Bad Companies, Bad Investments? On the Return and Risk Characteristics of Responsible and Sin Investing☆

(1)

1

Bad Companies, Bad Investments?

On the Return and Risk Characteristics of Responsible and Sin Investing

☆

Pieter Jan Trinks

University of Groningen, Faculty of Economics and Business

__________________________________________________________________________________

Abstract:

Does it make financial sense to invest in socially responsible or sustainable companies? Or should we aim for ‘bad’, ‘sinful’, or socially controversial investments? To answer these and related questions, we investigate the risk and return characteristics of portfolios of both ‘sin’ and ‘virtue’ stocks. We consider various reward-to-risk ratios, and employ four-factor time-series mean and median regression techniques using a uniquely constructed sample of 1,763 sin stocks and several Virtue portfolios based on Thomson Reuters’ Asset4 ESG data over the period 01/1991 to 12/2012. Our findings suggest that both sin and virtue investing pays off in terms of positive abnormal returns. However, as sin stocks generally outperform virtue stocks, and screened (SRI) portfolios underperform unscreened portfolios, investors may sacrifice returns when investing responsibly instead of ‘sinfully’. These conclusions require some reservations. While sin stocks beat the market and virtue stocks during the recent financial crisis, they fail to do so during the pre-crisis period. Similarly, while responsible investments appear to outperform the market during the crisis, they do so at the cost of underperforming before the crisis. Findings are furthermore in part contingent on data specification, return averaging methods, and estimation techniques.

Keywords: Socially Responsible Investing (SRI), sin stocks, virtue stocks, time series analysis, Least Absolute Deviation (LAD) estimation, downside risk

JEL codes: G11, G12, G15, M14

__________________________________________________________________________________

MSc Thesis Finance

Pieter Jan Trinks, s1789740

University of Groningen, January 10th_{, 2014}

Supervisor: Prof. Dr. L.J.R. Scholtens

___________________________________

☆ _{I would like to thank Prof. Dr. Bert Scholtens for his supervision and critical guidance during the entire process that led to this}

(2)

2

List of Tables

Table 1: Sin sample clearing ... 13

Table 2: Composition of screened S&P portfolios ... 15

Table 3: Descriptive statistics on excess returns, 01/1991-12/2012 ... 26

Table 4: Return performance of Sin portfolios, 01/1991-12/2012 ... 31

Table 5: Return performance of zero-investment portfolios, 01/1991 (01/2002)-12/2012 ... 32

Appendix C: Evolution of Sin and Virtue portfolios Table A.1: Constitution of Asset4 portfolio ... 51

Table A.2: Constitution of Sin portfolios at year-end ... 51

Appendix D: Sensitivity analysis: Virtue portfolios based on lagged Asset4 scores Table A.3: Return performance of zero-investment portfolios using lagged Asset4 scores, 2002-2012 ... 53

Appendix E: Results using equal-weighted returns Table A.4: Descriptive statistics on excess returns, 01/1991-12/2012 ... 54

LAD regressions Table A.5: Return performance of zero-investment portfolios, 01/1991 (01/2002)-12/2012 ... 55

Table A.6: Return performance of Sin Portfolios, 01/1991-12/2012 ... 56

Table A.7: Return performance of zero-investment portfolios using lagged Asset4 scores, 2002-2012 ... 57

Appendix F: Sensitivity analysis: OLS regressions Value-weighted returns Table A.8: Return performance of zero-investment portfolios, 01/1991 (01/2002)-12/2012 ... 58

Equal-weighted returns Table A.11: Return performance of zero-investment portfolios, 01/1991 (01/2002)-12/2012 ... 61

Appendix G: Sensitivity analysis: crisis and pre-crisis performance Pre-crisis descriptives Table A.14: Descriptive statistics on excess returns, 01/1991-11/2007 ... 64

Pre-crisis LAD regressions Table A.15: Return performance of zero-investment portfolios, 01/1991-11/2007 ... 65

Table A.16: Return performance of Sin portfolios, 01/1991-11/2007 ... 66

Pre-crisis OLS regressions Table A.17: Return performance of zero-investment portfolios, 01/1991-11/2007 ... 67

Crisis descriptives Table A.19: Descriptive statistics on excess returns, 12/2007-12/2012 ... 69

Crisis LAD regressions Table A.20: Return performance of zero-investment portfolios, 12/2007-12/2012 ... 70

Crisis OLS regressions Table A.22: Return performance of zero-investment portfolios, 12/2007-12/2012 ... 72

(4)

1

1. Introduction

Social responsibility is a fast growing concern of both individual and institutional investors. Recent global environmental and financial crises (e.g. the 2009 Gulf of Mexico oil spill, the 2011 Fukushima Daiichi nuclear disaster, and the various crises connected with the Global Financial Crisis of 2007/2008) have highlighted the relevance of ethics in the global economy, as well as the need to rethink the purpose and governance of financial institutions (Richardson, 2013). The notion of Corporate Social Responsibility (CSR) has hereby received growing attention in economic, political and philosophical literature. In brief, CSR is concerned with the ascription and implementation of moral responsibilities toward (constituent groups in) society, which extend beyond narrow economic and legal requirements (Davis, 1973; Melé, 2008). On the other –financial– side of the coin are an increasing number of investors that attempt to incorporate ethical considerations into their investment decisions. In Europe, this instance of Socially Responsible Investing (SRI) currently comprises about one sixth of all Assets under Management (AuM) (Eurosif, 2012). Almost half of Europe’s AuM has policies in place to avoid making investments in companies that are involved in the manufacturing of certain controversial weapon types, e.g. cluster munitions, or chemical and nuclear weapons (Eurosif, 2012). Moreover, during the last few years numerous countries have implemented strong legislation against financing these controversial weapons1_{. Clearly, SRI}

no longer constitutes a niche in the financial world2_.

But what is SRI exactly? While there are elements of ambiguity and subjectivity to answering this question, SRI is generally thought of as a form of investing that integrates both financial criteria as well as nonfinancial ethical and Environmental, Social, and Governance (ESG) concerns into analysis processes and investment decisions (Renneboog et al., 2008a). It thereby comprises commonly used concepts such as ‘ethical’, ‘social’, ‘green’, ‘sustainable’, and ‘impact’ investing (comprising well-known instances of micro-finance). SRI can be seen as an ambitious attempt to reconcile ethics with traditional finance theory and practice (Kurtz, 2008; Richardson, 2013).

SRI (as the concept of CSR) has strong roots in religious beliefs, which remain an important driving force today, especially in the United States (Kurtz, 20083_{; Peifer, 2011}4_{). John Wesley’s (1703-1791)}

sermons are often indicated as the first explicit blueprint for SRI, and formed an important source of inspiration for the first modern SRI mutual funds in the 1970s (Kurtz, 2008). These mutual funds excluded investments in ‘sinful’ or socially controversial companies and practices, commonly encompassing the alcohol, tobacco, weapons, and gambling industries (Renneboog et al., 2008a). Landier and Nair (2009)

1_{A comprehensive overview of national legislation on controversial weapons can be found online:}

http://www.stopexplosiveinvestments.org/legislation.

2_{Landier and Nair (2009) even predict that the share of SRI capital will double within the next few years, partly because of the}

standardization of information by initiatives (such as the CDP, GRI, and UNPRI), and the professional rise of women. An interesting observation here is that women form an important driving force for SRI, as they constitute about 60% of all SRI investors (Schueth, 2003) and it is well-known that their participation in conventional investing is rather low.

(5)

2 and Richardson (2013) list various successes of SRI since its conception, among which the 1970s global divestment campaign against companies profiting from South Africa’s apartheid regime.

Since its conception SRI practitioners have developed various ways or strategies to achieve their stated goals to improve CSR aspects and invest in accord with their values. Today, SRI is implemented by a great mixture of strategies, i.a. investing or divesting according to ethical or ESG criteria (i.e. positive and negative screening), exercising shareholders rights, and engaging or dialoguing with companies on ESG issues (Kurtz, 2008; Eurosif, 2012). A main strategy however remains the original form of sector-based

(exclusionary or product-related) negative screening, i.e. the exclusion of so called ‘sin stocks’ from the investment universe (Renneboog et al., 2008a). Almost a quarter of total AuM uses some kind of exclusionary screening (Eurosif, 2012). In the U.S., more than 95% of SRI mutual funds use negative screening criteria (Renneboog et al., 2011). Investors increasingly tend to avoid ‘sinful’ companies either because of the potentially unethical nature of their products or their vulnerability to social norms and public opinion (Hong and Kacperczyk, 2009; Salaber, 2013). Salaber (2013) lastly notes that a large part of SRI funds and investors avoid sin stocks while also employing positive screening criteria.

Sector-based negative screening should not be confused with the increasingly popular strategy of norm-based screening, which refers to the practice of benchmarking company activities against universally accepted ethical principles. An important instance is the 2006 UN Principles for Responsible Investing initiative, which more than 1,200 companies (mostly investment managers and large asset owners), totaling more than $34 trillion in AuM, nowadays have signed. Richardson (2013) states that such international initiatives could be a valuable tool to improve ethics and governance issues in financial markets.

(6)

3 (KPMG, 2013). Company CSR aspects are therefore of increasing interest for non-value driven investors as well, be it as a mere strategy to reduce overall portfolio risk (Kiymaz, 2012).

This paper concentrates on the above described instances of positive and negative (sector-based) screening. We in particular are interested in the financial performance of sin stocks (which are avoided under negative screening strategies). The main question throughout this paper is whether it ‘pays to be good’, i.e.: Do socially responsible or sustainable investments make financial sense, or should we rather aim for ‘bad’, ‘sinful’, or socially controversial companies? This question is relevant to both the individual investor, who desires to allocate her assets in a way that is optimal relative to her (financial and nonfinancial) preferences, as well as institutional investors (such as pension funds) or fund managers, which have a fiduciary duty to invest their clients’ capital in line with their best interests (generally meaning investing their capital in the most profitable way given a certain level of risk)5_{. Moreover, the growth of socially conscious investments}

might largely depend on the profitability of these investments (Landier and Nair, 2009).

In an attempt to find establish the (non-)existence of a ‘’Holy Grail’’, i.e. the greatly desired objective of SRI to combine values with superior returns, empirical studies have been providing with opposing and inconclusive answers (Margolis et al., 2009). On the one hand, a large body of literature has found some positive relationship between measures of CSR or SRI and stock performance (see Landier and Nair, 2009; Margolis et al., 2009). On the other hand, the past decade has evidenced an increasing interest in sin stocks. Books have been written that promote the potential of sin stocks (particularly during economic downturns) (see Ahrens, 2004; Waxler, 2004), there now is a special Vice Fund that performs very well, and websites have emerged that provide investors with ready-to-invest lists of sin stocks6_{. On the academic side,}

a number of recent empirical studies confirm that portfolios comprised of sin stocks exhibit a significant outperformance relative to the market and industry comparable stocks (see Hong and Kacperczyk, 2009 and related studies). In this respect investors might face a ‘’Faustian choice’’ between values and profits (Landier and Nair, 2009). Also, responsible investing might be incongruent with fiduciary duty law of financial institutions (Richardson, 2013). These suggestions are reinforced by recognizing the evident reduction of diversification opportunities associated with SRIs screening practices (Heinkel et al., 2001; Adler and Kritzman, 2008), and increased monitoring costs from SRI fund managers (see Chegut et al., 2011).

One nevertheless has to recognize that research on sin and responsible (or ‘virtue’) investing is at an early stage. A lot of empirical research on sin stocks maneuvers around the work of Hong and Kacperczyk (2009) and is rather limited in scope, considering a small or fairly specific time period, mostly concentrating on three rather classical sin sectors, showing strong bias towards U.S. contexts. Furthermore, previous studies have often taken the concepts of CSR and SRI as homogeneous, while these are in fact multidimensional and constantly changing concepts, which are understood and practiced very differently across investors (see

5_{SRI investors might therefore make investment decisions by considering for a major part their financial utility. For instance,}

‘sustainable’ banks (e.g. Triodos Bank) and pension or mutual funds may be incentivized not to exclude alcohol stocks from their investment universe, as these may be generally well-performing stocks. In this line of reasoning, this paper may also shed some light on the potential motives of SRI investors to divest certain sectors while keeping others (see also Bénabou and Tirole, 2010).

(7)

4 Chatterji et al., 2009; Bénabou and Tirole, 2010; Derwall et al., 2011 for interesting discussions). Chatterji et al. (2009) cast doubt on the usefulness and accuracy of CSR measures and ratings, as these ratings are found to have a disappointingly limited ability to predict CSR performance. Chegut et al. (2011) state that poor data (its integrity and collection process) would be a main driver of measurement errors. Scholtens (2014) observes a lack of sound definitions and metrics for SRI, which probably forms an impediment for an adequate assessment of the size of SRI and its value to investors, companies, and society as a whole.

In addition, SRI practices and conceptions of ‘sin’ clearly appear to differ per continent, country, and even in the composition of SRI funds within one country. For instance, in Europe, trends may be observed towards norm-based screening (especially for Scandinavian countries), positive screening (popular in the French SRI market), and increased sector-based screening practices (a trend typical for Southern countries) (Eurosif, 2012). Besides, empirical findings suggest that the conception and implementation of SRI may be influenced by cultural mediating factors, investors’ personal beliefs, values, and motivations (e.g. Kumar et al., 2011; Hood et al., 2012; Scholtens and Sievänen, 2013), religion, and legal and financial system architecture (Salaber, 2013; Scholtens and Sievänen, 2013). Lastly, recent studies suggest that while particular exclusionary screens, such as pornography and abortion, might be fairly specific to religious investors (Boasson et al., 2006), even within the well-established realm of religious SRI there are substantial differences in the implementation of SRI (Hood et al., 2012) and disagreement about the moral implications of religious doctrines for investing (Porter and Steen, 2006).

As a last set of limitations Margolis et al. (2009) explains that there is a lack of good measures for financial performance and risk attributes, problems with necessary causality, omitted variables, and endogeneity. Kiymaz (2012) adds that study periods are largely diverging between studies and moreover are relatively short to produce reliable conclusions. As a result of the above shortcomings, we should interpret available empirical research with some caution and with recognition for the fact that findings might be specific to i.a. the country, culture, time period, and methodological choices of each study.

The aim of this paper is to build on current empirical research regarding sin stocks and responsible investing, and to improve on at least part of the above mentioned shortcomings. We investigate the risk and return characteristics of a uniquely constructed sample of 1,763 sin stocks divided over fourteen potentially controversial sectors (1,634 non-duplicate stocks), as well as several Virtue portfolios based on Thomson Reuters’ Asset4 ESG data, over the period 01/1991 to 12/2012.

(8)

5 a much wider range of sin sectors, i.e. fourteen, relative to the three commonly studied and rather classical ‘Triumvirate of Sin’ stocks, consisting of alcohol, tobacco, and gambling stocks.

With respect to the virtue stocks, the directly constructed portfolios using high quality ESG data will constitute a more objective and reliable measure of ‘virtue’ or social responsibility, as well as circumvent potential biases related to studying SRI funds (biases may emerge due to differences in management skill, fees, and other fund-specific items). It also enables us to compare portfolios based on ESG subscores, e.g. analyzing the portfolio of companies that have a high score at Environmental performance indicators. In all, these analyses may provide with an improved understanding of which (combination of) sin sectors and for which part of SRI we might observe desirable return- and risk characteristics.

Besides, to our knowledge this study is unique in its direct comparison between sin and virtue stocks. As both responsible and sin investing has been promoted in literature, we might break this duality, by establishing abnormal return performance for either of those investment strategies.

Furthermore, we complement the above analysis with a pragmatic approach of applying negative and positive screening strategies to a broad market portfolio. To this end, we exclude our Sin portfolios from the S&P500 as well as include (or select) the virtue stocks in this index. This analysis might provide us with an understanding of whether and when we can find ‘Holy Grails’ or ‘Free lunches’ (higher returns when taking the same risk) in responsible investing.

Lastly, we consider two methodological improvements. By using Least Absolute Deviation (LAD) estimation, a semi-parametric median regression approach, we avoid potential problems of using simple Ordinary Least Squares (OLS) estimation, i.e. delivering biased results for non-normally distributed return series. We furthermore complement traditional risk and return measures with measures of downside risk, which might better reflect investors’ actual concerns and preferences.

This paper is structured as follows: Section 2 provides an overview of the relevant background literature and develops our main hypotheses. Section 3 will then describe the sin, virtue, and screened (SRI) portfolio construction methods. The research methodology is described in Section 4. Results will be presented and discussed in section 5. Section 6 concludes.

2. Background

(9)

respon-6 sible (screened) portfolios respectively, and develop our main hypotheses regarding the relation between each of investment strategy and stock performance. We end this chapter with a (more qualitative) discussion on the value of negative screening.

2.1.Sin stock returns

Hong and Kacperczyk (2009) study a classical Triumvirate of Sin portfolio of 156 stocks during 1965-2006. They find that these controversial stocks have higher expected returns than stocks with comparable characteristics (on the basis of Fama and French (1997) industry groups 2 (food), 3 (soda), 7 (fun), and 43 (meals and hotels)) in the U.S. This result is explained in a Merton (1987) theoretical framework, basically stating that sin stocks are undervalued because of their being neglected by norm-constrained investors (e.g. pension funds), which forces share prices below companies’ fundamental value due to decreased demand, as well as decreased risk-sharing opportunities. Consequently, sin stocks should have higher expected (required) returns than industry-comparable stocks or the market as a whole7_{. Heinkel et al. (2001)’s theoretical model}

moreover shows that when there is a sufficiently large share of norm-constrained investors the cost of equity capital will rise for shunned companies (meaning higher expected returns), and a lower cost of equity capital for non-controversial stocks. This neglect-effect (or ‘neglect premium’) associated with sin stocks might be the result of (institutional) investors’ vulnerability to societal norms and public opinion (as Hong and Kacperczyk (2009) suggest), but might also be explained by the negative affect of the average investor towards sin stocks, or moral motivations (Derwall et al., 2011).

The findings of Hong and Kacperczyk (2009) have been confirmed by Durand et al. (2013b) for the U.S (1990-2008), and by Visaltanachoti et al. (2009) and Durand et al. (2013a) for China and Hong Kong (1995-2007), and Australia and New-Zealand (1990-2009) respectively. Salaber (2013) confirms the outperformance of sin stocks for the European market, using a sample of 158 Triumvirate of Sin stocks over the period 1975-2006. She moreover adds that Protestant countries tend to display higher risk premia for controversial stocks than Catholic countries do. Sin stocks furthermore have higher risk-adjusted returns when they are located in countries with high excise taxation. This leads her to conclude that sin stock returns depend both on legal and religious environments. A study by Fabozzi et al. (2008) expands the classical sin sectors with the Defense, Biotech (comprising animal testing, genetic engineering, and ordinary stem cells), and Adult entertainment sectors. They find substantial positive excess returns (relative to the risk-free rate as well as the market) for their global sample of 267 sin stocks for 1970-2007. Lobe and Walkshäusl’s (2011) global 1995-2007 study constructs a portfolio based on the so-called ‘Sextet of Sin’. They however find no significant difference in returns between sin indexes and conventional benchmarks. Lastly, Salaber (2009) suggests that the ‘sin label’ is not a crucial determinant of excess returns. She finds that 183 U.S. Triumvirate of Sin stocks display excess returns when compared to the market, yet these disappear when sin stocks are compared to a control group of industry-comparable stocks.

7_{Hong and Kacperczyk (2009) add the suggestion that the price of sin companies’ stocks may also be lowered by their conservative}

(10)

7 A last strand of literature investigates the performance of the Vice Fund. Jo et al. (2010) find this fund to outperform the market as well as the responsible (positively screened) DS400 Index for the long-term. Chong et al. (2006) similarly find the Vice Fund to outperform both the Domini Social Equity Fund and the S&P500. The Sharpe ratio of the Vice Fund is higher, suggesting the fund can be employed to boost the reward-to-risk performance of investment portfolios. Liston and Soydemir (2010) find positive alphas for Sin portfolios and negative alphas for Faith-based portfolios, consistent with the ‘norm-neglect effect’ of the former and the ‘norm-conforming effect’ of the latter portfolios (Hong and Kacperczyk, 2009). Sin stocks are also found to have a higher Sharpe ratio than Faith-based stocks. Lastly, Durand et al. (2013b) find that sin stocks outperform their virtue counterparts. These findings do in fact suggest that in the long-run U.S. investors sacrifice returns by investing responsibly.

Considering the above empirical literature, we hypothesize that the risk-adjusted performance (in terms of reward-to-risk measures) of sin stocks differs from the market (following e.g. Fabozzi et al., 2008; Lobe and Walkshäusl, 2011), and secondly from virtue stocks as well (compare with Jo et al., 2010; Durand et al., 2013b). We expect sin stocks to generate significant abnormal returns8_{when controlled for common}

risk factors, and when benchmarked against the market and virtue stocks (ibid). In addition, we expect higher downside risks for sin stocks relative to the market portfolio, due to a potentially increased probability of negative publicity or litigation events (Fabozzi et al., 2008).

2.2.Virtue stock returns

If socially conscious investors gain nonfinancial utility from responsible investing (as suggested by Bollen (2007) and Ballestero et al. (2012)), demand for virtue stocks would –just as with sin stocks– be influenced by factors other than well-documented risk factors. Responsible investments are often expected to have higher risk-adjusted returns due to the fact that investors’ expectations are not fully adapted to all responsibility-related information, and its potential influence on companies’ expected future cash flows (and hence their fundamental value) (Boutin-Dufresne and Savaria, 2004; Derwall et al., 2011). Derwall et al. (2005) suggest that investors might systematically underestimate the value of ESG information.

Empirical research on the performance of virtue stocks is however mixed. Durand et al. (2013b) finds no outperformance for virtue stocks in the U.S. during the period 1990-2008. Studies on the performance of SRI mutual funds also provide with mixed results (Chegut et al., 2011; Kiymaz, 2012). Aside from the methodological issues highlighted in Chapter 1, Galema et al. (2008) and Lam et al. (2012) suggest that ESG performance could be priced in B/M ratios, explaining the absence of four-factor alphas. We briefly consider some notable studies on SRI funds. Bauer et al. (2005) compares 103 SRI funds with 4,384 non-SRI funds over the 1990-2001 period, and finds that SRI funds underperform conventional funds in the U.S., yet they outperform in the UK. Renneboog et al. (2008b) extends this type of analysis to the global market for the period 1991-2003, studying 440 SRI funds compared with 16,036 non-SRI funds. They find that the risk-adjusted returns of the average SRI fund is about 5% smaller relative to benchmark

8_{As current empirical literature has not wholly established the size and sign of the abnormal performance, we do not include a}

(11)

8 portfolios. The authors hereby suggest that SRI might adversely affect performance. Edmans (2011) nonetheless studies the Social pillar item of employee satisfaction, and finds that ‘’the 100 Best companies to Work for’’ have an alpha of 4%, which suggests that at least some positive screens may improve risk-adjusted stock performance. Derwall et al. (2005) furthermore concludes that stocks with a high score on their eco-efficiency perform about 3% better than stocks with a low score during 1995-2003.

Considering these mixed results, our hypothesis concerning virtue stocks is that they generate a statistically significant abnormal return relative to the market (compare e.g. Renneboog et al., 2008b; Jo et al., 2010 and Derwall et al., 2005; Edmans, 2011). We expect no differences in effects of separate ESG measures, as literature provides with opposing arguments to support each of the so-called ESG factors (Capelle-Blancard and Monjon, 2012), and considering the global nature of our sample. Lastly, we expect a lower downside risk for the Virtue portfolios, as Nofsinger and Varma (2012) argue that the nature of responsible investments may dampen their downside risk. Companies with high ESG scores have a lower probability to suffer large, negative events related to ESG aspects (e.g. environmental disasters, employee-related lawsuits, or agency costs) (Nofsinger and Varma, 2012). As a consequence, these companies have the potential advantage of more stable stakeholder relationships, which efficient markets would react upon in advance (Nofsinger and Varma, 2012; Gangi and Trotta, 2013).

2.3. Responsible (screened) portfolio returns

Considering the performance of portfolios that implement SRI by means of positive or negative screening, we might have an expectation that, if sin stocks outperform non-sin stocks or the market, an investors that shuns sin stocks from a broad market portfolio would hurt her return performance (Adler and Kritzman, 2008; Fabozzi et al., 2008). Similarly, given some evidence for the outperformance and reduced risks of virtue stocks (see the above discussion), we might expect positive screening strategies to improve return performance. With respect to screening, Durand et al. (2013b) state that if social norms provide a spur to norm-compliant investment activities (negative screening), we may ceteris paribus expect an opposing effect, i.e. positive abnormal returns, for responsible investments (positively screened portfolios). We could furthermore state that the implementation of negative screening is simpler and results in lower costs relative to positive screening portfolios, due to e.g. no need for extensive monitoring and managing of information on numerous ESG indicators (see Fabozzi et al., 2008).

Empirical results on screened SRI portfolios or funds however provide inconclusive results. Statman and Glushkov (2009) suggest that positive and negative screens employed by SRI funds may have opposing effects on performance, leading to conclude that SRI in general has no effect. Kiymaz (2012) adds that findings might moreover be specific to study periods. In this sense, Jo et al. (2010) find the responsible DS400 Index to outperform the market for the long-term (1990-2009), yet it underperformed the market when considering the 5 and 10 year-returns.

(12)

9 performance for SRI unit trusts relative to conventional unit trusts. Liston and Soydemir (2010) find a statistically higher Sharpe ratio for faith-based SRI portfolios relative to Sin stock portfolios.

A notable study by Humphrey and Tan (2013) considers negative screening and positive screening strategies for Triumvirate of Sin plus Defense stocks. They use benchmark indexes (e.g. the S&P500) as investment universe to mimic SRI funds’ practices, and find that excluding sin stocks or including virtue stocks from these indexes does not damage portfolio performance. Lobe and Walkshäusl (2011) and Lam et al. (2012) confirm this result. In a comparison between U.S. Sin and Faith-based portfolios Lyn and Zychowics (2010) find that faith-based funds mostly outperform the market, and moreover that faith-based funds do better than SRI funds in general. However, Jo et al. (2010) find the responsible DS400 Index to underperform the market when considering 5 and 10 year-returns. Chong et al. (2006) consider the Domini Social Equity Fund to be underperforming the S&P500. Moreover, in a meta-analysis on SRI mutual fund research, Renneboog et al. (2008a) concludes that literature so far hints that SRI investors are willing to accept suboptimal financial performance to pursue social or ethical objectives.

Lastly, there is some evidence regarding the impact of the intensity of screening. However, most studies consider only positive screens (e.g. Barnett and Salomon, 2006), while our interest lies with negative screening. Renneboog et al. (2008b) show that the number of ethical or sin screens do not have any significant impact. Capelle-Blancard and Monjon (2012) however find that the intensity of sector-screening reduces the risk-adjusted return, while norm-based negative screens have no impact.

Considering the above diverging results, we expect the negatively screened SRI portfolio to generate a significant abnormal return relative to the unscreened market portfolio (compare e.g. Liston and Soydemir, 2010 and Humphrey and Tan, 2013). A second hypothesis would be that the abnormal return of the negatively screened portfolio is statistically different from that of the positively screened portfolio (ibid). We furthermore expect the positively screened portfolio to display a significant abnormal return relative to the unscreened portfolio. Lastly, in the same line of reasoning as virtue stocks, we expect screened SRI portfolios to have lower downside risk relative to the unscreened market portfolio (see Nofsinger and Varma, 2012).

2.4.Responsible and sin investing during crisis periods

It would be interesting to test whether the recent financial crisis has affected the relationship between return-performance on the one hand and responsible and sin investing on the other hand. In particular, we could test the promise of recession-proof Sin portfolios (as suggested by e.g. Ahrens, 2004; Waxler, 2004; Salaber, 2009). Jo et al. (2010) state that the superior performance of sin stocks during crisis periods might be due to the fact that people might drink, smoke, or gamble more during tough economic times than good times. These statements however need some reservations, be it only because of diverging empirical results. Campus (2013) and Richey (2013) e.g. only establish significant alphas for Sin portfolios during bull markets.

(13)

10 propose that the decreased downside risk for virtue stocks and positively screened investments might mostly be noticed during bad economic times. Investors would prefer this countercyclical feature of responsible investments. Nofsinger and Varma (2012) lastly find that SRI portfolios using negative screens do not outperform during crisis periods.

In light of the inconclusiveness of the literature so far, we hypothesize that sin stocks, virtue stocks, and the responsible (screened) portfolio display lower downside risk and show significant abnormal returns relative to the market during a crisis period.

2.5.The value of negative screening

Screening controversial sectors might be controversial too. A first important concern of avoiding sin stocks is its possible incongruence with fiduciary duty law, due to the sacrifice of financial performance, as mentioned earlier. Secondly, De Colle and York (2009) argue that sector-based negative screening practices of SRI fund managers are not justified as these fail to accurately reflect investors’ values and moral orientations. Besides, ethical screening may thereby not be as ‘ethical’ as its name suggests (see Scharwz, 2003; De Colle and York, 2009; De Bruin, 2013). De Bruin (2013) most notably highlights issues of democratic legitimacy and effectiveness. Negative selection criteria should only be relevant to society-wide funds when its underlying moral norms are sufficiently universally accepted. This could imply a shift towards norm-based screening practices. However, SRI investors should preferably concern themselves with alternative means to achieve their stated ends (to reflect their personal values and ethics and encouraging corporate social performance), for instance focusing on an approach of engagement based on companies’ external impacts on their stakeholders (De Colle and York, 2009; De Bruin, 2013; Dimson et al., 2013).

Some ESG raters or SRI funds may however be committed to ban particular industries on the basis of their mission, values, or ethical beliefs regarding those industries. In these cases it would be problematic to oppose these practices; nonetheless, they should provide with transparent arguments and asset allocation decisions. Hawken (2004) states that practices frequently fall short of this capacity, e.g. the Dow Jones Islamic Index Fund did not exclude companies involved in corruption profiteering and war-mongering, and the Global Eco Growth Fund included companies with poor environmental records such as Exxon Mobil. Moreover, while exclusionary screening practices would be allowed for the above specific (e.g. faith-based) SRI funds, they should not be allowed to be adopted by society-wide institutional investors (such as pension funds) when their moral basis is not supported by a sufficiently wide consensus.

(14)

Capelle-11 Blancard and Monjon (2012) find that SRI funds that follow a Best-in-class approach sometimes are barely distinguishable from conventional funds. Moreover, Scalet and Kelly (2010) find that being dropped from a CSR ranking hardly encourages firms to address ESG concerns. We would add that any form of screening (both positive and negative) has to withstand the above mentioned critiques, as the practice of screening by definition implies the selection of some companies and the exclusion of others. We can define negative screening in a ‘positive sense’, i.e. as selecting companies that are not active in sectors or activities regarded as immoral, and positive screening in the ‘negative sense’ of excluding companies that do not meet certain standards. Still, negative screening remains problematic for practical reasons too. Woodbridge (2011) argues that, in the pornography sector, negative screening is very difficult to apply to distribution issues. A whole range of companies could potentially be excluded on the basis of their role in distributing adult content, such as TVs, laptops, mobile phones, and game consoles. Therefore, the most sensible SRI strategy might be engaging with companies, e.g. to establish opt-in basis accessibility to adult material. Additionally, if there is a sufficiently wide consensus, negative screens on producing companies could be employed to cause higher financing costs and constraints, making it more difficult for business in the industry to sustain their operations, and possibly induce firms to reorganize their activities and behave more responsibly (see Heinkel et al., 2001). The historical successes of SRI (discussed by Landier and Nair, 2009) moreover highlight the potential of negative screening strategies to establish societal changes.

In conclusion then, we can maintain that (especially norm-based) negative screening accompanied by other forms of SRI, such as engagement, has the potential to fulfill the objectives of responsible investors to reflect their values and exercise a strong voice in influencing company behavior.

3. Data

This chapter describes the data that we use to test our hypotheses. We will respectively explain the process of constructing the Sin, Virtue, and screened (SRI) portfolios.

3.1. What is ‘sin’?

(15)

12 We construct the Sin portfolios as follows. Using the ORBIS company database, we employ industry type (using NAICS and NACE codes, the latter being a major European industry standard classification system)) as well as elaborate search methods to retrieve potential sin stocks for fourteen different sectors9_{. We study the following ‘sin’ sectors: Abortion/abortifacients, Adult entertainment, Alcohol, Animal}

testing, Contraceptives, Controversial weapons, Fur, Gambling, Genetic engineering, Meat, Nuclear power, Pork, (Embryonic) Stem cells, and Tobacco. These sectors are selected and defined analogous to commonly employed definitions of ‘sin’ by ESG raters and SRI funds (see Renneboog et al., 2008a; Fabozzi et al., 2008; or MSCIs elaborate ‘’Business Involvement Screening Research’’, 2013)10_{. Appendix A provides our exact}

definitions. Sin sectors that are completely new to the literature are Abortion/abortifacients, Contraceptives, Fur, Meat, Pork, and Embryonic Stem cells. We would like to emphasize that we do not contend that investors should regard these sectors sinful or immoral, but rather merely that they in practice frequently do

(see Renneboog et al., 2008a; Lobe and Walkshäusl, 2011). See e.g. Viviers and Firer (2013) for a description of the perceived ‘sinfulness’ of most of the screens.

We refine the retrieved sin stock lists on a manual basis. We check general, history, and activity descriptions of companies on whether they are really ‘sinful’ according to the definitions as provided in Appendix A, to end up with a preliminary sample of sin stocks for each of the fourteen sectors. A detailed description of our identification process is included in the last Appendix H. To circumvent survivorship bias, we include in our sample all dead (delisted) stocks.

We match the sin lists from ORBIS with Thomson Reuters’ DataStream end-of-the-month data on returns11_{-, market capitalization (price times the number of outstanding common shares), and}

Dollar-Euro exchange rates, by removing the parts for which either the data on returns or market value is not available. For practical purposes we use DataStream’s Euro data. Additionally, although uncommon in the literature, the treatment of 0-returns in stocks’ return series needs to be addressed. As it will involve some subjectivity, 0-returns call for a clear and uniform treatment method. We require a stock’s return series to cover at least twelve months of continuous data (which is in line with Salaber (2009, 2013)). Furthermore, the 0-returns period must not persists longer than three months or occur more than three times in a possible twelve month series. If these conditions are satisfied we replace these ‘incidental’ 0-returns with the market return. When the 0-returns are non-incidental (i.e. they do not satisfy the conditions), the series are deleted up to the point for which conditions are satisfied, keeping the longest available valid series, in some cases implying deleting the stock altogether. After clearing the sample for available return data (as provided by DataStream), sufficient available return data (considering 0-returns), and available data on market value (DataStream), we are left with a total of 1,763 sin stocks across fourteen sectors, 1,634 after removing

9_{A potential shortcoming of using the ORBIS dataset would be our inability to control for changed company activities, since}

searches had to be made within product descriptions which are updated to the most recent activities. However, potential biases will be marginal as company involvement in particular sin industries generally does not change radically over time, and company history descriptions would capture these changes as well.

10_{Another commonly applied screen is e.g. interest-based financial institutions (as commonly avoided by Islamic investors), yet this}

screen was problematic i.a. to properly identify all companies involved in this business.

11_{We measure returns as the natural logarithm of a stock’s}_{Total Return index}_{on t=0 divided by the index on t=-1. The Total}

(16)

13 duplicates, which is considerably larger than previous studies. The complete list of sin stocks is included in Appendix B. Table 1 shows the sample clearing process with per sector information on the number of stocks in the remaining investable sample.

Table 1: Sin sample clearing

Ab or tio n / ab or ti fa ci en ts A du lt en ter ta in men t A lc oh ol A n ima l test in g C on tr ac eptives C on tr ov er si al wea po n s Fu r in du st ry Ga mb li n g Gen etic en gi n ee ri n g M eat N u cl ear po we r Po rk Stem ce lls To ba cc o To ta l

Initial list (ORBIS) 6 21 494 118 110 22 46 144 76 708 372 179 20 128 2,444

Avail. return data 6 21 399 113 104 21 25 133 75 502 344 119 20 90 1,972

Avail. sufficient return data 4 20 377 112 96 20 25 129 74 433 328 113 18 80 1,829

Avail. sufficient MV data 4 20 361 112 93 20 22 126 74 403 327 107 18 76 1,763

Final sample 4 20 361 112 93 20 22 126 74 403 327 107 18 76 1,763

Non-duplicate: 1,634

Aside from analyzing individual Sin portfolios we will consider various combinations, including the classical Triumvirate of Sin (as commonly studied), the so called ‘4Bs’ portfolio of booze, bets, bombs, and butts

(Alcohol, Gambling, Controversial weapons, and Adult entertainment), as proposed by Ahrens (2004), the ‘Sextet of Sin’ portfolio (Alcohol, Tobacco, Gambling, Controversial weapons, Adult entertainment, and Nuclear power), as studied by Lobe and Walkshäusl (2011) and identified by Sparkes and Cowton (2004), and lastly potentially controversial Medical stocks, comprising the Abortion, Animal testing, Contraceptives, Genetic engineering, and Embryonic Stem cells sectors. Table A.1 in Appendix C shows the evolution of the number of stocks and their total market value for the main individual and combined Sin portfolios.

3.2.What is ‘virtue’?

Contrary to most previous research, we directly construct Virtue portfolios by using ESG scores from Thomson Reuters’ Asset4 dataset. The Asset4 data set is the most comprehensive ESG dataset in existence, with information on 4,000+ global companies and 750+ ESG data points covering every aspect of sustainability reporting (Asset4, 2012a). Asset4 ‘’strictly sources publicly available information, including sustainability / CSR reports, company websites, annual reports, proxy filings, NGO as well as news of all major providers’’ (Asset4, 2012a). ESG data are updated on a bi-weekly basis, yet company data only appear to differ per year. Since the process of positive screening is a delicate one, which can be (and has been) criticized for its subjective character (ESG raters may conceive and implement SRI very differently), and its various methodological problems (see e.g. Chatterji et al., 2009), we require measures of social responsibility that are maximally objective and reliable. The use of information from Asset4, a reliable rater with well-understood and sophisticated scoring techniques, would be a major advantage in this respect. The study of Lam et al. (2012) also uses the Asset4 data set and elaborately discusses the characteristics of the dataset.

(17)

14 scores, e.g. analyzing the portfolio of companies that have a high score on Environmental performance indicators. Previous research suggest that CSR and SRI are multidimensional concepts for which different subparts and strategies might have different effects on financial measures (see e.g. Hoepner et al., 2010; Edmans, 2011; El Ghoul et al., 2011). In all, these analyses will therefore provide with improved understanding of which (combination of) sin sectors and for which part of SRI we might observe certain risk and return characteristics.

Virtue portfolios are constructed by selecting stocks in the upper quintiles and deciles of Asset4’s overall/integrated ESG rating (A4IR), and three ESG pillar (sub)scores (ENV, SOC, and GOV) (see Asset4, 2012b for descriptions of these general scores). The analyses are based on the longest available period of 01/2002 to 12/2012. Table A.1 in Appendix C shows the evolution of the number of stocks and their total market value of the main Virtue portfolios.

The ESG score data are not free from problems when used in empirical analysis. First, companies may have different fiscal years and may publish ESG-related information at different dates. Second, ESG information may be published with a considerable delay to both company practices and their actual CSR reports. The first issue however should be marginal when considering the fact that nearly all Asset4 constituents have their fiscal year ending at the calendar year end. Based on the advice of Thomson Reuters, we will use the end-of-the year ESG scores which we take to be valid for the past year. In a sensitivity analysis we will also use scores at the end of period t=1 and let it be valid for the whole t=0 period. We do this to mitigate the second issue mentioned above, i.e. to capture the possible effects of delay in information processing with respect to the publication of CSR reports as well as the fact that Asset4 would need maximally two months to process these reports. Since scores only (tend to) vary on a 12-month period basis, the one-year lagged scores will capture potential delay issues12_{. Appendix D, Table A.3 provides the main}

results for the lagged Asset4 scores.

3.3.Implementation of positive and negative screening

In line with Humphrey and Tan (2013) we analyze the risk-adjusted return performance of a portfolio that implements SRI by means of positively and negatively screening a broad market portfolio. Because of data availability reasons, we use the S&P500 index (which is also employed by Humphrey and Tan, 2013). We retrieve beginning and end-of-the-year constituent list, returns, and market value data for the S&P500 from DataStream. In order to construct the negatively screened portfolio we delete the stocks belonging to the TotalSin portfolio (all fourteen sectors) from the end-of-the-year S&P500 constituents13_{. We redo the}

analysis by excluding only Triumvirate of Sin (Alcohol, Tobacco, Gambling) stocks. In this way, we might get some preliminary insights into whether increasing the number of screens actually hurts return

12_{To clarify, when a company has changed with respect to their sustainability policy and practice as of summer 2011, the company}

might fully disclose this information in their annual CSR report, which is published somewhere at the end of 2011. Asset4 then may have incorporated this information in their aggregate scores by the end of January 2012 or possibly at the very end of 2011. This means that information about 2011 may potentially be incorporated in the 2012 ESG scores of Asset4. Taking the score at t+1 then would accommodate for this potential issue of delay.

(18)

15 performance. We construct a positively screened portfolio by selecting on a year-to-year basis the A4IR Quintile stocks present in the S&P500. Table 2 depicts the number of stocks excluded from or include in the S&P500 market portfolio. Contrary to Humphrey and Tan (2013)’s statements there are a reasonably large number of sin stocks to be shunned from a broad market index.

Table 2: Composition of screened S&P portfolios*

Negatively screened S&P Positively screened S&P

Excluding TotalSin Excluding Triumvirate of Sin Including A4IR Quintile

Year # stocks excluded # stocks left # stocks excluded # stocks left # stocks included # stocks left

1991 36 464 7 493 NA NA 1992 37 463 7 493 NA NA 1993 37 463 7 493 NA NA 1994 37 463 7 493 NA NA 1995 39 461 7 493 NA NA 1996 39 461 7 493 NA NA 1997 40 460 7 493 NA NA 1998 41 459 7 493 NA NA 1999 43 457 7 493 NA NA 2000 44 456 7 493 NA NA 2001 45 455 7 493 NA NA 2002 47 453 8 492 29 471 2003 46 454 8 492 41 459 2004 47 453 8 492 58 442 2005 49 451 9 491 67 433 2006 50 450 9 491 69 431 2007 50 450 9 491 74 426 2008 52 448 10 490 95 405 2009 52 448 9 491 106 394 2010 52 448 9 491 101 399 2011 52 448 9 491 97 403 2012 51 449 10 490 24 476

* More details on the screened market portfolio are available at request.

4. Methodology

This chapter describes the methodology that is used to test our hypotheses. First we delineate our return performance measures, focusing on both traditional as well as downside risk-related measures. Next, we describe and motivate the regression model that we use to explain risk-adjusted returns, along with the choice of each of its inputs. Sections 4.3-4.6 discuss our preference for using value-weighted returns, in a time series regression, using median regression techniques, and by means of zero-investment portfolios respectively. The last section delineates our interest in studying responsible and sin investing during the Great Recession.

4.1.Risk and return performance measures

(19)

16 which might better reflect investors’ actual concerns and preferences (Plantinga and De Groot, 2001). Aside from the traditional and Adjusted Sharpe ratio, we consider three common Downside Risk measures, namely the Semi Deviation, Sortino Ratio, and Upper Potential Ratio. A study of Viviers and Firer (2013) on South African SRI unit trusts is –to our knowledge– the only SRI-related study that also employs Downside Risk measures.

4.1.1. Sharpe ratios

In the traditional CAPM setting, investors are only able to improve performance by increasing the slope of the Capital Market Line, which is known as the Sharpe ratio (Sharpe, 1966)14_{. An important property is}

that the Sharpe ratio can be used as the objective function in mean-variance optimization, where the portfolio with the highest Sharpe ratio is the optimal portfolio of risky assets. The Sharpe ratio is defined as follows:

𝑆𝑝=

𝐸(𝑅_𝑝)−𝑅_𝑓

𝜎_𝑝 (1)

Pézier and White (2006) propose an improvement on the above ratio, considering the effects of non-normal Skewness and Kurtosis:

𝐴𝑆𝑅𝑝= 𝑆𝑝[1 + ( 𝑆𝑘𝑒𝑤 6 ) 𝑆𝑝− ( 𝐾𝑢𝑟𝑡−3 24 ) 𝑆𝑝 2_] ₍₂₎ 4.1.2. Semi deviation

While the above performance measures are widely used by practitioners and academics, they increasingly have to give way for performance measures based on Downside Risk (DR) (e.g. Chaudhry and Johnson, 2008). The relevance of these DR measures lies in their attempt to represent (or measure) the risk that is included in the ‘actual’ utility functions of investors, i.e. investors should primarily care about the risk of negative or undesirable outcomes. Whereas risk is commonly measured as the standard deviation of return series, DR defines risk in terms of ‘bad volatility’, i.e. the risk of returns below a certain Minimal Acceptable Rate of return (MAR) (Sortino and Van der Meer, 1991). Secondly, the DR approach incorporates part of the investor’s preference function by introducing reference rates (Plantinga and De Groot, 2001).

An important starting point for DR-based performance measures is the Semi Deviation of a return series, which is a simple measure of the standard deviation above or beneath a certain reference level. We calculate this measure as the standard deviation of the negative values in our return series. This is in line with Viviers and Firer (2013).

14_{The Sharpe ratio can be interpreted as a t-test for the hypothesis that the return on the portfolio is equal to the risk free rate}

(20)

17

4.1.3. Sortino ratio

Probably the most common risk-adjusted performance measure using DR is the Sortino Ratio, as developed by Sortino and Van der Meer (1991). It is calculated as follows:

𝑆𝑜𝑟𝑡𝑝= 𝐸(𝑅𝑝)−𝑅𝑀𝐴𝑅 𝛿𝑝 (3) 𝛿𝑝= √ 1 𝑇∑ 𝜄 −_(𝑅 𝑝,𝑡− 𝑅𝑀𝐴𝑅) 2 𝑇 𝑡=1

Where 𝛿𝑝 is the downside deviation of the portfolio, T the total number of periods, 𝜄− a dummy variable

with 𝜄−=1 for 𝑅𝑝,𝑡≤ 𝑅𝑀𝐴𝑅 and 𝜄−=0 for 𝑅𝑝,𝑡> 𝑅𝑀𝐴𝑅, 𝑅𝑝,𝑡 is the return of the portfolio in period t, and

𝑅𝑀𝐴𝑅 stands for the Minimal Acceptable Rate of return, which we take to be the risk-free rate (keeping the

numerator identical to the numerator of the Sharpe ratio).

The Sortino ratio can be viewed as a refinement of the Sharpe ratio. It favors investments with the highest return for a given level of downside risk, which is the risk where investors do generally build their utility (or value) function on (Plantinga and De Groot, 2002). Chaudhry and Johnson (2008) moreover find that the Sortino ratio is more powerful and results in less bias when compared to the Sharpe ratio when excess returns are skewed.

4.1.4. Upside potential ratio

Sortino et al. (1999) propose to replace the expected return by the Upside Potential Ratio (UPR), which is a measure of the potential reward of an investment opportunity. The UPR is calculated as follows:

𝑈𝑃𝑅𝑝= 1 𝑇∑𝑇𝑡=1𝜄+(𝑅𝑝,𝑡− 𝑅𝑀𝐴𝑅) 1 𝑇∑ 𝜄−(𝑅𝑝,𝑡− 𝑅𝑀𝐴𝑅) 2 𝑇 𝑡=1 (4)

Where 𝜄+=1 for 𝑅𝑝,𝑡> 𝑅𝑀𝐴𝑅, 𝜄+=0 for 𝑅𝑝,𝑡 ≤ 𝑅𝑀𝐴𝑅, 𝜄−=1 if 𝑅𝑝,𝑡 ≤ 𝑅𝑀𝐴𝑅, and 𝜄+=0 if 𝑅𝑝,𝑡> 𝑅𝑀𝐴𝑅. The

UPR better addresses the risk preferences of investors, as it favors investments that seek upside potential along with strong downside protection (see Shefrin, 1999). Plantinga and De Groot (2002) add that the UPR has as an important advantage over the Sortino ratio in that it is consistent in its use of the reference rate for evaluating both profits and losses.

(21)

18 not valid for non-normal data. As a last note, we should interpret results from the above ratio-based analysis with some caution, as these performance ratios generally measure returns adjusted for the risk-free rate. They do not control for common risk factors (such as the Size and B/M ratio of portfolios), and hence our regression analyses (described in Section 4.1) which focus on outperformance or abnormal returns (abnormal relative to well-documented risk factors as well as benchmarks), might yield different conclusions.

4.2. Model specification

Following related literature, we estimate a portfolio’s risk-adjusted performance using the Carhart (1997) four-factor model, specified as follows:

𝑅𝑖,𝑡−𝑅𝑓,𝑡= 𝛼 + 𝛽𝑀𝐾𝑇(𝑅𝑚,𝑡−𝑅𝑓,𝑡) + 𝛽𝑆𝑀𝐵𝑆𝑀𝐵𝑡+ 𝛽𝐻𝑀𝐿𝐻𝑀𝐿𝑡+ 𝛽𝑊𝑀𝐿𝑊𝑀𝐿𝑡+ 𝜀𝑖,𝑡 (5)

This equation expands the traditional Fama-French (1993) three-factor model by the Jegadeesh and Titman (1993) momentum factor. According to the four-factor model, the excess return on a particular security or portfolio can be measured by four common ‘risk factors’. 𝑅𝑚,𝑡−𝑅𝑓,𝑡, the market risk premium, is

represented by the excess return on a value-weighted market portfolio. 𝑆𝑀𝐵𝑡 is measured as the difference

in monthly return between a portfolio of small market capitalization stocks and a portfolio with big market capitalization stocks. The 𝐻𝑀𝐿𝑡 factor is represented by the difference in monthly return between a high

book-to-market portfolio minus a low book-to-market portfolio. 𝑊𝑀𝐿𝑡 is the difference in monthly return

between a portfolio of winners minus a loser portfolio. Lastly, 𝛼 represents the variable of interest, namely the abnormal (out- or under) performance of a particular investment portfolio, which cannot be explained by the above four risk factors.

While the Fama-French (1993) and the Carhart (1997) factor models have been proven to be empirically very strong, they nevertheless can be criticized for their theoretical soundness15_{. While there}

appears to be a clear match between the data (measurement results) and the model, we cannot establish a link between the observable phenomena (financial performance) and the relevant theoretical structure. We however need not go as far as assenting to our model’s full truth. We can be agnostic about the real existence of our model’s inputs, and embrace a view of empirical adequacy only (see Van Fraassen (2008) for an interesting discussion).

4.2.1. Fama-French factor data

Data on the regression factors are most often based on the Fama and French (1993) U.S. factors. Yet, as these factors might be inaccurate to specific study samples, it would be desirable to construct the factors ourselves. Unfortunately, we are unable to retrieve data on the yearly constituents for the MSCI ACWI (as contact with Thomson Reuters made clear). To avoid survivorship bias issues, we are therefore forced to use

15_{E.g., Fama and French (1993) state the factors to be no ‘anomalies’ but rather}_{risk factors}_{common to all stocks. The authors}

(22)

19 the factors as available on Kenneth French’s website16_{. We will use the recently published Global factors, as}

these most adequately represent the elements of our total global sample17_{. The Fama-French factor data are}

transformed into Euro terms using DataStream’s Dollar-Euro exchange rates.

There are two further limitations of using these factors. One is that the global factors are only available as of 01/1991, which reduces our study period by three years. A second restriction is that the global factors are constructed for developed countries only, which may distort the ‘true’ values of the factors (when emerging markets would have been included as well).

4.2.2. Market portfolio and risk-free rate

Our population as well as our final sample consists of companies in developed and emerging markets across the world. To adequately represent (or proxy for) the market portfolio we use the MSCI All Country World Index (ACWI), as this index consists of more than 2,400 large- and mid-cap stocks across developed as well as emerging markets. Lam et al. (2012) uses the MSCI ACWI index as well. The MSCI ACWI furthermore shows returns very similar to the commonly employed MSCI World index (a correlation of 99.8% during our study period). As explained above, we use the commonly employed S&P500 in our analysis on the negatively and positively screened broad market portfolio (which is similar to Humphrey and Tan, 2013). We refrain from utilizing national market indexes as this would imply adherence to the somewhat ambiguous assumption of completely home-biased investors, i.e. investors that do not invest a single euro in stocks which are not listed on their domestic stock exchange. We will follow mainstream finance literature in assuming (semi-strong) globally efficient and well-diversified markets.

With similar line of reasoning, we choose a global risk free rate proxy. We will use the U.S. one-month Treasury-bill rate provided by Kenneth French’s website, which is generally recognized as the conventional proxy for the risk-free rate (Fama and French, 1997). This is in line with the related studies that provide information on their data sources (Galema et al., 2008; Liston and Soydemir, 2010; Nofsinger and Varma, 2012; Campus, 2013)18_.

4.3.Value-weighted vs. equal-weighted returns

Empirical studies frequently use equally-weighted average portfolio returns. This method makes portfolio returns easy to calculate, and moreover gives a balanced view of the particular portfolio, showing the return an investor may expect when she considers to invest in the particular (sin) sector, irrespective of a presupposed asset allocation. Portfolios constructed using value-weighting, on the other hand, provide with a more pragmatic picture of sin investing: SRI or sin investors and funds typically select stocks out of a broad

16_{http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html}

17_{Fama and French (2012) recommend using the global factors in applications to explain the returns on global portfolios, e.g. to}

evaluate the performance of mutual funds that holds a global portfolio of stocks, as long as the portfolio does not have a strong tilt toward microcaps or towards the stocks of a particular region, which our sample does not.

18_{There are some critical notes to place with these data, as the same rate when downloaded from DataStream is found to be about}

(23)

20 market portfolio as their potential investment universe (usually the S&P500), which consists of large-caps, and may moreover be subject to the habit of investing predominantly in well-known companies. The study of Humphrey and Tan (2013) takes such a pragmatic approach. Moreover, equal-weighting might not be feasible or desirable for institutional investors, which make up for a large part of the total SRI market (about 95% in Europe (Eurosif, 2012)). When allocating only 1/1000 percent of their capital to a group of small companies, they will end up as complete owners of these small companies, which probably will not be their intention or preference. Lastly, Fama (1998) argues that value-weighted average returns more accurately capture the total wealth effects experienced by investors. Considering the above arguments, practice in related literature (e.g. Lobe and Walkshäusl, 2011; Salaber, 2013), and the fact that our market portfolio proxy is weighted for market value, we rely on value-weighted average returns in our analysis. As a sensitivity analysis, we also base our main analyses on equally-weighted returns. The outcomes of these analyses are depicted in Tables A.4-A.7 of Appendix E. Section 5.2.2 discusses the differences in outcomes between the two weighting methods.

4.4.Unbalanced panel and choice for time series regression

Since the monthly return for the portfolio is taken to be the weighted average return on the total of the stocks present at time t, the amount of return observations to base portfolio returns on varies across time (usually the number of observations is lower for earlier years)19, 20_{. To illustrate, the return series for our}

sector Contraceptives consists of 35 stocks at the beginning of our study period, 63 stocks in 2000, and at the end of our study period the amount of stocks in the Contraceptives portfolio totals 92 stocks.

A cautionary note should be placed on this procedure. First, it can create a bias towards certain sectors, as the fraction of a particular Sin portfolio to the TotalSin portfolio varies across time. The classical Triumvirate of Sin portfolio might for example be biased towards tobacco in earlier years and towards gambling for more recent years. We however checked for this effect (using Table A.2 in Appendix C), and it is generally marginal, as the fraction of the sin sectors’ market value relative to the total sin sample typically change no more than 1% during the study period. A possible way to circumvent this problem would be by using sophisticated panel regression methods. All available literature however leaves the issue untreated (most studies do not even recognize it) and employs ordinary time-series regressions. Possibly, the above mentioned issue might not constitute a problem at all, as the analysis will provide us with a ‘real’ picture of the potential sin universe in earlier years. An investor that chooses to invest in the classical sin stocks would come across more alcohol and tobacco stocks in her Sin portfolio for earlier years, and more gambling stocks as sin investment opportunities in more recent years. Furthermore, the use of an unbalanced panel for both the sin and the positively screened sample has as an advantage that it decreases the potential for survivorship

19_{We check the fraction of the amount of stocks in the particular portfolio at the beginning of the study period relative to the}

amount at the study period's end. On average about one-third of the total amount of stocks is present at the beginning of our study period. Typically the fraction is above the 25% figure, with as evident exceptions Adult Entertainment, Animal testing, and Stem cells (for which the number of (listed) companies involved has increased particularly from the 2000s onwards).

20_{Note that the total amount of stocks at the end of the study period may be lower than the initial figures mentioned earlier. This}

Bad Companies, Bad Investments? On the Return and Risk Characteristics of Responsible and Sin Investing☆