and conclusions

In light of growing environmental, social and governance awareness in society and increasing ESG integration for investments, this study tries to add evidence to the CSR-CFP literature. With this, CMEs and LMEs are distinguished in the second part of this study. It can be concluded that there is a mixed evidence for the CSR-CFP relationship in this study. There exists a positive significant relation between ESG and ROA, whereas no existent relation between ESG and Tobin’s Q is found, surprisingly. In line with growing environmental, social and governance awareness and increasing use of ESG, one would expect a growing CSR-CFP relationship during the years. Nevertheless, the opposite is found in this study: the relation between ESG and ROA is slightly stronger before the financial crisis than after the financial crisis. On a 2002-2018 timeframe, the relation between ESG and Tobin’s Q is, as already mentioned, not existent. However, the results showed some small significant effects before the financial crisis, implying a weaker ESG-CSR relationship throughout years. To address institutional theory, the CSR-CFP relationship is run for the two types of economies. Concerning this, the results contradict each other. A higher ESG rating has a smaller effect on Tobin’s Q in LMEs than in CMEs, which accepts hypothesis 5b. This is in line with the results from the study done by Walker et al (2019). However, the interaction variable of economy with ESG has a positive effect on ROA implying that LMEs benefit more from engaging in CSR activities than CMEs. This comes as a big surprise, since the conventional wisdom is pointing towards an opposite conclusion. One would expect this effect to be stronger in CMEs where stakeholder theory is prevalent, compared to the LMEs with the shareholder view. So, according to this study, a firm from an LME will increase its profitability more when ESG increases, than firms from CMEs. Although the coefficient estimates are significant, they are rather small. Given the contradictive results and these small estimates for the CME-LME taxonomy, it is hard to build fruitful conclusions on these test results. Maybe

Does it financially pay off to act responsible? A comparison between two capitalist economies in the Western world.