ESG ratings and stock performance: Do high ESG portfolios perform differently within and between industries? M. Terpstra S3203980

(1)

1

ESG ratings and stock performance: Do high ESG portfolios perform

differently within and between industries?

M. Terpstra S3203980 prof. dr. L.J.R. Scholtens Supervisor Abstract

In this study, I investigate whether twenty industry portfolios consisting of the best ESG performers within an industry perform differently compared to the industry and other industries by estimating the alpha in the Fama French 5-factor model. In the main model, I find that the ESG portfolio performs significantly worse than the industry within two industries. However, after considering different criteria, I only find a significantly negative alpha in the Healthcare Services industry. Furthermore, two industry portfolios perform significantly different from the market ESG portfolio in the main regression. Nevertheless, this finding is not consistent when different criteria are considered. Therefore, in specific situations the relative ESG performance and industry can be considered for investment decisions. However, given the size of the alphas, the effect on performance will be marginal.

(2)

2

Introduction

In this thesis, I study the portfolio performance of portfolios consisting of the industry’s best CSR performers. A firm’s CSR performance is measured by the Environmental Social and Governance (ESG) pillars. Furthermore, I investigate whether these portfolios perform differently between the industries. Different perspectives are used to explain how and why CSR is able to influence the corporate financial performance (CFP) of a firm. In this study, the CFP is measured by the firm’s stock returns. First, the neo-classical view argues that a firm’s main goal is to maximize its profits. In order to maximize its profits, a firm invests in the right amount of CSR. Therefore, increasing investments in CSR related activities is a waste of resources for the shareholders, since it does not directly increase profits (Friedman, 1970). However, a different perspective on the effect of CSR on CFP is the stakeholder theory. This view argues that a firm should focus on the relationship with their internal and external stakeholders, and that a firm is able to maximize its profits by optimizing their relationships with the stakeholders. For example, Mackey et. al (2007) argue that the firm’s market value does not only depend on the present value of the cash flows but also on the relationships with stakeholders. Therefore, if a firm purely focuses on maximizing the present value of its cash flows, the firm is not maximizing its market value. The authors argue that a firm is able to maximize its market value by focusing on optimizing the relationships with the internal and external stakeholders, even if this leads to a lower present value of the cash flows. These two theories are the main starting points for studies investigating a potential relationship between CSR and CFP. From a neo-classical perspective additional investment in CSR eventually leads to losses and from a stakeholders’ perspective it creates gains.

(3)

3 If these results are valid, I would expect a larger value for a firm’s stock with a higher level of CSR when I use traditional valuation models like the dividend discount model (DDM) since the denominator becomes smaller if CSR increases and other components stay constant. These findings imply that there is a possible relationship between CSR and CFP. Furthermore, if there is a relationship between CSR and CFP, I would not expect it to be a uniform relationship between industries. Industries differ from each other in many perspectives (e.g., business strategies, competition, differentiation, impact on environment, stakeholder relationships, etc.), therefore a uniform effect is unlikely. Some studies confirm this statement. For example, Hull and Rothenberg (2008) find that CSR has a larger effect on return on assets (ROA) in undifferentiated industries and in industries with a lower degree of innovation. Furthermore, Hoepner et. al (2010) find that portfolios consisting of the 10 leaders in CSR of the 10 Global Industry Classification Standard (GICS) industries yield a significantly positive excess return for the health care industry and consumer discretionary industry. Hence, a uniform relationship between CSR and CFP is not present between industries.

In my study, I investigate whether the effect of CSR on CFP at the industry group level, which is one level lower than the 10 GICS sectors, is different within and between the industries. Hoepner et. al (2010) argue that there are significant differences between the industry characteristics within the 10 GICS business sectors. Therefore, investigating whether such effects exist on the industry group level might give an interesting additional insight on the results of previous written literature. Hence, the main research question is: Does a portfolio consisting of high CSR stocks perform differently within and between industries?

(4)

4

2. Literature review

2.1 Relationship CSR - CFP

Many papers have been written on the relationship between CSR performance and CFP. Friede et. al (2015) studied over 2000 empirical papers to aggregate the findings of these papers. The authors find that roughly 90% of the studies report a nonnegative relationship between the two subjects. This finding implies that the majority of these studies find a positive relationship between the CSR and a firm’s CFP, hence a firm with a better CSR performance would result in a higher CFP. Nevertheless, a single positive relationship between CSR and CFP is probably not a valid expectation. Berman and Rowley (2000) discuss in their paper that CSR performance is not a viable construct. Therefore, different measurements of CSR are present in the literature, hence interpreting and comparing results between different studies is difficult. Moreover, this argument implies that findings of studies vary if different approaches and measurements are used to measure a firm’s CSR performance. Barnett (2007) argues that the influence of CSR on CFP can vary over time since the financial outcomes of CSR investments vary over time, for example, due to changes in the influence of stakeholders.

Berman and Rowley’s argument and Barnett’s argument are indirectly confirmed by the results of empirical studies, which I discuss in the next section. These arguments imply that a relationship between CSR and CFP can be expected. However, it might be hard to compare findings with other studies as the data or methodology might be different from other studies. Moreover, results can also differ over time, which makes it difficult to compare results between different time periods.

2.2 Empirical findings

In this section, I discuss empirical studies and their findings on the relationship between CSR and CFP to construct hypotheses. I focus on the market based studies of CFP and CSR since these studies use similar methods and models compared to my study.

(5)

5 However, a few critical arguments can be made on these papers. First, Hoepner et. al (2010) mention that heterogeneity within the sector classes is likely, hence results might be driven by industries with an overall better performance compared to the sector. Secondly, even though the alphas are significant on a statistical level, the impact of CSR on CFP is marginal considered the size of the coefficients. Third, Kempf and Osthoff (2007) construct their portfolio using the best-in-class strategy for each of the ten Standard Industrial Classification (SIC) codes in order to construct a portfolio which is not biased towards one industry. Even though they do take into account that CSR performance might be different across these sectors, their strategy still does not take into account different effects within the industries itself.

A different approach to measure the effect of CSR on market based CFP is taken by Bauer et. al (2002) and Hartzmark and Sussman (2019). Both studies track the performance of ethical and conventional funds based on funds data provided by Morningstar in a Carhart (1997) four factor model. Neither find significant results, which implies that ethical funds do not outperform conventional funds in the datasets of the studies. However, the approach by the authors is probably not the best method to test whether ethical stocks perform better than conventional stocks since the performance of funds heavily depends on the characteristics of the fund manager (Chevalier & Ellison, 1999). Therefore, the fund performance of either ethical or conventional funds might be driven by different effects than the level of CSR these funds represent. Hence, it would be a better strategy to study the individual constituents of the funds separately.

Baird et. al (2012) use a linear mixed model (LMM) to investigate the relationship between CSR and the stock price for the 10 GICS industries. They find an overall negative effect of CSR on the natural logarithm of the stock price in their study of -0.029, which implies that the better performing CSR firms yield lower returns. Furthermore, in their extended model they use a product term between the GICS classification and the CSR variable. These product terms yield significant results for different industries with varying signs. Therefore, the effect of CSR on the stock performance of firms is different across industries.

(6)

6 In other studies, which are not directly mentioned in this paper, more databases for CSR and ESG related data are reported.

In the data section, I explain which database I use for my research and why that might influence the results and comparability of this study. Furthermore, the mentioned studies investigate different time periods and do not mention time fixed effects in their models. Therefore, the differences between results of these studies might be driven by the changing effect of CSR on CFP over time (Barnett, 2007). For example, more recent studies like Baird et. al (2012) and Hartzmark and Sussman (2019) provide insignificant or negative results compared to older studies like Kempf and Osthoff (2007).

A different effect of CSR on the stock performance of a firm can be expected across industries. For example, Cai et. al (2012) argue that investors demand a premium for what they label sinful industries1_{. They find that in the sinful industries a better CSR performance results} in a better CFP. Salo (2008) studies whether there is a correlation between corporate governance and environmental performance. Salo does not find significant correlation between these two factors. However, Salo finds that the disclosure of corporate governance and environmental performance is a common predictor for these two measures. Salo argues that if the degree of disclosure increases, firms will manage these areas closer and therefore increase the social performance. Different levels of CSR disclosure are present between industries (Ali et. al, 2017). The degree of disclosure within an industry could lead to a different effect of CSR on CFP across industries, because firms operating in industries with a lower degree of disclosure would have a lower level of CSR on average. Therefore, if a firm could increase its disclosure of CSR and thus increase its CSR performance, the impact would be larger in these industries compared to industries with a higher degree of disclosure. Hull and Rothenberg (2008) find that in industries which are more undifferentiated and have a lower degree of innovation the CSR has a larger effect on the ROA. Therefore, in these industries a different effect of CSR on CFP might be present compared to differentiated industries and higher degree of innovation.

As the neoclassical perspective and the stakeholder theory suggest a relationship is likely between the CSR performance and the CFP. However, literature does not support a uniform relationship between the CSR performance and CFP. Either positive, negative or insignificant results are reported. These findings can be caused by many effects and may result in different outcomes if other methodologies, data, models and other relevant research components are used. Furthermore, literature suggests that the relationship between CSR and CFP is not uniform across industries.

(7)

7 Therefore, to test whether the high CSR assets perform differently compared to the industry market, I construct the following hypothesis since the relationship and sign are ambiguous:

H0: High CSR stocks do not perform significantly different compared to the industry H1: High CSR stocks do perform significantly different compared to the industry

Many empirical studies investigating a relationship between CSR and CFP use industry fixed effects as one of the control variables in their model, which suggests that the relationship between CSR and CFP is not uniform between industries. Some studies investigating the relationship between CSR and CFP find that there is a significant different performance between industries (see Hoepner et. al, 2010; Baird et. al, 2012). In my study, I investigate whether the constructed industry ESG portfolios perform differently compared to each other. Therefore, I construct the following hypothesis to test whether there is a difference in performance between industry portfolios in my sample:

H0: There is no difference in portfolio performance between industries H1: There is a difference in portfolio performance between industries

3. Research method and data

3.1 Research method

For my research, I use a similar method as Kempf and Osthoff (2007) and Hoepner et. al (2010). I construct equally weighted portfolios consisting of the top 25 percent CSR performers within in an industry based on their average ESG ratings. I use the top 25 percent instead of a fixed number of observations since the number of firms across each industry is not similar. Furthermore, the ESG data is refreshed once every year, leading to a varying number of firms in the industry over the sample’s time period. To partially overcome this problem, I rebalance the portfolio every year according to the new ratings. This results in a varying number of constituents in the portfolio each year since some firms start to report data and some stop to report data; however, it does not lead to significant differences. After the portfolios of the best performers are formed, I use a Fama French 5-factor model to track their monthly performance compared to their related industry index of Thomson Reuters. I use this model since the standard CAPM and its extensions by Fama French (1993; 2015) and Carhart (1997) are still considered to explain most excess returns of stocks and portfolios. Furthermore, using this model eases the comparability of results with other studies since most of them use a similar model.

The model in general is:

(8)

8 In this model, Rpt-Rft measures the excess return of the ESG portfolio relative to the risk free rate Rft. In my research, α is the most interesting variable as it measures the performance of the ESG portfolio compared to the industry or market. If a significantly positive (negative) alpha is present for the CSR portfolios, the portfolio does perform better (worse) than the related industry index. The industry index is measured in the RMt variable, which is the return of the industry market a firm operates in when testing for different performance compared to the industry. For testing different performance between industries RMt is the overall market return.

The β1 measures the portfolio’s exposure to the market risk. The other considered variables are control variables. SMB, HML, CMA, and RMW are independent variables which control for the firm size (SMB, Small minus Big), value (HML, High minus Low), investment (CMA, Conservative minus Aggressive), and return (RMW, Robust minus Weak). The β2,3,4,5 are the estimated coefficients of the control variables. Significant coefficients for these betas imply that the portfolio’s excess return are exposed to the risk factors by the size of the coefficient. For example, if the size (SMB) factor increases by one percent the portfolios excess returns increase or decrease by the size of the coefficient. Therefore, these betas indirectly tell us the characteristics of our portfolio composition depending on the sign. A positive (negative) coefficient implies the first (second) part of the risk factor.

However, due to a high degree of correlation between RMW and HML (Fama and French, 2015) it would be better to drop one of the variables to increase the robustness of the model. In my model, I drop the HML factor since the HML factor does not improve the description of average returns since the return of HML is captured by the exposure of HML to other factors (Fama and French, 2015). Therefore, my model becomes.

Rpt−Rft=αit+β1(RMt−Rft)+β2SMBt+β3CMAt + β4RMWt + ϵit (2)

I use this model to test for difference within the industries. I add dummy variables to this model in order to test how the ESG portfolios perform compared to each other using a common benchmark after being fixed for industry effects. The model becomes:

Rpt−Rft=αit+β1(RMt−Rft)+β2SMBt+β3CMAt + β4RMWt + β5I1+ β6 I2 + β7 I3 + β8 I4 + β9 I5+

(9)

9 In formula (3) the I1, I2, …, I20 represent the dummy variables for the twenty industries in this study. For the estimation of the coefficients I use an OLS regression. In order to check whether this estimation can be used, I test for homoscedasticity2 and autocorrelation3.

The relevant tests show no sign of heteroscedasticity, however, autocorrelation is detected. Therefore, I chose to use the procedure of Newey and West (1994) in order to create autocorrelation and heteroscedasticity consistent standard errors.

3.2 Data

I use Thomson Reuters Eikon to collect financial data for firms and to compile the industries according to the TRBC classification. Furthermore, I use the Asset4 database by Thomson Reuters for the relevant ESG data. I use the Kenneth French data library4_{for the relevant} research factors. In my research, the selected time period is 2016-2019 and the selected market is the US market. The main reason for this time period and region is that the overall coverage of ESG data on the Asset4 database significantly increases from 2016 onwards compared to 2015. Moreover, in the US there is more ESG data available in this database compared to other regions, which helps to construct portfolios and industries of respectable sizes.

The TRBC consists of 32 business sectors5. Unfortunately, not every business sector is usable as some lack observations or miss relevant ESG and financial data. First, four business sectors were removed as there is no financial information available for these sectors, hence 28 remain. Subsequently, I removed business sectors which consist of less than ten observations per business sector and do not share similarities with other business sectors or consist of conglomerates and collective holding and investment companies, hence I am left with 24 business sectors. Finally, five of the 24 business sectors are relatively small and portfolios consisting of the top 25 percent firms would lead to a portfolio consisting of 6 firms or less. Therefore, I used the TRBC Economic sector of these industries and the correlation matrix in table A1 (appendix) to combine the industries into one industry. The Renewable Energy business sector showed a high degree of correlation with other Economic sectors compared to the Fossil Fuels sector. I do not believe that another Economic sector would explain the returns of the Renewable Energy sector better than the energy sector. Furthermore, the Renewable Energy sector has a different business strategy and environmental impact compared to the Fossil Fuels sector. Therefore, I decided to leave that sector on its own. I reduced the number of industries from 32 to 20 in order to create a dataset which is suitable for my research. In table A2. the industries of this study are summarized and labeled for the remainder of the paper, this table is presented in the appendix.

2_{White (1980) test for heteroscedasticity. The test values are mentioned in the appendix}

3_{Breusch-Godfried (1978) test for autocorrelation. The test values are mentioned in the appendix} 4_{https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html}

(10)

10 The process of removing and merging business sectors is visualized in table A3. In table 1, the descriptive statistics of each industry’s monthly ESG data are presented. The descriptive statistics of the industry specific top 25 percent ESG portfolios are summarized in table 3.

A few arguments on the data of this study, and studies on this topic in general, need to be discussed. Different aspects of the dataset might influence the results and complicate the comparison with findings from other studies. First, in my study I use the Asset4 database by Thomson Reuters for my ESG scores, since I do not have access to other sources of reliable ESG data. Many studies use different datasets and sources for the CSR or ESG performance of a company. These sources each use different measurements and components for the individual ESG pillars or CSR scores. This could lead to inconsistent and divergent ratings for firms across those sources. Secondly, the data coverage is still limited. In many industries which I investigated in this study, data availability increased significantly from 2016 onwards compared to earlier years. This can indirectly influence the findings of this study since it could imply that the ESG ratings are not considered (yet) by investors, consumers or the management given that the ratings are not widely available and are thus irrelevant for making decisions. Third, the industry classification can be a major factor for different results. In my study, I use the TRBC as industry classification because the Eikon database provides easier access to the relevant firm and industry data for this classification compared to other classifications. However, many studies use the GICS as classification system, which would lead to different business sectors and different firms within the business sectors. This complicates comparing my findings with findings of prior studies. Finally, the quality of ESG data can be a concern in this study and similar studies. The ESG data of the Asset4 database6 (and many other databases) rely on the non-financial disclosure of firms and their coverage in news reports and suchlike. Therefore, the ESG data is not independent and firms can indirectly influence their rating without necessarily increasing their actual performance in one or multiple pillars. This problem can be fixed if there exists a dataset which objectively measures a firms CSR performance without relying on their reports and news sources. Regarding these concerns, the findings of this paper should be considered with some degree of conservatism since many aspects of the data might drive the findings.

(11)

11

Table 1. Descriptive statistics per industry

This table contains the descriptive statistics for each industry’s average monthly ESG rating, furthermore in the last column the monthly ESG rating for the entire dataset is summarized. Each industry is labelled by a number, the corresponding industry for each number is presented in Table A2. Industries. For each industry this table contains the total number of firms in the entire dataset from 2016 till 2019, the average monthly ESG score of these industries, the median monthly ESG score for the industries, the standard deviation of the ESG score, the minimum and maximum monthly ESG score, the level of skewness and the level of kurtosis of the average monthly ESG scores.

(12)

12

Table 2. Summary statistics top 25 percent ESG portfolios.

This table contains the summary statistics of each industry´s top 25 percent portfolio per year. It contains the minimum ESG score, maximum ESG score, average ESG score, median ESG score, the standard deviation of the ESG score and the number of firms in the portfolio. The industries are mentioned in table A2. Industries.

Min Max Average

2016 2017 2018 2019 2016 Industry 1 47.1 45.6 46.0 48.6 Industry 2 56.6 54.3 52.5 54.9 Industry 3 49.9 49.0 52.5 46.2 Industry 4 43.7 45.8 50.8 49.0 Industry 5 34.3 32.8 40.5 31.0 Industry 6 33.5 33.1 33.6 31.6 Industry 7 43.0 40.5 45.0 38.8 Industry 8 43.4 47.7 48.9 39.0 Industry 9 62.2 63.0 61.3 61.8 Industry 10 58.1 51.5 42.0 46.8 Industry 11 55.1 52.6 51.2 47.1 Industry 12 64.5 62.1 63.1 56.3 Industry 13 29.7 29.5 31.2 32.4 Industry 14 38.1 37.0 37.9 41.6 Industry 15 44.1 48.9 46.6 53.1 Industry 16 52.6 49.7 52.0 46.3 Industry 17 46.4 47.9 50.7 41.0 Industry 18 57.1 60.5 62.9 50.1 Industry 19 49.3 41.5 48.7 49.3 Industry 20 50.2 55.1 54.6 56.3 2017 2018 2019 2016 2017 2018 2019 90.5 90.0 91.3 90.4 58.7 58.8 60.7 60.2 89.6 86.7 88.4 89.3 67.4 66.6 67.1 68.5 82.9 83.8 80.9 80.4 63.3 63.4 65.6 62.4 88.4 86.2 86.7 85.8 60.9 63.2 66.5 60.8 88.3 87.2 89.7 90.0 58.7 58.0 56.7 55.9 86.9 88.8 87.8 88.2 48.0 48.3 50.2 45.3 85.8 84.6 85.6 85.9 62.8 62.0 63.7 58.8 84.4 89.8 86.7 72.1 58.4 61.0 62.7 51.1 89.3 90.5 87.1 89.6 75.4 75.5 75.2 75.0 85.7 89.5 88.0 88.6 69.7 69.2 68.4 67.0 78.2 81.1 77.3 76.7 65.7 64.9 62.5 58.1 89.4 85.9 89.7 76.1 72.8 72.7 71.9 65.2 90.6 88.0 88.1 78.4 47.3 46.6 46.6 44.7 78.9 83.6 73.4 79.4 50.0 51.4 52.9 54.6 84.0 87.0 85.9 87.6 59.7 64.0 66.6 65.9 79.6 87.4 83.7 86.7 64.7 62.8 66.2 62.7 83.7 78.7 84.1 84.8 59.9 60.7 63.3 56.7 77.0 78.3 85.1 82.0 66.3 67.2 72.1 67.6 68.0 72.9 71.4 69.5 58.6 55.8 56.9 56.2 81.1 85.3 84.5 82.6 63.2 65.6 67.8 65.0

Median Standard deviation Number of firms

(13)

13

4. Results

4.1 Results within industries

This section provides the results of the regressions and the discussion of these results. The results of the comparison between the performance of the ESG portfolios and the industry portfolio are presented in table 3. In table 3, the reported coefficients are the respective betas of each variable from formula (2). The alpha, presented in percentages, measures the under or over performance of the ESG portfolio compared to the industry portfolio. The results in this table and other tables are the result from an OLS regression following the procedure of Newey and West (1984) in order to create heteroscedasticity and autocorrelation consistent standard errors. Furthermore, STATA is used for estimating the results of this study.

For most industries, I do not find a significant alpha in the regression from table 3. For a few industries I find a significant alpha, which implies that for these industries the ESG portfolio’s excess return of the risk free rate cannot be explained by one of the risk factors, and that the portfolio performs better or worse than the related industry index depending on its sign. For industry 7 (Healthcare Services) and industry 19 (Renewable Energy), I find a negative alpha, each significant at the one percent level. Therefore, I can reject the null hypotheses for these two industries of H0: High CSR stocks do not perform significantly different compared the industry. and accept the alternative hypothesis of H1: There is a difference in stock performance between business sectors at the one percent significance level.

(14)

14 Therefore, the social responsible investor should consider whether social responsible investing is more valuable than the average loss per month.

For the Renewable Energy, I find a negative alpha coefficient of -2.35 percent, similar to the Healthcare Services, this means that the ESG portfolio of this industry performed worse in this time period than the industry portfolio. Therefore, this portfolio provides -2.35 percent less monthly returns than expected based on the risk factors. Similar arguments for this industry can be made on the data as for the Healthcare Services industry. Nevertheless, comparing results for this industry with other studies is difficult, since this industry is not mentioned in the other papers. In the paper of Baird et. al (2012) there is no significant result for the GICS industry which is closest to the Renewable Energy industry. The industry consists of 16 firms and the portfolio consists of two or three firms. This could be an explanation why a significant result is present for this industry as the portfolio size is relatively small and the industry itself is small and volatile. In this study, I find a large significantly positive coefficient for the RMW factor, which implies that the returns of the ESG portfolio behave like the returns of highly profitable firms. Since this industry is a relatively new and booming industry it is not surprising that the portfolio shows such characteristics. Furthermore, this industry’s characteristic could explain why the performance of the ESG portfolio is significantly worse compared to the industry portfolio. For example, this new industry could be considered sustainable since it invests in “sustainable” and “green” energy compared to traditional energy markets. Therefore, the relatively high performance of ESG firms might not be that important since the industry itself is considered sustainable and additional investment in CSR might destroy firm value.

(15)

15

Table 3. Regression results top 25 percent portfolio

This table reports the results of the OLS regression from formula (2) with the corresponding Newey-West standard errors. The betas corresponding to each factor of the Fama-French model are reported for each industry. The coefficient for RMKT-RF is the market beta for the top 25 percent ESG portfolio to its corresponding industry index, the SMB, RMW and CMA coefficients are the betas for the control variables of size, profitability and investment. The alpha, reported in percentages, is the constant and the excess return of the portfolio which cannot be explained by one of the risk factors. The total number of observations is the number of months used for this regression. Each industry in this table is labelled in Table A2. On the next page the remainder of the industries is reported.

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Variables Industry 1 Industry 2 Industry 3 Industry 4 Industry 5 Industry 6 Industry 7 Industry 8 Industry 9 Industry 10

(16)

16

Table 3. continued

RMKT-RF 1.077*** 1.036*** 1.080*** 1.048*** 1.275*** 1.099*** 1.007*** 0.972*** 1.184*** 0.977*** (0.068) (0.038) (0.133) (0.042) (0.123) (0.063) (0.084) (0.049) (0.112) (0.044) SMB 0.016 0.059 0.913*** 0.423*** -0.117 0.362* 0.392*** 0.372** 0.264 0.486** (0.182) (0.078) (0.190) (0.143) (0.367) (0.201) (0.137) (0.166) (0.579) (0.227) RMW 0.234 0.040 -0.712* -0.142 -0.377 -0.253 0.466** 0.059 2.618*** 0.315 (0.228) (0.136) (0.406) (0.214) (0.548) (0.387) (0.204) (0.241) (0.947) (0.189) CMA -0.145 -0.034 -1.031*** -0.076 0.711 -0.049 -0.091 0.016 0.695 0.190 (0.175) (0.070) (0.204) (0.166) (0.425) (0.158) (0.107) (0.192) (0.544) (0.193) Alpha (%) -0.267 0.027 -0.006 -0.125 0.190 -0.424 -0.176 0.141 -2.352** -0.287 (0.245) (0.134) (0.359) (0.159) (0.703) (0.281) (0.242) (0.255) (0.892) (0.306) Observations 48 48 48 48 48 48 48 48 48 48 Adjusted R-squared 0.898 0.929 0.847 0.888 0.824 0.912 0.806 0.849 0.708 0.883

(17)

17

Table 4. Results top 25 ESG portfolios between industries

This table presents the coefficients of the OLS regression from formula (3). For this table RMKT-RF is the overall excess market return and the coefficient is the beta of the ESG portfolios, the SMB, RMW and CMA coefficients are the betas for the control variables of size, profitability and

investment. The alpha, presented in percentages, is the constant for the performance of ESG portfolios compared to the market portfolio. The industry specific dummy variables present the performance of this industry relative to the performance of the market ESG portfolio. A significant dummy implies that the industry ESG performs better (worse) if it has a positive (negative) sign compared to the market. The observations are the number of observations for each industry portfolio times the number of industries in this sample. Each industry in this table is labelled in Table A2.

VARIABLES Top 25 ESG portfolio

(18)

18 4.2 Results between industries

In table 4, the results of the regression from formula (3) between industries are presented. The main difference between table 3 and table 4 is that table 4 presents the relative performance of the ESG portfolios compared to the market ESG portfolio. The regression in this table is equal to the regression in table 4, except that the formula is different. The overall performance of high rated ESG stocks in this sample is slightly worse compared to the market return. I find a negative alpha of -0.62 percent, which is statistically significant at the ten percent level. This finding means that the aggregate monthly returns of the market’s ESG portfolio are -0.62 percent lower than the market portfolio. A significantly negative alpha for the entire sample is consistent with the findings of Baird et. al (2012), other studies either find a positive effect of CSR on stock performance or an insignificant effect of CSR on stock performance. However, the coefficient is relatively small. If I use the same example as before, I can show the impact of this finding. An investment of 100 Euro would result in a gain of 10 Euro for the market portfolio and a return of 9.94 for the ESG portfolio. Therefore, if an investor wants to invest in high rated ESG stocks, the investor’s expected return for the taken risk is slightly lower. If the investor considers social investing to be more valuable than the average loss, the investor should still make the investment. Two industry dummies are significant in the regression from table 4. Industry 15 (Mineral Resources) and Industry 19 (Renewable Energy) are significant in this regression, Mineral Resources is significant at the 10 percent level and Renewable Energy is significant at the five percent level.

For the Mineral Resource industry, I find a significantly positive coefficient of 0.93, this means that the assets attributed to this industry perform 0.93 percent better than the market ESG portfolio of the industries combined. One possible explanation for this result could be the paper of Cai et. al (2012), who find that investors overall demand a premium for industries that have the most impact on the environment and the community. Furthermore, the authors argue that a better CSR performance led to a higher CFP in the industries they investigated. The Mineral Resource industry has a major impact on the environment, hence investors would demand a premium for this industry according to Cai et. al (2012). The effect of this theory can explain the positive alpha for this industry in the regression from table 4. Since the performance of this industry’s ESG portfolio does not perform significantly better or worse to the industry portfolio, I can only conclude that this ESG portfolio performs better than the market ESG portfolio.

(19)

19 The Renewable Energy industry is a relatively young industry, hence the relative poor performance of this industry’s portfolio might be driven by the fact that firms tend to have a poor stock performance after an IPO for a couple of years (Ritter, 1991). The ESG portfolio of this industry performed significantly worse than the industry portfolio (table 3), which could be due to the fact that this industry can be considered “sustainable” and “green” in general. Therefore, the firm’s relative ESG performance could be irrelevant and value destroying.

Other industry dummy variables are not significantly different from zero. This implies that these industry ESG portfolios do not perform significantly different from the market ESG portfolio. Furthermore, using a Wald test for testing whether the industry dummy variables are equal to each other, I retrieve a F-value of 1.76 with a p-value of 0.023. Two of the industry dummy variables are significantly different from zero and the Wald test provides evidence for different coefficients. Therefore, I can reject the null hypothesis of the second hypothesis H0: There is no difference in portfolio performance between industries. and accept the alternative of H1: There is a difference in portfolio performance between industries at a 5 percent significance level. In the next section, I explain the sensitivity tests and the findings.

4.3 Sensitivity analysis

In the introduction and literature review several arguments are made on why the relationship of CSR on CFP might differ. For example, the time period, measurement and methodology might influence the findings. Empirical studies provide results which indirectly confirm these arguments. In my study, I use two sensitivity analyses to test whether different selection criteria might provide consistent or different findings.

First, I check whether the number of firms in the portfolio makes a difference. I used the top 25 percent performers of ESG in my research; however, as the descriptive statistics of the portfolios suggest there are still firms in the portfolio who perform below average if I use the description of the Thomson Reuters ESG scores7. Reducing the number of firms in the portfolio increases the average ESG rating of the portfolio and might reflect ‘the best’ performers better than the top 25 percent. Furthermore, other studies often use a smaller number of observations for the portfolio of the best CSR performers. Therefore, I decide to test whether a different finding is present if I use the top ten percent of an industry instead of the top 25 percent of an industry.

Another sensitivity test is testing whether the individual components of the ESG score affect the portfolio performance differently compared to the average scores. Bird et. al (2007) argue that investors consider the individual components as well as the entire CSR score in evaluating a firm’s CSR performance.

(20)

20 Multiple studies investigated the effect of one of the components of CSR, mostly environmental component, on CFP and find different results. For example, Makni et. al (2008) did not find significant results for the overall CSR score of firms and its effect on CFP; however, they did find that the environmental component of their CSR measurement yielded significantly negative results. Furthermore, studies find that the eco-efficiency score of firms positively affects the firms stock performance (Derwall et. al, 2005; Guenster et. al, 2011). Such effects might also be present for two other pillars (S and G). Therefore, I decide to test the impact on portfolio performance when the individual pillars ESG are considered.

The tables of the sensitivity tests are presented in the appendix. In Table A4, the results of the top ten percent portfolio are presented. For industry 19 (Renewable Energy), I do not find a significant result compared to table 3; nevertheless, this portfolio now consists of one firm due to the limited number of observations for this industry, which might explain the different result and the increased standard deviation. In table A4, I do find a significantly negative alpha at the ten percent level for industry 6 (Banking & Investment Services), which was not present in the regression of table 3. Therefore, for this industry the portfolio consisting of the top ten percent best ESG performers performs worse compared to the portfolio consisting of the top 25 percent best ESG performers in this sample and time period. The results imply that a better CSR performance does not increase the firms stock performance in this industry. Other industries still yield no significant differences between the industry portfolio and the top ten percent ESG portfolio. Therefore, the first hypothesis can only be rejected for the Healthcare Services industry if I consider the findings of table 4 and table A4.

Table A5, A6 and A7 provide the results of each of the ESG pillars separately. In these regressions, different findings for industries are present. I list them in order of the tables.

(21)

21 For the Chemicals & Applied Resources industry, I find a significantly positive alpha, which implies that this environmental portfolio does perform significantly better than the industry market and average ESG portfolio. Eco-efficiency positively affects a firm’s stock performance (Derwall et. al, 2005; Guenster et. al, 2011). Therefore, if the resource component of the environmental pillar represents some degree of eco-efficiency, the findings would be in line with this theory. However, the findings of these authors do not apply to other industries.

Table A6 provides the results for the social portfolio. This portfolio is constructed by selecting the top 25 percent performers of the Social pillar from each industry. The social pillar score measures the firm’s performance on the following matters: the workforce, human rights, community and product responsibility, benchmarked against the TRBC industry group7. Different alphas are found for industry 1 (Cyclical Consumer Products) and industry 16 (Energy - Fossil Fuels). Both industries yield significantly negative alphas, while they did have an insignificant alpha in table 3. This implies that for these industries the portfolio consisting of the top 25 percent best social performers performs significantly worse than the industry portfolio and the aggregate top 25 percent ESG portfolio. For the Fossil Fuels industry this finding is surprising if I look at the components of the Social pillar. This industry does have an impact on the community due to emissions and suchlike. Furthermore, the product responsibility could be considered an important issue in this industry, due to its impact on the environment and community. However, the other two components of this pillar, which have a larger weight in the Social score, might not be that relevant for the industry. Hence, the result might be affected by the weighting of the pillar components. For the Cyclical Consumer Products industry, the individual components might influence the satisfaction of the consumer. However, the effect on the business performance might be smaller since this industry’s financial performance particularly depends on the state of the economy. Therefore, a better social pillar performance might not be that relevant for the financial performance.

(22)

22 After performing multiple sensitivity tests, I find that for eleven of the twenty industries the conclusion is consistent with the conclusion from table 3. For one industry, the Renewable Energy industry, the sensitivity tests provide only insignificant alphas instead of the significantly negative alpha in the main regression. This finding could be present since the industry and related portfolios consist of very few firms. Furthermore, other unidentified industry and business factors could play a role, given that the adjusted R-squared varies significantly between the regressions. The factors are hard to identify since it is a relatively new industry and little is written on the industry. For eight of the twenty industries, either the portfolio’s size or constituents provide significant different results. These findings imply that the performance of social responsible portfolios depends on different ESG factors and varies per industry. The Health Care Services industry is the only industry with a significantly negative alpha for each test, hence for this industry each considered aspect of the ESG score influences the stock performance negatively.

Table A8 provides the results of the performance between the industry specific portfolios for each sensitivity check. In column 1 the findings of the top ten percent portfolio are presented, in column 2 the findings of the environmental portfolio are presented, in column 3 the findings of the social portfolio are presented, and in column 4 the findings of the governance portfolio are presented. Each portfolio is constructed by selecting the top ten percent ESG performers from an industry, or selecting the top 25 percent best performers from each one of the ESG pillars from an industry separately. The alpha in each column is negative; however, in contrast to the findings in table 4, the alpha is not significantly different from zero. Therefore, the constructed market ESG portfolios do not perform significantly worse compared to the market portfolio. For most industries, results from the sensitivity checks, in table A8, provide similar findings compared to the regression in table 4. However, a few differences can be noted.

(23)

23 In each column, no significant coefficient is observed for the Mineral Resources industry. Column 4, the governance portfolio, does provide a similar coefficient and finding compared to table 4. This implies that the combination of the pillars does have a larger impact on the stock performance than the individual pillars or that firms with a larger governance pillar score have a higher average ESG score in this industry.

Industry 18 (Chemicals & Applied Resources) provides one different result in column 2 compared to table 4. The industry did not have a significant coefficient in table 4, in this regression the coefficient of the environmental portfolio is significant at the ten percent level. The industry is one of the industries with a relatively large impact on the environment. Furthermore, eco-efficiency and CSR performance has a larger effect on the stock performance of a firm in sinful industries (Cai et. al, 2012). Therefore, firms from this industry might yield additional excess returns compared to other industries if the environmental performance increases. Hence, the assets attributed to this industry can perform significantly better than the market ESG portfolio if the environmental aspect is considered.

Industry 19 (Renewable Energy) reports a different finding in each column compared to table 4. I find a significantly negative coefficient in column 2. The coefficient has a smaller magnitude and lower level of significance compared to table 4. If assets are selected on their environmental score, the assets attributed to this industry perform significantly worse than the market portfolio. In table 4, I find that the top 25 percent ESG performers from this industry perform significantly worse compared to the market ESG portfolio. Since this industry has a low number of observations, the constituents of the portfolio might overlap during some stages within the time period. As the coefficients differ across every table, I do not expect the different portfolios to consist of the same firms during the investigated time period. The different findings compared to table 4 imply that the best average performers of this industry perform worse than the best performers selected on one of the individual pillars. This is a relatively new “green” industry, hence an individual pillar might have a larger influence on the CFP than the average ESG score since firms might differentiate due to the individual components of the pillars.

(24)

24

Conclusion

In this paper, I studied whether the portfolios consisting of the top 25 percent ESG firms within an industry perform differently than the industry or not. Furthermore, I studied whether these ESG portfolios perform differently compared to the market ESG portfolio or not, in order to compare their performance. In my research, I measured the relative performance of the ESG portfolios by estimating the alpha of the Fama French 5-factor models in formula (2) and (3). In eleven of the twenty industries, I did not find a significant different performance between the industry portfolio and the top 25 percent ESG portfolio. I did find significant alphas after performing sensitivity checks for eight of the remaining industries. For one industry, Healthcare Services, I did find consistent results between the main regression and sensitivity tests. Therefore, investing in ESG portfolios will not lead to a significantly different performance within most industries. Hence, a social responsible investor would not harm its portfolio performance if it constitutes its portfolio with high performing ESG stocks from most industries compared to the industry portfolio. Nevertheless, the social responsible investment would not yield additional excess returns compared to the industry portfolio, given the insignificant alphas. In two industries, Healthcare Services and Renewable Energy, the social investment would harm the performance of the portfolio since both industries yield significantly negative alphas. In the Renewable Energy industry, the significant alpha disappears in most sensitivity checks, which implies that different aspects of ESG should be considered when selecting stocks from this industry based on their ESG ratings. The size of the industry could also play a role given that the adjusted R-squared is relatively low. The overall market ESG portfolio does perform significantly worse than the market portfolio in the main regression. Therefore, constructing portfolios based on the ESG data leads to a lower return than expected based on the risk factors from model (3). However, this finding disappears in the sensitivity checks. Eighteen of the twenty industries do not perform significantly different from the market ESG portfolio. The Mineral Resources industry does perform slightly better compared to the market ESG portfolio and the Renewable Energy industry does perform slightly worse compared to the market ESG portfolio, given their significant coefficients. Therefore, a portfolio consisting of ESG stocks from the Mineral Resources industry would provide the highest return for an investor if the assets are compared to each other. Nevertheless, this finding is not consistent in every sensitivity check.

(25)

25 In most industries an insignificant alpha is present, which could imply that a firm invests in the right amount of CSR to maximize profits, which is part of the neoclassical view. Social responsible investing and related research is a relatively new and developing sight in finance. Many aspects might drive the results of research on this topic, which can be confirmed by comparing my findings to other papers. Hence, many research possibilities arise. I studied the effect of the business sectors according to the TRBC classification, other classifications could be used to find similarities or differences compared to my findings. I studied an equally weighted ESG score and the pillars separately. In the future, when the data availability and the reliability increases, it could be interesting to test whether a weighted ESG score or individual components of the ESG pillars provides different findings. Moreover, the social and political environment can influence the importance of CSR for firms in the future. For example, a government may demand, by law, a higher degree of CSR within certain industries or for firms in order to reach climate goals or other political agendas. Such research could help to answer the question when or how social responsible investing can be beneficial for the investor.

My study finds statistically significant results for the ESG portfolio consistently within one industry. Furthermore, I find that a significant difference in performance between industries is present. Nevertheless, this is not consistent in every sensitivity test. Therefore, the answer to the main research question “Does a portfolio consisting of high CSR stocks perform differently within and between industries?” is: Different performance of high CSR stocks is present within and between industries. For most industries there is no significant difference in performance within and between the industries. Many aspects (i.e. industries, selection criteria of high CSR stocks, methodology, etc.) can influence the relative performance of such portfolios, which is confirmed by the regressions, the sensitivity tests and other papers. Therefore, individual aspects of firms and CSR might lead to different portfolio performance hence a uniform answer is not available for this question. Consequently, not enough evidence is provided that social responsible investing will harm the portfolio performance within and between industries. Furthermore, if the alpha is significant, they are relatively small. Therefore, it should be encouraged if an investor considers ESG performance to be valuable.

References

Ali, W., Frynas, J. G., & Mahmood, Z. (2017). Determinants of corporate social responsibility (CSR) disclosure in developed and developing countries: A literature review. Corporate Social Responsibility and Environmental Management, 24(4), 273-294.

Baird, P. L., Geylani, P. C., & Roberts, J. A. (2012). Corporate social and financial

performance re-examined: Industry effects in a linear mixed model analysis. The Journal of Business Ethics, 109(3), 367-388.

(26)

26 Bauer, R., Koedijk, K., & Otten, R. (2005). International evidence on ethical mutual fund performance and investment style. Journal of Banking & Finance, 29(7), 1751-1767. Berman, S., Rowley, T. (2000) A Brand New Brand of Corporate Social Performance. Business & Society, 39 (4): 397-418.

Bird, R., Hall, A. D., Momentè, F., & Reggiani, F. (2007). What corporate social

responsibility activities are valued by the market?. The Journal of Business Ethics, 76(2), 189-206.

Breusch, T. S. (1978). Testing for autocorrelation in dynamic linear models. Australian Economic Papers, 17(31), 334-355.

Cai, Y., Jo, H., & Pan, C. (2012). Doing well while doing bad? CSR in controversial industry sectors. The Journal of Business Ethics, 108(4), 467-480.

Carhart, M. M. (1997). On persistence in mutual fund performance. The Journal of Finance, 52(1), 57-82.

Cheng, B., Ioannou, I., & Serafeim, G. (2014). Corporate social responsibility and access to finance. Strategic management journal, 35(1), 1-23.

Chevalier, J., & Ellison, G. (1999). Are some mutual fund managers better than others? Cross‐sectional patterns in behavior and performance. The Journal of Finance, 54(3), 875-899.

Derwall, J., Guenster, N., Bauer, R., & Koedijk, K. (2005). The eco-efficiency premium puzzle. Financial Analysts Journal, 61(2), 51-63.

El Ghoul, S., Guedhami, O., Kim, H., & Park, K. (2018). Corporate environmental responsibility and the cost of capital: International evidence. The Journal of Business Ethics, 149(2), 335-361.

Fama, E. F., & French, K. R. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics, 33(1), 3-56.

Fama, E. F., & French, K. R. (2015). A five-factor asset pricing model. Journal of Financial Economics, 116(1), 1-22.

Friede, G., Busch, T., & Bassen, A. (2015). ESG and financial performance: aggregated evidence from more than 2000 empirical studies. Journal of Sustainable Finance & Investment, 5(4), 210-233.

Friedman, M. (2007). The social responsibility of business is to increase its profits.

In Corporate Ethics and Corporate Governance (pp. 173-178). Springer, Berlin, Heidelberg. Godfrey, L. G. (1978). Testing against general autoregressive and moving average error models when the regressors include lagged dependent variables. Econometrica: Journal of the Econometric Society, 1293-1301.

(27)

27 Hartzmark, S. M., & Sussman, A. B. (2019). Do investors value sustainability? A natural experiment examining ranking and fund flows. The Journal of Finance, 74(6), 2789-2837. Hoepner, A. G., & Yu, P. S. (2010). Corporate social responsibility across industries: when can who do well by doing good?. Available at SSRN 1284703.

Hull, C. E., & Rothenberg, S. (2008). Firm performance: The interactions of corporate social performance with innovation and industry differentiation. Strategic Management

Journal, 29(7), 781-789.

Kempf, A., & Osthoff, P. (2007). The effect of socially responsible investing on portfolio performance. European Financial Management, 13(5), 908-922.

Mackey, A., Mackey, T. B., & Barney, J. B. (2007). Corporate social responsibility and firm performance: Investor preferences and corporate strategies. Academy of Management Review, 32(3), 817-835.

Makni, R., Francoeur, C., & Bellavance, F. (2009). Causality between corporate social performance and financial performance: Evidence from Canadian firms. The Journal of Business Ethics, 89(3), 409.

Newey, W. K., & West, K. D. (1994). Automatic lag selection in covariance matrix estimation. The Review of Economic Studies, 61(4), 631-653.

Ritter, J. R. (1991). The long‐run performance of initial public offerings. The Journal of Finance, 46(1), 3-27.

Salo, J. (2008) "Corporate Governance and Environmental Performance: Industry and Country Effects." Competition and Change, 12 (4): 328-354.

Sharfman, M. P., & Fernando, C. S. (2008). Environmental risk management and the cost of capital. Strategic Management Journal, 29(6), 569-592.

(28)

28

Appendix

1. White test for heteroscedasticity

This test is created by White (1980) in order to detect heteroscedasticity in the variance of the standard errors. This Lagrange multiplier (LM) test statistic is the product of the R2_{value and} the sample size. It follows a chi-squared distributions. The null hypothesis for the test is H0 = Homoscedasticity and the alternative is H0 = Unrestricted heteroscedasticity. The test is executed in STATA with the command imtest, white. The test statistic for my main regression is.

Since the p-value is above usual levels of statistical significance I cannot reject the null hypothesis and should conclude that the standard errors are homoscedastic.

2. Breusch-Godfried test for autocorrelation

The Breusch-Godfrey test (1978) is a test to detect autocorrelation in the errors in a regression. The null hypothesis of the test is H0 = There is no serial correlation of any order up to (p) and the alternative Ha = There is serial correlation of any order up to (p). In this test (p) is the number of lags used. The test is executed with STATA and the command estat bgodfrey, lags(p). I find that average returns of the portfolios show sign of autocorrelation. The test values are presented below

lags Chi^2 df P-value

1 3.870 1 0.049** 2 5.323 2 0.070* 3 5.348 3 0.148 4 6.676 4 0.154 5 6.716 5 0.243 6 7.373 6 0.288 *** p<0.01, ** p<0.05, * p<0.1

I find significant serial correlation up to lag two, therefore I reject the null hypothesis and need to adjust my standard errors in the regression in order to create autocorrelation consistent standard errors

Table A1. Correlation Matrix (next page)

In this table the correlation matrix of the ESG data of the sectors is presented before merging the sectors. This correlation matrix is used to combine the sectors. Each sector is combined by using the related Economic sector of the TRBC8.

8 https://www.refinitiv.com/content/dam/marketing/en_us/documents/quick-reference-guides/trbc-business-classification-quick-guide.pdf

Chi^2 df P-value

(29)

(30)

30

Table A2. Industries

This table presents each industry used in this study. Furthermore, it presents each label of how the industry is classified in the remaining tables of the paper. Each industry is based on the TRBC business sector and for industry 5, 9, 10, and 18 it based on its TRBC economic sector and the correlation matrix as these industries are combined industries due to lack of observations.

Label Industry

Industry 1= Cyclical Consumer Products

Industry 2= Technology Equipment

Industry 3= Industrial Goods

Industry 4= Real Estate

Industry 5= Software & IT &Telecommunications Services Industry 6= Banking & Investment Services

Industry 7= Healthcare Services

Industry 8= Industrial & Commercial Services Industry 9= Consumer Non-Cyclical Products & Services Industry 10= Retailers and Food & Drug Retailing

Industry 11= Transportation

Industry 12= Utilities

Industry 13= Pharmaceuticals & Medical Research

Industry 14= Insurance

Industry 15= Mineral Resources

Industry 16= Energy - Fossil Fuels

Industry 17= Cyclical Consumer Services Industry 18= Chemicals & Applied Resources

Industry 19= Renewable Energy

(31)

31

Table A3. Business sector overview

In this table the TRBC business sectors are identified in column 1 to create industries in column 2 for this paper. Furthermore, this table shows which business sectors are omitted or merged with another business sector. The red business sectors did not provide financial data and are removed. The yellow business sectors did have a low number of observations and did not share direct similarities with other business sectors. Finally, the green business sectors are merged with another business sector based on the correlation matrix A1 and the Economic sector.

(1) (2)

Business sectors Obs. Industry Obs.

Energy - Fossil Fuels 119

Renewable Energy 16

Uranium 1

Chemicals 56

Mineral Resources 47

Applied Resources 29

Cyclical Consumer Products

Industrial Goods 167 114

Industrial & Commercial Services 145 Technology Equipment 162

Transportation 66 Industrial Goods 167

Automobiles & Auto Parts 39 Real Estate 215

Cyclical Consumer Products 107 Software & IT &Telecommunications Services 259 Cyclical Consumer Services 136 Banking & Investment Services 395

Retailers 101 Healthcare Services 184

Food & Beverages 67 Industrial & Commercial Services 145

Personal & Household Products &

Services 23 Consumer Non-Cyclical Products & Services 88

Food & Drug Retailing 21 Retailers and Food & Drug Retailing 120

Consumer Goods Conglomerates 8 Transportation 66

Banking & Investment Services 407 Utilities 67

Insurance 88 Pharmaceuticals & Medical Research 304

Collective Investments 2 Insurance 84

Investment Holding Companies 1 Mineral Resources 47

Healthcare Services & Equipment 200 Energy - Fossil Fuels 115 Pharmaceuticals & Medical Research 384 Cyclical Consumer Services 132

Technology Equipment 164 Chemicals & Applied Resources 83

Software & IT Services 244 Renewable Energy 16

Financial Technology (Fintech) &

Infrastructure 0 Automobiles & Auto Parts 38

Telecommunications Services 29

Utilities 67

Real Estate 216

Institutions, Associations &

Organizations 0

Government Activity 0

(32)

32

Table A4. Regression results top 10 percent portfolio

This table reports the results of the OLS regression from formula (2) with the corresponding Newey-West standard errors. The betas corresponding to each factor of the Fama-French model are reported for each industry. The coefficient for RMKT-RF is the market beta for the top 10 percent ESG portfolio to its corresponding industry index, the SMB, RMW and CMA coefficients are the betas for the control variables of size, profitability and investment. The alpha, reported in percentages, is the constant and the mispricing of the portfolio which cannot be explained by one of the risk factors. The total number of observations is the number of months used for this regression. On the next page the remainder of the industries is reported. This table can be compared with table 3 in the main text.

(33)

33

Table A4. continued

RMKT-RF 1.122*** 0.996*** 0.988*** 1.063*** 1.624*** 1.095*** 1.168*** 0.940*** 1.208*** 0.824*** (0.085) (0.078) (0.084) (0.070) (0.186) (0.045) (0.089) (0.050) (0.158) (0.122) SMB -0.231 0.097 0.442** 0.426* -0.377 0.169 0.257 -0.116 -0.149 0.410 (0.269) (0.160) (0.164) (0.242) (0.583) (0.117) (0.229) (0.164) (0.970) (0.275) RMW -0.209 -0.033 0.231 -0.008 -0.275 -0.293 0.128 -0.261 3.038** -0.196 (0.288) (0.284) (0.192) (0.239) (0.706) (0.219) (0.259) (0.265) (1.236) (0.312) CMA -0.000 -0.184 -0.565*** -0.268 1.272* -0.131 -0.367* -0.242 0.636 0.299 (0.194) (0.145) (0.202) (0.221) (0.689) (0.133) (0.185) (0.200) (0.867) (0.233) Constant -0.076 0.055 -0.153 -0.472 0.902 -0.108 -0.133 0.446 -0.939 -0.216 (0.444) (0.287) (0.249) (0.319) (0.859) (0.211) (0.388) (0.336) (1.084) (0.456) Observations 48 48 48 48 48 48 48 48 48 48 Adjusted R-squared 0.829 0.728 0.841 0.799 0.775 0.958 0.730 0.788 0.553 0.746

(34)

34

Table A5. Results top 25 percent ENV. Performers

This table reports the results of the OLS regression from formula (2) with the corresponding Newey-West standard errors. The betas corresponding to each factor of the Fama-French model are reported for each industry. The coefficient for RMKT-RF is the market beta for the top 25 percent Environmental score portfolio to its corresponding industry index, the SMB, RMW and CMA coefficients are the betas for the control variables of size, profitability and investment. The alpha, reported in percentages, is the constant and the mispricing of the portfolio which cannot be explained by one of the risk factors. The total number of observations is the number of months used for this regression. On the next page the remainder of the industries is reported. This table can be compared with table 3 in the main text.

RMKT-RF 1.027*** 0.997*** 0.995*** 1.036*** 0.890*** 0.980*** 1.115*** 0.983*** 0.838*** 0.704*** (0.050) (0.049) (0.033) (0.035) (0.066) (0.021) (0.055) (0.072) (0.084) (0.066) SMB 0.512*** 0.362** 0.329** 0.356*** 0.161 0.307*** 0.163* 0.724*** 0.417*** 0.967*** (0.147) (0.167) (0.128) (0.117) (0.120) (0.076) (0.095) (0.243) (0.148) (0.210) RMW 0.278** -0.492* 0.167 -0.027 -0.329 0.286*** 0.265* 0.632*** 0.240 0.956*** (0.124) (0.283) (0.158) (0.138) (0.199) (0.104) (0.135) (0.227) (0.147) (0.323) CMA 0.187 0.196 0.195 0.058 0.266 -0.042 0.116 0.544** 0.142 0.635*** (0.161) (0.143) (0.131) (0.078) (0.165) (0.073) (0.133) (0.220) (0.139) (0.181) Alpha (%) -0.209 -0.164 0.071 -0.255* -0.157 -0.098 -0.798*** -0.259 -0.033 -0.562 (0.250) (0.325) (0.207) (0.131) (0.190) (0.125) (0.179) (0.245) (0.157) (0.503) Observations 48 48 48 48 48 48 48 48 48 48 Adjusted R-squared 0.914 0.890 0.915 0.932 0.848 0.976 0.935 0.828 0.724 0.689

(35)

35

Table A5. continued

RMKT-RF 1.067*** 1.014*** 1.037*** 1.034*** 1.175*** 1.077*** 1.043*** 0.977*** 1.203*** 1.068*** (0.075) (0.038) (0.107) (0.046) (0.092) (0.060) (0.094) (0.051) (0.126) (0.056) SMB -0.069 0.040 0.621*** 0.434*** -0.123 0.243 0.420*** 0.269* 0.371 0.169 (0.170) (0.086) (0.217) (0.136) (0.292) (0.157) (0.144) (0.159) (0.676) (0.261) RMW -0.186 0.092 0.008 -0.071 -0.115 -0.311 0.390** -0.018 2.115* 0.211 (0.233) (0.124) (0.307) (0.192) (0.426) (0.225) (0.184) (0.227) (1.206) (0.223) CMA -0.041 -0.044 -0.843*** -0.142 0.305 -0.267** -0.172 -0.043 0.631 -0.114 (0.135) (0.060) (0.128) (0.155) (0.341) (0.124) (0.107) (0.186) (0.596) (0.152) Alpha (%) -0.154 -0.078 -0.956*** -0.160 -0.380 -0.145 0.046 0.373* -1.830 -0.380 (0.284) (0.118) (0.326) (0.169) (0.658) (0.177) (0.260) (0.214) (1.167) (0.400) Observations 48 48 48 48 48 48 48 48 48 48 Adjusted R-squared 0.911 0.936 0.826 0.896 0.834 0.951 0.808 0.850 0.615 0.831

(36)

36

Table A6. Results top 25 percent SOC. Performers

This table reports the results of the OLS regression from formula (2) with the corresponding Newey-West standard errors. The betas corresponding to each factor of the Fama-French model are reported for each industry. The coefficient for RMKT-RF is the market beta for the top 25 percent Social score portfolio to its corresponding industry index, the SMB, RMW and CMA coefficients are the betas for the control variables of size, profitability and investment. The alpha, reported in percentages, is the constant and the mispricing of the portfolio which cannot be explained by one of the risk factors. The total number of observations is the number of months used for this regression. On the next page the remainder of the industries is reported. This table can be compared with table 3 in the main text.

RMKT-RF 0.971*** 0.938*** 0.968*** 1.067*** 0.870*** 0.962*** 1.053*** 0.945*** 0.780*** 0.788*** (0.068) (0.050) (0.035) (0.035) (0.069) (0.023) (0.051) (0.040) (0.072) (0.059) SMB 0.727*** 0.447* 0.433*** 0.284** 0.208* 0.258*** 0.196* 0.713*** 0.289* 0.826*** (0.190) (0.227) (0.126) (0.119) (0.117) (0.067) (0.113) (0.159) (0.149) (0.227) RMW 0.651*** -0.169 0.095 -0.060 -0.373 0.183* 0.182* 0.569*** 0.384** 0.775** (0.232) (0.287) (0.141) (0.125) (0.239) (0.102) (0.098) (0.193) (0.171) (0.375) CMA 0.077 0.360** 0.195** 0.058 0.171 -0.126 0.033 0.295 0.256* 0.774*** (0.146) (0.153) (0.095) (0.073) (0.178) (0.079) (0.122) (0.194) (0.135) (0.255) Alpha (%) -0.475** -0.176 -0.035 -0.133 -0.103 -0.091 -0.450*** -0.124 -0.058 -0.713 (0.197) (0.292) (0.186) (0.130) (0.203) (0.103) (0.156) (0.251) (0.151) (0.475) Observations 48 48 48 48 48 48 48 48 48 48 Adjusted R-squared 0.889 0.870 0.931 0.942 0.844 0.978 0.943 0.837 0.700 0.710