Does ESG performance matter from a market perspective?

Does ESG performance matter from a market

perspective?

MSc Organisation Economics Master’s Thesis

Abstract

Corporate Social Responsibility is a concept that received a lot of attention over the last thirty years. In spite of the growing importance of Environmental, Social and Corporate

Governance performance ratings in investors’ decisions, up to now no clear understanding exists on the effect of CSR performance on firm value. This study examines this effect through the implementation of three different methodologies. The proxies for CSR (ESG rating) and firm value (Tobin’s Q) are obtained from a European dataset for 2005-2013 that is provided by ASSET4 of Thomson Reuters. Although the results point towards an overall negative effect, an additional analysis on potential moderators shows that this relationship varies among firms with initial low and high scores and hints at a positive relationship for the latter. Furthermore, the results suggest that firms operating in environmentally sensitive industries would also experience a positive relation, as opposed to their non- environmentally operating counterparts.

Thesis by: Tessa Kimmel 10666214 Supervised by: Mr. Thomas Buser

Preface

For approximately three years ago, while participating in an exchange program at the University of Surrey in the UK, I firstly encountered the notion of Corporate Social

Responsibility through a subject called Sustainable Tourism. As you may have noticed from the abstract, this has greatly inspired me as it is now the topic of this Master’s Thesis.

This thesis has enabled me to integrate this continuous interest with a new, recent interest in Finance which I discovered through the choice of Corporate Finance as an elective. As a consequence, I very much enjoyed writing this thesis. Especially the interviews, which I carried out as a follow-up of the quantitative analysis, were very interesting as it brought the topic very much to life.

Evidently, my supervisor Mr. Buser has been of great support to me. I would like to thank you for your (very!) quick responses, new insights and especially in enabling me to make progress with the methodology and Stata.

Furthermore, I would like to thank K-W Rademakers, H-J Pietersen and J-W Vosmeer for your participation in the interviews. It has been very helpful and interesting to obtain insights from a business perspective.

Then, by handing in the last piece of work of my Master’s in Organisation Economics, the only thing left for me to say is:

Enjoy reading!

Tessa Kimmel

Statement of Originality

Table of Contents

8 8 11 16 19 Methodology

Introduction to the data Variables Validity issues Method 20 20 21 23 25 Empirical results Summary statistics Evidence 30 30 31 Discussion Methodology analysis Interpretation of results 36 36 42 Conclusion 48 References 50 Appendix

Interviews PostNL, Ahold and Heineken

54 62

Introduction

For already over thirty years, we have witnessed increased attention for the role of Corporate Social Responsibility (CSR) (Widiarto Sutantoputra, 2009). Although CSR has been defined in numerous ways, it may be best explained as serving stakeholders such as “people,

communities, society and the environment” (Jiao, 2010 p.1) as such that it exceeds what is demanded by law(Cai et al., 2012). According to the European Commission’s definition this ‘serving’ is implemented through “the voluntary integration of social and environmental issues in a firm’s business activities and in their interactions with various stakeholder groups”1. This definition is often extended by the inclusion of corporate governance issues, as these concepts tend to coincide and are both instrumental to the sustainability of a business (Money and Schepers, 2007; Renneboog et al., 2008). So CSR in general can be thought of as an activity that supports the managers’ endeavor in advancing the firm’s corporate governance (Cai et al., 2012). Illustrating examples of CSR are donating to charity, encouraging a diverse workforce, protecting the environment by bringing down CO2 footprint, locally sourcing of materials, re-using production input etc. (Nelling and Webb, 2009).

Through this growing awareness strategic managers are now confronted with an increasingly demanding environment in which investors play a major role (Waddock and Graves, 1997). Eurosif2 highlights this trend in their report on Strategic Responsibility Investing (SRI) of 2014 for the European capital market. The report shows that SRI has been growing heavily since 2011 (Eurosif report, 2014). From the variety of SRI strategies, the most

well-established one is ESG integration which is defined as “the explicit inclusion by asset managers of Environmental Social and corporate Governance (ESG) risks and opportunities into traditional financial analysis and investment decisions”3. More specifically, investors and financial analysts use ESG performance ratings which are provided by private firms that assess the Environmental, Social and Corporate Governance performance of public

companies (Bouten, 2013). These independent firms, of which KLD, ASSET4 and

1 _{Green paper ‘Promoting a European framework for corporate social responsibility’ European Commission}

2001 P. 8

2 _{“Eurosif is the leading non-for-profit pan-European sustainable and responsible investment (SRI) membership}

organization whose mission is to promote sustainability through European financial markets” http://eurosif.org/about/mission/

Sustainalytics are the most common, score companies’ performance regarding the three ESG pillars and subsequently sell this information to investors (Bouten, 2013). In the period of 2011- 2013 ESG integration has grown with 65.4 percent which shows increased popularity especially over the last couple of years as the average annual growth since 2005 was 30% (Eurosif report, 2014).

This increasing ESG integration shows that the company’s performance cannot be solely demonstrated by the conventional way of financial reporting. According to Bassen and Kovacs (2008) there exists in fact a growing gap between this way of reporting and firm valuation, as the conventional reported information no longer reflects all information necessary for assessing the value of a firm due to the increased importance of intangibles4. Firm valuation is the assessment of a firm by investors based on future prospects of the firm, which can be thought of as the price if the company were bought (Marsat and Williams, 2003). The use of so called ‘extra-financials’ like ESG could improve the investors’ and managers’ assessment of a company by communicating information about the value of workforce diversity, environmental strategies, support of local businesses, recycling operations, brand reputation amongst customers etc. (Bassen and Kovacs, 2008). This facilitates an improved evaluation of the company’s future prospects as investors are now supplied with better knowledge on potential risks and opportunities (Bassen and Kovacs, 2008). Consequently, the use of ESG performance ratings is found to become an increasingly powerful tool in the valuation process and investment decisions (Bassen and Kovacs, 2008).

In spite of this trend it remains unclear how this CSR performance of companies would affect their financial results (Waddock and Graves, 1997). A large body of research exists on this relationship, which can be classified into studies measuring financial performance from a market or accounting based perspective, of which the majority tend to find a positive

relationship. However, the results remain mixed and most of this research does not warrant a full causal interpretation due to endogeneity problems like reversed causality. Although the majority of research studied the effect of CSR on financial performance (Flammer, 2013), the reversed causation seems to exists as well (Nelling and Webb, 2009; Anderson, 2014).

4 _{According to a worldwide study, intangibles, such as the company’s reputation and corporate culture,}

The goal of this study is to contribute to a better understanding of the relationship between CSR and financial performance. It aims to test the causal relationship between CSR performance, for which the ESG rating is used as a proxy, and a market- based measure of financial performance; firm value which I proxy as Tobin’s Q5. The dataset studied, is provided by ASSET4 of Thomson Reuters and contains annual ESG ratings from 2005 – 2013 for 428 European-based companies. The financial information on these firms is obtained from Datastream. This market – based measure, Tobin’s Q, reflects the value of a firm based on the expectations of future performance, taking into account the investors’ valuation of intangible assets (Derwall, 2007). Therefore, firm value is not restricted by measuring the effect of tangible aspects of ESG as is the case with accounting based performance measures (Jiao, 2010).

A critical analysis of the related literature suggests that the endogeneity problems when analyzing this relationship remain unsolved. Some authors do not even discuss the existence of these issues (Marsat and Williams, 2013; Stonski et al., 2014). Ideally, this research would propose a complete new model in overcoming all these issues, but due to complexity of the problem I could not solve them entirely, if possible at al. This study contributes to the existent literature through combining and improving on several methodologies (multiple OLS, Granger Causality and Instrumental Variable 2sls) used by previous research within the same dataset. Consequently, this implementation allowed for critical evaluation of the results and to identify whether these methods provide consistent results. In addition, three follow-up interviews with sustainability managers from companies within this dataset (PostNL, Ahold and Heineken) facilitated a more practical discussion of the results. Another contribution is, that this study meets the increasing interest in moderating factors (Dixon-Flower et al., 2013) through building on research of Derwall (2007), Reverte (2011) and Flammer (2013) by investigating whether the relationship would be more pronounced for firms with initial low ESG performance levels and firms that operate in environmentally sensitive industries. Finally, the database used is objective, original and allows an analysis of the effect of country level fixed effects.

This study finds consistent evidence from all three methods for a negative overall effect of ESG performance on firm value, that would mainly emerge through better performance on

5 _{Tobin’s Q is a ratio has often been used in the literature (specifically in finance and accounting) as a proxy for}

environmental issues. Nevertheless, as the initial ESG rating of the firm and the

environmental sensitivity of the sector in which the firm operates have shown to moderate the relationship, this overall result may not be generalized. These moderating factors hint at a positive relationship for firms with initial high ESG scores and those operating in

environmentally sensitive industries. A possible interpretation of the former moderating result could be a high expectation level among investors. The latter finding suggests that CSR may become beneficial through the high level of stakeholder pressure and media attention these types of firms experience. In addition, investors may perceive their CSR performance as a direct improvement of the sustainability of their core business as this would be

threatened by changing environmental conditions.

Future research should build on the limitations of the current study by investigating the effect of the moderating variables through other methods and samples. For example including non-publicly traded companies and using an industry or country specific sample. Also, a further investigation on the lagged nature of the relationship as well as the varying directions found in the previous literature would be useful. Furthermore, research should continue

investigating the effect of other possible moderators. Doing this would enhance the utilization of empirical findings at business level.

The remainder of the study unfolds as follows. Section one reviews the literature which includes the main theories on this topic as well as the major empirical findings from a market perspective and introduces the hypotheses. The second section presents the methodology. In section three the results are analyzed and subsequently discussed in section four. Finally, section five summarizes the results and limitations and provides suggestions for further research.

Literature review

The different findings of past studies, on how CSR and financial performance would be related to each other, very much correspond to the contradicting theories on this matter (El Ghoul et al., 2011). This section will first discuss these existing theories as well as recent empirical findings. Subsequently, the hypotheses will be introduced.

Theory

According to recent literature mainly four possible scenarios and even more theories exist on the relationship between CSR and financial performance (Salzmann et al., 2005; Makni et al., 2009). For the sake of clarity, an overview of these management theories is presented in the following table.

Table 1. Overview of the theories on the CSR – financial performance relationship

(based on Salzmann et al., 2005)

Causal sequence Direction of relationship Negative link Positive link

CSR leads to FP Trade-off theory Agency costs theory

Stakeholder theory Instrumental theory Good management theory FP leads to CSR Managerial opportunism theory

Slack resource theory

These management theories will now be discussed into more detail and form a basis for the construction of the hypotheses, later on reinforced with financial theory.

CSR causes firm value; a negative effect

According to neoclassicists like Friedman, the investment in CSR can be considered as a trade- off between possible benefits arising from serving stakeholders and the associated ‘ serving costs’ through the waste of resources used for the investment and reduced efficiency (Derwall, 2007; Stonski et al., 2014). These neoclassicists believe CSR investment is a net cost (costs> benefits) and is equivalent to giving up profits (Waddock and Graves, 1997; Jiao, 2010; Marsat and Williams, 2013). Consequently, investors would not promote these kinds of

initiatives (Schleifer, 2004; Stonski et al., 2014) as it may in fact reduce the value for shareholders, which contradicts the main objective of a firm from a financial point of view (Schleifer, 2004). Likewise, CSR can be considered an agency cost as managers may pursue such investments motivated by their personal liking for ‘doing good’ or their wish to upgrade their reputation amongst stakeholders (Flammer, 2013). This will eventually decrease firm value, through a degradation of stock by investors, when this is brought to light (Jiao, 2010; Cai et al., 2012).

CSR causes firm value; a positive effect

Alternatively, when evaluating CSR from a stakeholder point of view it could be considered a potential value enhancing investment (Salzmann et al., 2005; Flammer, 2013). As an

extension of the stakeholder theory, the so called instrumental theory, suggests that through serving stakeholders the value of the firm can be increased (Makni et al., 2009). Following this theory, CSR can be implemented as an instrument to enhance financial performance (Flammer, 2013). More specifically, CSR might give rise to direct and indirect benefits (Margolis et al., 2007). Directly, according to the good management theory, CSR can be treated as a resource “as it reflects the company’s ability to respond to long term trends and maintain competitive advantage” (Bassen and Kovacs, 2008, p184). For example, employee productivity might increase through enhanced employee satisfaction, new markets may be entered through the support of local businesses and innovation might improve through the implementation of ‘cleaner’ technologies (Waddock and Graves, 1997; Margolis et al., 2007; Anderson, 2014). Similarly, the companies may enjoy ‘risk insurance- like’ benefits through being able to efficiently adapt to changing conditions in the future, like increased

environmental taxes or stricter laws as well as through the avoidance of “environmental disasters, financial lawsuits and consumer boycotts” (Bassen and Kovacs, 2008; Marsat and Williams, 2013 p.4). Accordingly, these firms may now “offer a lower risk profile and enhanced return opportunities to their shareholders compared with competitors that do not adequately prepare for these developments” (Bassen and Kovacs, 2008, p.185). The indirect effect of CSR emerges through the increased value of intangibles such as corporate

reputation, brand equity and human capital (Waddock and Graves, 1997; Orlitzky et al. 2003; Jiao, 2010). This might increase the demand for stock and products and the supply of better employees; subsequently paving the way for increased competitive advantage (Jiao, 2010; Flammer, 2013). These investments in intangible assets enhance the value of the firm over the long term and may prevent short sightedness (Jiao, 2010).

Firm value causes CSR; a negative and positive effect

Although the majority of studies, as will be described in the next section, has focused on the ‘CSR leads to FP’ relationship (Flammer, 2013) theory suggests the existence of a reversed causality and even a virtuous or vicious circle (Waddock and Graves, 1997; Nelling and Webb, 2009; Makni et al., 2009; Jiao, 2010). Whilst the focus of this study is not the reversed causation, a brief overview is needed to fully understand this potential endogeneity problem when studying this relationship. It might be the case that managers decide whether to implement CSR based on their knowledge on how their firm was or would be valued (Jiao, 2010). Although managers may not know the exact firm value at the moment they determine the CSR actions, past firm values may provide comparable information when the firm value does not undergo major adjustments over time (Jiao, 2010). This either leads to a negative or positive reversed effect. The former can be explained by the so called managerial

opportunism hypotheses which expects investment in CSR following low firm values, as managers try to counterbalance these bad results that fall short the expectations, and less CSR following high firm values so that managers can “maximize their own short term private gains” (Makni et al., 2009 p.410). Conversely, according to the slack resource theory a firm would have ‘slack resources’ at hand following periods of great financial performance, this way enabling the firm to make CSR investments in the first place (Makni et al., 2009; El Ghoul et al., 2011)

Virtuous or vicious circle

According to some theorists, a vicious or virtuous circle might exist. Following the positive synergy hypotheses, based on the stakeholder and slack-resource theory, improved CSR performance would cause greater financial performance which in turn provides resources to make further investment in CSR possible (Makni et al., 2009). The negative synergy

hypothesis, based on the trade-off and managerial opportunism theory, predicts lower

financial performance following great CSR investment and subsequently more investment in CSR (Makni et al., 2009; Jiao, 2010).

Empirical findings

This review will keep in mind the market perspective of this study and thus focuses on this area of research only. Leading up to this overview, I will first discuss the results of studies that primarily focused on the direction of causality of the relationship.

Reversed causality

The theoretical overview proposed a potential reversed causality or even a virtuous/ vicious circle within the relationship of CSR and financial performance (FP). As a consequence of this possible reversed causality which remains problematic in studying the relationship (Salzmann et al., 2005) some studies tried to shed more light on this problem (Waddock and Graves, 1997; Nelling and Webb, 2009; Flammer, 2013; Anderson, 2014). Waddock and Graves (1997) tested and found support for the existence of a positive virtuous circle using a multiple OLS regression model. Nelling and Webb (2009) improved on this research by taking into account firm fixed effects to control for firm characteristics that might influence CSR engagement and find that the relation is not as strong as earlier studies suggested. Additionally, they further support these results by a (fixed effects) Granger Causality analysis. However, Granger causality is not equivalent to true causality, but in fact indicates whether some variable helps forecasting future values of some other variable (Flammer, 2013). Similarly, Anderson (2014) has addressed reversed causality by means of a Panel Vector Autoregressive Regression (PVAR) that he used to investigate the relationship between CSR, FP, Earnings Management, Corporate Governance and Management Compensation in both directions and found some evidence for a negative reversed effect between CSR and FP. Although their methodology controlled for firm and year-fixed effects and a PVAR treats all variables as endogenous, no causal interpretations can be made as an Autoregressive model, just like Granger Causality, is a model to forecast multiple variables by lagged values (Stock and Watson, 2010). Flammer (2013) argues to be the first to surpass the correlation analysis by using a discontinuity approach. He exploits exogenous variation in CSR in the form of ‘close-call shareholder proposals’ on CSR. This means that the effect of these proposals “that pass or fail by a small margin of votes” (Flammer, 2013 p.17) can be analyzed as if CSR was randomly assigned which allows the results to be interpreted as causal. Although the results suggest CSR to increase shareholder value, it is not entirely clear what percentage of passed proposals is in fact implemented and thus yields doubt on whether the effect is truly the result of the implemented CSR program.

A market perspective overview

A large body of research has studied the relationship between CSR and FP but according to Flammer, (2013) the majority of results may not be interpreted as causal. Although most of them point towards a positive association (Pava and Krauz, 1996; Makni et al., 2009; Cai et

al., 2012; Dixon – Flower et al., 2013; Anderson, 2014), the results remain mixed, hard to

compare and still not provide a clear outcome (Derwall, 2007; Cai et al., 2012). These different findings are often ascribed to different research methods, samples, CSR and FP proxies and the time period analyzed (Cai et al., 2012; Dixon – Flower et al., 2013)6. In this literature review, I will focus primarily on studies that used market based performance measures; stock returns, fund returns and firm value, as opposed to accounting based measures7, as these may be more helpful in interpreting the results of the current study.

Stock returns

First of all, a lot of event studies that investigate the immediate market reaction according to CSR related disclosure find a significant effect (Derwall, 2007). For example, the analysis of Klassen and McLaughlin (1996) illustrates that announced environmental performance awards were followed by positive abnormal returns whereas an environmental related failure of a firm caused negative returns. In addition, the former effect was found to be weaker than the latter effect. Griffin et al., (2012) concluded that companies’ market value lifts up after voluntary disclosure of greenhouse gas emission reductions. Event studies’ drawback is that the results “are valid only when the researcher has truly identified the abnormal returns associated with the event”, “the markets are efficient, the event was not anticipated and there are no confounding effects during the event window” (Curran, 2005 p.530).

Others studied differences in the long term between SRI and non SRI stock returns using regression methods. The majority of this type of research does not illustrate any significant distinctness between the returns of SRI and non SRI stock (Renneboog et al., 2008).

However, Brammer et al., (2005) observed that, in fact, CSR caused stock returns to decline and that “considerable abnormal returns are available when holding socially least desirable stocks”(Brammer et al., 2005 p.97). This is supported by Hong and Kacperzyk (2009) who

6 _{For example small firms, US based firms, and market based measures of FP are found to benefit to a greater}

extent from environmental practices (Dixon – Flower et al., 2012).

found that sin- stocks8 are often priced at less than then their true value and show greater returns compared to their non-sin counterparts. This can be theoretically motivated (and supported by a study of Galema et al., 2008) as SRI stocks experience excess demand (due to more complete information and investors’ social preferences) which causes overpricing of these stocks compared to underpricing of non SRI/ sin stocks. Also, due to this shortage of people willing to invest in this non SRI stock the risk can only be shared among few

stockholders who thus need to be compensated with greater return premiums (Galema et al., 2008). Contrary, the results of Kempf and Osthoff (2007) suggests higher return for SRI stocks.

Fund returns

Some studied this relationship on a portfolio level by comparing the performance of SRI mutual funds to conventional funds. Some evidence exist for the efficient market theory which predicts investors to disadvantage of removing sin (non-SRI) -stocks as this reduces portfolio efficiency and risk diversification (Brammer et al.,, 2005). However, Hamilton et al., (1993) suggests that there is no difference in the returns of these funds. Renneboog et al (2008; p. 1) concludes that “existing studies hint but do not unequivocally demonstrate that SRI investors are willing to accept suboptimal financial performance to pursue social or ethical objectives and derive non-financial utility from investing in SRI funds”.

Firm value

Another stream of research, to which the current study belongs, exists on the link between CSR and firm value. For this reason, this section provides a more critical review of previous studies than the former discussed areas of research.

Flammer (2003) found that firm value improved immediately in the year in which the CSR9 proposal was accepted, which seems to be the result of a reaction by the stock market close to that date. As the firm value did not change much in the years following, Flammer interprets this as being consistent with the argument that a firm may benefit from CSR in the long-term. Similarly, the OLS results of Dowell (2000) suggest that higher firm values are realized in case a firm chooses to meet stricter environmental norms instead of reverting to more flexible norms. Although both of these studies create some sort of exogenous shock the study of

8 _{Sin stocks are “stocks that are excluded from a portfolio because of negative ethical issues (such as alcohol,} tobacco and gaming)” (Galema et al., 2008 p2647)

Flammer, using a discontinuity approach, is superior in allowing a causal interpretation. The ‘shock’ studied by Dowell may in fact be correlated with unobservable firm-level variables such as the quality of management or the strategy of the firm for which they do not control. Also, their ‘shock’ may even itself be a consequence of current financial performance and may not take properly into account reversed causality. Dowell (2000) tried to overcome this problem by running a regression on the residuals and lagged residual values from two

regressions: one of Tobin’s Q and another of an environmental standard variable regressed on the same set of control variables which then should be composed of the part of Tobin’s Q and the environmental standard that is not driven by the controls. However, the results are not significant, which Dowell (2000) ascribed to the possibility of too little shocks or that the market responds to such a change within a one year period.

The results of a time series analysis10 by Derwall (2007) illustrates that eco-efficiency, defined as “the ability to create more value while using fewer environmental resources” (Derwall, 2007 p.10), is positively associated with firm value. They tested for robustness by regressing on a varying dependent variable; Tobin’s Q, industry- adjusted Tobin’s Q and trimmed Tobin’s Q, this way reducing difficulties that may arise in case the variable does not have a normal distribution. However, they did not control for any fixed effects and did not discuss the possibility of reversed causation in their model.

Marsat and Williams (2011) and Stonski et al., (2014) are closest to the current study in a way that they both studied this relationship using ESG rating (provided by KLD and ASSET4 respectively) as a proxy for CSR performance. Marsat and Williams (2011) used cross

section data for 2005-2009. They controlled for years, industry and region fixed effects through the inclusion of dummy variables. Contrary to previous discussed research, the results from their multiple OLS analysis illustrates that “a one-step increase in ESG rating decreases ceteris paribus Tobin’s Q by 0.03” (Marsat and Williams, 2011 p.10). They conclude that the potential benefits of CSR may not be valued by the market or that the market is not aware of the true value of CSR (Marsat and Williams, 2011). They checked for robustness through the inclusion of extra controls as well as regressing their model using the individual ESG scores instead. Stonski et al., (2014) used a very similar OLS regression (for 2009-2011) but studied the ESG pillars independently. They found that the environmental

10 _{They conducted this time series regression at five specific dates: December 1996, 1998, 2000, 2002 and}

performance negatively affects Tobin’s Q whereas the social and Corporate Governance pillar increase firm value. Although they both address the omitted variable concern by including fixed industry, region and year fixed effects, they do not control for firm-level fixed effects which may be important as these unobserved firm level characteristics are found to alter results (Nelling and Webb, 2009). Moreover, they do not discuss the possibility of reversed causality in their models while making conclusions about ESG impacting firm value.

Jiao (2010) studied the impact of stakeholder welfare, which they proxy by the ESG score provided by KLD, on Tobin’s Q by an Instrumental Variable regression which is often used to overcome endogeneity problems of omitted variable bias and reversed causality and suggest a positive effect of the score on firm value. Although a drawback of this method is that the instruments used need to be both relevant to the independent and exogenous to the dependent variable, I believe their choice of instruments (past positive earnings and equity ownership by activist public pension funds) are valid.

Finally, Cai et al., (2012) studied the CSR- Tobin’s Q relationship for firms in controversial industries by using a system of equations (3sls), which can be used to analyze a relationship in which there are two variables that simultaneously determine each other within the system, and found a positive effect. Such a system of equations is an extension of the Instrumental Variable regression. As Cai et al., (2012) claims, this way they properly take into account endogeneity problems that would arise when a specific type of firms (with greater resources, greater public interest and operating in a certain industry) may implement CSR more often than others. A limitation of this method is that it does not allow controlling for any

unobservable variable by firm level fixed effects (Cai et al., 2012). Furthermore, despite the fact that in such a system both equations need to be predetermined by exogenous instruments (Cheng et al., 2013) they do not specify any instruments to predict values of Tobin’s Q. Also, the exogenous validity of their instruments; operation cash flow to assets (to indicate the resources available), number of analysts following a firm (to proxy public attention) and finally industry (as CSR is found to differ among industries) can be called into question. The first one is similar to Tobin’s Q in a sense that Tobin’s Q incorporates the discounted future cash flows. Also, it is found that the number of analysts improves firm valuation and that more analysts tend to follow firms that are highly valued by investors (Lang et al., 2004). The final instrument is an industry dummy, which is in fact often included when investigating the effect of CSR on firm value (Marsat and Williams, 2013; Stonski et al., 2014).

Hypotheses

Like the existent literature points to a positive relationship between CSR and financial performance this study predicts a similar positive relationship between the current proxies; ESG performance and Tobin’s Q. Our prediction is based on the previously described potential direct and indirect beneficial effects of ESG performance according to the

stakeholder theory. As firm value (Tobin’s Q) is assumed to be “an unbiased estimate of the present value of its future cash flows” and thus “should include any adjustments the market has made to incorporate the expected valuation effects associated” with CSR (Derwall, 2007 p.2553) it is important to understand the underlying mechanisms explaining how these direct and indirect effects of ESG performance translate into higher firm value.

ESG performance can affect firm value through three mechanisms; ‘market process effects, cash flow effects and investor discount rate11 effects’ (Richardson, 1999). The first

mechanism implies that any extra information that is pertinent to the valuation process, like ESG extra financial information, would decrease investors’ uncertainty about future business performance and would attract more analysts. This would subsequently increase the demand for stock as well as decrease the discount rate through a lower level of firm specific risk (Richardson, 1999; Jiao, 2011). This effect does not take into account whether ESG would benefit or damage the firm (mechanism two) and investors’ social preferences (mechanism three). The second mechanism of cash flows effects takes into account the “net present value (NPV) evaluations of socially responsible projects, expected future regulatory costs and product market effects” (Richardson, 1999, p.18). ESG performance, from a stakeholder perspective, results in the earlier described benefits of for example improved productivity, cost reductions through risk-insurance like benefits and increased value of intangibles resulting in increased demand for products which all increase present value of future cash flows 12. In addition, in case the market prices CSR, the cost of capital will be influenced (El Ghoul et al., 2011). Investors may demand a lower return on equity capital because of lower perceived, undiversifiable risk according to a more stable future perspective and so driving up the value of the firm (Margolis et al., 2007; Soana, 2011). This is supported by the study

11 _{Discount rate = “cost of equity capital, the internal rate of return that the market applies to a firm’s future} cash flows to determine its current market value” (El Ghoul et al., 2011 p 2389).

12 _{Monetary CSR Value Added = ∑ 1}𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵 𝑛𝑛𝐶𝐶𝐶𝐶𝐶𝐶− 𝐶𝐶𝐶𝐶𝐵𝐵𝐵𝐵𝐵𝐵 𝑛𝑛𝐶𝐶𝐶𝐶𝐶𝐶

(1+𝐵𝐵)𝑛𝑛

∞

𝐵𝐵 in which i is the cost of capital/ discount factor

of El Ghoul et al.,(2011) who concluded that a better ESG performance results in a lower cost of equity capital. The third mechanism implies that some investors have social

preferences for responsible firms (Richardson, 1999). This social consciousness may result in increased demand for their stock followed by overvaluation of these firms as opposed to firms not investing in ESG (Marsat and Williams, 2013). Consequently, the risk of the stock may decrease through a growing number of investors and hence greater liquidity (El Ghoul et

al., 2011; Marsat and Williams, 2003). According to Heinkel’s theory (Heinkel et al., 2001),

the cost of capital, then, of firms investing in ESG will be lower than those not investing in ESG in case enough social conscious investors choose not to invest in the latter type of firms (and thus risk cannot be easily shared) (El Ghoul et al., 2011). According to this reasoning I will test the following hypothesis:

H1 ESG performance positively affects firm value (Tobin’s Q)

Although a confirmation of this first hypothesis would suggest that ESG performance is valued by the market, this relationship might be moderated by specific firm characteristics.

First of all, an increase in ESG performance may affect firm value differently for firms exhibiting very low levels of ESG compared to firms exhibiting very high levels of ESG. This is sometimes referred to as reaping ‘low hanging fruits’ of CSR, which means that "initial efforts to increase ESG performance may yield substantial operational benefits" (Flammer, 2013 p2). The results of Flammer (2013) illustrate such a diminishing effect of the accepted close call proposals on firm value in the US. In addition, the market may value the benefits and costs of the ESG improvement incorrectly (Marsat and Williams, 2003). This is supported by the study of Manescu (2011) who found evidence for the mispricing

hypothesis meaning that investors may not be able to price CSR efficiently due to a lack and intangibility of information. This is in line with the results of Klassen and McLaughlin (1996) who found that the positive stock market reaction is greater the first time a firm receives an award for great environmental performance than the following awards. Hence, a

disequilibrium might exist (Derwall, 2007) which raises the possibility that investors may experience increases in ESG performance from firms with lower initial scores as more valuable than those from firms with initial high scores. This is tested by the following hypothesis:

H2 For firms with low initial ESG scores the positive relationship between ESG performance and Firm value (Tobin’s Q) is more pronounced than for firms with high initial ESG scores.

In addition, the sector in which the firm operates may moderate the relationship as well. Firms that operate in sectors that are commonly known for their operations that harm the environment may be coerced to a higher extent and potentially by a larger number of

stakeholders to do something about their environmental performance. Additionally, they may receive more attention from the media “resulting in the potential for greater gains in

organizational legitimacy through better environmental performance” (Reverte, 2011; Dixon-Flower et al., 2013 p357). Also, as the risks and impact of otherwise bad environmental performance are greater than in non-environmentally sensitive industries, a good

environmental performance would be more valuable to stakeholders and investors. Although the results of a meta-analysis by Dixon –Flower et al., (2013) suggests no difference in the results of studies using samples from industries that harm the environment and samples covering all industries, the study by Reverte (2011) on Spanish firms suggests that the environmental sensitiveness is in fact an important moderator. He found a greater negative effect of CSR disclosure quality on the cost of equity capital for those firms. The result of Dixon- Flower et al., (2013) might have to do with a major limitation of meta – analysis as they rely on the methods used by other studies or that they compare ‘bad to general’ and not ‘bad to not bad’ industries. To verify whether Environmental performance has a greater positive relationship with firm value in case of environmentally sensitive industries we test the following hypothesis:

H3 For firms that operate in environmentally sensitive industries the positive relationship between ESG performance and Firm value (Tobin’s Q) is more pronounced than for firms operating in non-environmentally sensitive industries. Note: Instead of ESG performance, this hypothesis uses Environmental performance as the dependent variable as this is more directly related to the moderating variable.

Contribution to literature

According to the previous overview multiple methods have been used to investigate the single relationship between CSR and firm value. All these methods have their advantages and drawbacks and there is, to my best knowledge, not some ideal model left out there to apply to this situation which is applicable within the scope and time frame of this study. Therefore this study aims to improve on these individual methods in ways that are later on discussed in the methodology section. Subsequently, combining and acknowledging the specific

boundaries of the different methods allowed for critical analysis of their outcomes. This let me identify whether these methods provide consistent results within the same dataset and within one study. Also as a follow up of this quantitative analysis three interviews have been carried out with sustainability managers of firms within this data set (PostNL, Ahold and Heineken) through which I gained insight in real-life evaluations of the relationship. Also, as stated by Margolis et al., (2007) future research should test conditions that may moderate the relationship. This interest in moderating factors (Dixon-Flower et al., 2013) is met through building on research of Derwall (2007), Reverte (2011) and Flammer (2013) by investigating whether the relationship varies by the initial ESG performance level (Low vs. High) and by the environmental sensitivity of the sector in which the firm operates. Finally I use an objective database on ESG performance provided by ASSET4 from Thomson Reuters with data on a long time period (2005-2013) and European firms only. To my best knowledge this region has not been analyzed before in this specific context. This database enables an

investigation on the effect of country characteristics on this relationship, which would be interesting as CSR engagement tend to vary among country (Doh and Guay, 2005;

Renneboog et al., 2008; Lapinskienė and Tvaronavičienė 2012; Veenstra and Shi, 2014) and there exists a lack of studies taking into account cross-country effects (Salzmann et al., 2005).

Methodology

This section provides an overview of the methodology used in the current study. First the data is introduced and the variables of interest are discussed. Then the main methodological issues of previous research are illustrated after which the current method and equations are explained.

Introduction to the data

The ESG performance data used in this research as a proxy for CSR performance is taken from the ASSET4 database provided by Thomson Reuters. ASSET4 is a company that sells objective and easily comparable environmental, social and governance information on firms to (institutional) investors and corporations. This information is provided as a rating (between 0-100%) per company which is an equally weighted assessment of a company’s performance benchmarked against all the other firms in the dataset (called the ASSET4 universe) and based on approximately 280 performance indicators13. These KPI’s can be grouped in 18 categories and subsequently four main pillars; Environmental , Social, Corporate Governance and Economic, see Appendix 1. As can be noticed, the Economic pillar is left out of this study as the main interest is the E,S and G score. The information needed for assessment, which is executed by over 100 analysts, is derived from publicly available sources only (like annual reports, NGO websites and CSR reports, Carbon Disclosure project data)14. Although these sources are often disclosed by the firms itself, ASSET4 critically and sometimes more strictly assesses this information.

The ASSET4 universe exists of more than 4000 publicly traded companies. Although lots of companies report on ESG information nowadays, this is not standardized and therefore not comparable information. The comparability of the score provided by ASSET4 is therefore a major advantage of the data.

The current study utilizes the ASSET4 Europe –subset database which provides information on companies based in 20 European countries. The number of companies rated by ASSET4 has expanded from 353 in 2002, in which they started providing the scores, to approximately 950 companies in 2014. The final sample of this study, after the deletion of outliers and

13

Thomson Reuters ASSET4 data information

reducing the number of missing data points, is a panel data set with annual ESG and financial information from year 2005 to 2013 on 428 firms. For these companies at least 2 years of information is available.The financial data on firm value and the control variables are taken from DataStream.

Variables

Dependent

Tobin’s Q- Tobin’s Q is a ratio that has often been used in the literature (specifically finance and accounting) as a proxy for firm value (Chung and Pruitt, 1994):

Q= 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝐵𝐵𝐵𝐵 𝑣𝑣𝑀𝑀𝑣𝑣𝑣𝑣𝐵𝐵 𝐶𝐶𝐵𝐵 𝑀𝑀 𝐵𝐵𝐵𝐵𝑀𝑀𝑓𝑓 𝑀𝑀𝐵𝐵𝑟𝑟𝑣𝑣𝑀𝑀𝑟𝑟𝐵𝐵𝑓𝑓𝐵𝐵𝐵𝐵𝐵𝐵 𝑟𝑟𝐶𝐶𝐵𝐵𝐵𝐵𝐵𝐵 𝐶𝐶𝐵𝐵 𝑀𝑀𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵.

However, as this ratio is very complex and contingent on a lot of variables, a more

convenient estimate can be used instead (Gompers et al., 2001), called the market-to-book-ratio, which is used in this study =

𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝐵𝐵𝐵𝐵 𝑣𝑣𝑀𝑀𝑣𝑣𝑣𝑣𝐵𝐵 𝐶𝐶𝐵𝐵 𝑀𝑀𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵 𝑏𝑏𝐶𝐶𝐶𝐶𝑀𝑀 𝑣𝑣𝑀𝑀𝑣𝑣𝑣𝑣𝐵𝐵 𝐶𝐶𝐵𝐵 𝑀𝑀𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵 =

𝑏𝑏𝐶𝐶𝐶𝐶𝑀𝑀 𝑣𝑣𝑀𝑀𝑣𝑣𝑣𝑣𝐵𝐵 𝐶𝐶𝐵𝐵 𝑀𝑀𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵 +𝑓𝑓𝑀𝑀𝐵𝐵 𝑣𝑣𝑀𝑀𝑣𝑣𝑣𝑣𝐵𝐵 𝐶𝐶𝐵𝐵 𝑟𝑟𝐶𝐶𝑓𝑓𝑓𝑓𝐶𝐶𝐵𝐵 𝐵𝐵𝑒𝑒𝑣𝑣𝐵𝐵𝐵𝐵𝑒𝑒−𝑏𝑏𝐶𝐶𝐶𝐶𝑀𝑀 𝑣𝑣𝑀𝑀𝑣𝑣𝑣𝑣𝐵𝐵 𝐶𝐶𝐵𝐵 𝑟𝑟𝐶𝐶𝑓𝑓𝑓𝑓𝐶𝐶𝐵𝐵 𝐵𝐵𝑒𝑒𝑣𝑣𝐵𝐵𝐵𝐵𝑒𝑒 𝑏𝑏𝐶𝐶𝐶𝐶𝑀𝑀 𝑣𝑣𝑀𝑀𝑣𝑣𝑣𝑣𝐵𝐵 𝐶𝐶𝐵𝐵 𝑀𝑀𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵𝐵

Independent

ESG – The major independent variable in the current study is the overall ESG score, except for hypothesis 3 where only the individual Environmental score is used. As no theory exists on a specific way to calculate an overall (perhaps weighted) ESG score from the individual scores, as provided by ASSET4, I calculated this score by taking the average of the three sub scores; Environmental score, Social score and Corporate Governance score. Although the overall ESG score is the main interest of this study, the three sub scores are used individually as a robustness check.

Controls

Thanks to panel data, there will be controlled for unobserved year and firm fixed effects. The firm fixed effects control for factors such as corporate culture and management influence that differ between firms but remain constant over time, whereas the year fixed effects holds constant changes in for example regulation and environmental situation that are similar across firms but change over time. Additionally, some observable variables that are likely to

impact our dependent and independent variable will be included in the model. All these controls are important to reduce the omitted variable bias problem(Stock and Watson, 2010) .

Most commonly used control variables are Size, Risk and industry fixed effects. Other variables that have been controlled for are Sales growth, Capital expenditure (Capex), Return on Equity (ROE), Return on Assets (ROA), Firm age and Research and Development

intensity (R&D intensity). Some authors motivated their choice of control variables by their common appearance in previous literature only (Waddock and Graves, 1997;Cai et al., 2012; Makni et al., 2009; Stonski et al., 2014). I believe a more critical evaluation is necessary to decide on which controls to include. To prevent the estimates to be a spurious correlation between ESG performance and Tobin’s Q, there should be controlled for those factors that affect Tobin’s Q and ESG. Importantly, these variables should not be itself anoutcome of ESG performance.

From our correlation matrix (see appendix 2) we see that Risk, Size, Sales Growth and Return on Assets are significantly correlated to both Tobin’s Q and ESG whereas R&D intensity, Firm age, Capex and ROE are only significantly correlated with one of the two.

Despite the fact that Sales Growth is found to positively affect firm valuation through signaling future growth (Cai et al., 2012), it will not be controlled for in the current analysis. Because it is not found to determine ESG performance (Marsat and Williams, 2013; Cai et al 2012), the correlation shown in the matrix between ESG and Sales Growth might be the result of a reversed effect. This mechanism is touched upon in the literature review; that ESG may increase demand for products through the increased value of intangibles such as

reputation (Marsat and Williams, 2013). This suggests that Sales Growth may itself be an outcome of ESG. Also, this study will not control for ROA and ROE as these measures of firm profitability are found to be an outcome of ESG(Waddock and Graves, 1997; Nelling and Webb, 2009). Firm age and Capex will also not be controlled for in this analysis. These control variables have only been used once by Derwall (2007) and Jiao (2010) respectively. As, to my best knowledge, no clear theory exists on how these controls would affect my variables of interest I do not include them in my analysis.

There will be controlled for Risk, Size, and R&D intensity based on the following theoretical arguments. Risk, measured as long-term debt divided by total assets, is a ratio that illustrates the risk of default by not being able to make interest payments. This is found to be negatively

related to Tobin’s Q (Dowell, 2000; Nelling and Webb, 2009) and may negatively affect ESG performance. That is, the level of risk toleration of the firm’s management may affect their perceived value of ESG as these possibly create risk-insurance like benefits as explained before (Waddock and Graves, 1997). In order to avoid confusion, these risk- insurance like benefits are not equivalent to the risk of default. The control variable Risk is therefore not an outcome of ESG performance. Firm size controls for the fact that large firms may have more resources at their disposal to make ESG investing possible and for the fact that larger firms may be subject to a more demanding environment when it comes to CSR (Waddock and Graves, 1997; Dowell, 2000; Jiao, 2010). Also, Size is included as it is found to influence Tobin’s Q. In fact, Dowell (2000) found that Size decreases firm valuation. R&D intensity tends to have a positive effect on Tobin’s Q which may be caused through signaling

increased innovation and greater opportunities in the future (Dowell, 2000; Derwall, 2007). Although we do not find a significant correlation between R&D intensity and ESG it has been suggested that firms who spend more on R&D may heavily depend on their reputation and human capital, and hence may decide to increasingly invest in CSR (Jiao, 2010).

Finally, I control for industry and country effects either through firm-level fixed effects or through the inclusion of industry and country dummies. This is important as the valuation process by investors is based on their beliefs about the future growth of a certain industry, its level of competition and industry-specific shocks (Waddock and Graves, 1997; Dowell, 2000). Moreover, previous research found that the level of CSR performance and its relationship with financial performance is affected by industry and country characteristics (Doh and Guay, 2005; Renneboog et al., 2008; Ioannou and Serafeim, 2012; Veenstra and Shi, 2014).

Validity issues

Internal validity

As can be concluded from the empirical findings on the relationship between CSR and Tobin’s Q is overcoming the possible endogeneity problems is a major challenge. The two most commonly acknowledged issues of endogeneity are the ones that emerge from omitted unobserved variables and reversed causality (Jiao,2010). Although different methods have been used to overcome these problems, still limitations exist as discussed in earlier analysis. This study acknowledges the fact that it would be difficult if not impossible to find the perfect solution. Therefore it will improve and combine several individual methods used in

previous research. Our methods will include an Ordinary Least Squares (OLS) analysis in which will be controlled for firm fixed effects as an improvement on the industry dummies used by Stonski et al., (2014) and Marsat and Williams( 2013), a Granger Causality including more lags than the one-year lag used by Dowell (2000), Nelling and Webb (2009) and Makni et al., (2009) and an Instrumental Variable (IV) regression for which the instruments

previously used by for example Cai et al (2012) and Jiao (2010) are critically assessed on their validity. This IV will also be extended through the analysis of a one – year lagged value of ESG performance that, to my best knowledge, has not been included before in an IV regression in this context. Studying this lagged relationship would be worthwhile as it is still not clear how long it takes to experience any benefits (if at all) of CSR investments (Dixon-Flower et al., 2013).

Although the OLS will address the problem of omitted variables it won’t allow a fully causal interpretation of results. The Granger Causality (including fixed effects as well) is interesting to say something about the ability of ESG to forecast future values of the firm. However, this method measures predictability rather than causality. Although IV is often used to address both reversed causality and unobserved variables, this study acknowledges the fact that it is hard to find a good instrument that is both relevant and exogenous. However, in case the results of these three analyses are reflecting a consistent message this would enhance the validity of the results. On the other hand, if no consistent message arises from the same dataset, it will still be interesting to analyze these critically.

External validity

In addition to these endogeneity problems the samples most commonly used, include firms based in the US or the entire world. Surprisingly, no study controls for state/ home country of these firms, except for studies conducted on a single country (Makni et al., 2009; Galbreath, 2010) or using firm level fixed effects regressions (Nelling and Webb, 2009). Although some control for continent it is found that differences on country level like regulation, culture and macroeconomic trends influence the level of CSR performance (Doh and Guay, 2005; Renneboog et al., 2008). For example, Veenstra and Shi (2014) conducted a study on the moderating effect of cultural values (from Hofstede’s cultural model) and found that social performance would decrease financial performance in very individualistic and short term oriented countries whereas in countries that are more focused on the long term and

collectiveness would benefit from increased social performance. By narrowing the scope of the study to Europe only and control for home country through fixed effects makes the results

more precise and decreases heterogeneity within the sample (Orlitzky et al., 2003), and in turn may increase external validity. It would also be interesting to analyze the effect of the in- or exclusion of country dummies in the analysis.

Also, through analyzing an extensive time period (2005-2013) instead of a short time frame like Waddock and Graves (1997) and Makni et al., (2009) the potential risk of temporal errors influencing the data is mitigated.

Measurement validity

“The validity of a measure is the degree to which any measurement approach or instrument succeeds in describing or quantifying what it is designed to measure.” (Gethin et al., 2015 p212). As Margolis et al., (2007) conclude that the effect of proxies like revealed

irresponsible actions, observers’ perceptions, charity donations and self-disclosed information are larger than for more objective proxies like ratings of independent companies and fund performance, suggests that these former measures pick up appearance effects of the specific CSR proxy (Margolis et al,. 2007). They suggest that future studies should proxy CSR with objective measures like the rating provided by independent companies as their evaluation process is transparent, can be verified and hence will be most objective.

Method

Hypothesis 1

This study utilizes three different methods in order to analyze the first hypotheses; fixed effects OLS, fixed effects Granger Causality and Instrumental variable analysis. For an overview of the variables used in the following section, see Appendix 3.

Fixed effects OLS

Firstly, the relation between ESG performance and Tobin’s’ Q is analyzed through OLS. This is a common method in this field of research (Waddock and Graves, 1997; Derwall, 2007; Nelling and Webb, 2009; Marsat and Williams, 2013; Stonski et al., 2014). As said before, the advantage of panel data is that fixed effects can be included. Although I am aware of the fact it has been suggested that it probably takes time to reap the benefits of ESG performance, I do not (like some others did) test lagged values of the dependent variable in this OLS

method because of the possibility of reversed causality (Waddock and Graves, 1997; Nelling and Webb, 2009; Anderson, 2014).

The following equation is regressed to test whether ESG is positively related to Tobin’s Q:

𝑄𝑄𝐵𝐵𝐵𝐵 =

𝛼𝛼𝐵𝐵+ 𝛽𝛽1𝐸𝐸𝐸𝐸𝐸𝐸𝐵𝐵𝐵𝐵+ 𝛽𝛽2𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑠𝑠𝐹𝐹𝑠𝑠𝑠𝑠𝐵𝐵𝐵𝐵+ 𝛽𝛽3𝑅𝑅&𝐷𝐷𝐹𝐹𝑛𝑛𝑡𝑡𝑠𝑠𝑛𝑛𝑠𝑠𝐹𝐹𝑡𝑡𝑡𝑡𝐵𝐵𝐵𝐵+ 𝛽𝛽4𝑅𝑅𝐹𝐹𝑠𝑠𝑅𝑅𝐵𝐵𝐵𝐵 + 𝛼𝛼𝐵𝐵+ ∑8𝐵𝐵=1𝑌𝑌𝑠𝑠𝑌𝑌𝐹𝐹𝐵𝐵𝐵𝐵+ 𝜀𝜀𝐵𝐵 (1)

Where the firm value for a firm i in year t (Qit) is a function of ESG, the overall score the company received on ESG performance; Firm size, the firms log of Total Assets; R&D intensity, the ratio of a company’s Research and Development investment to its total sales; Risk; the leverage ratio of a company’s long term debt to assets; α, firm fixed effects; Year dummies for the year fixed effects and a robust error term, ε.

This analysis is extended with Granger Causality and Instrumental Variable regression to get more insight in the one-way causality which is of particularly interest to this study.

Granger Causality

Next, we will use a fixed effects Granger Causality model to test whether ESG performance ‘Granger causes’ Tobin’s Q. This approach, involves a regression model of the form: 𝑄𝑄_{𝐵𝐵𝐵𝐵} = 𝛼𝛼_𝐵𝐵 + 𝛽𝛽₁𝑄𝑄_{𝐵𝐵,𝐵𝐵−1}+ 𝛽𝛽₂𝑄𝑄_{𝐵𝐵,𝐵𝐵−2}+ 𝛽𝛽₃𝐸𝐸𝐸𝐸𝐸𝐸_{𝐵𝐵,𝐵𝐵−1}+ 𝛽𝛽₄𝐸𝐸𝐸𝐸𝐸𝐸_{𝐵𝐵,𝐵𝐵−2}+ 𝛽𝛽₅𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑠𝑠𝐹𝐹𝑠𝑠𝑠𝑠_{𝐵𝐵𝐵𝐵}+ 𝛽𝛽6𝑅𝑅&𝐷𝐷𝐹𝐹𝑛𝑛𝑡𝑡𝑠𝑠𝑛𝑛𝑠𝑠𝐹𝐹𝑡𝑡𝑡𝑡𝐵𝐵𝐵𝐵+ 𝛽𝛽7𝑅𝑅𝐹𝐹𝑠𝑠𝑅𝑅𝐵𝐵𝐵𝐵+ 𝛼𝛼𝐵𝐵+ 𝜖𝜖𝐵𝐵 (2)

Where the firm i in year t (Qit) is a function of two lagged values of Tobin’s Q and two lagged values of ESG performance, Firm size, the firms log of Total Assets; R&D intensity, the ratio of a company’s Research and Development investment to its total sales; Risk, the leverage ratio of total debt to total assets; α, firm fixed effects and a robust error term, ε. This regression is extended through the inclusion of three year and four year lagged values of Tobin’s Q and ESG as well.

Two stage least square

The Instrumental Variable method focuses on finding a variable (or variables) that influences ESG performance, but does not influence Tobin’s Q (and thus is not correlated with the random error term in the second stage equation). We estimate the following Two-stage-least-square (2sls) regressions:

𝑄𝑄_{𝐵𝐵𝐵𝐵} = 𝛼𝛼_𝐵𝐵 + 𝛽𝛽₁𝐸𝐸𝐸𝐸𝐸𝐸_{𝐵𝐵,𝐵𝐵} + 𝛽𝛽₂𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑠𝑠𝐹𝐹𝑠𝑠𝑠𝑠_{𝐵𝐵𝐵𝐵}+ 𝛽𝛽₃𝑅𝑅&𝐷𝐷𝐹𝐹𝑛𝑛𝑡𝑡𝑠𝑠𝑛𝑛𝑠𝑠𝐹𝐹𝑡𝑡𝑡𝑡_{𝐵𝐵𝐵𝐵} + 𝛽𝛽₄𝑅𝑅𝐹𝐹𝑠𝑠𝑅𝑅_{𝐵𝐵𝐵𝐵}+ 𝛼𝛼_𝐵𝐵 + ∑8𝐵𝐵=1𝑌𝑌𝑠𝑠𝑌𝑌𝐹𝐹𝐵𝐵𝐵𝐵+ 𝜖𝜖𝐵𝐵 (3)

𝐸𝐸𝐸𝐸𝐸𝐸𝐵𝐵,𝐵𝐵 = 𝛼𝛼𝐵𝐵+ 𝛽𝛽1𝑍𝑍 + 𝛽𝛽2𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑠𝑠𝐹𝐹𝑠𝑠𝑠𝑠𝐵𝐵𝐵𝐵+ 𝛽𝛽3𝑅𝑅&𝐷𝐷𝐹𝐹𝑛𝑛𝑡𝑡𝑠𝑠𝑛𝑛𝑠𝑠𝐹𝐹𝑡𝑡𝑡𝑡𝐵𝐵𝐵𝐵+ 𝛽𝛽4𝑅𝑅𝐹𝐹𝑠𝑠𝑅𝑅𝐵𝐵𝐵𝐵+ 𝛼𝛼𝐵𝐵 + ∑8𝐵𝐵=1𝑌𝑌𝑠𝑠𝑌𝑌𝐹𝐹𝐵𝐵𝐵𝐵+ 𝜖𝜖𝐵𝐵 (4)

Where (for equation 3) the firm i in year t (Qit) is a function of the predicted values of ESG performance from the first stage, Firm size, the firms log of Total Assets; R&D intensity, the ratio of a company’s Research and Development investment to its total sales; Risk, the leverage ratio of total debt to total assets; α, firm fixed effects; Year dummies for the year fixed effects and a robust error term, ε. Equation 4 is similar except that now ESG is a function of the instrument Z. This IV model is extended by the analysis of a first stage in which ESG is a function of the lagged value of the instrument Z.

Important for the choice of instrument is that it is both relevant (corr Z, X≠ 0) and exogenous (corr Z,u = 0) (Stock and Watson, 2010). Prior research has used the following instruments for CSR performance; average industry-mean performance (Cai et al., 2012), average CSR score for each country-sector subset and country – year subset excluding the CSR

performance of the focal firm (Cheng et al., 2014), lagged values of CSR performance (Schrek, 2011), Firm age (Jo and Harjoto, 2011), past positive earnings and level of

shareholder activism (Jioa, 2010) and finally a combination of operation cash flow to assets, number of analysts following and industry dummies (Cai et al., 2012). A prior discussion on the latter instruments illustrated doubt on their exogenous character. Furthermore, firm age and lagged values of CSR are found to be directly related to Tobin’s Q (Derwall, 2007; Nelling and Webb, 2009; Makni et al., 2009; Anderson, 2014). Whilst the shareholder activism instrument of Jiao seems to be valid, the database of the current study does not provide such information. Validity for the first three instruments is met through the idea that a firm’s decision on how much to invest in CSR is affected by the level and activities of CSR shown by other firms within their own industry and the CSR level within their home country that might both change over time (Dowell, 2000; Iannou and Serafeim, 2012). Also, to my best knowledge, the average ESG performance of such a subset is not likely to impact Tobin’s Q directly. The past positive earnings instrument I believe is valid as it has been experienced that a firm’s decision to invest in CSR investment depends on their previous business results, being favorable or not (Jiao, 2010). Moreover, as this would be measured as a dummy variable (1 if last year’s earnings were positive and 0 otherwise) it will not

represent information about the degree of cash flows thereafter; thus not expected to be a determinant of Tobin’s Q directly. For example, a firm which is expected to grow a lot in the future may experience a high Tobin’s Q irrespective of his business results being positive or

negative the prior year. The relevance of these four instruments is tested for before making any conclusions on the results.

Hypotheses 2 and 3

Although I test the first hypothesis using three different methods, the second and third hypothesis are only analyzed by OLS due to time constraints and the fact that the first hypothesis is of main interest to this study.

I will test the second hypothesis, whether firms with initial low ESG scores experience a greater positive relationship between their ESG performance and firm value than those with initial high ESG scores, with a similar OLS regression as the first hypotheses. First, the coefficients for firms exhibiting initial low levels of ESG (scores lower than 25) and for firms exhibiting initial high values of ESG (scores higher than 75)15 are estimated by the following equation:

𝑄𝑄𝐵𝐵𝐵𝐵 = 𝛼𝛼𝐵𝐵+ 𝛽𝛽1𝐸𝐸𝐸𝐸𝐸𝐸𝐵𝐵,𝐵𝐵+ 𝛽𝛽3𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑠𝑠𝐹𝐹𝑠𝑠𝑠𝑠𝐵𝐵𝐵𝐵+ 𝛽𝛽4𝑅𝑅&𝐷𝐷𝐹𝐹𝑛𝑛𝑡𝑡𝑠𝑠𝑛𝑛𝑠𝑠𝐹𝐹𝑡𝑡𝑡𝑡𝐵𝐵𝐵𝐵+ 𝛽𝛽5𝑅𝑅𝐹𝐹𝑠𝑠𝑅𝑅 + 𝛼𝛼𝐵𝐵+ ∑8𝐵𝐵=1𝑌𝑌𝑠𝑠𝑌𝑌𝐹𝐹𝐵𝐵𝐵𝐵+ 𝜀𝜀𝐵𝐵 (5)

Where the firm value for a firm i in year t (𝑄𝑄_{𝐵𝐵𝐵𝐵}) is a function of ESG, the overall score the company received on ESG performance; Firm size, the firms log of Total Assets; R&D intensity, the ratio of a company’s Research and Development investment to its total sales; Risk, the leverage ratio of total debt to total assets; α, firm fixed effects; Year dummies for year fixed effects and a robust error term, ε.

Then, it is tested whether the difference between the coefficients is significantly different from zero. As such a test is not supported for fixed effects regressions by Stata, this test is executed using a similar equation that includes dummy variables, to control for fixed industry and country effects, instead. (Firm dummies could not be included as too many base levels would be specified). This problem may have been solved by bootstrapping, however this is not considered worthwhile for such a small detail of this analysis. It is kept in mind that the results of this test is only meaningful in case both regressions (fixed and dummy) show similar results.

15 _{This classification is based on that of Derwall (2007) who tested for non-linearity in the effect of} eco-efficiency.

Similarly, the third hypotheses; whether firms that operate in environmentally sensitive industries experience a greater positive relationship between their ESG performance and firm value than their non-environmentally sensitive operating counterparts16, is tested using the following equation:

𝑄𝑄𝐵𝐵𝐵𝐵 = 𝛼𝛼𝐵𝐵+ 𝛽𝛽1𝐸𝐸𝑛𝑛𝐸𝐸𝐹𝐹𝐹𝐹𝐸𝐸𝑛𝑛𝐹𝐹𝑠𝑠𝑛𝑛𝑡𝑡𝑌𝑌𝐸𝐸𝐵𝐵𝐵𝐵+ 𝛽𝛽3𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑠𝑠𝐹𝐹𝑠𝑠𝑠𝑠𝐵𝐵𝐵𝐵+ 𝛽𝛽4𝑅𝑅&𝐷𝐷𝐹𝐹𝑛𝑛𝑡𝑡𝑠𝑠𝑛𝑛𝑠𝑠𝐹𝐹𝑡𝑡𝑡𝑡𝐵𝐵𝐵𝐵 + 𝛽𝛽5𝑅𝑅𝐹𝐹𝑠𝑠𝑅𝑅𝐵𝐵𝐵𝐵+ 𝛼𝛼𝐵𝐵 + ∑8𝑗𝑗=1𝑌𝑌𝑠𝑠𝑌𝑌𝐹𝐹𝐵𝐵,𝑗𝑗+ 𝜀𝜀𝐵𝐵 (6)

Where the firm value for a firm i in year t () is a function of Environmental, the score the company received on its environmental performance; Firm size, the firms log of Total Assets; R&D intensity, the ratio of a company’s Research and Development investment to its total sales; Risk, the leverage ratio of total debt to total assets; α, the firm fixed effects, Year dummies to for year fixed effects and a robust error term, ε.

Again, as a test on the difference of coefficients is not supported for fixed effects regressions by Stata the test is ran using an equation including dummy fixed effects instead. However, this equation deviates from the test of Hypotheses 2 as the industry dummies are left out. Otherwise too many base levels would be specified and a test on the coefficients would again be impossible. This has to do with the fact that the dummy character of the environmental sensitivity classification is itself a sub – specification within the industry classification. However, it will be checked whether including the industry dummies would yield similar results in order to assure the test is executed on meaningful coefficients.

Robustness

As the implementation of a variety of methods (for hypotheses 1) in fact checks for

robustness, I only analyze some additional regressions on the individual pillars instead of the overall ESG score. This way it can be verified whether their effect on Tobin’s Q differs from each other.