Corporate Environmental Performance versus Stock Returns
Marien de Haan1 July 2011
Abstract
Is there a relationship between corporate environmental performance (CEP) and stock returns? And if so, what drives this relationship: changes in corporate risk exposure or mispricing because of investors’ taste for high CEP stocks, based on personal values or social norms? To answer these questions, I use a new and comprehensive ranking that measures the environmental performance of the 500 largest publicly traded United States corporations. The methodology in this study is based on the Fama-French-Carhart four factor asset pricing model. In addition, I incorporate a fifth factor to capture common CEP related risks. The results point to a negative relationship between CEP and stock returns; partially driven by common CEP related risks. At the same time though, the influence of taste cannot be ruled out. The relationship appears to be influenced by investors’ taste for high CEP stocks.
Keywords: corporate environmental performance; ranking; stock return; taste; risk; multifactor model JEL classification: A13; G11; G12
1
Marien de Haan, student number: 1322435, e-mail: mariendehaan@gmail.com. University of Groningen, Faculty of Economics and Business, Master’s Thesis BA Finance.
1 Introduction
Green seems to become more mainstream. The performance of the latest generation of environmentally friendly hybrid cars matches that of their conventional counterparts. But what about corporations? How do green corporations measure up to the less responsible ones? More specifically, how does their corporate financial performance (CFP), as measured by stock returns, compare?
This issue is particularly important in today’s world where investors like pension funds adopt sustainable investment policies, whether or not encouraged by public opinion or political pressure. Such sustainable policies seem responsible from an environmental perspective, but are they responsible from a financial point of view as well? I seek to answer the following research question: Is
there a relationship between corporate environmental performance (CEP) and stock returns? This
study contributes to the existing literature by differentiating between potential drivers of this relationship. I therefore formulate a second (follow-up) research question: What drives this possible
relationship: changes in corporate risk exposure or mispricing of stocks (not based on economic fundamentals) due to social norms or personal values (i.e. investors’ taste)?
Another contribution of this study is that uses a new and comprehensive composite ranking to measure the environmental performance of the 500 largest publicly traded United States corporations, the Newsweek Green Rankings 2009.2 In this ranking, a corporation’s overall score is based on sub-scores for environmental impact, policies and reputation. Furthermore, the methodology in this study is based on the asset pricing model of Fama and French (1993) and Carhart (1997). So the returns of low and high CEP stocks are compared only after adjusting for common risk factors associated with market activity, corporation size, book-to-market value and stock momentum.
In order to answer the first research question, I follow a two step approach. The first step is to estimate four factor betas for each corporation over the period from March 2004 until September 2008. These betas control for the common risk factors in a second step cross-section regression with CEP (sub-)scores as independent variables and average stock returns as dependent variable. The results point to a negative relationship between CEP and stock returns, that is robust to sample period and major credit crisis events.
To answer the second research question, I extend the four factor model with a fifth factor to capture any common risks related to CEP. I estimate this five factor model for nine stock portfolios sorted on overall CEP score. The fifth beta of several portfolios is statistically significant, indicating that the relationship between CEP and stock returns is at least partially risk-driven. At the same time though, the inclusion of the fifth CEP factor does not push alpha to zero, leaving room for the influence of taste. I perform a second test to determine whether the five factor alphas actually
2
represent compensation for investors’ taste for high CEP stocks (as opposed to compensation for latent risk factors).
The second test compares the five factor alphas of a group of stock portfolios sorted on environmental impact with those of a group sorted on environmental reputation. The series of Reputation alphas increases with deteriorating portfolio reputation, while the series of Impact alphas does not increase with growing environmental impact. This pattern indicates that the relationship between CEP and stock returns is influenced by investors’ taste for high CEP stocks. The increase of alpha with worsening environmental reputation, reflects extra compensation to induce investors to hold portfolios of dubious reputation. Such extra compensation is required, as the reasonably assumed visibility of environmental reputation, relative to that of environmental impact, forces investors to more openly violate tastes based on social norms and personal values.
This paper is organized as follows. Section 2 discusses the literature and introduces two testable hypotheses. Section 3 presents the data, it discusses in detail the CFP measure (stock returns), the CEP measure and the control variables. Furthermore, section 4 explains the methodology. Section 5 then analyses and interprets the results. Finally, section 6 concludes.
2. Literature review and hypotheses
2.1 The relationship between CEP and CFP
Capturing a broad concept such as CEP into a single definition is challenging. For this study, I adopt the European Commission’s definition of corporate social responsibility, narrowed down to cover only
environmental responsibility (Europese Commissie, 2001). So, I define CEP as the extent to which
corporations voluntarily, i.e. beyond legal obligations, integrate environmental concerns in their business operations and interactions with stakeholders (Europese Commissie, 2001). It is natural to base the definition of CEP on the definition of corporate social responsibility, since the first is usually considered an important part of the latter (see e.g. Orlitzky et al., 2003). Several studies on the relationship between corporate social performance and CFP actually use environmental performance as a proxy, Orlitzky et al. (2003). I include studies on corporate social performance in the literature discussion below as this helps understanding the relationship between CEP and CFP.
maximizing shareholder value (Sternberg, 1996;Jensen, 2002). Corporations that (also) pursue goals such as improving environmental performance are bound to lose shareholder value, as they will face increasing marginal costs and only limited additional benefits (Walley and Whitehead, 1994). Deviating from value maximization means imposing taxes on shareholders and deciding how it should be spent (Friedman, 1970). At the same time, shareholder theory does not prohibit corporations to allocate resources to environmental initiatives, as long as it ultimately raises the value of the firm. Indeed, such initiatives should be regarded only as a means to the end of profitability (Smith, 2003).
This viewpoint is not shared by those who claim a positive relationship between CFP and CEP (e.g. Porter and Van der Linde, 1995) based on stakeholder theory (Freeman, 1984; Donaldson and Preston, 1995; Jones, 1995). Stakeholder theory suggests that satisfying the interest of all stakeholders is key to the corporation’s continuity and is therefore regarded as an end in itself (Smith, 2003). In addition, such behaviour strengthens the competitive position of the corporation and, through that, may lead to financial gains (see Guenster et al., 2006). Porter and Van der Linde (1995) for instance, dismiss the notion of a trade-off between competitiveness and the environment. They make a case for proactively reducing corporate pollution because this may in itself generate innovations enhancing resource productivity and competiveness. Stakeholder theory differs fundamentally from neoclassical shareholder theory because it implies that cooperation, instead of opportunism gives corporations competitive advantages (Jones, 1995).
Critics of stakeholder theory emphasise that taking account of all shareholders is simply impracticable (Sternberg, 1996; Jensen, 2002). How is a manager to decide on conflicting interests? Or as Jensen (2002) puts it: “Multiple objectives is no objective”. Shareholder theory in turn, is criticized for too easily assuming that markets behave efficiently and do not cause external effects or distributional issues (Heal, 2005). These shortcomings can possibly be alleviated by combining elements from both theories: i.e. pursuing long term value maximisation without ignoring stakeholders or exploiting market failures (Jensen, 2002; Heal, 2005).
Apart from the theoretical discussion, many studies analyse concrete (positive) effects of environmental performance on different corporate areas. In this respect, Heal (2005)identifies six corporate areas alleged to benefit from CEP.3 The first area is corporate risk management. Heal (2005) himself suggests that a CEP programme helps to manage corporate litigation risks resulting from environmental conflicts with society. These conflicts likely arise when the costs of for instance polluting are much lower to the corporation than to society as a whole. A well executed CEP programme may mitigate such differences in private and social costs by internalizing (part of) the external costs of polluting. Heal (2005) believes higher CEP can prove profitable because it strengthens a corporation’s vital relationships. Feldman et al. (1996) develop a model for connecting
3
CEP to systematic (i.e. beta) risk. In this model, signalled improvements in environmental management convince investors that the company is less exposed to risks, like environmental disasters or increases in environmental legislation. This reduction in perceived risk diminishes the company’s beta.
Heal (2005) identifies waste reduction as the second corporate subject to CEP. As stated above, Porter and Van der Linde (1995) emphasise that waste (pollution) reduction triggers profitable innovations. The empirical study of Hart and Ahuja (1996) appears in line with this win-win point of view. From their results it seems the emission cuts of large US corporations in the late eighties are indeed followed by increased operating and financial performance. Hart and Ahuja (1996) also refer to the increasing marginal costs of reduction, corresponding to the aforementioned (sceptical) assertion of Walley and Whitehead (1994). King and Lenox (2002) confirm the results of Hart and Ahuja (1996). More specifically, they find that waste prevention (but not waste treatment) within US corporations in the first half of the nineties is associated with higher profits. In a more recent study, Guenster et al. (2006) reach similar conclusions on the effects of eco-efficiency within US corporations.
The third corporate area allegedly influenced by CEP is that of regulatory relationships. Corporations with sound environmental track records are probably in a better position to negotiate permission to expand their business. This is especially true in case of environmentally controversial expansions, like drilling for oil in the Arctic.4 (Heal, 2005; Ambec and Lanoie, 2008). Moreover, corporations with relatively high environmental performance may actually call for stringent environmental regulations in an attempt to lock in a competitive first-mover advantage (see e.g. Russo and Fouts, 1997; Ambec and Lanoie, 2008).
The fourth and fifth corporate area Heal (2005) distinguishes are related, so I combine them into one. CEP is alleged to have an effect on corporate relations with customers and employees. Branding, i.e. creating a desirable product image, is important in the marketing strategy of many corporations. Corporate environmental performance could add value to a brand (see e.g. Heal, 2005). Concretely, consumers might be willing to pay a premium for environmentally responsible products that appeal to their personal or social values (see e.g. Fombrun and Shanley, 1990;McWilliams and Siegel, 2001). Something similar could apply to the labour market. Employees tend to prefer working for companies they perceive as ‘good’; which may just as well be ‘good’ from an environmental responsibility perspective (see e.g. Heal, 2005).
Cost of capital is the sixth and final corporate area Heal (2005) considers to be influenced by CEP. I am particularly interested in any CEP effects on the cost of capital, or actually, on its mirror
4
image, stock returns, as I measure CFP in this study by stock returns. Therefore, I devote the rest of this subsection to (empirical) studies on the relationship between CEP and stock returns.
When it comes to this relationship, what matters is whether and how CEP is valued by financial markets. The aforementioned supposed effects of corporate environmental performance aside; ultimately, the relationship is determined by the extent to which investors factor CEP into stock prices (see e.g. Hamilton et al., 1993; Derwall et al., 2005; Dam, 2008). Various studies seek to examine the link between CEP and stock returns empirical; about twenty are included in the meta-analysis of Margolis et al. (2007) for instance. Comparing the studies is hampered by differences in study design. The measurement of CEP for example, varies across studies from environmental rankings to compliance expenditures and emission data (see e.g. King and Lenox, 2001; Wagner, 2001; Margolis et al., 2007). Broadly, studies can be grouped into three categories, including event studies, portfolio studies and multiple regression based studies (see Wagner, 2001).
The empirical study of Brammer et al. (2006) fits best in the multiple regression category. The authors use, among others, the familiar Fama and French (1993) three factor asset pricing model to examine the impact of environmental performance (as part of a composite measure of social performance) on the financial performance of United Kingdom corporations. They find that environmental performance is negatively associated with stock returns. Ziegler et al. (2007) also apply the three factor asset pricing model but report different results. The relative environmental performance of large European corporations within their industry does not influence risk adjusted stock returns. But the environmental performance of the industry to which a corporation belongs does affect returns positively.
Filbeck and Gorman (2004) employ a portfolio approach to study the link between CEP and stock returns. Their results point to a negative relationship between the environmental performance and stock returns of 24 large United States electricity companies. This negative relationship seems to hold for environmental compliance as well as for more pro-active corporate environment behaviour. Contrarily, Derwall et al. (2005) find that a portfolio of high CEP corporations in the United States provides higher risk adjusted returns than a portfolio of low CEP corporations does. Derwall et al. (2005) believe their findings are difficult to reconcile with the familiar risk-return framework and suggest there may be mispricing in financial markets, in this case underpricing of high CEP stocks.
a lag, by an increase in the stock returns of that corporation.5 An earlier event study of Klassen and McLaughlin (1996), including over 120 events relating to nearly 100 United States corporations, yields similar results. Public reports of corporation related environmental crises like spills, are associated with an increase in the corporation’s stock returns. The contrary also holds, positive news about winning an environmental award is followed by a stock return decrease.
I conclude this subsection by formulating a testable hypothesis about the relationship between CEP and stock returns (CFP). Above, I introduced the first research question as: is there a relationship between CEP and stock returns (CFP)? Based on previous theoretical and empirical literature, I hypnotize that:
H1: CEP and stock returns are negatively related.
2.2 What drives the relationship between CEP and stock return?
Why would corporations with high environmental performance generate low stock returns for investors? There seem to be two (not mutually exclusive) explanations for the underperformance of high CEP stocks, Renneboog et al. (2008). The first of these explanations is straightforward: the underperformance results from differences in risk exposure and the way they are measured. In particular, when high CEP stocks are truly less risky than low CEP stocks, but asset pricing models fail to capture this adequately, investors will perceive high CEP stocks as overpriced and underperforming low CEP stocks. This underperformance however would simply vanish if returns were properly adjusted for all relevant risk factors. So, high CEP stocks are indeed priced at their fundamental value, it is only that stock pricing models fail to recognize this.
According to the second explanation for underperformance, quite the opposite is true: high CEP stocks are actually not priced at their fundamental value, but are overpriced instead (Renneboog et al., 2008). High CEP stocks may become overpriced when many investors have a strong taste for environmental responsible corporations, based on their personal values or social norms; regardless of whether high environmental performance leads to lower financial risk. These investors derive utility from investing in responsible corporations as such, and may therefore settle for a lower expected rate of return. Accordingly, taste-augmented demand pushes high CEP stocks prices beyond the level that is justified by economic fundamentals.
Models linking CEP to stock returns typically assume a division between CEP-loving and other (e.g. CEP-neutral) investors. Such dichotomy leads to an excess demand for high CEP stocks and a short demand for low CEP stocks, causing the first to become overpriced and the latter to become underpriced (see Galema et al., 2008).6 Merton (1987) shows in a seminal paper on this topic
5
I interpret the results of event studies in a similar manner as Dam (2008); a stock price decrease in the event window corresponds to a stock return increase. This is because after the event, investing in the stocks is cheaper, which consequently, increases expected returns for investors (given that profits are unaffected by the event).
6
that neglected (e.g. low CEP) stocks are underpriced to compensate investors not only for giving up diversification opportunities (they must hold too many neglected stocks), but also for becoming exposed to idiosyncratic risks. Hong and Kacperczyk (2009) refer in this context to litigation as an example of idiosyncratic risks.
Fama and French (2007) propose an asset pricing framework in which investors are divided by the presence or absence of taste for assets as consumption goods. That is, some investors derive direct utility from merely holding certain investment assets, like domestic or employer stocks, but also socially responsible or indeed ‘green’ stocks. In the end however, these investors receive negative alphas (i.e. risk adjusted or abnormal returns, see Jensen (1968)), while the investors without taste receive positive alphas by holding the complementary portfolio. Fama and French (2007) conclude that price effects caused by tastes for assets as consumption goods are similar to those resulting from misinformation. Heinkel et al. (2001) show in an equilibrium model how investors with a distaste for pollution could influence corporate behaviour through cost of capital. As the group of these investors increases, so does the return on stocks of polluting corporations (their cost of capital) because diversification possibilities diminish. And if capital costs rise high enough, corporations may start changing their polluting behaviour by investing in cleaner technologies.
Heinkel et al. (2001) themselves find only little empirical evidence for ‘behaviour-changing’ differences in corporations’ cost of capital, but Hong and Kacperczyk (2009) do find evidence. Hong and Kacperczyk (2009) study the influence of social (not environmental, but the principle remains the same) norms on United States stock returns. They report evidence for a stable dichotomy among investors: pension plans and other ‘norm-constrained’ investors hold less so called ‘sin stocks’ as compared to individual investors, mutual funds and hedge funds. Furthermore, the results show that sin stocks indeed outperform their comparables to an extent predicted by Heinkel et al. (2001).
Galema et al. (2008) also find a negative relationship between corporations’ score on certain social responsibility criteria, including environmental criteria, and stock returns. They argue though, that this relationship is actually not fully captured by alpha as generated from regressions in the spirit of the Fama and French (1993) three factor asset pricing model, but also by decreases in the book-to-market ratio. Excess demand causes the stocks of responsible corporations to become overpriced and decreases their book-to-market ratio. This demand-driven decrease reduces the risk exposure of responsible stocks to the so-called ‘value factor’ (HML factor) of the three factor asset pricing model, which in turn inflates alpha (but not the actual risk adjusted return).
factor’ to verify whether overlooked differences in risk exposure cause the underperformance. The correction has only little influence on the returns, which, according to Renneboog et al. (2008), supports the hypothesis that investors pay for their taste for ethics.
To return to the second research question, raised at the beginning of this subsection: what drives the relationship between CEP and stock returns? Changes in the risk exposure of environmentally responsible corporations, or mispricing caused by the tastes of investors? I hypothesize that:
H2: The negative relationship between CEP and stock returns is partially driven by an overpricing of high CEP stocks because of investors’ taste, based on personal values or social norms.
3. Data
3.1 Measuring CEP
I measure a corporation’s environmental performance by its score in the Newsweek Green Rankings 2009, a new and comprehensive ranking on environmental performance, published in September 2009 (McGinn, 2009).7 The Green Rankings evaluates the 500 largest publicly traded United States corporations on various components of environmental performance.8 Newsweek and its research partners started the effort early 2008 and it took well over a year to complete (KLD, 2009).
A corporation’s overall score, called ‘Green Score’, is based on three sub-scores. First, there is the ‘Environmental Impact Score’, which captures the total costs of a corporation’s global environmental impact. These costs are normalized against annual corporate revenues to ensure the comparability of corporations of different industries and sizes. The second sub-score, the ‘Green Policies Score’, is based on an analytical assessment of a corporation’s environmental policies and initiatives (see also Makower, 2009). The third and final sub-score, the ‘Reputations Score’, results from a survey among corporate social responsibility professionals, CEO’s, academics, and other environmental experts. The three sub-scores are converted into z-scores, whereas the overall Green Score is in turn calculated as a weighted average of these z-scores.9 Both the Environmental Impact Score and the Green Policies Score receive a weight of 45%, leaving a 10% weight for the Reputations Score. Both this latter (industry-neutral) sub-score as well as the Green Policies Score are included to level the playing field for industries with an above average environmental impact, like the oil, gas or utility industry (see Makower, 2009). Finally, all scores are mapped on a 100 point scale.
7
The actual data series come from the accompanying website: http://greenrankings2009.newsweek.com (accessed January, 2011).
8 See appendix A for a more elaborate description of the Green Rankings. 9
The Green Rankings is a comprehensive and profound measure of CEP. Comprehensive because most measurement strategies known from literature are (separately) covered by the ranking (e.g. audits, reputational ratings, disclosures, see Herremans et al. (1993); Wagner (2001) and Orlitzky et al. (2003)). The ranking is profound too, being based on a vast database research project, rather than on anecdotal observations typically used in “green lists” (see KLD, 2009). Developing a sound methodology for the Green Rankings was challenging, especially with regard to weighting all different environmental components (see Makower, 2009). Data quality and comparability proved an issue too, since corporations were sometimes unable or reluctant to submit the appropriate data. But even despite these difficulties, the Green Rankings seem to provide a wealth of new CEP information, or as Makower (2009) puts it: “…it may well be the best effort yet to rigorously and comprehensively assess the mainstream corporate marketplace - at least in the U.S”.
Panel A in Table 1 summarizes the CEP scores of 462 corporations from the Green Rankings. I removed 38 ranked corporations from the sample because of incomplete financial data.10 On average, the corporations in the sample receive 70.41 out of 100 points on overall environmental performance. Relatively many scores are close to this mean, as indicated by the peaked distribution of the Green
10
One corporation, ConAgra, is removed from the sample because its score in the Green Rankings is obscured by an overstatement of its water usage, see http://greenrankings2009.newsweek.com (accessed January 2011).
Table 1 Descriptive statistics of corporate environmental performance Panel A CEP scores
Mean St. dev. Median Min. Max. Skewness Kurtosis Obs. Overall Green Score 70.41 10.19 70.74 1.00 100 -1.39 10.50 462 Environmental Score 49.43 28.65 49.10 0.20 100 0.04 1.86 462 Policy Score 39.92 18.29 38.78 1.00 100 0.45 3.15 462 Reputation Score 34.61 13.78 33.05 1.00 100 1.31 6.66 462 Panel B Correlations Overall Green Score Environmental Score Policy Score Reputation Score Overall Green Score 1.00
Environmental Score 0.30*** (6.75) 1.00 Policy Score 0.76*** (25.08) -0.07* (-1.51) 1.00 Reputation Score 0.44*** (10.51) -0.09** (-2.00) 0.48*** (11.34) 1.00
Notes: This Table presents the descriptive statistics for the CEP measure used in this study.
Panel A summarizes the Newsweek Green Rankings 2009 CEP scores received by the 462 United States corporations in the sample. The overall score of a corporation, the Green Score, is based on the Environmental Score (a corporation’s global environmental impact), on the Policy Score (environmental policies and initiatives) and on the Reputations Score (environmental reputation).
Panel B presents the correlations between the four CEP scores. The t-statistics are between parentheses and calculated as
(N 2) (1 r2)
r
t= − − , where rrepresents the correlation coefficient and N denotes the number of corporations in the
sample (N=462). *, **, *** signify the significance levels at the 10%, 5% and 1% thresholds, respectively.
Score (kurtosis of 10.50). At the same time, the negative skewness of -1.39 implies that a majority of corporations actually performs (just) above-average. The mean sub-scores on environmental reputation (34.61) and environmental policies (39.92) are much lower than on overall environmental performance. The distributions of these two sub-scores are also less peaked, as well as positively skewed. The third sub-score, the Environmental Impact Score, is relatively uniformly distributed with almost no skewness and a kurtosis of 1.86. None of the four scores is normally distributed.
It appears from panel B in Table 1 that environmental impact and reputation are actually negatively correlated. So corporations do not seem to live up to their reputations. On the contrary, a better reputation tends to be associated with a larger global environmental footprint. Yet, this supposed relationship should be interpreted with caution, since a correlation coefficient of -0.09 is not very strong, even though it is statistically significant.11 The environmental reputation of a corporation does appear to go well with its green policies and initiatives, as indicated by the positive correlation coefficient of 0.48.
3.2 Measuring CFP
Corporate financial performance is measured by stock returns. This forward looking market-based measure allows to study the influence of CEP on market expectations of future corporate financial performance (see e.g. Ziegler et al., 2007). In addition, stock returns are less prone to changes in (fiscal) rules than backward looking accounting-based measures, and they are more difficult to manipulate (see Herremans et al., 1993; Orlitzky et al., 2003; Scholtens, 2008).
I consider monthly stock returns from March 2004 until September 2008. So I focus on a few years period that extends well into the year of the ranking, like e.g. Ziegler et al. (2007). This implies the assumption that CEP is stable during the period under consideration, which seems reasonable (see Derwall et al., 2005; Guenster et al., 2006; Ziegler et al., 2007). The length of the period balances different interests. On the one hand, the desire to have a large sample of complete corporate financial data series requires a limited period that does not includes many disrupting events like bankruptcies and mergers. But on the other hand, the period should also not be too restricted, as to ensure that the data are sufficient to accurately estimate model parameters (Young and Lenk, 1998). What is more, the period ends in September 2008 to limit the influence of the financial credit crisis. Most major crisis events happened later, like the bankruptcy of Lehman Brothers, the $ 85 billion bailout of AIG and the political rejection of the United States $ 700 billion rescue bill.12 I seek to minimize the potentially large influence of the credit crisis since it may obscure any possible (subtle) relationship between CEP and stock returns. Even so, I also consider two additional periods for robustness checks. An extended period from March 2004 until September 2009 including all the major credit crisis events mentioned
11
See also http://janda.org/c10/Lectures/topic06/L24-significanceR.htm (accessed January 2011).
12
above. And a reduced period from March 2004 until February 2008 that excludes almost the entire credit crisis.13 nonetheless
Monthly total return indices of the 462 ranked corporations come from Thomson Financial Datastream. These total return indices
(
RIit)
, measured in the middle of the month t(
t=1,2,...,T)
,assume that any dividends associated with a specific stock i
(
i=1,2,...,462)
are reinvested in that same stock. I calculate monthly logarithmic stock returns( )
rit by taking the natural logarithm of theratio of this month’s total return index to the previous month’s total return index
(
rit =ln(
RIit RIit−1)
)
.And by subsequently substracting the monthly logarithmic risk free rate
( )
f tr , I arrive at the monthly logarithmic excess stock returns
(
f)
t it e
it r r
r = − . As a proxy for the risk free rate, I obtain the one month United States treasury bill return from Ibbotson and Associates Inc. from the website of Kenneth French.14
Panel A of Table 2 describes the monthly logarithmic excess stock returns of the 462 corporations during the main period under consideration, March 2004 until September 2008. The 25.410 monthly observations have a mean value of 0.48% and a standard deviation of 8.19%. The returns range from as low as -70.49% to as high as 47.68%. Panel A of Table 2 also presents the returns in individual years. Two observations stand out. First, the mean excess return decreased to a negative value in the first three quarters of 2008, while the associated standard deviation increased. This could well be the influence of the deepening credit crisis that year. The second observation is that
13 The reduced period of March 2004 until February 2008 excludes for instance the nationalisation of Northern Rock
and the bailout of Freddie Mae and Fannie Mac. 14
See http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html (accessed January 2011). Table 2 Descriptive statistics of corporate financial performance
Panel A Monthly logarithmic excess stock returns
Mean St. dev. Median Min. Max. Skewness Kurtosis Obs. Mar 2004 – Sep 2008 0.48% 8.19% 0.80% -70.94% 47.86% -0.49 5.95 25,410 2004 (Mar – Dec) 1.30% 7.81% 1.55% -51.23% 41.64% -0.37 5.80 4,620 2005 1.06% 7.20% 0.87% -56.39% 31.17% -0.11 4.86 5,544 2006 0.76% 7.29% 1.11% -40.81% 33.84% -0.55 5.32 5,544 2007 0.42% 7.49% 0.86% -40.23% 36.85% -0.37 4.50 5,544 2008 (Jan – Sep) -1.52% 11.04% -1.24% -70.94% 47.86% -0.37 5.18 4,158 Panel B Average monthly logarithmic stock returns
Mean St. dev. Median Min. Max. Skewness Kurtosis Obs. Mar 2004 – Sep 2008 0.74% 1.05% 0.72% -3.21% 4.66% 0.11 4.19 462
Notes: This Table presents the descriptive statistics for the CFP measure used in this study.
Panel A shows the monthly logarithmic excess stock returns (with a one month United States treasury bill as reference asset) of the 462 ranked United States corporations for the whole period of March 2004 till September 2008. Returns are also shown for several sub-periods.
Panel B summarizes the average monthly logarithmic returns of the 462 corporations from March 2004 till September
2008. Average returns are calculated as =
∑
T=t it
i T r
r (1 ) 1 , where rit is the logarithmic stock return of corporation i
the distribution of returns in 2005 is negatively skewed, despite the mean being considerably higher than the median. The reason is a local maximum in the left tail of the distribution; a cluster of (very) low returns tips the scales towards negative skewness.
I also calculate the average monthly logarithmic stock return
( )
ri for each corporation in thesample during the period under consideration (see Table 2, panel B). The mean of these average returns is 0.74% with a standard deviation of 1.05%. At the high end of the spectrum, the average return is 4.66% per month, while at the low end it is -3.21%. Comparing the standard deviation in panel B of Table 2 with those in panel A demonstrates how volatility cancels out by averaging individual stock returns.
Finally, Table B1 (appendix B) describes the stock returns during the two alternative sample periods and contains additional descriptive statistics. It appears that stock returns decrease and standard deviations increases with increasing period length and, accordingly, inclusion of the credit crisis.
3.3 Control variables; risk factors
The regressions in this study are based on the four factor asset pricing model of Fama and French (1993) and Carhart (1997), as I will explain shortly in the methodology section. In this model, stock returns are a function of four common risk factors, associated with: overall economic activity (the market), like in the famous CAPM (Sharpe, 1964), corporation size, the book-to-market (value) ratio and stock momentum. Each risk factor is mimicked by a specific portfolio; and the returns of these mimicking portfolios enter the four factor model as control variables (Fama and French, 1992, 1993; Carhart, 1997).
I obtain the data of relevant United States portfolios from the website of Kenneth French, see Table 3.15 The excess return on the value weighted market portfolio of stocks, with a one month United States treasury bill as risk free reference asset, proxies for the market risk factor. The SMB portfolio (‘small minus big’) mimics the common risk factor related to corporation size (market value) and is determined by the difference between the returns on a small stock portfolio and a large stock portfolio. These two latter stock portfolios have similar book-to-market ratios to filter out the influence of the book-to-market ratio on returns. In a comparable vein, the HML portfolio (‘high minus low’) mimics the risk factor associated with the book-to-market ratio and is the return on a high book-to-market stock portfolio minus the return on a similar sized low book-to-market stock portfolio. The last portfolio, the MOM portfolio (‘momentum’) suggested by Carhart (1997), proxies for the risk factor associated with stock momentum. This portfolio is determined by the return difference between a high one-year momentum stock portfolio and a similar sized low one-year momentum stock portfolio.
15
Panel A in Table 3 presents the monthly logarithmic returns of the four mimicking portfolios from March 2004 till September 2008. The excess returns of the market portfolio range from -10.64% to 4.66% and have a mean of -0.23%. Despite this negative mean, the median of 0.43% indicates that excess returns are positive in at least half of the months. The opposite holds true for the SMB portfolio, having a positive mean but a negative median return. Furthermore, the HML portfolio has the best Sharpe ratio, combining a relatively high mean return (0.46%) with a relatively low standard deviation (1.75%). The returns of this portfolio and those of the SMB portfolio correlate with the excess market portfolio returns at a statistically significant level (see panel B, Table 3). None of the correlations coefficients however, is so high as to lead to multicollinearity.
Table B2, panel A summarizes the monthly logarithmic returns of the four portfolios during the two alternative sample periods for robustness checks. It shows that the standard deviations of these returns increase with period length or inclusion of the credit crisis. The pattern seems less clear though for the means of the portfolio returns; while most decrease with period length, the mean return of the SMB portfolio increases. This suggests that small firms did relatively well during the credit crisis. Panel B of Table B2 shows the correlations between the portfolio returns.
Table 3 Descriptive statistics of mimicking portfolios Panel A Monthly logarithmic returns
Mean St. dev. Median Min. Max. Skewness Kurtosis Obs. Market portfolio (market factor) -0.23% 3.18% 0.43% -10.64% 4.66% -1.02 4.22 55 SMB portfolio (size factor) 0.06% 2.13% -0.13% -4.03% 5.20% 0.23 2.66 55 HML portfolio (value factor) 0.46% 1.75% 0.06% -3.05% 4.39% 0.37 2.49 55 MOM portfolio (momentum factor) 0.62% 3.27% 0.35% -8.18% 11.83% 0.32 4.85 55 Panel B Correlations Market portfolio (market factor) SMB portfolio (size factor) HML portfolio (value factor) MOM portfolio (momentum factor) Market portfolio (market factor) 1.00
SMB portfolio (size factor) 0.39***
(3.08) 1.00
HML portfolio (value factor) -0.27*** (-2.04) (0.02) 0.00 1.00 MOM portfolio (momentum factor) 0.03
(0.22)
0.15 (1.11)
-0.21*
(-1.56) 1.00
Notes: This Table presents descriptive statistics of the market portfolio and the factor mimicking stock portfolios.
Panel A summarizes the monthly logarithmic excess returns on the United States market portfolio (with a one month treasury bill as risk free reference asset) and the monthly logarithmic returns on the three other United States common risk factor mimicking portfolios from March 2004 till September 2008. The excess returns on the market portfolio proxy for the common market risk factor. The SMB portfolio mimics the risk factor related to corporation size and is the difference between the returns on a small stock portfolio and a large stock portfolio with similar book-to-market ratios. In the same respect, the HML portfolio mimics the risk factor associated with the book-to-market ratio and is the return on a high book-to-market stock portfolio minus the return on a similar sized low book-to-market portfolio. Finally, the MOM portfolio proxies for the risk factor associated with stock momentum and is the return difference between a high one-year momentum portfolio and a similar sized low one-year momentum portfolio. Panel B presents the correlations between the returns on the factor mimicking portfolios from March 2004 till September 2008. The t-statistics are between parentheses and calculated as t=r (N−2)(1−r2), where r is the correlation coefficient and N the
number of observations (N=55). *, **, *** signify the significance levels at the 10%, 5% and 1% thresholds, respectively.
4 Methodology
The methodology in this study is based on the Fama-French-Carhart four factor asset pricing model. So I compare the returns of low and high CEP stocks only after adjusting for common risk factors associated with market activity, corporation size, book-to-market value and stock momentum. This is an improvement over many earlier studies that used non-adjusted returns or returns adjusted only for market risk, e.g. by means of the CAPM (Derwall et al., 2005; Bauer et al., 2005; Ziegler et al., 2007).
4.1 The relationship between CEP and CFP
I follow a two step approach similar to that of Ziegler et al. (2007) to test for the negative relationship between CEP and stock returns predicted by hypothesis 1. The first step involves ordinary least-squares time-series regressions of the ranked corporations’ excess stock returns against the returns on the four mimicking portfolios. For each of the 462 corporations in the sample I estimate
it t MOM i t HML i t SMB i e t MKT i i e it r r r r r =α +β1 , +β2 , +β3 , +β4 , +ε (1) where e it
r is the logarithmic excess stock return of corporation i in month t and e t MKT
r , is the
logarithmic excess return on the market portfolio in month t . Furthermore, rSMB,t rHML,t and rMOM,t
represent the logarithmic return in month t on the SMB, HML and MOM portfolio, respectively. αi
is the intercept, or Jensen’s alpha, associated with corporation i and εit is the error term. The other
parameters, the betas or ‘factor loadings’, β1i, β2i, β3i and β4i, measure the exposure of corporation
i to the each of the four common risk factors.
In the second step, I regress the average monthly logarithmic stock returns of the ranked corporations on the estimated first step betas and a vector of CEP scores. This ordinary least-squares cross-section regression allows to distinguish the influence of CEP on average stock returns, as the betas control for the common risks. I estimate the following equation
i i i i i i i r =α+γ1βˆ1 +γ2βˆ2 +γ3βˆ3 +γ4βˆ4 +k′cep +ε (2)
where r is the average logarithmic stock return of corporation i i during the period under
consideration, March 2004 till September 2008. βˆ1i, βˆ2i,βˆ3i and βˆ4i represent the betas of
corporation i as estimated in equation (1), whereas the gammas, or risk premiums, γ1i, γ2i, γ3i and i
4
γ are the corresponding parameters. cepi symbolises an m×1 vector that contains m CEP scores
have a negative sign in order not to reject hypothesis 1, or a negative relationship between CEP and stock returns.
Next to estimating a single full-period cross-section regression in the second step, I follow a Fama and MacBeth (1973) procedure and estimate separate regressions for each month.16 This alternative procedure tends to reduce autocorrelation (and therefore biases) in second step regression residuals (see e.g. Fama and French, 2004). The dependent variable in the regressions is the monthly stock return of the ranked corporations. For each month in the period under consideration, I estimate
it i t i t i t i t i t t it r =α +γ1βˆ1 +γ2 βˆ2 +γ3βˆ3 +γ4 βˆ4 +k′cep +ε (3)
These estimations yield time-series of estimated parameters. In order not to reject hypothesis 1, the sign of the average parameter kˆ =[κˆ1,...,κˆm] should be significantly negative.
17
A known issue of two-step regression procedures is the so-called ‘error-in-variables’ problem (see Fama and MacBeth, 1973). This problem arises because the betas used as control variables in the second step regression are not true betas, but estimates from the first step. The inherent measurement error in these betas causes normal standard deviations and t-statistics to overstate the precision of the estimated parameters in the second step (see e.g. Shanken, 1992, Pasquariello, 1999 and Fama and French, 2004). To correct for this overstatement of precision, I follow Cheng et al. (2005); Brammer et al. (2006) and Chung et al. (2006)and adjust the second step standard deviations in the way suggested by Shanken (1992).18
4.2 What drives the relationship between CEP and stock return?
According to hypothesis 2, the negative relationship between CEP and stock returns results at least partially from an overpricing of high CEP stocks because of investors’ taste, based on social norms and personal values. To test this hypothesis, I expand the Fama-French-Carhart four factor model with a fifth factor to capture any common risks associated with CEP. It follows that if this five factor model turns out to be able to explain the variation in stock returns solely on the basis of differences in common risk exposure, hypothesis 2 should be rejected.
I construct a mimicking portfolio for the fifth ‘CEP’ factor using the approach of Fama and French (1993) and Carhart (1997). The portfolio, called NMG (‘non-green minus green’), is the return on a non-green stock portfolio minus the return on a green stock portfolio (i.e. analogous to the SMB,
16 See also Cavanaile et al. (2009) for a discussion on different estimation procedures for the second step. 17
I calculate the time series averages as κˆ1=
∑
Tt=1κˆ1t, where κˆ1t is the estimated coefficient for the first CEP score inmonth t . Furthermore, I apply a t-test to determine the significance of the time-series average ( )k
(
(
se T)
)
t =κˆ1×1/ (κˆ1)×1/ , where k represents the degrees of freedom T−1 and se(κˆ1) is the standard deviation of the time series average of the estimated coefficient κ1 (see e.g. Cavanaile et al., 2009).
18
HML, MOM portfolios). To obtain these two latter stock portfolios, I sort the 462 corporations in the sample on size (market value) in the fourth month of each year during the period under consideration, March 2004 till September 2008.19 Corporations in the top half receive the label ‘big’, while those in the bottom half receive the label ‘small’. I then sort the corporations again, but now on CEP as measured by their overall Green Score. I call the top 30% of corporations ‘green’ and the bottom 30% ‘non-green’. All remaining corporations are considered ‘neutral’. Subsequently, I allocate each corporation to one of six portfolios based on their labels: green&small, green&big, neutral&small, neutral&big, non-green&small, or non-green&big. These portfolios may differ from year to year in accordance with the re-evaluation of corporation size. Finally, I calculate the NMG portfolio returns by subtracting the simple average of the value-weighted returns on the green portfolios (green&small, green&big) from the simple average of the value-weighted returns on the non-green portfolios (non-green&small, non-green&big).20 These NMG returns enter the five factor model as control variables.
To obtain the dependent variable for the five factor model regressions, I group the 462 corporations in the sample into nine portfolios according to their overall Green Score. These CEP based portfolios are especially suitable to determine whether the NMG portfolio captures common risks related to CEP.21 For each of the nine portfolios, I estimate
Pit t NMG Pi t MOM Pi t HML Pi t SMB Pi e t MKT Pi Pi e Pit r r r r r r =α +β1 , +β2 , +β3 , +β4 , +β5 , +ε (4) where e Pit
r is the logarithmic value-weighted excess return of stock portfolio Pi
(
i=1,2,...,9)
in month t .22 In addition, rNMG,t symbolises the logarithmic return on the NMG portfolio in month t .The five parameters, or factor loadings, β1Pi, β2Pi, β3Pi, β4Pi and β5Pi, measure the exposure of
portfolio Pi to the common risk factors.
Statistically significant loadings on the fifth CEP factor indicate that portfolio Pi is exposed to common risks associated with CEP. Accordingly, this means that the relationship between CEP and stock returns is at least partially risk-driven (as opposed to taste-driven). There is reason to doubt the influence of taste altogether, and thus to reject hypothesis 2, if inclusion of the fifth CEP factor pushes
19
The market values are obtained from Thomson Financial Datastream.
20
The value-weighted return on each of the six size&CEP portfolios is calculated as:
∑
= − − − × = ni it it it t CEP Psize it t CEP Psize RI RI RI MV MV ret 1 1 1 , & ,& , where MVit is the market value of corporation i in month t ,
t CEP Psize
MV / , is the market value of the size/CEP portfolio in month t and RI is the total return index of the stock of it
corporation i in month t . Furthermore, I subtract
(
retPsmall&green,t+retPbig&green,t)
2 from(
retPsmall&non−green,t+retPbig&non−green,t)
2to obtain the return on the mimicking portfolio.
21
See Fama and French (1993) for an analogy with dependent portfolios based on size and the book-to-market ratio.
22
I calculate the logarithmic value-weighted excess return as follows: f t Pit e Pit r r r = − , with 100 1 ln 1 1 1 × + − × =
∑
= − − n i it it it Pit it Pit RI RI RI MV MVr . Where MVPit is the market value of portfolio Pi (i=1,2,...,9) in
alpha (i.e. abnormal or risk adjusted returns) to zero. Estimated alphas of (near) zero signify the five factor model’s ability to explain stock returns on the basis of common risk exposure, leaving little room for tastes based on personal values or social norms. If the alphas are non-zero though, the influence of taste should not be ruled out, as alpha may then represent compensation for the exposure to ‘latent common risk factors’ (i.e. those factors not captured by the five factor model) but also for the taste for high CEP stocks (i.e. the less popular low CEP stocks are underpriced to induce investors to hold them).
In a second test of hypothesis 2, I compare the five factor model alphas of a group of stock portfolios formed on environmental impact with a group of portfolios formed on environmental
reputation. The reason for this comparison is that differences in the development of alpha related to
the environmental performance of the portfolios in both groups, may actually indicate that the relationship between CEP and stock returns (alphas) is influenced by tastes. To form the first portfolio group, I sort the ranked corporations into nine equally sized portfolios based on actual environmental footprint (their Environmental Impact Score). For the second group, I use a similar procedure, except that I now sort corporations on environmental reputation (their Reputations Score), rather than on actual environmental impact.
Estimating the five factor model (equation 4) for each portfolio in both groups yields two series of nine alphas, ranging from low CEP portfolios alphas (measured by environmental impact or reputation) to high CEP portfolios alphas. The development of the two alpha series relative to each other may follow one of three patterns. First, both series may develop in the same direction (increasing, decreasing or staying constant). Second, the series of ‘impact alphas’ increases with increasing environmental impact, while the series of ‘reputation alphas’ does not increase with deteriorating environmental reputation. Third, the pattern may be the other way around; the series of reputation alphas increases with deteriorating environmental reputation, whereas the series of impact alphas does not increase with increasing environmental impact.
subject to social scrutiny, like pension funds and bank, than by other, more anonymous investors, like independent investment advisors.
There is another reason why alpha signals the influence of taste on stock returns if it indeed follows the abovementioned third pattern of increasing with deteriorating environmental reputation and not increasing with increasing environmental impact. Suppose that returns are determined by risk alone, this would mean that a worsening of corporate environmental reputation, increases the exposure to latent risk factors (increasing alpha), whereas an increase in environmental impact does not (constant alpha). The opposite seems more likely: if actually having a large environmental footprint does not add to latent risk factor exposure, neither does only having a reputation of being environmentally unconscious. It follows then that if alpha increases with deteriorating environmental reputation anyway, this has to be because it represents additional compensation for holding the less popular dubious reputation stocks.
Finally, note that if alpha follows the two other development patterns, it becomes difficult (if not impossible) to distinguish between the influence of investors’ taste and latent risks. In both alternative scenarios, alpha may consist entirely of compensation for latent risk factors without violating any of the above arguments.
5 Results
5.1 Results on the relationship between CEP and CFP
Table 4 presents the results of the first step time-series regressions of the Fama-French-Carhart four factor asset pricing model. The Table summarizes the series of betas, or factor loadings, generated by estimating equation (1) for each of the 462 ranked corporations. It shows that the MKT betas
( )
βˆ1ihave the highest mean (0.65). So on average, the corporations are relatively highly exposed to risks
Table 4 Summarized results of first step time-series regressions
Mean St. dev. Median Min. Max. Skewness Kurtosis Obs.
i α) (Alpha) 0.35 1.01 0.35 -2.60 4.63 0.23 4.35 462 i 1 ˆ β (MKT) 0.65 0.40 0.61 -1.00 2.26 0.57 4.46 462 i 2 ˆ β (SMB) -0.02 0.55 0.01 -2.15 1.86 -0.21 4.04 462 i 3 ˆ β (HML) 0.21 0.70 0.23 -2.45 3.18 0.19 4.56 462 i 4 ˆ β (MOM) 0.30 0.39 0.27 -1.18 1.75 0.50 4.19 462
Notes: This Table summarizes the results of the ordinary least-squares time-series regressions of equation (1)
it t MOM i t HML i t SMB i e t MKT i i e it r r r r
r =α+β1 ,+β2 ,+β3 , +β4 ,+ε . I estimate one regression for each of the 462
corporations in the sample for the period of March 2004 till September 2008. e
it
r is the logarithmic excess
stock return of corporation i in month t , e
t MKT
r , is the logarithmic excess return of the market (MKT) factor
mimicking portfolio. Moreover, rSMB,t, rHML,t and rMOM,t are the logarithmic return on the mimicking
associated with overall economic activity. In contrast, the negative mean SMB beta
(
βˆ2i =−0.02)
indicates that corporations are, on average, negatively exposed (or hedged) to the common risk factor associated with size. This negative size-risk exposure probably results from the sample’s bias towards large corporations (see also Derwall et al., 2005). Finally, none of the four estimated betas series is normally distributed.These series are the independent variables in the second step cross-section regressions; along with the vector of CEP scores. Table 5 shows the correlation coefficients between the independent variables. Several coefficients are statistically significant but their limited sizes do not signal multicollinearity. The dependent variable of the second step regressions, the average monthly logarithmic stock returns of the ranked corporations, contains outliers. I follow the approach of Semova and Hassel (2008) and remove observations that are more than 1.5 ‘Interquartal Ranges’ below the first quartile or above the third.23 Fourteen corporations are removed, leaving 448 in the sample.24
The results from the second step cross-section regressions in Table 6 give no reason to reject hypothesis 1 of a negative relationship between CEP and stock returns. I estimate five specifications of equation (2) that depend on the inclusion of the four CEP (sub-)scores. The estimated parameters of the CEP scores are modest in size, but highly significant and of negative sign. That of the overall Green Score
( )
κˆ1 is -0.017 and significant at the 1% level. This implies that a one point increase in acorporation’s Green Score is associated with an approximate 20 basis points decrease in yearly average stock returns. The negative relationship between returns and CEP measured either by the Green Policies (sub-)Score or the Reputations (sub-)Score is smaller, but still highly significant. The
23Interquartal Range (IQR): the difference between the first and third quartile of a data set. 24
Without the outliers, the sample of average monthly logarithmic stock returns is normally distributed with a skewness of 0.06 and a kurtosis of 2.97. Whereas with the outliers, the sample is not normally distributed, see Table 2. Table 5 Correlations of first step betas and CEP scores
Overall Green Score Environmental Score Policy Score Reputation Score βˆ1i(MKT) βˆ2i(SMB) βˆ3i (HML) i 1 ˆ β (MKT) -0.03 (-0.64) 0.10** (2.16) -0.09** (-1.94) -0.13*** (-2.81) 1.00 i 2 ˆ β (SMB) 0.19*** (4.15) 0.28*** (6.26) 0.10** (2.16) -0.04 (-0.86) -0.32*** (-7.24) 1.00 i 3 ˆ β (HML) -0.19*** (-4.15) -0.31*** (-7.00) -0.07* (-1.51) -0.05 (-1.04) 0.16*** (3.48) -0.41*** (-9.64) 1.00 i 4 ˆ β (MOM) -0.04 (-0.86) -0.20** (-1.65) -0.02 (-0.43) -0.04 (-0.86) 0.26*** (5.78) -0.27*** (-6.01) 0.23*** (5.07)
Notes: This Table presents the correlations between the betas estimated in the first step time-series regressions (estimated over
the period from March 2004 till September 2008) and the four CEP scores. The t-statistics are between parentheses and
calculated as t=r (N−2) (1−r2), where r represents the correlation coefficient and N denotes the number of corporations
estimated CEP parameter of both scores (κˆ3 and κˆ4 ) is -0.009. If CEP is finally measured solely by
the Environmental (sub-)Score, the parameter
(
κˆ2)
is -0.003 and only significant at the 10% level. Table 6 shows that the alphas exceed the expected (monthly) risk free rate based on the specification of the model in equation (2). So, the alphas could represent, for instance, additional compensation for latent (i.e. uncaptured) risk factors. What is more, the negative value of the estimated SMB parameters is unexpected, as they are effectively the premium for size-related commonTable 6 Results of second step cross-section regressions
α (Alpha) 1.94 0.82 1.08 1.07 1.37 (6.54)*** (8.64)*** (7.75)*** (7.50)*** (7.95)*** [6.24]*** [8.25]*** [7.37]*** [7.10]*** [7.61]*** 1 γ (MKT) -0.20 -0.16 -0.24 -0.26 -0.21 (-1.60) (-1.27) (-1.85)* (-2.00)** (-1.61) [-1.52] [1.21] [-1.76]* [-1.89] [-1.54] 2 γ (SMB) -0.48 -0.49 -0.50 -0.56 -0.48 (-5.10)*** (-5.12)*** (-5.33)*** (-6.00)*** -4.95)*** [-4.87]*** [-4.89]*** [-5.07]*** [5.68]*** [-4.73]*** 3 γ (HML) -0.02 -0.01 0.01 0.00 -0.03 (-0.20) (-0.11) (0.09) (0.01) -0.39) [-0.19] [-0.11] [0.09] [0.01] [-0.38] 4 γ (MOM) 0.56 0.52 0.56 0.54 0.51 (4.59)*** (4.11)*** (4.53)*** (4.22)*** (4.22)*** [4.38]*** [3.93]*** [4.31]*** [3.99]*** [4.04]*** 1 κ (Green Score) -0.017 (-4.46)*** [-4.26]*** 2 κ (Envir. Score) -0.003 -0.003 (-1.89)* (-2.44)** [-1.81]* [-2.34]** 3 κ (Policy Score) -0.009 -0.008 (-3.65)*** (-2.88)*** [-3.47]*** [-2.76]*** 4 κ (Repu. Score) -0.009 -0.005 (-3.34)*** (-1.62) [-3.16]*** [-1.55] Adjusted R2 0.19 0.16 0.19 0.18 0.20 N 448 448 448 448 448
Notes: This Table summarizes the results of the ordinary least-squares cross-section regressions of equation (2)
i i i i i i i
r =α+γ1βˆ1+γ2βˆ2 +γ3βˆ3 +γ4βˆ4 +k′cep +ε for the period of March 2004 till September 2008. ri is the average logarithmic
stock return of corporation i during this period. β represents the beta of corporation i associated with the MKT (market) risk ˆ1i
factor, as estimated in the first step time-series regressions. In a similar respect, β , ˆ2i β and ˆ3i β represent the estimated betas of ˆ4i
corporation i associated with the SMB (size), HML (value) and MOM (momentum) risk factors, respectively. cepi is an m×1
vector that contains m CEP scores of corporation i. Fourteen corporations (observations) are removed because they are outliers in
average return, leaving a sample of 448 corporations. White (1980) heteroskedasticity consistent t-statistics are in parentheses,
Shanken (1992) adjusted t-statistics are in square brackets. *, **, *** signify the significance levels at the 10%, 5% and 1%
risk. Ziegler et al. (2007) find similar, negative SMB parameters and argue that this may result, among other things, from the large firm bias in the sample. A final observation from Table 6 is that the error-in-variables problem seems limited in this particular two-step regression procedure. The conclusions about significance are indeed unaffected by using Shanken (1992) adjusted t-statics, as opposed to White (1980) heteroskedasticity consistent t-statistics.
The main results from the two-step procedure are robust to changing the sample period and to including major credit crisis events. Table D1 (appendix D) presents the results of the first step time-series regression for the two alternative periods; March 2004 until February 2008 and March 2004 until September 2009. The results resemble those of the main period. A noticeable difference is the negative MOM beta (-0.03) in the extended period. This negative exposure to common stock momentum risk, reflects previous stock price decreases in the sample; possibly caused by a deepening credit crisis. Like in the main period, the correlation coefficients between the estimated first step betas in the alternative periods do not point towards multicollinearity (see Table B3). The second step cross-section regression results are in Tables D2a and D2b. The negative relationship between CEP and stock returns increases in size and significance with decreasing period length. For the limited period from March 2003 until February 2008, a one point decrease in the overall Green Score is associated with a 28 basis points decrease in average yearly stock returns
(
κˆ1=−0.023)
.The alternative Fama and MacBeth (1973) procedure of estimating separate monthly regressions in the second step also points towards a negative relationship between CEP and stock returns. Table 7 shows that the results are less significant though, than those of the above single second step regression procedure. The significance of the overall Green Score parameter
( )
κˆ1 falls justbelow the 5% significance level using Shanken (1992) t-statistics. And if all three sub-scores are included in the regression, only the Policy Score is significant at a 5% level. Table D3 presents results of Fama and MacBeth (1973) robustness checks based on the two alternative sample periods. It shows that the procedure yields insignificant parameters in the extended period from March 2003 until September 2009. But, similar to the above single second step regression procedure, the significance of the negative relationship between CEP and stock returns increases with decreasing period length.
