• No results found

E, S or G? A study of ESG score and financial performance

N/A
N/A
Protected

Academic year: 2021

Share "E, S or G? A study of ESG score and financial performance"

Copied!
77
0
0

Loading.... (view fulltext now)

Full text

(1)

E, S or G? A study of ESG score

and financial performance

YRR AHLKLO

CARIN LIND

KTH

(2)
(3)

E, S or G? A study of ESG score and

financial performance

by

Yrr Ahlklo

Carin Lind

Master of Science Thesis TRITA-ITM-EX 2019:12 KTH Industrial Engineering and Management

Industrial Management SE-100 44 STOCKHOLM

(4)

E, S eller G? En studie om ESG score,

lönsamhet och avkastning

av

Yrr Ahlklo

Carin Lind

Examensarbete TRITA-ITM-EX 2019:12 KTH Industriell teknik och management

Industriell ekonomi och organisation SE-100 44 STOCKHOLM

(5)

and financial performance Yrr Ahlklo Carin Lind Approved 2019-01-29 Examiner Hans Lööf Supervisor Christian Thomann Commissioner

Erik Penser Bank

Contact person

Jonas Thulin

Abstract

Sustainability is not a new concept to the financial markets, but its popularity and wider use have increased as people have grown more concerned about the future of this planet. However, the relationship between sustainable investments and financial performance is not clear. One of the most used measures of sustainability is the concept of ESG score, where E, S and G stand for environmental, social and governance. In this study, we investigate the relationship between ESG score and financial performance, both market and accounting based. We also separate the score into its individual parts E, S, and G, and try to distinguish which factor has the strongest relation to financial performance. To evaluate the relationship, a regression analysis was performed on a sample of Nordic stocks and the Sustainalytics ESG rank. Our findings concluded no significant relationship between ESG score and financial performance, neither market nor accounting based. The environmental factor (E) showed the strongest relation to financial performance, however slightly negative and only significant to one dependent variable out of three. Our results indicate that based on the ESG score used in this study, no conclusions can be drawn about financial performance. Since our research does not indicate a significant relationship, our recommendation is to invest in the highest ESG ranked stock in case of choosing between two otherwise similar stocks.

Key-words

ESG score, financial performance, regression, return, firm value, sustainable investment, corporate social responsibility, stakeholder theory.

(6)

lönsamhet och avkastning Yrr Ahlklo Carin Lind Godkänt 2019-01-29 Examinator Hans Lööf Handledare Christian Thomann Uppdragsgivare

Erik Penser Bank

Kontaktperson

Jonas Thulin

Sammanfattning

Hållbarhet är inget nytt koncept inom finans, men dess popularitet och användning har ökat kraftigt. Dock är det fortfarande oklart hur hållbara investeringar förhåller sig till lönsamhet och avkastning. En av de mest använda hållbarhetsmåtten är ESG, som står för environmental, social and governance. I denna studie undersöker vi relationen mellan ESG-mått och lönsamhet, både marknads- och resultatbaserad. Vi delar också upp ESG i sina tre komponenter E, S, och G för att undersöka vilken faktor som har den starkaste relationen till lönsamhet. Detta görs genom en regressionsanalys med

paneldata från ett urval av nordiska aktier och Sustainalytics ESG-mått. Vårt resultat visar ingen signifikant relation mellan ESG-mått och lönsamhet. Komponenten E visar den starkaste relationen till lönsamhet, ett signifikant och något negativt samband, men endast till en av tre responsvariabler. Vårt resultat indikerar således att inget samband verkar finnas mellan lönsamhet och dessa ESG-mått. Eftersom vår studie inte visar på något signifikant samband, blir vår rekommendation att investera i den aktien med högst ESG-mått, om man skulle välja mellan två annars lika aktier.

Nyckelord

ESG, finansiell lönsamhet, avkastning, regressionsanalys, hållbara investeringar, rörelsevärde, CSR, Intressentmodellen.

(7)

1 Introduction 1

1.1 Background . . . 1

Sustainable investing - a debated topic in finance . . . 1

Measuring sustainability is complex . . . 1

ESG - the dominating concept for measuring sustainability 2 Regulation pressure banks to take interest in ESG . . . 2

1.2 Problematization . . . 3

1.3 Purpose of thesis . . . 4

1.4 Research questions . . . 4

1.5 Defining and delimiting the scope . . . 4

1.6 Contribution to research field and industry . . . 5

2 Literature Review 6 2.1 Definitions of sustainability . . . 6

2.2 Corporate Social Responsibility . . . 7

What is it? . . . 7

Should CSR affect financial performance? . . . 7

2.3 Measuring CSR by different scores . . . 8

Scores and actors . . . 8

ESG score . . . 8

2.4 Screening methods in sustainable investing . . . 9

Negative screening . . . 9

Positive screening . . . 9

Combined approach example . . . 10

2.5 Links between CSR activity and financial performance . . . 10

Accounting based financial performance . . . 10

Market based financial performance . . . 11

Risk . . . 13

2.6 Literature review summary . . . 13

3 Method 15 3.1 Scientific approach . . . 15

Research design and approach . . . 15

Literature review method . . . 15

Validity and reliability . . . 16

(8)

Researchers’ bias . . . 17

3.2 Designing the regression analysis . . . 17

Regression variables . . . 18

Preliminary regression models . . . 20

3.3 Data collection . . . 21

Data collection method . . . 21

Sample selection . . . 21

Exploring the data . . . 23

3.4 Development of the final regression models . . . 26

Statistical tests used in model diagnostics . . . 26

Modifications of the regression models . . . 26

Final regression models . . . 27

4 Empirical results 29 4.1 Regression summary . . . 29 4.2 ROA regressions . . . 29 Results . . . 29 Model diagnostics . . . 29 4.3 Tobin’s q regressions . . . 32 Results . . . 32 Model diagnostics . . . 32

4.4 Stock return regressions . . . 33

Results . . . 33

Model diagnostics . . . 34

5 Discussion & Analysis 35 5.1 Total ESG score and financial performance . . . 35

5.2 Effect of individual ESG components . . . 36

5.3 Critical review . . . 38

5.4 Implications for academia and industry . . . 39

6 Conclusion 40 6.1 Answering the research question . . . 40

6.2 Recommendations . . . 40

6.3 Future research . . . 40

Bibliography 42 A Regressions with ESG score 46 A.1 Response variable: ROA . . . 46

(9)

A.2 Response variable: Tobin’s q . . . 51 A.3 Response variable: Stock returns . . . 56

(10)

3.1 Sustainalytics rank sample histogram . . . 23 3.2 Correlation matrix . . . 25 4.1 Residuals vs fitted values for the regression with ROA as

response variable and ESG score as variable of interest. . . 31 4.2 Residuals vs fitted values for the regression with log(Tobin’s

q) as dependent variable and ESG score as variable of interest. 33 4.3 Residuals vs fitted values for the regression with stock returns

as dependent variable and ESG score as variable of interest. . . 34

(11)

2.1 Summary of literature review. . . 14

3.1 Keywords used in literature research. . . 16

3.2 Summary of variables. . . 18

3.3 Overview of sample. . . 23

3.4 Descriptive statistics of sample . . . 24

3.5 Descriptive statistics of processed sample . . . 24

3.6 Summary of all regression variable combinations. . . 28

4.1 Regression results summary . . . 30

4.2 Confidence intervals of ESG effect . . . 31

(12)

to write our thesis in collaboration with Erik Penser bank, and with a great deal of flexibility regarding the scope of the thesis. Your input has been very helpful and contributed to a more interesting scope.

We also want to thank our supervisor Christian Thomann, for always an-swering our questions straight away and for the advice on econometrical issues. Additionally, we appreciate all the discussions and seminars that contributed to improving our thesis.

(13)

Introduction

1.1 Background

Sustainable investing - a debated topic in finance

Sustainability is not a new concept to the financial markets, but its popularity and wider use has increased as people have grown more concerned about the future of this planet. Since the world strives to be more environmentally friendly in order to reduce global warming (TT 2018), the financial sector must keep up with the trend. It is our view that sustainability and investing is a debated combination which attracts the attention of the general public. For example, a Swedish newspaper published an interview with professor Robert G Eccles whose research shows that sustainable companies are more profitable (Johansson 2015). But some critical voices mean that in the financial sector, sustainability is just a buzzword to sell new (or slightly changed old) products at a higher price (Petersson 2018).

One reason for the critique might be that the question remains of what a sustainable investment really is and how it is defined. The lack of clear definition and standard contribute to a more doubtful market and can make investors unsure about the whole concept. Sustainability is such a broad and complex concept, and there are many definitions and applications of the word itself. We believe it is common to foremost associate sustainability with reducing carbon emissions.

Measuring sustainability is complex

The problem with one dimensional measures such as only measuring carbon emissions, is that low carbon emissions do not necessarily make a business sustainable. A company could have low carbon emissions, but a rather unsustainable business or product. For example, a company that produce single use items like disposable plastic straws could use 100% renewable energy, but their products may still end up polluting the seas. On the other

(14)

hand, a company can have high carbon emissions but fill a sustainable purpose, for example a public transport company. It probably has relatively high carbon emissions but is still an environmentally sustainable business, since buses transport a lot of people that would otherwise travel in individual cars.

ESG - the dominating concept for measuring sustainability

Because of the complexity of measuring sustainability, a number of actors and measures have evolved (see section 2.3). One of the most used concepts is the ESG score. E, S and G stand for environmental, social and governance which are three factors commonly used for measuring the sustainability of an investment (PRI 2018). Unfortunately there is no clear and well defined way to calculate an ESG score and there is no explicit standard for exactly what is included in the E, S and G factors. Different actors can use their own methods for calculating ESG scores, and there is little to no transparency. What constitutes an acceptable ESG score is therefore subjective. This uncertainty has increased pressure on the authorities to decide on a standard for the financial markets.

Regulation pressure banks to take interest in ESG

Soon, sustainable investing might not only be a concern of the individual investor. In May 2018, the European Commission published a new proposal for sustainability regulation regarding the financial markets, which will affect institutional investors when it takes effect. Since sustainable investments are subject to such uncertainty, this regulation should be welcomed by the market. However, institutional investors might also be worried about having to disclose a sustainability measure if they do not agree with the measuring method.

The proposal from the European Commission contains three parts, which are summarized below (European Commission 2018).

1. Establishing a unified EU classification system of sustainable economic activities (taxonomy).

2. Improving disclosure requirements on how institutional investors inte-grate ESG factors in their decision-making process.

3. Creating a new category of benchmarks which will help investors compare the carbon footprint of their investments.

(15)

This proposed regulation shows that sustainability measures in some form will most likely be mandatory in the future. Since the regulation is not reality yet, it may change before implementation and hence the uncertainty about defining sustainable investments currently remains. The fact that the EU Commission is so far in the process makes banks start to investigate carbon and ESG measures, to be able to influence the way the regulation is formulated. We imagine that most banks take interest in having a definition of sustainability that also allows them to make profitable investments. This thesis is written in collaboration with Erik Penser Bank, a smaller institution, with a goal of making both profitable and sustainable investments.

1.2 Problematization

The focus on measuring sustainability leads to the question of whether sustainability measures can be used to improve financial performance. Some studies, but maybe foremost marketing from the financial sector, say that sustainable portfolios give higher returns and outperform index (Thompson 2018). One example of an ESG portfolio that actually does well would be the World SRI (Socially Responsible Investment) Index by MSCI, a provider of stock/fixed income indices, which has performed slightly better than the corresponding general world index since 2007 (MSCI 2018). The selection of stocks is based on ESG data, but is it really sustainability that drives the performance? As can be seen in section 2.5 in our literature review, the picture is not clear of whether there exist a link or not, and in that case why. Today, asset managers, private investors and smaller institutions rely on external analysis for ESG scoring, since most do not have ESG analysts of their own. The ESG scoring process is not always very transparent, so if there is a link, the lack of transparency makes it hard to judge where the returns really come from. Maybe investors could increase financial performance if they knew how to pick the best out of the ESG rated companies? In that case, the question is what part of the ESG score is related to returns. One hypothesis of why a world SRI index have higher returns could be that sustainable companies ought to perform better in the long run, since sustainable businesses and products could be subject to higher demand in the future as concern for the environment grows. Another hypothesis is that ESG portfolios show better performance because investors respond to a temporary ESG trend, which drives up the stock prices for companies with good ESG practices. With different types of ESG scores, little transparency

(16)

and an undefined scope of measuring sustainability, it is hard to know which factors that really drive the performance.

1.3 Purpose of thesis

The purpose of this thesis is to investigate the relation between ESG score and financial performance. Because of the uncertainty of how ESG and performance could be linked, the aim is to investigate both market based performance that relates to the stock market, and accounting based perfor-mance which relates to profitability. With these two kinds of measures, we want to add to the knowledge of why there might be a relationship. Further, we want to investigate what part of the ESG score is the most connected to financial performance.

1.4 Research questions

The research questions are developed in collaboration with Erik Penser Bank, in order to be relevant and useful to actors in the financial sector. The research will be focused on the following questions.

• What is the relationship between ESG score and financial performance? • What part of the ESG score shows the strongest relation to financial

performance?

1.5 Defining and delimiting the scope

Since this thesis is written in collaboration with Erik Penser Bank, a small Swedish institutional investor, we have chosen a sample of Nordic stocks as our universe. We have also chosen to concentrate on the relationship between ESG score and financial performance, and not causal links. This delimitation has been made due to our choice of method, since it is impossible to prove any causal links with a regression analysis. One way to investigate a causal link better would have been to conduct an event study, such as the study by Hartzmark and Sussman (2018) which investigates the effect of the introduction of an ESG rating of mutual funds. The most similar option for us would have been an event study of the introduction of an ESG score for stocks. However, for single stocks the problem is isolating the effect of the ESG score from the effect of company fundamentals or news reporting, which was not suitable for our time frame.

(17)

1.6 Contribution to research field and industry

The area of sustainability measures and investments is important to many investors, and risk management and return potential are the leading factors for considering sustainable investing according to a recent survey (Morgan Stanley 2018). Even though there is much written on this subject, we believe that the development of the area is faster than the research has been. We want to contribute to a more complete picture of sustainability in financial products, by investigating the ESG relation to return with several measures. We also use a different ESG score than other similar studies (see section 2.5). Furthermore, we want to add a piece to the puzzle of what part of the ESG score has the strongest relation to financial performance, as previous studies have showed varying results. Our study will hopefully contribute to a slightly clearer picture for institutional as well as sustainability conscious private investors.

(18)

Literature Review

2.1 Definitions of sustainability

In this paper, we do not attempt to judge the sustainability of companies nor of stock portfolios. Nevertheless, it is important to know different definitions of sustainability used in the literature. One of the most common and quoted definitions comes from the UN World Commission on Environment and De-velopment. It says "sustainable development is development that meets the needs of the present without compromising the ability of future generations to meet their own needs" (United Nations 1987). In finance, sustainabil-ity and sustainable investments is an umbrella term for investments that pursue to have a long term positive effect on society and environment. Fur-thermore, sustainability conscious investors also seek returns and financial performance, like most investors do. Under the umbrella term sustainable investments, we find terms like socially responsible investing (SRI), ethical investing, green investing and ESG score. There exits different schools and opinions about how to define these terms. According to the European Com-mission European CoCom-mission (2018), sustainable finance is defined as "the provision of finance to investments taking into account environmental, social and governance considerations". The European Commission also states that sustainable finance includes a strong green finance component that aims to support economic growth while:

• Reducing pressures on the environment

• Addressing green-house gas emissions and tackling pollution

• Minimizing waste and improving efficiency in the use of natural re-sources

It also encompasses increasing awareness of and transparency on

• The risks which may have an impact on the sustainability of the finan-cial system

(19)

• The need for financial and corporate actors to mitigate those risks through appropriate governance

We will use the European Commissions definition of sustainable finance and especially look at the measurement ESG score.

2.2 Corporate Social Responsibility

What is it?

The European Commission defines Corporate Social Responsibility (CSR) as "a concept by which companies decide voluntarily to contribute to a bet-ter society and a cleaner environment by going beyond compliance and investing more into human capital, the environment and the relations with stakeholders" (Arvidsson 2010). Compared to sustainable finance and mea-suring terms like ESG, CSR is more an organizational policy which must be aligned to the business model to be successful. CSR is a broad concept, and it depends on industry and business how it takes form. As much as CSR is important for the society, it ought to be equally valuable for the company itself.

Should CSR affect financial performance?

According to stakeholder theory, CSR should affect financial performance. In reverse, shareholder theory claims it should not. Stakeholder theory chal-lenges the classical shareholder view, which indicates that a company should solely run its business for the shareholders benefit. Shareholder theory is based on the premise that management is hired to run the company for the shareholders and serve their interest. The theory was originally presented by Milton Friedman (Friedman 1970), but is now challenged by the stakeholder theory. Instead of just shareholders, the stakeholder theory includes other groups as important to a company’s well being, for example employees, cus-tomers, unions and many more. The book Strategic Management: A Stakeholder Approach by Freeman (1983) is mentioned by many as the foundation of the stakeholder theory. Several articles use stakeholder theory as the theoretical base for theories about CSR and financial performance, for example Brooks and Oikonomou (2018) and J. Kim, Chung, and Park (2013).

(20)

2.3 Measuring CSR by different scores

Scores and actors

There are plenty of actors and types of scoring in CSR. For example, Sustain-alytics, Thomson Reuters and Morningstar all provide similar types of ESG scores. Both Sustainalytics and Thomson Reuters provide ESG scores and other relevant CSR related data on particular firms and Morningstar provides an ESG score on mutual funds in collaboration with Sustainalytics. Other actors are MSCI, an index provider of SRI (socially responsible investments) indices, and Bloomberg, although their score is a disclosure score. There is a difference between disclosure and performance, where disclosure is how well companies do sustainability reporting and performance is how good they are doing environmentally or socially, for example how much carbon emissions they have. Global reporting initiative (GRI) is another organiza-tion, whose purpose is to create a standardized system for ESG reporting. For a more detailed overview of some of the well-known actors, see Huber and Comstock (2017).

ESG score

As this thesis will focus on ESG score and its implications for financial performance, ESG score is explained below. According to the Principles of Responsible Investing (PRI), ESG is defined as a term that helps investors to better manage risk and sustainable long-term returns (PRI 2018). ESG scores are used by investors to estimate the ability of companies to be sustainable, in addition to other analysis tools for assessing future financial performance. The Sustainalytics ESG rank (the ESG score used in this thesis) ranges from 0 to 100, where 100 is the best score a firm can achieve. As earlier mentioned, ESG stands for environmental, social and governance. We explain the factors below (PRI 2018).

The environmental factor

The environmental factor consists of how a company handle for example pollution, waste, deforestation, carbon dioxide emissions and climate change.

The social factor

The social factor includes how a company treats people and the community. Here employee relations, working conditions, local community, diversity,

(21)

conflict management, health and safety are of great importance.

The governance factor

The governance factor considers how a company is led. The factor includes different policies, tax strategies, donations, lobbying, corruption and bribery.

2.4 Screening methods in sustainable investing

Negative screening

There are two types of negative screening, which are processes where the worst companies are removed. The first type of negative screening is where you remove the worst individual stocks, for example by setting a minimum ESG score allowed. If a company is involved in too much negative practises, and therefore has too low ESG score, it will not be considered when choosing the sustainable investments. Negative practises include for example poor pollution records, bad employment records and inadequate health and safety records.

The second type of negative screening is excluding whole sectors, called "sin sectors". Typical industries and business to be ruled out during this type of negative screening is oil, tobacco, mining and pornography (The Guardian 2001). Interestingly, Statman and Glushkov (2009) finds that excluding whole sectors comes at a disadvantage. The social responsible portfolio that just excluded "sin" sectors had lower returns than the conventional portfolio. The social responsible portfolio has lower diversification, and it therefore becomes costly to exclude whole sectors when measuring return.

Positive screening

Positive screening is a screening method where the best companies are selected. Companies with strong records of social and environmentally friendly activities such as good working conditions, energy efficient solutions and good recycling policies are favored. These companies usually have higher ESG scores. As for negative screening, one method is to pick the best individual stocks, for example by setting a lowest allowed ESG score for the portfolio. This type of positive screening is often referred to as the best-in-class concept. Another method is to invest in sectors which are deemed beneficial for sustainability. Examples of industries or businesses that typically are recognized in a positive screening are waste management,

(22)

public transport, education, environmental technology and renewable energy (The Guardian 2001).

Combined approach example

An example of a combined approach to positive and negative screening can be seen in the study by Statman and Glushkov (2009). By analyzing the returns of stocks related to social responsibility from 1992–2007, they find that a combination of both types of screening does not achieve higher returns. Using positive screening and investing in socially responsible companies gave a an advantage in return over conventional portfolios. Using negative screening by ruling out stocks from "sin" sectors gave a disadvantage for the socially responsible portfolio compared to the conventional portfolio, as mentioned above. Hence, the two types of screening neutralized each other to a zero net effect. The study therefore concluded that a best-in-class (i.e. selecting among the top ranked stocks) positive screening and not using negative screening ought to be the best way to go in terms of returns (Statman and Glushkov 2009).

2.5 Links between CSR activity and financial

performance

To summarize, several studies have investigated the relation between CSR and financial performance in different ways. The conclusion from reading the studies included in our literature review is that there seems to be a positive correlation on a company level, but not on an aggregated portfolio and fund level.

Accounting based financial performance

The state of research today seems to be that there is a small but statistically significant positive correlation between CSR activity and accounting based financial performance. By accounting based financial performance we mean profitability, for example return on assets (ROA).

According to a study by Eccles, Ioannou, and Serafeim (2014) using 180 Amer-ican companies which they ranked as high or low sustainability companies, the high sustainability companies significantly outperform their counterparts in the long run, both in accounting and market based performance. Another

(23)

study by Friede, Busch, and Bassen (2015) that compiled data from 2200 studies also found a positive link to accounting based performance. Contrary to these findings, Dahlberg and Wiklund (2018) found no link between ESG score and accounting performance based performance in their recent study of the Nordic stock market.

In a study of the German market, Velte (2017) examines the relationship between ESG score and financial performance for companies listed on the German Prime Standard during the years of 2010-2014. The results of his research and regression analysis show that there is a positive relationship between ESG performance and accounting based financial performance. However, the study concludes that there is no positive relationship between ESG and market based financial performance. Velte also investigated which of the factors E,S and G showed the strongest correlation to accounting based financial performance. The factors where included separately in the regression model. All factors showed a positive correlation to the accounting based financial performance, and the governance factor (G) showed the strongest correlation.

Market based financial performance

As can be seen below, there are several studies showing that there exists a slightly positive relationship between CSR activity and market based finan-cial performance. By market based finanfinan-cial performance we mean measures based on market value, for example stock return or Tobin’s q ratio. But there are also studies that show no significant relationship. As earlier mentioned, Velte (2017) found no significant relationship between ESG performance and market based financial performance.

As mentioned above, Friede, Busch, and Bassen (2015) present a study that concludes empirical results from over 2000 previous studies of ESG factors and their effects on financial performance. The research contains both portfo-lio and nonportfoportfo-lio-based studies. The authors conclude that there is, on average, a positive relationship between market based financial performance and ESG factors. The conclusion also includes financial performance of dif-ferent asset classes and regions. Another meta-study that has compiled the result of over 200 previous studies shows a similar result. The authors found that 80 % of the analyzed studies showed a positive correlation between ESG performance and market based financial performance (Clark, Feiner, and Viehs 2015). When large American non-financial firms were investigated, a

(24)

significantly positive relationship was found between corporate sustainabil-ity and market based financial performance (Lo and Sheu 2007). The authors also found a positive relation between corporate sustainability and sales growth. Furthermore, there was evidence presented that being sustainable causes firm value to rise.

In a study of the Korean market from 2013, it is found that a firm’s ESG score measured by MSCI is correlated with both stock returns and Tobin’s q (J. Kim, Chung, and Park 2013). Another study of the Korean market shows a slightly different result (Han, H. J. Kim, and Yu 2016). This empirical study of 94 firms listed on the Korean Stock Exchange shows that different ESG factors have different relations to market based financial performance. The governance factor showed a positive relation while the environmental factor showed a negative relation towards financial performance. The social factor turned out to be neutral as it showed no significant relationship.

A master thesis from Stockholm School of Economics shows that firms in-cluding sustainability in their business model outperform companies that work with sustainability as philanthropy (Broman and Lundqvist 2016). The authors have collected data from Swedish companies listed on the Nasdaq OMX Stockholm and divided the companies into different groups according to how they work with sustainability and their ESG performance. Despite not finding any connection between ESG score and accounting based fi-nancial performance, Dahlberg and Wiklund (2018) found a link between ESG score and market based financial performance measured by Tobin’s q. This indicates that investors in Nordic stocks value ESG factors and that the stakeholder theory is supported.

Sahut and Pasquini-Descomps (2015) investigate the connection between a news-based ESG score and financial performance in the UK, Switzerland and the US. They find that there is a slightly negative relation, but only statistically significant in the UK. When they investigate sub-category ratings of GRI (governance, economic, environment, labor, human rights, society and products), they find significant relations but only for some sectors and time periods, which were not the same between the three countries. This study also finds through non-parametric kernel regression that the link between financial performance and changes in this news based ESG score is probably not linear.

Lee, Faff, and Rekker (2013) corroborate the "no link theory" (i.e. that there is no link between ESG and financial performance) in their study of US stock

(25)

portfolios, which shows that there is no difference in financial performance between investing in a high ESG portfolio compared to investing in a con-ventional portfolio or a low ESG portfolio. Neither Hartzmark and Sussman (2018) find any relationship between high Morningstar globe rating (another sustainability measure, based on data from MSCI) and market based finan-cial performance, although the study shows that investors prefer funds with the highest sustainability rating compared to the lowest (Hartzmark and Sussman 2018). This finding indicates that investors value sustainability even though no relation between high scores and financial performance was found.

Risk

Verheyden, Eccles, and Feiner (2016) investigate the relationship between ESG, risk, return and diversification. They create four different universes of stocks based on ESG screening, and find that the screening adds 0.16% on average to annual returns. The risk is also lower, measured by volatilty, draw-downs and conditional value at risk (CVaR). In three out of four universes they also find that the added return outweighs the sacrificed diversification, going against the earlier mentioned "no link theory". The study by Clark, Feiner, and Viehs (2015) mentioned above that compiled the results from over 200 previous studies about ESG score and financial performance presents the result that sustainable companies achieve higher profits and are less risky.

2.6 Literature review summary

To summarize the literature review, most previous studies indicate a positive relation between sustainability and financial performance. However, there are also studies which show no relationship or a slightly negative one. For accounting based marked performance we have reviewed five studies indi-cating a positive relationship and one with no relation. Regarding market based financial performance, we have summarized eight previous studies that show a positive relationship, three studies indicating no relation and two with a negative relation. Note that some of these studies are review studies. All the previous studies reviewed are summarized in table 2.1 on page 14.

(26)

Author Time-frame Sample descrip-tion No. obs Research method ESG rating method Corr fin. perf. Corr acc. perf. Broman (2016) 2012– 2015 Large and Mid Cap Nasdaq OMX Stockholm 4992 Cahart four-factor model Thomson Reuters Asset4 + + Clark (2015) 1976– 2011 World 200 Review study Several + n/a Dahlberg (2018) 2006– 2016 Nordic 108 Regression with ROE, Tobin’s q Thomson Reuters Eikon + 0 Eccles

(2014) 1993–2009 US firms 180firms Propensityscore matching n/a + + Friede (2015) 1970– 2014 Global 2200 stud-ies Review study n/a + + Han (2016) 2008– 2014 Koreanstock market 94

firms with ROE,Regression Tobin’s q Bloomberg disclosure score +/- n/a Hartzmark (2018) 2016– 2017 US based open-end funds n/a Event Study Morningstar Sustainalyt-ics 0 n/a Kim (2013) 2011 Korean stock market 96

firms RegressionMarket adjusted return, Tobin’s q Morgan Stanley Capital Int. + n/a Lee (2013) 1999– 2007 US stocks portfolios n/a Fama French model Sustainability Asset Man-agement Group(SAM) 0 n/a Lo (2007) 1999– 2002 US non financial firms 349 firms Regression with To-bin’s q n/a + + Sahut (2015) 2007– 2011 UK, US and Swiss firms 200 firms Four factor model News based score, Cova-lence -/0 n/a Velte (2017) 2010– 2014 German Prime Standard 80 firms, 412 obs Regression with ROA, Tobin’s q Thomson Reuters Asset4 0 +

(27)

Method

3.1 Scientific approach

Research design and approach

Since the aim of this study is to investigate the relation between ESG score and financial performance, a quantitative study seems to suit our purpose well (Blomkvist and Hallin 2015). To make the problematization researchable, we use an explanatory research design which means that our study aims to find for example correlation between potential cause and effect. According to Blomkvist and Hallin (2015), when choosing a research design it is im-portant to know what type of empirical material (explanans) will help you understand the studied phenomenon (explanandum). Our explanandum is whether sustainable investments, i.e. stocks with higher ESG score, have higher financial performance than those with low ESG score. In order to understand the explanandum, we gathered empirical material in the form of different types of financial data and ESG score, which have been our explanans.

The research approach can be inductive or deductive, which describes the relationship between research and theory (Blomkvist and Hallin 2015). In an inductive approach, it is the empirical material that indicates which theory is of interest. It may lead to new theories and other frameworks than you first started from. Using a deductive approach, you will start reading and relate your research to existing theory. Since this study aims to and analyze and contribute to existing theory, a deductive approach is an obvious choice.

Literature review method

Continuously during the research process, a literature review has been con-ducted. The search engines used were Google Scholar, Diva Portal and KTHB Primo, where we have found relevant books, journals and articles of different kinds. Keywords used in the literature search can be found in table 3.1 below.

(28)

Keywords

ESG score Firm value

Sustainable investment ESG and return

ESG regression

Financial performance Stock market performance

Corporate social responsibility (CSR) Social responsible investing (SRI) Ethical investing

Green investing Stakeholder Theory

Table 3.1:Keywords used in literature research.

Validity and reliability

Validity entails studying the right thing (Blomkvist and Hallin 2015). We have to make sure our regression analysis measures what we want and can answer our research questions. Internal validity concentrates on the issue of causality (Greener 2008). This is an important issue, especially for us, since correlation can easily be mistaken for causality and vice versa. We do not attempt to find any causal links between ESG score and financial performance, we rather investigate the relationship and aim to find the correlation between them. External validity, more commonly known as generalizability, concerns whether the results of the study can be generalized to other contexts and situations (Greener 2008). Since our data sample does not include all Nordic companies and just one type of ESG score (i.e the Sustainalytics rank), the limitations of the generalizability of this study must be considered when applying the results to other contexts.

The validity of the research is obviously connected to the reliability. Reliabil-ity ensures studying in the right way and the research should be conducted so the reader can trust the result. High reliability ensures that the study is done with a method of high transparency and allows for consistency (Greener 2008). It indicates that the results of the study should be repeatable to be considered reliable. As we use secondary data, but from a reliable source (Bloomberg/Sustainalytics), the data could be considered reliable. However since ESG score are subjective and different actors use different methods to

(29)

rank companies, the result can not be guaranteed to be the same when a different ESG source is used. To maximize reliability, we have tried to be transparent and include all data manipulation.

Researchers’ bias

Since we use a quantitative method and data from a reliable source, the bias is minimized. While we have tried to be as impartial as possible in our anal-ysis, we as authors could still be biased about the subject. To acknowledge this matter, we here present our thoughts on sustainability and financial performance so the reader can take our opinions into consideration. We hope that there exists a positive correlation between sustainability, i.e. ESG score, and financial performance. We also hope that investing in companies with higher scores can encourage better environmental, social and governance contributions to the society in the future. If investors knew that sustain-able investments gave higher performance, more would invest in them and contribute to a more sustainable world.

3.2 Designing the regression analysis

In order to investigate whether there exists a relationship between ESG score and financial performance, we have used a multiple regression model. As a starting point, we used an ordinary least squares (OLS) multiple regression analysis, which is an estimation technique that makes quantitative estimates of an econometric relationship by calculating the coefficients that minimize the sum of the squared residuals (Studenmund 2014). In our case, the regres-sion is used to describe the relation between financial performance and ESG scores, and find a model that gives estimated dependent variables (financial performance) as close as possible to the observed data for every stock in the sample. The variables used in our model are explained in the next section below.

A multiple regression analysis requires certain assumptions to be fulfilled. In line with Hair et al. (2014), Studenmund (2014) and previous studies (Velte 2017; Dahlberg and Wiklund 2018) we check if the model meets the assump-tions of linearity, normality of residuals, homoskedasticity, multicollinearity and independence of residuals. Note that a regression analysis model can not prove a causal relationship between different variables, no matter how accurate the results are (Studenmund 2014). For example, a relationship between ESG and financial performance could depend on correlation with

(30)

an unknown variable which is not accounted for in the model. Hence, this study solely aims to find the relationship between ESG score and financial performance and therefore no additional causality test will be executed. After collecting relevant data, the regression model was evaluated using a series of statistical methods and tests to assess the assumptions mentioned above. The tests are described in section 3.4 on page 26. We used the programming language R along with some specific R packages for data manipulation and statistical tests. Based on available data and test results, the model was adapted into the final regression model, which is found in section 3.4 on page 26.

Regression variables

A summary of the regression variables can be found in table 3.2.

Dependent Variables

Return on Assets (ROA) Net Income / Total Assets

Tobin’s q Market Value of Equity and Liabilities/ Book value of Equity and Liabilities

Return Yearly stock returns

Independent Variables

ESG Sustainalytics rank, i.e. ESG score

E Environmental score

S Social score

G Governance score

Control Variables

BETA The beta factor, calculated as explained below on page 20, (Systematic risk)

Leverage (LEV) Total debt/total assets, (unsystematic risk) Size of firm (SIZE) Natural logarithm of total assets

R&D expenses (R&D) R&D expenses from financial statements

2016 Variable for year 2016

2017 Variable for year 2017

2018 Variable for year 2018

Industry Group (IND) Dummy variable for branch of industry (GICS) Country of stock (COUNT) Dummy variable of country of trade

(31)

Dependent variables

As dependent (or response) variables, i.e. variables representing the financial performance, we have chosen three different measures. We use Return on Asset (ROA) as the variable for accounting based performance and Tobin’s q and stock return as the variables for market based performance. As can be seen in the literature review summary in table 2.1, several earlier studies with similar methods use the same variables as dependent variables (Velte 2017; Dahlberg and Wiklund 2018; Han, H. J. Kim, and Yu 2016). Both ROA and Tobin’s q are financial ratios. ROA is the profit a firm earns in relation to its overall assets. Tobin’s q represents the ratio between market value and book value, it is a valuation estimator that expresses if a stock is under or overvalued. If the ratio is below one, it indicates that the cost to replace the value of a firm’s assets is greater than its stock value, i.e. the stock is undervalued. A ratio over one indicates the opposite and a ratio of one indicates that the market value is the same as the asset value. Since we are also interested in whether there exists any direct relationship to the stock returns as well, we added yearly stocks returns as the last dependent variable. The stock returns are calculated from the mid price of the last trading day in December, year over year. All the dependent variables are defined in table 3.2.

Independent variables

The variables of interest in this thesis are the Sustainalytics ESG score and its components, so these are used as independent variables in the regressions. The first independent variable is the total environmental, social and gover-nance (ESG) performance score. The separate environmental (E) score is used as the second independent variable, the social (S) score as the third and the governance (G) score as the fourth. The scores are further described in sec-tion 2.3 on page 8. All scores and informasec-tion are collected from Bloomberg Terminal and Sustainalytics.

We do not believe that the impact of the ESG data on the financial perfor-mance will occur instantly and therefore we have included a time lag. We have chosen to lag the data one year in line with previous studies (Velte 2017) and because it makes the data comprehensible. The financial performance from year t + 1 will be hence matched with the ESG score from the year t.

(32)

Control variables

As control variables we use several variables that have been selected to be in line with earlier studies, for example Velte (2017) and Dahlberg and Wiklund (2018). We use the beta factor and total debt over total assets, i.e. leverage, as the firm risk. The beta factor was calculated for every stock and year with the R function CAPM.beta from the package Performance Analytics. As input we used a risk free rate of zero, the daily price of every stock and the stock index corresponding to the country the stock trades in (OMXS30 for Sweden, OMXC20 for Denmark, OMXH25 for Finland and OMXO20 for Norway). The firm size is given by the natural logarithm of total assets in line with Velte (2017). R&D expenses are reported by companies in their financial statements. The same way as the ESG data, we lag the control variables one year so the financial performance from year t + 1 will be matched with the control variable from the year t. Lastly, we add year variables for 2016, 2017 and 2018. Industry and country are used as dummy variables.

Preliminary regression models

The preliminary regression models are determined to be

ROAit =↵ + 1ESGit+ 2SIZEit+ 3LEVit+ 4BET Ait

+ 5R&Dit+ 62016 + 72017 + 82018

+ 9IN D + 10COU N T + ✏it

T obin0s qit =↵ + 1ESGit+ 2SIZEit+ 3LEVit+ 4BET Ait

+ 5R&Dit+ 62016 + 72017 + 82018

+ 9IN D + 10COU N T + ✏it

Returnit =↵ + 1ESGit+ 2SIZEit+ 3LEVit+ 4BET Ait

+ 5R&Dit+ 62016 + 72017 + 82018

+ 9IN D + 10COU N T + ✏it

where all variables can be found in table 3.2 above, ↵ is a constant, 1 10are

(33)

3.3 Data collection

Data collection method

Sustainalytics

Sustainalytics is a company that provides ranking of listed companies based on their ESG performance (Sustainalytics 2018). The companies are ranked from 0–100, with 100 being the best score. Sustainalytics use a industry indicator for each company, so the rankings are within a specific industry groups. This is the ranking we use as ESG score in this thesis.

Bloomberg Terminal

Bloomberg Terminal is a computer software system where you can find for example real time market data and trade with electronic orders. It also keeps you updated with the latest news and research (Bloomberg 2018). Erik Penser Bank uses the Bloomberg Terminal as their everyday tool. Because of that, and our perspective from a small institution’s point of view, we have decided to use data available in Bloomberg Terminal in order to make this study as authentic as possible.

Accessing the data from Bloomberg Terminal

In Bloomberg Terminal we have accessed a world full of data but the most im-portant for our study is the Sustainalytics rank (i.e. ESG score). In Bloomberg Terminal we also found the individual components of the Sustainalytics rank: an environmental rank, a social rank and governance rank. This meant that we could separate the E, S and G in our study.

Important to know is that Bloomberg Terminal did not provide the whole Sustainalytics universe, just a part of it. For our specific choice of sample, Nordic stocks, the data found in Bloomberg Terminal was estimated to approximately two thirds of the Sustainalytics universe. In addition to the Sustainalytics rank, we also accessed all kinds of market and company reported data that we needed for our regression analysis.

Sample selection

Population

We decided to select a sample from the population of Nordic stocks, since we found few studies conducted on such a sample. Other reasons to

(34)

in-vestigate Nordic stocks is that this study was done in cooperation with a small, Swedish bank with an interest in the Nordic market and that we are interested in how the Nordic market compares to other studies with different samples. For example, we can compare our results with Velte (2017) who has done a similar study on the German market or Han, H. J. Kim, and Yu (2016) who have done a comparable study on the Korean market. We can also compare our results to Dahlberg and Wiklund (2018) who have also used the Nordic market as their sample but with a different source of data and ESG score.

The population was determined to be all the stocks which are part of the indices:

• OMX Stockholm All Shares (OMXSPI) • OMX Copenhagen All shares (OMXCPI) • OMX Helsinki All shares (OMXHPI) • Oslo Børs All-share Index (OSEAX).

Sample construction

We selected the stocks from the indices above which had both an ESG score from Sustainalytics reported in the Bloomberg terminal and data for the independent and control variables. However, during this stage of the data collection we discovered that very few companies reported their R&D ex-penses, and hence we had to exclude R&D as a control variable. The sample can be found in AppendixB.

Financial firms were excluded from the sample since their properties are very different regarding for example the control variable leverage (LEV). A leverage level that is normal for a financial firm could be a sign of distress for a non financial firm. Removing financials is a common practice in quanti-tative research and also in line with previous studies (Velte 2017; Dahlberg and Wiklund 2018; Han, H. J. Kim, and Yu 2016). Since the data in this study is of panel structure, we also removed stocks with less than three years of data, to obtain a more balanced panel.

Our final sample contains 267 stock-year observations. Table 3.3 shows the number of stocks per year in the sample, and how the different modifications affected the sample size.

(35)

Year 2014 2015 2016 2017 2018

Dataset before lag 83 86 90 86 83

Added lag 1 year 82 85 85 83

Removed financials 68 69 69 67

Removed few obs stocks 66 68 68 65

Table 3.3: Overview of sample.

Exploring the data

First, we want to know how the companies in our sample are ranked in general. The distribution of Sustainalytics rank (i.e. the ESG score) for all stock-years in the final sample is shown in figure 3.1. Note that the sample is very skewed to the right, with more than 80 of the stock-year observations being of companies which are ranked 100, and an absolute majority being ranked above 80.

Figure 3.1:A histogram of the Sustainalytics rank observations in the sample

Descriptive statistics

The descriptive statistics of all dependent and independent variables before processing the data sample are presented in table 3.4 below. R&D is excluded

(36)

Min q25 Median Mean q75 Max SD ROA -14.01 3.30 5.65 8.65 10.18 97.23 12.70 Tobin’s q 0.89 1.36 1.67 2.30 2.35 12.02 1.81 Return -0.5 -0.1 0.1 0.8 0.3 1.1 0.2 ESG 15.4 76.7 87.4 83.3 95.7 100.0 16.3 E 6.4 67.2 82.2 77.3 91.9 100.0 19.9 G 14.4 68.8 85.4 79.4 93.1 100.0 18.6 S 26.3 78.0 86.5 82.6 92.7 100.0 15.6 Size 9.11 10.78 11.35 11.37 12.00 13.82 1.01 Lev 0.00 0.15 0.24 0.24 0.32 0.85 0.13 Beta -7.84 0.00 0.06 0.13 0.19 10.87 1.30

Table 3.4: This table shows descriptive statistics of the sample before processing for the analysis. q75 and q25 are the quartiles, i.e. the values that delimit the 25% largest and smallest value. SD is the standard deviation of the variables.

Min q25 Median Mean q75 Max SD

ROAw -14.01 3.22 5.59 7.51 9.42 42.31 8.56 log(Tobin0s q)w -0.06 0.30 0.51 0.66 0.85 2.31 0.53 Returnw -0.35 -0.09 0.05 0.07 0.25 0.69 0.23 ESG 15.4 76.7 87.4 83.3 95.7 100 16.3 ESG 6.4 67.2 82.2 77.3 91.9 100 19.9 S 26.3 78 86.5 82.6 92.7 100 15.6 G 14.4 68.8 85.4 79.4 93.1 100 18.6 Sizew 9.15 10.78 11.35 11.36 12 13.75 1.01 Levw 0.01 0.15 0.24 0.24 0.32 0.7 0.13 Betaw -5.63 0 0.06 0.1 0.19 3.5 0.85

Table 3.5: This table shows descriptive statistics of the processed sample used in the regression analysis. q75 and q25 are the quartiles, i.e. the values that delimit the 25% largest and smallest value. SD is the standard deviation of the variables. Superscript w means the data is winsorized using levels (0.01, 0.99). The description of the data processing is found in section 3.4.

and financial stocks were removed, as mentioned in section 3.3. The ESG, E, S and G score scale is from 0 to 100. The beta factor was calculated as mentioned above in section 3.2 on page 20. The descriptive statistics of the processed sample used in the regression analysis could be found in table 3.5 on page 24. The data manipulation of the sample is explained in section 3.4.

(37)

Correlation Matrix

The correlation matrix of the non dummy variables is presented in figure 3.2 on page 25. As expected, the social, governance and environmental scores are highly correlated with the total ESG score, which is natural since ESG is the aggregated score of its components. But the correlations are not one, which means it is meaningful to test the components individually. The correlation between the social/environmental score and the total ESG score is higher than between the governance and ESG score. This indicates that the environmental and social factors are more in line with the total ESG score than the governance factor is.

Figure 3.2:Correlation matrix. The correlations in colored squares have a p-value of <0.05, and the correlations in white squares are non significant.

We also notice that the size is negatively correlated to Tobin’s q, i.e. market based financial performance, and slightly positively correlated to the ESG score, which could suggest that larger firms have higher ESG scores. We also note that the variables Beta and Return do not have any significant correlation with any variable. This is a bit surprising since Return ought to have a correlation to both Beta and Tobin’s q.

(38)

3.4 Development of the final regression models

Statistical tests used in model diagnostics

In order to do a multiple regression analysis, we need to check certain assumptions. In line with Hair et al. (2014), Studenmund (2014) and previous studies (Velte 2017; Dahlberg and Wiklund 2018) we check if the model meets the assumptions of linearity, normality of residuals, homoskedasticity, multicollinearity and independence of residuals.

We check the linearity assumption by assessing the residuals vs fitted values plot for each regression, looking for indications of nonlinear patterns. We use the same plot to check the homoskedasticity assumption by looking for a widening of variance pattern (cone shape). We also use the Breusch-Pagan heteroskedasticity test from the lmtest R package. The normality of the residuals is checked by assessing the normal Q-Q plot of residuals and a histogram of residuals visually. We also use the Shapiro-Wilk test of normality from the stats R package.

For serial correlation of errors, we first visually examine the sequential plot of residuals for unusual patterns or drift of errors. Then we use two statistical tests as well, namely the Durbin-Watson test for panel models and the modified Bhargava/Franzini/Narendranathan (BFN) Panel Watson test (both from plm R package). The modified BFN Panel Durbin-Watson test does not provide a p-value, instead Bhargava, Franzini, and Narendranathan (1982) advises to look for test statistics < 2 to avoid serial correlation in larger samples.

Lastly, correlation is checked by making a correlation matrix of the model variates and the residuals. Collinearity is checked with the VIF test together with assessing the correlation matrix of the data (figure 3.2).

Modifications of the regression models

In this section, we summarize what changes were made from the initial regression model to the final regression model.

Earlier studies, for example Velte (2017), used R&D as a control variable, which we also wanted to do. However, as mentioned in section 3.3, there were not enough data available for this variable and R&D was therefore excluded from the model.

(39)

differences in variance between countries and industries and the fact that we had to exclude R&D as a variable, we believe there is unobserved hetero-geneity in the sample. Therefore we used a fixed effects regression model. We believe that this type of regression model is the most suitable for our data and type of problem. A fixed effects model is used also because we are only interested in analyzing the impact of variables that vary over time. To explain fixed effect regression briefly, it measures the the relationship between predictor and outcome variables within a unit (i.e. within a stock). Each stock has its own particular characteristics that may or may not affect the predictor variables (Torres Reyna 2007). Since fixed effects models do not take variables into account if they do not change over time, the dummy variables for country and industry are no longer needed in the model. To reduce heteroskedasticity in the regressions with Tobin’s q as dependent variable, we decided to use log(Tobin’s q) instead. We also used winsorizing with levels 0.01, 0.99. This reduced, but did not eliminate, the heteroskedas-ticity evident in the residuals (see figure 4.2). We then decided to use robust, individually clustered standard errors to account for heteroskedasticity and the panel structure of the data. For the dependent variables ROA and stock returns, winsorizing them and the control variables (we used the quantiles 0.01 and 0.99) to reduce outliers proved to be the most effective technique for better model fit. For the ROA regressions, we also decided to exclude five very extreme observations of ROA > 70%, which we knew were due to tech-nical accounting issues because of company splits or sales. The observations were removed before winsorizing.

Final regression models

The final fixed effects regression models for the total ESG score are presented below.

ROAitw =↵i+ 1ESGit+ 2SIZEitw+ 3LEVitw+ 4BET Awit+

+ 52016 + 62017 + 72018 + "it

log(T obin0s q)wit =↵i+ 1ESGit+ 2SIZEitw+ 3LEVitw+ 4BET Awit+

+ 52016 + 62017 + 72018 + "it

Returnitw =↵i+ 1ESGit+ 2SIZEitw+ 3LEVitw+ 4BET Awit+

(40)

For each dependent variable we test E, S and G individually as well as the total ESG score, which means we do in total twelve fixed effects regressions. Table 3.6 summarizes the different regressions performed. All regression variables are described in table 3.2 above, w means winsorized, ↵i are the

individual fixed effects constants, 1 7 are the coefficients and "it is the

clustered robust standard error term.

Combinations of dependent and independent variables:

ROAwand ESG

ROAwand E

ROAwand S

ROAwand G

Log(T obin0sq)w and ESG

Log(T obin0sq)w and E

Log(T obin0sq)w and S

Log(T obin0sq)w and G

Returnw and ESG

Returnw and E

Returnw and S

Returnw and G

Total number of regressions: 12

(41)

Empirical results

4.1 Regression summary

In summary, our regressions do not indicate any significant relationship be-tween financial performance and the total ESG score, nor bebe-tween financial performance and any of the E, S or G components. Only the E component showed a slightly negative significant relationship with log(Tobin’s q) (esti-mate -0.0031, p-value < 0.05). For a full summary of the results, see table 4.1. The confidence intervals for the coefficients for the total ESG score, E, S and G are found in table 4.2. In Appendix A, the full R output of the regression with total ESG score can be found, including all plots and test statistics not presented in this chapter.

4.2 ROA regressions

Results

With ROA as the dependent variable, we have performed four regressions for the variables of interest ESG, E, S and G. Neither of them had an effect on ROA significantly different from zero. ESG, E and S have slightly positive but insignificant coefficient estimates, and G has a slightly negative insignificant coefficient estimate. Only the variable beta has a significant effect on ROA in these regressions, with an estimated coefficient of approximately 0.7 for all of the regressions.

Model diagnostics

The model diagnostics tests show the same results for the four ROA regres-sions, why the comments below apply to all four of them. First, we look at figure 4.1 which shows residuals plotted against the fitted values. In the plot, we do not see clear signs of heteroskedasticity, and the Breusch-Pagan test does not find significant heteroskedasticity either. There is a clear cluster of residuals around the fitted ROA value of five. We do not think this pattern

(42)

ROA log(Tobin’s q) Stock returns ESG 0.0456 (0.0429) -0.0017 (0.0029) -0.0069 (0.0038) Size -4.5464 (3.7903) -0.3252 (0.1412)* -0.2448 (0.2823) Lev 5.8152 (13.9436) 0.3759 (0.4694) 0.7589 (0.6236) Beta 0.7396 (0.1059)*** -0.0020 (0.0099) -0.0300 (0.0069)*** 2016 -0.1246 (0.6189) -0.0023 (0.0200) -0.0129 (0.0514) 2017 0.4283 (0.8826) 0.0875 (0.0269)** 0.0877 (0.0501) 2018 0.7424 (1.1111) 0.0983 (0.0346)** -0.0570 (0.0451) E 0.0144 (0.0371) -0.0031 (0.0015)* -0.0015 (0.0026) Size -4.5360 (3.8343) -0.3043 (0.1379)* -0.2500 (0.2773) Lev 5.8605 (14.2160) -0.2873 (0.4721) 0.7270 (0.6815) Beta 0.7101 (0.1019)*** -0.0047 (0.0104) -0.0246 (0.0096)* 2016 -0.0868 (0.6173) -0.0040 (0.0196) -0.0186 (0.0520) 2017 0.5710 (0.8887) 0.0823 (0.0249)** 0.0657 (0.0510) 2018 0.9103 (1.1165) 0.0905 (0.0326)** -0.0814 (0.0472) S 0.0116 (0.0033) -0.0011 (0.0025) -0.0052 (0.0033) Size -4.3945 (3.7313) -0.3296 (0.1407)* -0.2631 (0.2792) Lev 6.0829 (14.1080) -0.3735 (0.4627) 0.7816 (0.5972) Beta 0.6919 (0.0889)*** -0.0004 (0.0106) -0.0238 (0.0073)** 2016 -0.1169 (0.5961) -0.0011 (0.0216) -0.0062 (0.0518) 2017 0.5057 (0.8593) 0.0884 (0.0306)** 0.0949 (0.0502) 2018 0.8579 (1.0944) 0.0976 (0.0350)** -0.0573 (0.0452) G -0.0212 (0.0393) 0.0008 (0.0017) 0.0017 (0.0028) Size -4.4481 (3.6487) -0.3272 (0.1394)* -0.2581 (0.2736) Lev 5.7136 (14.6185) -0.3746 (0.4800) 0.7163 (0.6895) Beta 0.6775 (0.0909)*** 0.0003 (0.0116) -0.0214 (0.0109) 2016 -0.0706 (0.6243) -0.0044 (0.0190) -0.0199 (0.0521) 2017 0.6541 (0.9292) 0.0792 (0.0243)** 0.0596 (0.0542) 2018 1.0040 (1.1562) 0.0892 (0.0351)* -0.0873 (0.0517)

Table 4.1:Summary of the results of the twelve fixed effects regressions. The first row shows which dependent variable the results below are associated with. First number presented is the estimate of the coefficients for each regression. The numbers presented within parentheses is the clustered and robust standard errors. The symbols *, ** and *** indicate significance levels of 5%, 1% and 0.1% respectively.

(43)

ROA log(Tobin’s q) Stock returns 5% 95% 5% 95% 5% 95% ESG -0.025 0.117 -0.006 0.003 -0.014 -0.001 E -0.047 0.076 -0.006 -0.001 -0.006 0.003 S -0.042 0.065 -0.005 0.003 -0.011 0 G -0.086 0.044 -0.002 0.004 -0.003 0.006

Table 4.2: This table shows the confidence intervals of the regression estimates of how the ESG score and its components affect the financial performance measures ROA, Tobin’s q and stock returns.

−10 −5 0 5 10 0 10 20 30 40 fitted_values_ROA residuals_R O A

Figure 4.1: Residuals vs fitted values for the regression with ROA as response variable and ESG score as variable of interest.

indicates non-linearity, because it corresponds to the distribution of the ROA data. For comparison, the median ROA is 5.7% and the distribution can be found in section 3.3, see table 3.4.

By inspecting the Q-Q plot and histogram of residuals, we can see that the distribution of residuals has too heavy tails to be normally distributed. The Shapiro-Wilk test also rejects the hypothesis of normality.

The Durbin-Watson and modified BFN Durbin-Watson tests do not indicate any serial correlation of residuals. The sequential plot of residuals show that some residuals form lines. Due to this and the panel structure of the data,

(44)

we have used individual stock clustered errors.

Lastly, none of the independent variables are correlated with the residuals, and the VIF test does not indicate multicollinearity.

4.3 Tobin’s q regressions

Results

When log(Tobin’s q) was used as dependent variable (see section 3.2 for motivation for the log-transform), four regressions were performed for E, S, G and the total ESG score. No significant effect was discovered for either of them except for the environmental (E) score, which had a slightly negative effect on Tobin’s q with a significance level of 5%. The other insignificant coefficients estimates are also slightly negative except the governance factor (G), which seems to have a rather small positive effect, although it is not significant. Other significant coefficient estimates are the variables size and years 2017 and 2018. The size variable seems to have a slightly negative im-pact on Tobin’s q, with an estimated significant coefficient of approximately -0.3 for all regressions. This indicates that larger stocks seem to have lower

Tobin’q and hence be less overvalued compared to their asset values.

Model diagnostics

The model diagnostics tests show the same results for the four log(Tobin’s q) regressions, why the comments below apply to all four of them. First, we look at figure 4.2 with residuals plotted against the fitted values. In the plot, there is a classical cone shaped heteroskedasticity pattern, and the Breusch-Pagan test indeed rejects the homoskedasticity hypothesis. Hence, taking the logartithm of Tobin’s q did not completely remedy the hetero-skedasticity problem. However, we use robust standard errors which are heteroskedasticity consistent. In the plot there is also a clear cluster of residuals between the fitted log(Tobin’s q) values 0-1. We do not think this pattern indicates non-linearity, because it corresponds to the distribution of the log(Tobin’s q) since the Tobin’s q data is skewed towards lower values. For comparison, the distribution can be studied in the descriptive statistics in table 3.4.

By inspecting the Q-Q plot and histogram of residuals, we can see that the distribution of residuals is too peaked and has too heavy tails to be normally distributed. The Shapiro-Wilk test also rejects the hypothesis of normality.

(45)

−0.6 −0.3 0.0 0.3 0.0 0.5 1.0 1.5 2.0 fitted_values_tobins_q residuals_tobins_q

Figure 4.2: Residuals vs fitted values for the regression with log(Tobin’s q) as dependent variable and ESG score as variable of interest.

The Durbin-Watson and modified BFN Durbin-Watson tests do not indicate any serial correlation of residuals. The sequential plot of residuals show that some residuals form lines. Due to this and the panel structure of the data, we have used individual stock clustered errors.

Lastly, none of the independent variables are correlated with the residuals, and the VIF test does not indicate multicollinearity.

4.4 Stock return regressions

Results

The results for the return regressions follow almost the same pattern as for the log(Tobin’s q) regressions. This seems reasonable since both Tobin’s q and returns are related to market value. As can be seen in table 4.1 there are slightly negative but insignificant results for the total ESG score, E and S. For the governance factor (G), there is a slightly positive but insignificant relationship. This is in line with the results of the log(Tobin’s q) regressions. However, for the return regressions the beta variable and its coefficient estimates are slightly negative and significant for E, S and ESG, but not for G.

Figure

Table 2.1: Summary of literature review.
Table 3.1: Keywords used in literature research.
Table 3.2: Summary of variables.
Table 3.3: Overview of sample.
+7

References

Related documents

Submitted to Linköping Institute of Technology at Linköping University in partial fulfilment of the requirements for degree of Licentiate of Engineering. Department of Computer

1993/94 års debatt driver två uppfattningar där man dels menar att omvärlden är långt från stabil och att man ännu inte kan räkna bort hotet från det forna Sovjetunionen, dels

When investigating the relationship between turnover growth - ESG-score on a group basis, the small sample size led to the decision of us dividing the companies into three

This question is formulated with the aim of exploring the processes by which financial numbers produced and used for accounting purposes come to take on particular meanings

Ytterligare en skillnad är dock att deras studie även undersöker hur sentiment påverkar specifika aktie segment, det gör inte vår studie, vilket leder till att det

The control group for the interaction dummy variables is firms of ESG score A, during the same period shown by the time dummy variable.. 4.3.3

Fama and French (1992) found that the correlation between the B/M ratio and return performance had a positive sign, indicating that stocks with high B/M ratio (value

Therefore, the objective of this paper is to contribute to the literature on equity market integration through examining a relationship between the leading stock