Exploratory Research into Industry Characteristics and Their Effects on Sustainability Practices: Evidence from Different Industries

Exploratory Research into Industry Characteristics and

Their Effects on Sustainability Practices: Evidence from

Different Industries

June 12th 2019

By Manon Mulder (S2729024) University of Groningen Faculty of Economics and Business

m.mulder.22@student.rug.nl Supervised by prof. dr. L.J.R. Scholtens

14.744 words

Abstract

This study is an exploratory research into economic and business characteristics that determine industry classification and their influence on social performance. The goal is to delve deeper into industry determination and to investigate how industry classification relates to sustainability practices. In previous CSR literature, industry is a concept that is assumed to be common knowledge, but it is ambiguous what factors in fact determine industry allocation. I look at economic and business characteristics, such as operational, production and performance indicators instead of merely taking classification as a perspective for CSR. Using a sample of 2328 US publicly listed companies for which environmental, social and governance information is available at the end of fiscal year 2017 from 69 industries, I find that the extent of raw material usage disclosure by firms best explains industry differences, followed by the level of competition, the level of internationalisation, ROA, firm age, revenues and raw material usage intensity. Moreover, industries that disclose information about raw material usage, face low levels of competition, are internationally oriented, have better financial performance in terms of ROA, are older, have low revenues and rely on raw materials have higher social performance compared to other industries. Looking at the conventional sector classification, which is mainly based on revenues and earnings, these operational, production and performance indicators are much more satisfying and better performing in explaining the 69 industries and in addressing differences in sustainability practices.

1

1. Introduction

In the last half century, attention for corporate social performance (CSP) increased considerably. A growing number of firms across various industries voluntarily adopt sustainability practices to make good business and financial sense as well. Many studies control for industry when studying the link between corporate social performance (CSP) and corporate financial performance (CFP), either by analysing one particular industry (e.g. Ogden and Watson, 1999; Simpson and Kohers, 2002) or by treating industry as a moderating variable (e.g. Baron et al., 2009; Hull and Rothenberg, 2008). This seems warranted because of the evidence. Alternatively, Heal (2005) investigates CSR practices across industries from a resource-allocation perspective. He argues that private-social cost differentials in cases of market failure determine the resource-allocation for CSR programs. In sectors where the private-social cost differential is small, CSR initiatives do not play a significant role, while in sectors with large private-social cost differentials, CSP can play a valuable role in producing a social good. Furthermore, Ioannou and Serafeim (2017) investigate the adoption of sustainability practices over time and find that sustainability practices converge within an industry. The extent of convergence across industries is related to the implementation of CSR by the industry’s market leaders and the relative importance of environmental and social issues compared to governance issues.

But what economic factors and firm characteristics differentiate for example the technology sector from the tobacco, oil or car industry? And how does this relate to CSR? Naturally, production processes differ between industries, but this is not one-on-one with leverage, profitability and size. Firms in the tobacco industry earn money by producing and/or selling an addictive poison that is slowly killing their customers. Firms selling cars, or providing oil to car owners contribute to freedom and independence in terms of personal mobility, but they are stimulating environmental pollution and climate change at the same time. Not to mention cars causing traffic accidents. And what about the technology sector? On the basis of existing research, it is clear that CSR varies widely between sectors and that, therefore, the interaction between CFP and CSR varies as well. The aim of this study is to focus more on industry determination and to investigate which industry characteristics are associated with differences in CSR implementation.

2

concept that is assumed to be common knowledge, but the actual determination of industry classification is neglected. Presumably because industry allocation is based on non-financial information (e.g. production process and final product) and is, therefore, difficult to quantify. The Global Industry Classification Standard (GICS) grounds its industry classification on a company’s principal business activity. Specifically, revenues and earnings play an important role, as well as market perception as revealed by investment research reports. However, using financial factors alone to explain variation in sustainability practices is fairly limited as it is quite disappointing and very poorly performing in explaining the 69 industries. Therefore, this study will investigate the economic and business characteristics, such as operational, production and performance indicators instead of merely taking classification as a perspective for CSR. Especially, I will look at firm characteristics such as firm age, risk, size, liquidity, tangibility, the degree of internationalisation and the level of competition, and into the production process of firms, particularly to the production input function and relate these to CSR.

To the best of my knowledge, this research is the first that decomposes industry classification into the underlying driving factors and empirically examines the effects of industry determinants on sustainability practices instead of only considering the role of industry in general. Developing this understanding is meaningful, for three main reasons. Firstly, the accelerating amount of companies that issue corporate sustainability reports indicates that for companies across different industries sustainability has become a central practice. According to KPMG (2017), 93% of the world’s largest 250 companies report on CSR and 78% include information on sustainability practices in the annual financial reports. Secondly, the findings may provide interesting and important practical implications. Prior studies on CSR have debated on whether firms should implement CSR programs (Griffin and Mahon, 1997, Orlitzky and Benjamin, 2001). This study suggests that firms should consider firm characteristics that are associated with industry determination when deciding on the relevance and implementation of CSR programs. Thirdly, a rich literature about the importance of sustainability practices and its impact on financial performance exist in which the role of industry is underexposed (see Margolis and Walsh, 2001 or Aguinis and Glavas, 2012 for an overview of the literature) or assumed to be common knowledge. Therefore, this study aims to clarify the concept ‘industry’ and provides insights into the underlying factors that determine industry classification and their effect on sustainability practices.

3

Reuters Eikon, which is a comprehensive company fundamentals database with industry-specific estimates. A regression model is used to identify which firm characteristics determine industry classification and to explain differences in sustainability practices across industries. The results indicate that the extent of raw material usage disclosure by firms best explains industry differences, followed by the level of competition, the level of internationalisation, ROA, firm age and revenues. Industries that disclose information about raw material usage, face low levels of competition, are internationally oriented, have better financial performance in terms of ROA, are older, have low revenues and rely on raw materials seem to have better social performance. Compared to the conventional sector classification, these operational, production and performance indicators are much more satisfying and better performing in explaining the 69 industries and in addressing differences in sustainability practices.

The remainder of this study is organized as follows. Section 2 reviews the literature and presents the hypotheses, section 3 describes the model, the data sources, the variables and the research method that will be used to test the hypotheses, section 4 summarizes the findings, and section 5 set forth the conclusions.

2. Literature review

In this section I discuss empirical findings, the classification guidelines of GICS and some industry determinants that may influence sustainability practices. CSR can be defined as the internalization of external effects in the strategy and conduct of a firm (Heal, 2005).

Sustainability practices and the role of industry

4

industries (Derwall et al., 2005; Semenova and Hassel, 2008). Furthermore, the industrial level of product/service differentiation may also influence the CSP-CFP relation, because CSR can be employed as a differentiation strategy (McWilliams and Siegel, 2001; Chih et al., 2010; Fernández-Kranz and Santaló, 2007). However, Hull and Rothenberg (2008) state that only industries with less product/service differentiation benefit from CSR programs.

The main hypothesis of this research is of an exploratory nature. On the basis of existing research it is clear that CSR varies widely between sectors and that, therefore, the interaction between CFP and CSP also differs. Assuming that this also applies to my sample, I research industry characteristics and their effects on sustainability practices.

H0: sustainability practices are homogeneous across industries H1: sustainability practices are heterogeneous across industries

To study industry characteristics, I look at economic and business characteristics, such as operational, production and performance indicators instead of merely taking classification as a perspective for CSR. I first investigate the sector definition. Then I look at firm characteristics that may influence industry classification. As the GICS allocates firms into industries based on its principal business activity, I first look at the production process of firms and relate this to CSR. Next, I consider other firm characteristics that may determine industry classification and can be associated with variation in CSR, in addition to the common control variables in the meta-analysis of 95 studies on the CSP-CFP relation, such as size, age and risk (Margolis and Walsh, 2001).

Industry definition

The goal of this research is to focus on industry determination and to investigate which industry characteristics can be associated with differences in CSR implementation between industries. The first question to address: what is an industry? An industry exists to serve a market. The market is the group of customers who require the products or services provided by an industry. Therefore, industry can be seen as a component of a market.

5

policy requires a broader definition of a market than the analysis of a competitive effect of a merger between two coal suppliers (Tirole, 1988). This also holds for an industry.

GICS touches upon this issue, by using 11 sectors, 24 industry groups, 69 industries and 158 sub-industries, in which the definition of ‘the industry’ becomes narrower and more precise. GICS strives to provide an industry classification that is correct, complete and provides a durable view of the global investment universe from an industrial perspective (S&P Global and MSCI, 2018). The GICS methodology assigns each company to a sub-industry, according to the definition of its principal business activity. GICS uses revenue as the key factor in determining a company’s principal business activity along with earnings and market perception, as revealed by investment research reports. When a company is active in two or more substantially different business lines, none of which contributes to 60% or more of revenues, the company is classified in the industry that provides the majority of its earnings. Lastly, the industry classification will be determined based on market perception, requiring professional judgement (S&P Global and MSCI, 2018).

This research will consider revenues and earnings as key industry determinants. Companies with higher revenues and earnings have more means to invest in CSR than firms with low revenues and earnings. Campbell (2007) argues that firms in weak financial circumstances are less likely to perform CSR. Furthermore, attention to non-financial stakeholders who might value CSR is limited when firms face scarce resources (Watson, 2015). Therefore, I expect firm’s with higher revenues and earnings to have higher ESG scores (hypotheses 1 and 2).

Hypothesis 1: firms with higher revenues have higher ESG scores than firms with lower revenues

Hypothesis 2: firms with higher earnings have higher ESG scores than firms with lower earnings

Although revenues and earnings are GICS’ main classification measures, investigating financial factors alone to determine the principal business activity is fairly limited as it insufficiently explains the 69 industries. Therefore, I also study firms’ principal business activity from a non-financial perspective, especially I look at the production function of a firm.

The production function

6

is given by the production function of the form: q = f (k, l, m), where q represents a company’s output of a particular good during a period, k represents capital usage during the period, l stands for the hours of labor input and m represents raw materials used.

According to Aigner and Chu (1968), the production function sets the highest possible output level with a given combination of input factors at a given state of technical knowledge during a period. This applies not only to a particular firm, but the maximum attainable output level holds for all firms in a particular industry. Due to differences in technical efficiency, economic efficiency and labor input, not all firms are able to obtain the maximum output level given by the industry production frontier (Aigner and Chu, 1968). Although the exact amount of capital, labor and raw material inputs differs per firm, these factors can be considered as industry determinants.

Labor-intensive firms may use CSR as a means to improve employee satisfaction (Flammer and Luo, 2017), increasing labor productivity and sales growth (Flammer, 2015). Therefore, I expect labor-intensive firms to have higher social scores than less labor-intensive firms (hypothesis 3).

Looking at the opponent of labor-intensity, capital-intensity, Arora and Dharwadkar (2011) state that capital intensity places limitations on managerial discretion. Furthermore, according to Finkelstein and Boyd (1998), capital intensity creates rigidity in organizations, making it more difficult to adapt sustainability practices. Therefore, I expect capital-intensive firms to have lower environmental and social scores than less capital-intensive firms (hypothesis 4).

Lastly, material intensive firms heavily rely on material suppliers and raw-material suppliers are considered to be the most irresponsible actors along the supply chain (Tencati et al., 2008). The success of CSR implementation depends on the adherence of raw-materials suppliers to CSR standards (Lech, 2013). Given the dependence on raw-material suppliers, I expect raw-material intensive firms to have lower environmental scores than firms relying less on raw-materials (hypothesis 5).

7

Therefore, I expect firms that disclose information about water and energy usage to have higher environmental scores than firms that do not disclose this information (hypothesis 6).

Hypothesis 3: labor-intensive firms have higher social scores than firms with lower levels of labor-intensity.

Hypothesis 4: capital-intensive firms have lower environmental and social scores than firms with lower levels of capital-intensity

Hypothesis 5: raw-material intensive firms have lower environmental scores than firms which rely less on raw-materials.

Hypothesis 6: firms that disclose information about water and energy usage have higher environmental scores than firms that do not disclose this information

Additionally to the firm’s production characteristics, there may be more firm characteristics that determine industry classification and can be associated with variation in CSR. For example, size, age and risk are commonly used control variables in the meta-analysis of 95 studies on the CSP-CFP relation of Margolis and Walsh (2001). Next to size, age and risk, I exploratory investigate other business characteristics, such as liquidity, tangibility, the degree of internationalization and the level of competition and relate these to CSR.

Firm characteristics

According to Porter (1979), the theory of industrial organization has viewed industries as homogeneous units. Businesses in a specific sector are assumed to be ‘similar’ in all dimensions (Porter, 1979). Therefore, in addition to the production inputs mentioned above, firm characteristics, such as firm age, size, risk, liquidity, tangibility, the degree of internationalisation and the level of competition can be viewed as industry determinants that may influence sustainability practices.

Firstly, large firms have more means to invest in social responsibility than small firms (Gupta, 1969) and as they mature and grow, large firms attract more attention (Clark et al., 2008) and increasingly need to meet stakeholders’ demands (Burke et al., 1986). Similarly, CSR rating agencies have the tendency to concentrate most on the largest companies (Schäfer et al., 2006). Hence, the need for larger companies to engage in CSR is likely to differ from smaller, less publicly watched companies. Therefore, I expect larger firms to have higher ESG scores than small firms (hypothesis 7).

8

predictable performance and cash flows and may be better able to afford CSR. Peloza (2006) finds that, in the development stage, younger firms can use CSR programs to differentiate from competitors. However, due to the unpredictability of cash flows of younger firms, they do not always have the means to invest in CSR. Therefore, I expect mature firms to have higher ESG scores than younger firms (hypothesis 8).

Thirdly, I expect firm risk to influence sustainability practices. High-risk companies do not face negative financial consequences when implementing CSR initiatives, because a market reaction will stay out in the fear of increasing the firm’s risk position further (Orlitzky and Benjamin, 2001). Furthermore, Oikonomou et al. (2012) find that CSP reduces financial risk, which is valuable for shareholders (Smith and Stultz, 1985; Stultz, 2002). Given these findings, I expect high-risk companies to have higher ESG scores than low-risk companies (hypothesis 9).

Furthermore, I expect the level of liquidity to influence sustainability practices. According Casey and Grenier (2014), low-liquidity firms may engage in sustainability practices to reduce information asymmetries and to promote or restore confidence with investors to attract new capital. However, as CSR is costly, liquidity may be a constraining factor influencing a firm’s decisions on whether to act responsible. Dhaliwal et al. (2011) finds that low-liquidity firms do not have the financial means to invest in sustainability practices. Furthermore, Chan et al. (2017) finds that, in general, illiquid firms do not engage in any CSR activities. Therefore, I expect high-liquidity firms to have higher ESG scores than low-liquidity firms (hypothesis 10).

A firm’s level of tangible assets may also influence sustainability practices. Verwijmeren and Derwall (2010) state that firm’s with more tangible assets can use these assets as collateral, which will increase debt ratios. Firms with higher debt ratios have limited capacity to perform sustainability practices due to the crowding-out effect (McWilliams and Siegel, 2000). Therefore, I expect firms with more tangible assets to have lower ESG scores than firms with less tangible assets (hypothesis 11).

9

more is at stake, therefore, I expect international firms to have higher social scores than national firms (hypothesis 12).

Lastly, the level of competition is expected to influence sustainability practices. Sustainability practices strengthen a firm’s competitive position by differentiating it from rivals (Hull & Rothenberg, 2008; McWilliams & Siegel, 2001). Therefore, I expect firms facing high competition to have higher ESG scores than firms in less competitive environments (hypothesis 13).

Hypothesis 7: larger firms have higher ESG scores than small firms Hypothesis 8: mature firms have higher ESG scores than younger firms

Hypothesis 9: high-risk companies have higher ESG scores than low-risk companies Hypothesis 10: high liquidity firms to have higher ESG scores than low liquidity firms Hypothesis 11: firms with more tangible assets have lower ESG scores than firms with less tangible assets

Hypothesis 12: international firms have higher social scores than national firms

Hypothesis 13: firms in a more competitive-environment have higher ESG scores than firms in less competitive environments

3. Model, Data and Methods

To test the hypotheses, a cross-sectional study is performed, which is an observational study that analyses data on one or more variables collected at a single point in time (in this study; end of fiscal year 2017). In this section I first discuss the generic model. Secondly, I introduce the data and measures. Then, I provide a more specified model that is used to explain differences in CSR across industries. Finally, I discuss the methods used to estimate the models.

Model

In order to test the hypotheses, a regression model is used to identify which economic and business characteristics explain differences in sustainability practices across industries. The ordinary least squares (OLS) regression model is used as it appears to be the mostly used research method to investigate the social and financial performance relation (Margolis and Walsh, 2001; Orlitzky and Benjamin, 2001; Wang et al., 2015).

10

production and performance indicators instead of merely taking classification as a perspective for sustainability practices. It appears that financial performance precedes social performance in more cases than the other way around (Preston and O’Bannon, 1997; Scholtens, 2008). Good financial performance allows firms to make investments that improve social performance. Instead of testing certain theories (e.g. precedence), this study is based on a pragmatic exploratory model to analyze industry differences in CSR. The generic model of interest is: CSPi = f(CFPi, X), where:

- CSPi is a measure of firm i’s socially responsible performance, which includes the overall ESG score, the social score, the environmental score and the governance score - CFPi is a measure of firm i’s financial performance, which includes ROA and excess

return

- X is a vector of control variables, which includes firm i’s economic and business characteristics and the industry in which firm i operates.

I assume the relationship between CSP, CFP and firm characteristics to be linear, because linearity is typically assumed in CSR literature (Guney and Schilke, 2010). Among the firm-level control factors to be included in the X vector are measures of revenues, earnings, capital intensity, labour intensity and raw-material intensity, size, age, risk, liquidity, tangibility, internationalization and competition as specified in the literature.

Data

Data to estimate the model of interest and to test the hypotheses are taken from Thomson Reuters Eikon, which is a comprehensive company fundamentals database with industry-specific estimates. The focus of this study is on publicly listed companies that published environmental, social and governance (ESG) information at the end of fiscal year 2017, are distributed across 69 industries of which the GICS industry name is known and are located in the United States (US) based on the measure ‘primary country of risk’. This yields a number of 2328 firms. Next, these firms are linked to return, revenues, earnings, total assets, risk, production input variables and other company characteristics. To analyze the hypotheses, the best available proxy is chosen from Thomson Reuters Eikon.

11

2019). Thomson Reuters ESG Scores enhance and replace ASSET4® _{equal weighted ratings} and are a robust, data-driven assessment of companies’ performance on ESG metrics where company size and transparency biases are reduced to an acceptable minimum (Thomson Reuters, 2019). Recently, many other studies used Thomson Reuters Eikon for CSR research, such as Garcia et al. (2017) and Miralles-Quirós et al. (2018). CSP is measured by Thomson Reuters’ overall ESG score. The ESG score is an overall company score which measures relative environmental, social and corporate governance performance, commitment and effectiveness across 10 main dimensions (emissions, environmental production innovation, human rights, shareholders, etcetera) based on company-reported data in the public domain (Thomson Reuters, 2019).

Aggregating all ESG information into one combined CSP measure is problematic, because it may mask the individual pillars that are equally important and relevant as well as interesting information may disappear. Furthermore, the composing pillars (environment, social and governance) are imperfectly correlated. Therefore, this study will also consider the individual dimensions (environment, social and governance) separately. Although, the individual dimensions are also aggregates of various dimensions, it gives a more detailed picture of the interests of different stakeholder groups.

The environmental score is based on performance regarding resource use, emissions and innovation. The social score consists of information about workforce, human rights, community and product responsibility (Thomson Reuters, 2019). Lastly, the governance score is based on information about relations with management and shareholders, and the CSR strategy. The environment pillar score measures a company’s impact on living and non-living natural systems, including the air, land and water. The social pillar score measures a company’s capacity to generate trust and loyalty with its workforce, customer and society, through its use of best management practices. The corporate governance pillar measures a company’s systems and processes, which ensure that its board members and executives act in the best interests of its long term shareholders (Thomson Reuters, 2019). A detailed variable description about the CSP measures can be found in appendix A.1.

12

measures (Branch, 1983; Briloff, 1972). However, market-based measures may not reflect fair evaluation from investors in the case of asymmetric information (Scholtens, 2008) and relates only to financial stakeholders, ignoring non-financial stakeholders (McWilliams et al., 2006). Both measures have weaknesses, therefore, I use two profitability ratios: return on assets (ROA) as an accounting-based measure of CFP and excess stock market return as a market-based measure of CFP.

Both measures are taken from Thomson Reuters Eikon. ROA measures a company’s operating efficiency regardless of its financial structure and is calculated as income before tax for the fiscal period divided by the average total assets for the same period and is expressed as percentage. Two variables for the excess stock market return are calculated. One using the S&P 500 index and one using the S&P 500 sector index. The S&P 500index is the best benchmark of the US stock market since its release in 1957 (Arouri et al., 2011). The index is broken down into eleven sectors according to the global industry classification standard (GICS). The S&P500 sector indices are reviewed on a regular basis and show the state of various market sectors and industries. Based on the performance of these indices the makeup of the S&P 500 is determined (The Balance, 2019). The excess returns used in the analysis are calculated by subtracting the S&P 500 (sector) index of 2017 (Novel Investor, 2019) from the total return of the firm. According to Thomson Reuters Eikon, the total return incorporates the price change and any relevant dividends for the specified period and is calculated by using the compounded daily return (dividend reinvested total return methodology). A detailed variable description about the financial performance measures can be found in appendix A.1.

13

Revenues and earnings are considered as key industry determinants by the GICS. To measure revenues, a revenue ratio is calculated by dividing total revenues by total assets. Total revenue represents revenue from all of a company's operating activities after deducting any sales adjustments and their equivalents. Total assets are the total assets of a company. The Banks and Thrifts & Mortgage Finance industry (industry 23) does not report on revenues. For this industry, revenues is excluded from the analysis. To measure earnings, an earnings ratio is calculated by dividing earnings before interest and taxes (EBIT) by total assets. EBIT is computed as total revenues for the fiscal year minus total operating expenses plus operating interest expense, unusual expense/income and non-recurring items, and supplements for the same period (it excludes non-operating income and expenses).

The inputs of the production function are capital, labor and raw-materials. Capital is measured by the variable total capital, which is the sum of total equity, total debt and minority interest as of the end of the fiscal period. Furthermore, the variable labor and related expense is used to measure labor input. Labor and related expense consists of expenses paid to employees of a company in the form of salaries, wages, fees, benefits or any other form of compensation. Lastly, raw material inventory is used as a proxy for raw-material input. Raw material inventory represents raw materials acquired, but not yet used. It may also include raw materials in transit. I am aware that raw material inventory represents a stock variable, while the other variables are flow variables, but this is the best proxy available at Thomson Reuters that can represent raw material input. To measure the relative importance of the input factors, intensity variables are calculated. Capital intensity is calculated as, total capital divided by the sum of total capital, labor and related expenses, and inventories raw materials, multiplied by 100 (to get a percentage). Labor intensity and raw material intensity are calculated similarly, by adjusting the numerator accordingly.

Then, to measure transparency about water and energy usage, the variable ‘water and energy disclosure’ is calculated. It states "1" if company measures water use to revenues or total energy use to revenues and "0" if company does not report on water and energy usage. Water use to revenues is the total water withdrawal in cubic meters divided by net sales or revenue in US dollars. Total energy use to revenues is the total direct and indirect energy consumption in gigajoules divided by net sales or revenue in US dollars. 313 firms in the sample reported about energy and water usage, 2016 companies did not.

14

stock moves for a given move in the market. It is the covariance of the security's price movement in relation to the market's price movement. Firm size is measured by the total assets of a company and firm age is calculated by subtracting the organization founded year from the current year (2019), assuming no mergers and acquisitions have taken place.

I also exploratory investigate other business characteristics, such as liquidity, tangibility, the degree of internationalization and the level of competition and relate these to CSR. The liquidity ratio is calculated by dividing the amount of cash and short term investments over total asset. Cash and short term investments are the sum of cash, cash and equivalents and short-term investments. To measure tangibility, first the amount of tangible assets are calculated by subtracting the intangible assets from total assets. Intangible assets are the gross intangibles reduced by accumulated intangible amortization. Net intangibles are used when the company does not provide gross intangibles. To measure tangibility as a ratio, the tangible assets are divided by total assets.

To measure the degree of internationalisation, domestic taxes paid are divided by total taxes paid. Domestic (current) taxes reflects the portion of a company's current income tax provision attributable to its domestic tax jurisdiction. Total (current) taxes reflects the sum of current tax domestic, current tax foreign, current tax local and current tax others. Preferably, I would use foreign sales over total sales to measure the level of internationalisation, but Thomson Reuters Eikon has no information available regarding foreign sales. However, as taxes are based on income, this proxy is the best available. I am aware of country-level differences in corporate statutory tax rate regimes in the US (Newberry and Dhaliwal, 2001), but this should not matter for the ratio calculation.

As a proxy for the level of competition, the market share of a company is used. A low market share reflect the absence of market power and are generally associated with high levels of competition (Cottrill, 1990). The market share is calculated by dividing net sales over the sectors’ total net sales. Net sales represents sales receipts for products and services, less cash discounts, excises tax, and sales returns and allowances. Revenues are recognized according to applicable accounting principles. The Banks (industry 23) and Thrifts & Mortgage Finance industry (industry 24) do not report on net sales, therefore, total assets is used to calculate market share. A detailed variable description about the firm characteristics analyzed can be found in appendix A.1.

15

score. The main analysis is based on the total ESG score, but the sub-scores are also analysed to answer some of the hypotheses. However, the variables will not be included in a regression simultaneously. Furthermore, the variables excess return calculated based on S&P global index and excess return calculated using the S&P sector index are perfectly correlated, indicating that the variables cannot both be included in the regression. The model will include the excess return based on the S&P sector index, because this study aims to provide insights in sector differences. Therefore, the variable excess return based on the S&P sector index is most in line with the research objective. The earnings ratio is strongly positively correlated with one of the main CFP measures, ROA. Thus, I will not include the earnings ratio in the model. Lastly, the variables capital intensity, labour intensity and raw material intensity have too many missing variables, resulting in a significant sample size reduction from 1306 to 331. Therefore, I decided to exclude the production input variables in the main analysis and analyse them in the sensitivity analysis.

Taking the correlation matrix (appendix A.5) and the literature into account, a more detailed model is constructed to exploratory research sector characteristics and their effects on CSR, where 𝜀 is the error term and suffix i is for company:

ESGscore_𝑖 = 𝛼₀ + 𝛼₁𝑅𝑂𝐴_𝑖 + 𝛼₂Revenues_𝑖 + 𝛼₃FirmSize_𝑖 + 𝛼₄FirmAge_𝑖 + 𝛼₅FirmRisk_𝑖 + 𝛼₆Liquidity_𝑖 + 𝛼₇Tangibility_𝑖 + 𝛼₈LevelofInternationalisation_𝑖 + 𝛼₉LevelofCompetition_𝑖 + 𝛼10GICSIndustry𝑖 + 𝜀 _𝑖

16

Methods

To estimate the model and to analyse the stated hypotheses, ordinary least squares (OLS) regression analysis is used. To run an OLS regression, the Gauss-Markov assumptions need to be satisfied. Firstly, the errors should have zero mean, which is satisfied because a constant is included in the model. Secondly, the errors should have constant and finite variance (homoscedasticity). To deal with this, this study uses White’s heteroscedasticity consistent standard errors. Thirdly, the errors should be uncorrelated between observations. No time component is in the data, therefore, the residuals cannot be serially correlated. Lastly, the errors and independent variables should be uncorrelated. To meet this condition, several variables are excluded from the analysis because they were (im)perfectly correlated with other variables (appendix A.5).

Before analysing the exploratory hypotheses, the underlying hypothesis that sustainability practices are heterogeneous across industries, is tested. A one-way multivariate analysis of variance (MANOVA) is run to determine the effect of industry on social performance. A MANOVA is simply an ANOVA with several dependent variables (French et al., 2008). In this case four measures of social performance were assessed: the overall ESG-score, the environmental ESG-score, the social score and the governance score. The MANOVA essentially creates a linear combination of dependent variables to maximize group differences and tests whether the independent grouping variable (industry) explains a significant amount of variance in the dependent variables (Holton and Burnett, 2005).

To analyse the exploratory hypotheses and to determine which economic and business characteristics should be included in the further analysis, an OLS regression is run. Then, for all variables that significantly impacted CSP, new variables are created representing the ‘distance’ between the observed value and the mean of the total sample for a specific variable. In this step, I actually assume industry is not relevant to create an interesting benchmark to which the individual industries can be compared. Next, I assume industry is relevant. The total sample is divided into sub-samples based on industry and for all variables that significantly impact CSP, new variables are created representing the ‘distance’ between the observed value and the mean of the industry for a specific variable.

17

Household products Personal products) (N=54) is 50.55. Therefore, the newly created industry ‘distance’ variable (indicated by _2) shows that Energizer Holdings Inc. is underperforming on social performance compared to its direct competitors (48.43 – 50.55 = -2.12 = <0). The OLS regressions are run similarly as the one stated above, but now taking the newly created ‘distance’ variables (_1 and _2) into account, resulting in one regression for the ‘distance’ variables _1 and 33 regressions for the ‘distance’ variables _2 (one for each industry). The mean of the newly created ‘distance’ variables _1 are compared to the 33 industry means of the newly created ‘distance’ variables _2 to address industry differences on the variable level.

Lastly, the coefficients obtained from the 33 regressions (_2) are compared to the coefficients of the variables of all other industries to address differences in social performance across industries. I will perform the Wilcoxon rank-sum test (also known as Mann-Whitney test or Mann-Whitney-Wilcoxon test), which is a non-parametric statistical hypothesis test that is widely used for two-group comparisons for non-normal data (Oyeka and Ebuh, 2012). The Wilcoxon rank-sum test is a test for the equality of medians (Conroy, 2012) and is useful to identify significant differences between the industry groups. These statistically significant differences per variable and per industry will be counted and presented in a cross-tab. The results give valuable insights in industry differences and differences in sustainability practices, which will be dealt with in detail in the results section.

4. Results

In this section, I first discuss the outcomes of the MANOVA that is run to test the underlying hypothesis that sustainability practices are heterogeneous across industries. Secondly, I answer the exploratory hypotheses and investigate which economic and business factors that can potentially explain industry differences significantly influence CSR. Thirdly, I graphically present industry differences for each variable under consideration. Then, I rank the variables that best explain industry differences and variation in sustainability practices across industries. Lastly, I discuss the outcomes of the sensitivity analysis.

Main analysis

18

The minimum number of companies allocated to an industry group is 26, while the maximum number of companies allocated to an industry group is 206 (appendix A.2). A statistically significant MANOVA effect was obtained, Pillai’s Trace = 0.141, F(32, 2280) = 2.61, p< 0.000. Furthermore, also the other multivariate statistics (Wilks’ Lambda, Lawley-Hoteling trace and Roy’s largest root) show a statistically significant difference between at least two groups on the combined dependent variables. The results are in line with previous literature that CSP is heterogeneous across industries (e.g. Heal, 2005; Baron et al., 2009; Lee and Faff, 2009) and justify the exploratory nature of this research into industry differences and their effects on CSR.

To investigate industry differences in more detail, I exploratory research specific sector characteristics and relate these to different CSR measures. Table 1 reports the regression outcomes using the overall ESG score, environmental score, social score and governance score as independent variables in model 1.1 to 1.8 respectively. The results for the different measures of CSP are very similar, except for the governance score. However, the low explanatory power of this model compared to the other models may explain the differences.

First of all, model 1.1 confirms that industry significantly impacts the ESG score at a 5% level, which is in line with previous literature that took industry into account (e.g. McWilliams and Siegel, 2000). Looking at the measures of financial performance and controlling for industry, ROA and excess return positively influences CSP, but only the impact of ROA is significant, indicating that better financial performance in terms of ROA is associated with higher social performance. This result is in line with Waddock and Graves (1997), but contrast with McGuire et al. (1988) who find mixed results.

Furthermore, model 1.1 suggests that revenues have a significantly negative impact on the ESG score, implying that higher revenues are associated with lower social performance. Hence, I find no evidence for the first hypothesis that firms with higher revenues have higher ESG scores than firms with lower revenues. The finding contrasts with the finding of Campbell (2007), stating that firms with weak financial performance (measured by revenues) are less likely to engage in CSR.

Next, firms that disclose information about energy and water usage have significantly higher environmental scores than firms that do not disclose this information, which confirms the sixth hypothesis. This finding is in line with Tamimi and Sebastianelli (2017), who find significant differences in transparency and the ESG-score on both the social and governance dimensions between industries.

19

Table 1: corporate social performance regressions based on the ordinary least squares (OLS) approach, using the overall ESG-score and the individual pillar scores as independent variables

Model 1.1 Model 1.2 Model 1.3 Model 1.4 Model 1.5 Model 1.6 Model 1.7 Model 1.8 Variables ESGscore ESGscore E-score E-score S-score S-score G-score G-score

ROA 0.102*** 0.101*** 0.096*** 0.094*** 0.115*** 0.112*** 0.093** 0.092** (0.024) (0.024) (0.032) (0.032) (0.034) (0.034) (0.044) (0.044) Excess return* 0.286 0.284 0.279 0.277 0.295 0.291 0.306 0.305 (0.218) (0.217) (0.312) (0.312) (0.245) (0.244) (0.199) (0.198) Revenues -1.510*** -1.773*** -1.250* -1.451** -1.84*** -2.217*** (1.247) (1.381) (0.481) (0.478) (0.645) (0.627) (0.576) (0.570) (0.964) (0.952) Raw material usage

disclosure 20.855*** 20.741*** 29.23*** 29.14*** 23.11*** 22.94*** 9.600*** 9.539*** (1.136) (1.139) (1.453) (1.455) (1.573) (1.577) (1.567) (1.562) Size 0.000 0.000 0.000** 0.000** 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Age 0.088*** 0.083*** 0.058*** 0.054*** 0.066*** 0.058*** 0.143*** 0.140*** (0.015) (0.015) (0.019) (0.019) (0.017) (0.017) (0.023) (0.022) Risk -0.435 -0.659 -0.350 -0.523 -0.902 -1.221** -0.095 -0.211 (0.543) (0.537) (0.684) (0.672) (0.625) (0.621) (0.902) (0.889) Liquidity -2.298 -0.692 -0.359 0.855 -3.951 -1.719 -2.746 -1.937 (2.103) (2.010) (2.824) (2.684) (2.799) (2.696) (3.424) (3.279) Tangibility 3.720 3.501 1.008 0.842 0.411 0.106 8.987* 8.877* (3.244) (3.234) (4.389) (4.367) (4.013) (3.998) (5.292) (5.290) Level of internationalisation 0.049*** 0.051*** 0.070*** 0.071*** 0.046*** 0.049*** 0.033 0.034 (0.019) (0.019) (0.017) (0.017) (0.014) (0.015) (0.044) (0.044) Level of competition 71.693*** 71.494*** 83.84*** 83.715*** 91.13*** 90.896*** 30.244** 30.157** (14.823) (14.477) (16.470) (16.237) (19.662) (19.192) (14.942) (14.840) Industry 0.101** 0.077 0.142*** 0.051 (0.040) (0.054) (0.049) (0.063) Constant 36.116*** 38.462*** 34.06*** 35.851*** 39.47*** 42.767*** 36.215*** 37.409*** (2.960) (2.798) (3.936) (3.715) (3.631) (3.448) (4.860) (4.563) Observations 1306 1306 1298 1298 1298 1298 1298 1298 R-squared 0.413 0.411 0.398 0.397 0.357 0.354 0.089 0.089

This table reports the coefficients on the given variables estimated with an OLS regression model. The dependent variable is the overall ESG-score in model 1.1 and 1.2, the environmental score in model 1.3 and 1.4, the social score in model 1.4 and 1.6 and the governance score in model 1.7 and 1.8. Regression model 1.1, 1.3, 1.5 and 1.7 control for industry, whereas regression model 1.2, 1.4, 1.6 and 1.8 do not control for industry. A detailed description of the variables can be found in Appendix A.1. Robust Standard errors are reported in parentheses. ***, ** and * state statistically significance at the 1, 5 and 10% levels, respectively.

at all. According to Udayasankar (2008), who investigated economic motivations in the relation between CSR and firm size, there is a U-shaped relation between firm size and CSR. Large and small firms are equally motivated to invest in CSR, while medium-sized companies are the least motivated. The finding of this study (no relation) may be explained by the opposing effect stated by Udayasankar (2008).

20

considering human rights and product safety, which may explain the small effect when taking the combined ESG score into account.

Furthermore, I find no significant evidence that firm risk impacts the ESG-score (hypothesis 9). This result confirms and complements the finding of Waddock and Graves (1997), who controlled for firm risk by using the ratio total debt over total assets instead of the CAPM beta. Furthermore, I find no evidence for hypothesis 10, because liquidity does not significantly impact the ESG-score, which is in line with the findings of Casey and Grenier (2014) whose results are also not significant. The degree of tangibility (hypothesis 11) does also not yield a significant result, which explains the lack of research to this variable.

Furthermore, the level of internationalisation has a positive significant effect on the social score at the 1% level. Hence, higher levels of internationalisation increase social performance, supporting hypothesis 12. The results are in line with the findings of Watson and Weaver (2003) stating that firm-level internationalisation increases management attention to ethical behaviour. Lastly, market power has a large positive effect on CSR at the 1% significance level, indicating that higher market shares (lower levels of competition) result in better social performance, which rejects hypothesis 13 stating that higher competition results in better CSR. This finding is in line with Cottrill (1990) and contrasts with Campbell (2007). Campbell (2007) finds a curvilinear effect. High or low levels of competition result in less CSR, while moderate levels of competition tend to enhance CSR.

To summarize, industries that disclose information about raw material usage, face low competition, are internationally oriented, have better financial performance in terms of ROA, are older, have low revenues and rely on raw materials seem to have higher ESG scores. These variables are analysed in more detail to explain differences in social performance between industries. An overview of the findings can be found in table 3 at the end of the results section.

21

As can be seen from figure 1, industry 32 (Electric Utilities, Gas Utilities, Multi-Utilities, Independent Power and Renewable Electricity Producers) has the highest ESG-score, while industry 24 (Thrifts & Mortgage Finance), has the lowest score. Interesting, the most highly polluting industry according to Kassinis and Vafeas (2006) (Electricity, Gas & Oil Industry) has better social performance than an industry with low pollution potential (Financial Institutions). The leading ESG position of the oil and gas sector may be explained by increased attention resulting from public criticisms of oil and gas companies regarding environmental damage (Frynas, 2005). Furthermore, from the perspective of Heal (2005), the leading ESG position can be explained by the large private-social cost differential in the oil and gas sector, which may give CSR a resource-allocation role in cases of market failure.

The top-left scatterplot of figure 2 shows that industry 21 (Biotechnology) is performing very poorly regarding ROA compared to the total sample average. An explanation for this might be the statements of Donald Trump in 2017 criticising the biotechnology industry; ‘drug prices that are out of control’ and ‘drug companies are getting away with murder’, which decreased CFP (Reuters, 2017). Furthermore, in the top-right, middle-left and middle-right scatterplots, industries are scattered around the total sample average very well and no straightforward pattern is observed. The bottom-left scatterplot of figure 2 shows four surprising outliers, industry 4 (Construction Materials, Containers & Packaging, Paper & Forest Products, Metals & Mining), 7 (Machinery, Trading Companies & Distributors), 15 (Distributors, Internet & Direct Marketing Retail, Multiline Retail) and 30 (Semiconductors & Semiconductor Equipment) have negative levels of internationalisation, which seems unrealistic. This may be due to the sub-optimal proxy chosen as stated in appendix A.1. The bottom-right scatterplot of figure 2 shows

Figure 1: scatterplot of the relation between the overall ESG-score and GICS industry. A detailed overview of industries corresponding to each number can be found in appendix A.2. The straight line shows the mean of the total sample, while each diamond represents the mean of a specific industry.

3 5 4 0 4 5 5 0 5 5 E S G -s co re 0 10 20 30 40 GICS Industry

22 0 .5 1 1 .5 2 R e v e n u e s 0 10 20 30 40 GICS Industry

Mean Total Sample Mean per Industry

-6 0 -4 0 -2 0 0 2 0 R O A 0 10 20 30 40 GICS Industry

Mean Total Sample Mean per Industry

0 .1 .2 .3 .4 R a w M a te ri a l D is c lo s u re 0 10 20 30 40 GICS Industry

Mean Total Sample Mean per Industry

1 5 2 0 2 5 3 0 3 5 4 0 F ir m A g e 0 10 20 30 40 GICS Industry

Mean Total Sample Mean per Industry

-4 -2 0 2 L e v e l o f In te rn a ti o n a lis a ti o n 0 10 20 30 40 GICS Industry

Mean Total Sample Mean per Industry

0 .0 5 .1 .1 5 L e v e l o f C o m p e ti ti o n 0 10 20 30 40 GICS Industry

Mean Total Sample Mean per Industry

23

that industry 32 (Electric Utilities, Gas Utilities, Multi-Utilities, Independent Power and Renewable Electricity Producers) and industry 26 (Insurance) have a much lower competitive environment (higher market shares) than the total sample average. The low level of competition in both industries can be explained by large market power of a few players in the petroleum sector (Hubbard and Weiner, 1991) and the high market concentration in the insurance industry (Barresse et al., 2007).

Appendix B.9 shows the number of statistically significant differences, tested with the Wilcoxon rank-sum test, per variable and per industry. This table is based on the p-values displayed in appendix B3–B8. Industry 23 (Banks), 24 (Thrifts & Mortgage Finance), 26 (Insurance) and 32 (Semiconductors & Semiconductor Equipment) consist of too few observations to meet the minimum sample requirement of 14 (2SPV rule of Austin and Steyerberg, 2015) and are, therefore, excluded in the analysis. Table 2 shows the ranking of the industries for each variable based on the percentage of significant differences between industries, per industry. Furthermore, the last column (TOT) shows which variable has the highest number of significant differences compared to the maximum number of differences that can be detected (33x32). This gives insights in which variable has the most potential explaining industry differences and variations in social performance.

24

Table 2: industry ranking on variable level of the number of observed significant differences between industries with a p-value smaller than 0.05 per industry, expressed as a percentage of total number of differences that can be detected per industry (32). The total number of significant differences per variable for all industries, expressed as a percentage of the total number of differences that can be detected in total (1056) are expressed in the last column (TOT)

Industry 19 6 25 2 21 33 3 4 7 13 1 5 8 11 20 29 31 14 16 17 30 9 10 15 18 22 27 28 12 TOT ROA_2 91 88 88 53 44 25 19 19 19 19 16 16 16 16 16 16 16 13 13 13 13 9 9 9 9 9 9 9 6 21 Industry 33 21 2 20 13 31 19 29 30 5 15 28 25 7 1 14 3 4 6 8 9 10 11 12 16 17 18 22 27 TOT Revenues_2 38 34 31 25 22 22 19 19 19 16 16 16 9 6 3 3 0 0 0 0 0 0 0 0 0 0 0 0 0 10 Industry 2 4 5 7 8 10 17 18 19 21 25 29 30 33 6 12 13 15 16 20 22 27 28 31 1 9 11 14 3 TOT RawMaterialD isclosure_2 100 100 100 100 100 100 100 100 100 100 100 100 100 100 97 97 97 97 97 97 97 97 97 97 94 94 94 94 84 86 Industry 21 28 3 14 27 4 7 33 18 20 31 22 25 30 2 8 6 13 19 29 1 5 10 11 12 15 17 16 9 TOT Age_2 63 63 47 38 34 31 31 31 28 28 25 22 19 19 16 16 13 13 13 13 9 6 6 6 6 6 6 3 0 19 Industry 4 7 15 5 12 18 30 28 13 16 17 25 19 6 20 9 11 27 8 14 22 29 31 33 3 10 21 1 2 TOT LevelofInterna tionalisation_2 100 100 100 97 97 94 94 78 66 59 56 47 44 41 41 38 38 38 34 34 34 34 31 31 28 28 28 25 25 47 Industry 33 7 10 17 21 14 1 20 19 28 29 2 5 13 30 22 31 8 15 25 11 16 18 9 3 4 6 27 12 TOT LevelofCompe tition_2 97 94 94 91 84 81 78 75 72 72 72 69 69 69 69 66 66 63 63 63 56 56 56 53 50 50 50 50 22 59

25

and 28 (Software) differ from 63% of the other industries. Looking at the data in more detail, industry 21 (Biotechnology) and 28 (Software) appear to be the youngest industries (16.6 and 15.6 years respectively).

For the factors degree of internationalisation and level of competition, the percentage of differences vary greatly from 100 percent to approximately 20 percent. Industry 4 (Construction Materials, Containers & Packaging, Paper & Forest Products, Metals & Mining), 7 (Machinery,Trading Companies & Distributors) and 15 (Distributors, Internet & Direct Marketing Retail, Multiline Retail) differ from all other industries (100%) in terms of level of internationalisation, whereas industry 1 (Energy Equipment & Services) and 2 (Oil, Gas & Consumable Fuels) only differ from 25% of the other industries. As the oil and gas industry is mainly internationally oriented, I expect industry 4 (Construction Materials, Containers & Packaging, Paper & Forest Products, Metals & Mining), 7 (Machinery, Trading Companies & Distributors) and 15 (Distributors, Internet & Direct Marketing Retail, Multiline Retail) to be more nationally oriented. Lastly, in terms of competition, industry 33 (Equity Real Estate Investment Trusts, Real Estate Management & Development) differs from 97% of all industries while industry 12 (Leisure Products, Textiles, Apparel & Luxury Goods) only from 22% of all industries, indicating great variation in competition. To summarize, the factor raw material disclosure tend to best explain industry differences and variation in sustainability practices, followed by the level of competition, the level of internationalisation, ROA, firm age and revenues.

Appendix B.10 confirms this finding on the industry level. Appendix B.10 shows the number of significant differences per variable and per industry, displayed as a percentage of the total number of significant differences of an industry for all six variables (ROA, revenues, raw material disclosure, firm age, level of internationalisation and level of competition). For 72.4% of the industries, the variables raw material disclosure, level of internationalisation and level of competition explain most of the differences in the industries. Therefore, these variables are most likely the industry determinants that explain differences in social performance across industries.

Sensitivity Analysis

26

activity, which is the main industry determinant used by GICS, is closely related to the production function of a firm. Therefore, I will perform the same analysis (including the production input variables), but with a smaller sample. The small sample size decreases statistical power, but the results gives insights in the robustness of the findings of the main analysis. The intensity variables, capital intensity, labour intensity and raw material intensity, add up to 1, which results in an error when using OLS regression. In a two dimensional setting a firm is either labour intensive or capital intensive, indicating that labour intensity is a substitute for capital intensity. Furthermore, the correlation matrix shows that labour intensity is correlated with capital intensity. Therefore, labour intensity will be excluded in the regression model.

Appendix C.1 reports the regression outcomes using the overall ESG score, environmental score, social score and governance score as independent variables in model 1.1 to 1.8 respectively. The results for the different measures of CSP are similar, but less similar than in the main analysis. The results of model 1.7 and 1.8 are not taken into account when analysing the hypotheses, because these models have low explanatory power compared to the others. Looking at each variable in more detail, the ROA positively influences the overall ESG score at a 5% significance level, which is in line with the results of the main analysis. However, the results are not significant for the other models at a 5% significance level, indicating that ROA only determines overall social performance. The variable excess return does not yield significant results.

Revenues negatively and significantly impact social performance in all models, except for the social score, but this result is not significant at a 5% level. Furthermore, the result is in line with the result of the main analysis and does not provide support for the first hypothesis. Capital and raw material intensity positively impact all measures of social performance at a 5% significance level, except for the governance score. Thus, capital-intensive firms have higher environmental and social scores than less capital-intensive firms and raw-material intensive firms have higher environmental scores than less raw-material intensive firms. Hence, I find no support for hypothesis 4 and 5.

27

Furthermore, the level of internationalisation has a small negative effect on social performance at a 10% significance level in all models, except for model 5 and 6, indicating that internationalisation does not influence the social score, but does influence the overall score. This finding is, surprisingly, not in line with the finding in the main analysis, where internationalisation had a significant positive effect on social performance. However, it does support hypothesis 12. The contrasting result may be due to the small sample size. Lastly, market power has a strong positive effect on social performance. Therefore, lower levels of competition result in higher social performance. This result is in line with the main analysis and does not provide support for hypothesis 13. To summarize, ROA, revenues, capital intensity, raw material intensity, raw material disclosure, firm age, the degree of internationalisation and the level of competition significantly influence social performance and are analysed in more detail.

Appendix C.2 present scatterplots for the input variables. Interesting to observe is that industry 13 (Hotels, Restaurants & Leisure, Diversified Consumer Services) is very labour-intensive and capital-extensive compared to the other industries, which is plausible given the nature of the activities. Appendix C.3 shows the estimation results for the regression assuming industry is not relevant. Appendix C.4 displays the estimation results for the regressions considering industry. As can be seen, most of the industries consist of too few observations to meet the minimum sample requirement of 18 (2SPV rule of Austin and Steyerberg, 2015), except for industry 19 (Health Care Equipment & Supplies), 21 (Biotechnology), 22 (Pharmaceuticals, Life Sciences Tools & Services), 29 (Communications Equipment, Technology Hardware, Storage & Peripherals, Electronic Equipment, Instruments & Components) and 30 (Semiconductors & Semiconductor Equipment). Therefore, I will continue analysing only those industries that have sufficient observations.

Appendix C.13 shows the number of statistically significant differences, tested with the Wilcoxon rank-sum test, per variable and per industry. This table is based on the p-values displayed in appendix C5–C12. Appendix C.15 shows the ranking of the industries for each variable based on the percentage of significant differences per industry. Furthermore, the last column (TOT) shows the variables with the highest number of significant differences compared to the maximum number of differences that can be detected (33x32). Appendix C.14 shows the number of significant differences per variable and per industry, displayed as a percentage of the total number of significant differences of an industry for all 8 variables.

28

the level of competition explain most of the differences in the industries. In contrast, the level of internationalisation is less relevant than in the main analysis, but this can also be due to the small sample size. Capital intensity seems not to be explaining industry differences, but raw-material intensity does, as is in line with expectations. Raw raw-material intensity is related to raw material usage disclosure, which is one of the variables, according to the main analysis, that explains differences in CSR across industries. An overview of the findings of the hypotheses of the main analysis, including the findings about the production input variables of the sensitivity analysis, can be found in table 3.

Table 3: Overview of the testing results of the hypotheses stated in the literature review section

Hypotheses Variables Expected Sign Actual Sign Accept Hypothesis Reject Hypothesis Significant Result (5% level) Hypothesis 1 Revenues Positive Negative x Yes Hypothesis 2 (not tested) Earnings Positive Hypothesis 3 (not tested) Labour Intensity Positive Hypothesis 4 (Sensitivity Analysis) Capital Intensity Negative Positive x Yes Hypothesis 5 (Sensitivity Analysis) Raw-Material Intensity Negative Positive x Yes Hypothesis 6 Raw-Material Disclosure Positive Positive x Yes

Hypothesis 7 Firm Size Positive x No

Hypothesis 8 Firm Age Positive Positive x Yes Hypothesis 9 Firm Risk Positive Negative x No

Hypothesis 10 Liquidity Positive x No

Hypothesis 11 Tangibility Negative x No

Hypothesis 12 Internationalisation Positive Positive x Yes Hypothesis 13 Competition Positive Negative x Yes This table gives an overview of the results regarding the hypotheses stated in the literature review. Hypotheses 2 and 3 cannot be tested, because the variables are correlated with ROA and capital intensity, respectively. Hypothesis 4 and 5 are tested in the sensitivity analysis. The column expected sign presents the expected relation with CSR. For example, revenues are expected to be positively related to CSR, indicating that higher revenues result in higher ESG-scores. However, as can be seen under the column actual sign, this turned out not to be the case as higher revenues result in lower ESG-scores. Therefore, the first hypothesis is rejected. However, the negative coefficient for revenues is significant at a 5% level, thus the variable is analysed in more detail to explain industry differences.

5. Discussion and Conclusion

This study investigates how industry classification relates to sustainability practices. On the basis of existing research, it is clear that CSR varies widely between sectors and that, therefore, the interaction between CFP and CSR varies as well. However, in previous CSR literature, industry is a concept that is assumed to be common knowledge, but it is ambiguous what factors in fact determine industry allocation. I try to find out which factors do so and how they relate to sustainability practices.

29

also looks at other firm characteristics that may influence industry classification, such as firm age, risk, size, liquidity, tangibility, the degree of internationalisation and the level of competition, and into the production process of firms, especially to the production input function.

Previous literature did study the role of industry in sustainability practices. For example, Heal (2005) looks at CSR practices across industries from a resource-allocation perspective and Ioannou and Serafeim (2017) investigate the adoption of sustainability practices within industries over time. Heal (2005) finds that incorporating CSR initiatives results in increased CFP in sectors where private-social cost differentials are large, but not in industries where private and social cost are equal. Ioannou and Serafeim (2017) find that sustainability practices converge within an industry and that the extent of convergence across industries is related to the implementation of CSR by the industry’s market leaders and the relative importance of environmental and social issues compared to governance issues. However, to the best of my knowledge, this research is the first that decomposes industry classification into the underlying driving factors and empirically examines the effects of industry determinants on sustainability practices instead of only considering the role of industry in general.

30

Developing knowledge about underlying industry determinants and relating these to sustainability practices is meaningful, for three main reasons. Firstly, the accelerating amount of companies that issue corporate sustainability reports indicates that for companies across different industries sustainability has become a central practice. Secondly, the findings provide interesting and important practical implications. Prior studies on CSR have debated on whether firms should implement CSR programs (Griffin and Mahon, 1997, Orlitzky and Benjamin, 2001). This study suggests that firms should consider firm characteristics that are associated with industry determination when deciding on the relevance and implementation of CSR programs. Thirdly, a rich literature about the importance of sustainability practices and its impact on financial performance exist in which the role of industry is underexposed (see Margolis and Walsh, 2001 or Aguinis and Glavas, 2012 for an overview of the literature) or assumed to be common knowledge. This study clarifies the concept ‘industry’ and provides insights into the underlying factors that determine industry classification and their effect on sustainability practices.

31

6. References

Aecom, 2019. 40 Proposed U.S. transportation and water infrastructure projects of major economic significance. [online] Available at < https://www.treasury.gov/connect/blog/ Documents/final-infrastructure-report.pdf> [Accessed May 4th, 2019]

Aguinis, H., Glavas, A., 2012. What we know and don’t know about corporate social responsibility: a review and research agenda. Journal of Management, 38(4), 932-968.

Aigner, D.J., Chu, S.F., 1968. On estimating the industry production function. The American Economic Review, 58(4), 826-839.

Arora, P., Dharwadkar, R., 2011. Corporate governance and corporate social responsibility (CSR): the moderating roles of attainment discrepancy and organization slack. Corporate Governance: An International Review, 19(2), 136-152.

Arouri, M.E.H., Jouini, J., Nguyen, D.K., 2011. Volatility spillovers between oil prices and stock sector returns: implications for portfolio management. Journal of International Money and Finance, 30(7), 1387-1405.

Attig, N., Boubakri, N., El Ghoul, S., Guedhami, O., 2016. Firm internationalization and corporate social responsibility. Journal of Business Ethics, 134(2), 171-197.

Austin, P.C., Steyerberg, E.W., 2015. The number of subjects per variable required in linear regression analyses. Journal of Clinical Epidemiology, 68(6), 627-636.

Barrese, J., Lai, G., Scordis, N., 2007. Ownership concentration and governance in the US insurance industry. Journal of Insurance Issues, 30(1), 1-30.