Corporate Financial Performance in the
international setting
University of Groningen
Student number:
S2193124
Name:
Paul Newman
Study program:
MSc International Financial Management
Supervisor: Dr. Raymond Zaal
Deadline: 13
thof June, 2016
Abstract: This paper studies 605 reconstitutions of the S&P Environmentally and Socially
Responsible Index in order to extend our current understanding of the relationship between
corporate social performance (CSP) and corporate financial performance (CFP), and how it is
influenced by internationalisation, in the social index context. It applies the event study
methodology to news of the event of index reconstitutions, and then uses Cumulative
Abnormal Returns (CARS) as the dependent variable in further testing. For 152 firm
observations of deletions from the index, significantly negative returns are experienced in the
period surrounding the announcement, whilst for 453 observations of additions to the index
no significant impact is felt. Firm internationalisation, measured by the proportion of foreign
to total sales, is found to moderate the relationship between CSP and CFP, as international
firms experience more negative stock reactions to deletions from the index and more positive
stock reactions for additions to it. This supports the view that international firms are more
strongly punished for negative CSP than firms with a greater domestic focus.
Key words: Corporate social performance; Corporate financial performance; stock index
changes; Event study; International
2
1. Introduction
The role of companies in society is a hotly debated topic. On the one hand their responsibility is to generate wealth for shareholders, yet a narrow focus on doing so can generate undesirable externalities. With growing awareness of corporate behaviour, the notion that unethical corporate actions have no impact on firm value seems increasingly outdated. Numerous real world examples exist which serve to demonstrate that in many cases markets react negatively to news of unethical firm behaviour. In September 2015, news of the issuance of a notice of violation of the Clean Air Act to Volkswagen resulted in a one-thirds decline of the firms share price (US Environmental Protection Agency, 2015 ). Other examples of stock price reactions of similar magnitudes include the British Petroleum oil spill of 2010, the financial debacle triggered in 2011 at Olympus and of course the Enron and WorldCom scandals (for more detailed comparisons see Thomson Reuters, 2015). These are examples of large, internationally orientated firms whose actions in social arenas resulted in financial punishment.
In addition to the bountiful anecdotal evidence supporting the sentiment that corporate misdeeds are punished by financial markets, a large body of empirical work reflects it too. In the academic context, the analysis incorporates both the ‘good’ and ‘bad’ sides of corporate actions (Margolis, Elfenbein, and Walsh, 2007). Corporate social responsibility (CSR) has been defined as ‘situations where the firm goes beyond compliance and engages in actions that appear to further some
social good, beyond the interests of the firm and that which is required by law’ (McWilliams, Siegel,
and Wright, 2006). Such a conception of firm responsibility has intuitive appeal, but is hard to apply given the subjective nature of it. As an illustration of this difficulty, it is possible to imagine that people differ in their opinions of what constitutes actions that go beyond the interests of the firm. Scholars instead use CSP as a measure of CSR (McWilliams and Siegel, 2000). This paper follows suit and uses the concept of corporate social performance (CSP) to reflect a firm’s prior adoption of CSR practices and policies. Where reference is made to poor corporate behaviour it is in the sense that such behaviour runs contrary to efforts to maintain strong CSP (generally reffered to as unethical behaviour).
3 the firm’s CSP. The opposite is true for firms which are deleted from the index. Investors are
increasingly interested in ensuring their resources reach sustainable, responsible and impact investing (SRI) funds, which is reflected in increasing levels of assets under SRI management (The Forum for Sustainable and Responsible Investment, 2014). This fact illustrates that investors care about social performance, which combined with the above leads to a cursory cunjecture that the removal of expert endorsement from a social index will be negatively greeted by investors and vice versa for the provision of their endorsement.
Evidence on the relationship between CSP and corporate financial performance (CFP) is mixed, though a consensus that a small and positive relationship exists is starting to take hold (Margolis et al., 2007; Van Beurden and Gössling, 2008). A key issue in the literature surrounds the problem of understanding whether CSP drives CFP, or vice versa (Margolis et al., 2007). This endogeneity problem has been solved more recently by studies using reconstitutions of social indices as the basis for a market signal regarding CSP. The reconstitutions follow changes in CSP, and the reconstitutions themselves then provide the market information on which it reacts. Hence, in this setting the CSP clearly precedes CFP. Studies have since adopted it as a common methodology for event studies analysing the relationship between CSP and CFP.
Another branch of academic literature focuses on the relationship between internationalisation and CSP. Johanson and Vahlne (1977) view internationalisation as a ‘process in which the firms
gradually increase their international involvement’. Researchers have found evidence of a positive
relationship between internationalisation and CSP. Intuition for such a relationship can be partly derived from Stakeholder theory, as with internationalisation comes responsibility to increasing numbers of people that are affected by a firms behaviour (Brammer, Pavelin, and Porter, 2009; Chapple and Moon, 2005; Kang, 2013). Given such a link, one might expect that a firm’s degree of internationalisation strengthens the relationship between CSP and CFP. This leads to the question of whether internationalisation moderates the relationship between CSP and CFP.
Many studies have looked at the two distinct areas of CSP related study discussed above, namely the CSP-CFP link and the relationship between internationalisation and CSP. Few studies explicitly combine the areas of study, to understand whether internationalisation itself influences the relationship between CSP and CFP. This paper therefore analyses whether internationalisation strengthens the CSP-CFP link. It is to my knowledge the first paper to use the event study
methodology in conjunction with social index reconstitutions to analyse whether internationalisation moderates the CSP-CFP relationship.
4 indices because studies have concentrated on a limited number of indices. Ramchander, Schwebach, and Staking (2012) say there is a need to study other social indices, as they may have different effects on markets. Doh, Howton, Howton, and Siegel (2010) echo a similar sentiment and say their study of the Calvert social index lacks generalisability as a result. This paper helps answer these calls for the study of additional social indices by analysing an index which has been to date far less frequently studied than others. A second distinct area in which this paper looks to enhance academic knowledge is related to the influence of firm industry on studies using social index reconstitutions. Nitani, Carriere, and Bleackley (2015) specifically call for studies using social index reconstitutions to compare different industries. This paper incorporates an industry analysis, and therefore also expands our understanding in this area. Finally, this paper addresses the influence of internationalisation on the CSP-CFP relationship, where other studies tend to have a more strict focus on either the CSP-CFP relationship or the Internationalisation-CSP relationship. For example, Attig, Boubakri, El Ghoul, and Guedhami (2016) call for studies to consider other corporate outcomes of the internationalisation-CSP connection. Thus this paper further contributes by studying additional outcomes of the aforementioned link.
The study aims to improve academic knowledge related to the aforementioned areas, but also has practical implications. Some researchers point to the idea that CSP can be seen as an attribute which can be added to firm product offerings, and is something that will be popular among certain consumer groups (McWilliams and Siegel, 2001). In this context CSP can be incorporated in decision making in similar circumstances to any investment decision. Firms may then for example make choices on social initiatives to be adopted and promoted in conjunction with a product offering, in the same way that they choose aesthetical features to attract consumers. In fact examples of this are already common: Fairtrade products exist, Etnies promise to plant a tree for every pair of shoes sold (Etnies, 2015) and TOMS will help a person in need for each product purchased (TOMS, 2006). Hence knowledge of whether CSP is profitable and to whom (in terms of industries and multinational firms), can help improve organizational decision making in resource allocation choices related to corporate social practices. More generally, knowledge of the CSP-CFP relationship has implications for corporate social initiatives, specifically in the context of product offerings. Acting for the benefit of society could be considered profitable, which would further societal interests. Furthermore, if internationalisation does strengthen the relationship between CSP and CFP, then choices of social initiatives become more complex for multinational firms, as people from different countries may have heterogeneous preferences.
5 descriptive statistics. The results are then explained in detail, before the implications of the results are examined. Finally the paper considers the limitations inherent in the study and discusses avenues for future research.
2. Theory and hypotheses
In developing the theoretical framework upon which the analysis in this paper is based, this section builds upon two related strands of theory. The relationship between CSP and CFP is dealt with first, before attention turns to how the internationalisation of firm operations may influence the relationship.
2.1 CSP and CFP
The extant literature has tackled the issue of the relationship between CSP and CFP using different methods, and come to differing conclusions. Central to understanding the relationship is the idea of what we expect of firms in society. If we accept that firms have a duty to act responsibly, then one would anticipate that unethical behaviour, as defined previously, will be punished by society. The longstanding debate regarding the role of companies in society has developed from arguments with a primary focus on shareholders, to a more rounded view which encapsulates the demands of a greater number of interest holders. This section makes a division between theories which focus on CSP in the context of it being a burden to companies which reduces their profitability and stock prices, and theories which instead consider CSP to have value to the company. The recurring theme of
endogeneity is relevant to both perspectives. As is discussed shortly, advocates of CSP being a cost have often argued that strong CFP precedes CSP, and that the relationship is unidirectional. On the other hand, proponents of the view that CSP has value consider the relationship to at least be bidirectional.
2.1.1 CSP as a cost
Early arguments were primarily rooted in the work of Friedman (1962), whose contention was that the only social responsibility of companies was to focus on their bottom line. Accordingly, the argument considers CSP to be a cost, which reduces the final profit of a company. The trade-off hypothesis is based on the premise that CSP acts to syphon off the available resources of firms, and therefore puts them at a relative disadvantage to competitors (Preston and O'bannon, 1997; Aupperle, Carroll, and Hatfield, 1985). The trade-off hypothesis is therefore used to support the work of
6 find no evidence to support the trade off hypothesis. The trade off hypothesis of course ignores
potential benefits that may accrue to firms as a result of improved employee compensation, as will be discussed shortly.
An agency theory perspective can be used to advance our understanding of Friedman’s work. The theory suggests that self interested agents will take actions in their own interests which may be at the expense of the firm’s owners (Eisenhardt, 1989). The assumption is that managers investing in CSP to further some social cause, only do so to extract some personal benefit such as an improved image (McWilliams and Siegel, 2001). Accordingly, CSP can be seen as an unnecessary expense which certain managers may indulge in. Chin, Hambrick, and Treviño (2013) find that more liberal CEO’s tend to generate greater advances in the CSP than less liberal CEO’s, as they inject their own beliefs into the CSP policies they chose to adopt. This suggests the existence of individual managerial preferences is central to firm decision making, which could lead to the establishment of non optimal policies which are not solely designed to benefit the firm’s owners.
The managerial opportunism hypothesis further contends that better financial performance enables directors to “cash in” on firm resources and reduce CSP, thereby increasing their individual gains, whereas when profitability is low costly social initiatives can be blamed (Preston and O'bannon, 1997; Charlo et al., 2015). There is little support for this view however, and it is directly contradicted by the slack resource hypothesis, which conversely contends that better financial performance enables a stronger focus on social performance (Waddock and Graves, 1997). Seifert, Morris, and Bartkus (2004) and Melo (2012) among others find support for the thesis that increased CSP leads to improved CFP, and thus find no evidence for the managerial opportunism hypothesis. Orlitzky, Schmidt, and Rynes (2003) analyse previous studies of the CSP-CFP relationship and conclude that the relationship is at least partly bidirectional in that while CSP can enhance CFP, CFP can enable greater CSP. This issue of endogeneity is a core advantage of the use of reconstitutions to social indices. This point is advanced in section 3 in greater detail.
7
2.1.2 CSP as a strategic tool
The branch of thinking which currently dominates the academic literature is more closely related to the work of Freeman (1984). Freeman’s stakeholder theory emphasises the interests of a wider range of stakeholders than simply the shareholders, recognising their importance to the continued success of a business. According to Freeman (1984), stakeholders include anyone that can ‘affect or is affected by the achievement of the organization's objectives’. Freeman emphasised the cost of stakeholder conflicts. In the Freeman context, unlike Friedman, CSP is a tool whose use facilitates the smooth running of stakeholder relationships and reduces transaction costs for the firm, through for example reputational benefits accrued as a result of social initiatives that shield the firm from the risk of consumer activism.
According to the literature CSP can positively influence firm level financial performance in a number of ways. Nitani et al. (2015) argue that there are four basic means by which CSP can have a positive impact on financial performance. These are: 1) improved reputation 2) enhanced human capital 3) cost reductions and 4) risk reductions. First of all, firm level CSP developments provide a public signal which forms part of the public image of the company, and influences the firm’s relationship with most stakeholders (Fombrun and Shanley, 1990; Fombrun, Gardberg, and Barnett, 2000). To the extent that consumers care about ethical issues, the reputational impact will influence any number of firm level outcomes related to its financial performance (e.g. consumer boycotts). It is therefore not normative managerial judgements that matter, but instead the opinion of consumers which is important, as if they deem behaviour to be unethical they may retaliate against the firm (Shea, 2010).
Secondly, employees are a key stakeholder in any firm and with the emergence of global competition, the necessity for a highly talented workforce has increased. In a ‘global war for talent’ companies need to differentiate themselves to attract talent, and CSP can function as one such means (Beechler and Woodward, 2009). According to the efficiency wage and gift exchange literatures there is a relation between the real wage paid and employee productivity, as employers that pay a wage above the market clearing rate benefit from increased employee effort (Yellen, 1995; Akerlof, 1982). Firms that pay high real wages can be seen to transfer wealth to a key stakeholder, namely their employees. Thus in the context that high real wages represent positive CSP, it is the firms CSP
initiatives that are enhancing employee productivity and providing benefits to the firm itself. In a study of Spanish manufacturing firms, Sánchez and Benito-Hernández (2015) find that internal CSP policies lead to short term increases in employee productivity. This contradicts the argument supported under the Friedman way of thinking, where transfers to employees can be considered solely as a cost.
8 and Epstein, 2005; Flammer, 2013; Godfrey, Merrill, and Hansen, 2009). This amounts to cost savings in the event of such problems, and can therefore generate a profitable return on the firms CSP
investment. The fourth and final way that CSP can positively influence firm level financial
performance is that firms which invest a lot in CSP policies have a lower cost of equity capital, as they are associated with larger investor bases and lower perceived risk (El Ghoul, Guedhami, Kwok, and Mishra, 2011). The reduced risk thus generates cost savings on equity capital.
Instrumental stakeholder theory is a more recent conception of Freemans work on stakeholder theory, and sees CSP as an ‘instrument’ through which shareholder returns can be maximised. CSP can enable firms to build strong relationships with stakeholders who they repeatedly engage with, such as customers and suppliers (Jones, 1995). CSP initiatives can for example, facilitate trust building and thus enable the firm to develop more effective and profitable relationships with stakeholders.
Donating to environmental groups in response to concerns over a firms pollution, is one example of an initiative which can build trust amongst consumer groups, and enable the management of stakeholder relationships. The upshot of this is that firms may find it profitable to invest in CSP as a means of developing improved stakeholder relationships, and that CSP then becomes part of the firm’s corporate strategy (McWilliams and Siegel, 2001). McWilliams and Siegel (2001) further apply the resource based view to CSP, as means to illustrate the potential for it to be a source of competitive sustainable advantage.
It is important to return to the issue of endogeneity, but now in the context of studies which have focused on beneficial results of CSP. The social impact hypothesis is another version of stakeholder theory, but it argues for a bidirectional relationship between a firm’s social performance and their financial performance, which sees stakeholder relationships develop and financial results follow (Preston and O'bannon, 1997). As touched upon when considering CSP as a cost, other evidence already exists to support this form of the relationship, including work by Orlitzky et al. (2003) amongst others. The available funds hypothesis on the other hand considers financial
performance to lag corporate social performance, as firms with strong financing are able to fund social initiatives more easily. Treating CSP as the dependent variable has provided support for this
hypothesis (Margolis and Walsh, 2003; Preston and O'bannon, 1997). This study escapes the
endogeneity problems associated with a bidirectional relationship as firm level CSP, by construction, precedes the financial markets reaction to the firm’s addition to, or deletion from the S&P
Environmentally and Socially Responsible Index.
2.1.3 Current thinking on the relationship
9 of endogeneity which is inherent in many previous studies. The reconstitutions enable researchers to separate CSP from its financial repercussions to overcome the bidirectional issue (Kappou and Oikonomou, 2012). Many studies rely on developments to social ratings to provide an indicator of changing CSP, but Kappou and Oikonomou (2012) additionally argue that changes to social ratings can be hard to disentangle from changes to financial performance. The reconstitutions to social indices are an improvement, as changes in the companies CSP precede the firm’s addition to, or deletion from the index, and these reconstitutions provide the signal to which financial markets react.
In their review of 167 studies on the CSP-CFP relationship, Margolis et al. (2007) conclude that there exists a small, but positive effect. Van Beurden and Gössling (2008) also review the literature, but are stronger in their assertion that there is a positive correlation between social and financial performance. Studies have persisted, with the majority of scholars now finding support for the view that a broader firm focus incorporating a wider range of stakeholder groups can in fact have a positive impact on CSP (Flammer, 2013; Deng, Kang, and Low, 2013; Aktas, De Bodt, and Cousin, 2011). Studies looking specifically at reconstitutions to social indices as a means to study the CSP-CFP relationship tend to find a negative impact of firm deletion (Kappou and Oikonomou, 2012; Doh et al., 2010; Ramchander et al., 2012; Becchetti, Ciciretti, Hasan, and Kobeissi, 2012), while evidence on additions is mixed. The above illustrates that there appears to be an increasing agreement amongst scholars that CSP and CFP are positively related.
McWilliams and Siegel (2001) develop a model which shows that in equilibrium a firm investing in social initiatives should do no better or worse than one which makes no such investment, as whilst they incur higher costs they will also generate higher revenues. In their model the social initiative is just anther attribute which is demanded by consumers and priced into the product. They go on to suggest this may be a reason for divergent findings in the early literature on the influence of CSP on financial performance. If companies choose to attach different levels of CSP to their products, then the impact of changes in a given firms CSP depends on which firm is being studied. However the more recent convergence towards a general finding that a positive relationship exists between CSP and CFP, as found by both Margolis et al. (2007) and Van Beurden and Gössling (2008), would appear to undermine this idea.
10
Hypothesis 1. Firms added to the S&P environmental and socially responsible index, experience a positive stock price reaction in the days surrounding the announcement. Hypothesis 2. Firms deleted from the S&P environmental and socially responsible index, experience a negative stock price reaction in the days surrounding the announcement.
Several scholars that have used social indices as a means to study the CSP-CFP relationship have found that the magnitude of market reactions differ when firms are added to social indices, compared with when they are dropped from it. Doh et al. (2010), Becchetti et al. (2012) and Kappou and Oikonomou (2012) find that markets react negatively to deletions from social indices, but find no effect for additions. These findings suggest that the negative market reaction to deletions from the index is of a greater magnitude to the positive reaction of additions to the index. Additionally, Trudel and Cotte (2009) conclude that unethical behaviour weighs more heavily on consumers willingness to pay for a product, than ethical behaviour does. This implies that the underlying business of firms that behave unethically can suffer, as consumers will only pay a lower price for their products vis-à-vis competitors.
Doh et al. (2010) suggest that the psychology field and prospect theory can explain why there are differences in reactions to ethical and unethical behaviour, or put differently good and bad CSP. The psychology field has documented that people react more fiercely to negative information than they do to positive information (Diener, 2000). Prospect theory postulates that it is a person’s reference point which determines how they experience an event (Jawahar and McLaughlin, 2001). Because of the impact of a person’s reference point, it is possible that the psychological value people assign to actions and events can differ from the actual value of the event (Kahneman and Tversky, 1979). Thus even if additions and deletions in the social index context refer to actions of the same magnitude, people are likely to adopt a reference point which punishes negative CSP more strongly than it rewards positive CSP due to the psychological disposition found by Diener (2000). This leads to following hypothesis:
Hypothesis 3. Firms deleted from the S&P environmental and socially responsible index, experience a higher magnitude of stock price reaction in the days surrounding the announcement than the stock price reaction for firms which are added to the index.
2.2 Internationalisation, CSP and financial performance
11
2.2.1 Internationalisation and CSP
Internationalisation has received many definitions. An early definition from Johanson and Vahlne (1977) views internationalisation as a ‘process in which the firms gradually increase their international involvement’. They consider internationalisation to occur through ‘increasing
involvement of the firm in the individual foreign country, and successive establishments of operations in new countries’. This section first extends stakeholder theory to the international context, in which the firm has an increased number of parties that are affected by their activities, before applying the strategic view of CSP and considering several other relevant perspectives common in the literature. Before considering the influence of internationalisation on the CSP-CFP relationship, it is necessary to look at the direct impact of internationalisation on CSP.
To understand how internationalisation influences the firm’s level of CSP, Stakeholder theory again proves relevant. Previously, stakeholder theory was applied in order to illustrate how CSP can facilitate the smooth functioning of relationships, which can have a myriad of benefits to the firm. In this international context, geographic diversification results in an increase in the number of
stakeholders a firm has (Kang, 2013; Brammer et al., 2009). When a firm adds locations abroad, it immediately creates a new layer of stakeholders in the form of foreign governments, customers, suppliers and other interest groups. Chapple and Moon (2005) call this a stakeholder multiplier effect. Each of these groups will have special interests, which intensifies the demands on the individual firm to engage in CSP as a means of satiating them.
The strategic view of CSP is in line with stakeholder theory in assessing the importance of CSP for international firms. The strategic view of CSP takes the earlier discussion on the costs and benefits of CSP a step further, by advocating that an analysis of these costs and benefits form the basis for the determination of a firm’s optimal level of CSP (Orlitzky, Siegel, and Waldman, 2011). The implication of this is that with internationalisation, the firm’s best strategic response is to increase investment in CSP in order to meet the increased demands placed on the firm (Attig et al., 2016).
The literature tends to focus on several additional arguments to support the idea that
12 possible negative events which can hit them increases, meaning the goodwill and trust generated by CSP become increasingly valuable (Kang, 2013; Laudal, 2011).
2.2.2 CSP as an avoidable distraction to international firms
An alternative view does exist regarding the impact of internationalisation on CSP, which considers internationalisation as a means to avoid CSP pressures. The pollution haven hypothesis supports the notion that firms may locate operations in countries whose CSP standards are low (Dam and Scholtens, 2008). Additionally. a short term focus on profit maximisation could also result in the internationalisation-CSP relationship breaking down as the benefits of CSP are likely to be more long term (Kang, 2013). While these arguments have a solid base there is, as will be discussed next, strong evidence within the literature to show that internationalisation and CSP are in fact positively related.
2.2.3 Implications for the CSP-CFP relationship
Studies that consider the impact of internationalisation on CSP have invariably concluded that more international firms do have better CSP. Attig et al. (2016) find that firm internationalisation has a positive influence on CSP ratings, and that the relationship holds for a sample spanning 44 countries. Orlitzky, Louche, Gond, and Chapple (2015) also confirm this by stating that the international scope of a company may be a key antecedent to its CSP. According to Laudal (2011), internationalisation results in greater differences in cost levels and labour standards, which both generate a high potential for CSP. Laudal (2011) goes on to say that differences in norms and values across countries influence demands on workplace conditions and environmental actions, which further influence the CSP demands on international firms. Chapple and Moon (2005) find that multinational companies have higher CSP than domestic counterparts, and their CSP is likely to correspond more closely to the profiles of the host country than the country of origin. Kang (2013) hypothesises, and finds support for the idea that level of firm diversification is positively related to their CSP. Other studies, including those by Strike, Gao, and Bansal (2006) and Brammer, Pavelin, and Porter (2006) find further evidence to support the relationship.
13 This makes intuitive sense when one considers that firms ‘invest’ in CSP when they believe it to be profitable, which is in line with the view on CSP as a McWilliams and Siegel (2001) model where CSP can be commoditised. Equally, poor CSP should then be punished more fiercely when conducted by international firms. In line with this, firm internationalisation can be said to moderate the CSP-CFP relationship as more international firms are likely to be more highly rewarded for positive CSP, and more heavily punished for negative CSP. This line of argument leads to the following hypotheses:
Hypothesis 4. Internationalisation positively moderates the reaction of stock prices in the days surrounding the announcement of the addition of firms to the S&P environmental and socially responsible index.
Hypothesis 5. Internationalisation negatively moderates the reaction of stock prices in the days surrounding the announcement of the deletion of firms from the S&P environmental and socially responsible index.
3 Methodology
To test the aforementioned hypotheses this paper uses the event study methodology to analyse the impact on short term stock returns of additions to, and deletions from the S&P Environmentally and Socially Responsible Index. The methodology is found to be well specified and powerful when daily stock data are used as in this case (Brown and Warner, 1985). This section describes the methodology.
3.1
Event window
14
Estimation window
-260
-1
Event window
Event date (t=0)
-2
Event time (t)
(-1, 0) event window, that is an event window starting one day prior to the event and including the day of the event. This event window is also common in the specific area where social index reconstitutions are the basis for the events (Becchetti et al., 2012; Doh et al., 2010). Although the window is short and restricts the ability to draw long term conclusions, Doh et al. (2010) state that it helps isolate the specific impact of the event. For robustness, and to check whether other effects may occur outside this window, the analysis is conducted for three extra event windows, namely the (-3, 3), (-1, 3) and (-1, 1) windows.
3.2
Estimation window
While the aforementioned event window covers the time period during which one expects a reaction to the event, the estimation window is used for estimating the parameters needed for the market model. This paper uses a years worth of trading days prior to the event, to estimate the parameters. The estimation window runs up to the start of the event window, but does not overlap it. This is common in the literature as it helps ensure that the estimators derived from the market model are not affected by the returns in the event window, which would otherwise result in both the abnormal and normal returns incorporating the impact of the event (MacKinlay, 1997). The estimation window used here is sufficiently long to allow for the assumption that the sampling error is zero, and therefore does not influence the variance of the abnormal returns (MacKinlay, 1997). Figure 1 is used to illustrate the set-up of the study in accordance with the main (-1, 0) event window described above. It shows that the estimation window starts a year prior to the event and does not overlap the event window.
Figure 1: Time line of the event study
15
3.3
Abnormal returns
Abnormal Returns (ARs) are used as a measure of the impact an event has on the firm’s stock returns. It is the realised return of the company minus its ‘normal return’ during the event window. As is common in the literature this paper makes use of the market model to calculate normal returns (MacKinlay, 1997). The market model is based on the assumption that the mean return of a stock has a stable linear relationship with the return of the market. Whilst simple, this model provide very reliable estimates of the abnormal returns, which more complex models fail to do while simultaneously only scarcely improve the goodness of fit (Campbell, Lo, and MacKinlay, 1997). The normal return for firm i at time t, can then be calculated as follows:
Rit= α i + βi∗ Rmt+ εit (1)
𝑅it is the expected return of firm i on day t, Rmt is the return on the market portfolio on day t. α i and βi are the regression coefficients of each individual firm and εit is the corresponding
disturbance term. More intuitively, α i is the return on security i when the market portfolio return is zero, and βi measures the sensitivity of the security to the market return. As the paper employs data from the S&P Environmentally and Socially Responsible Index it makes sense to use the S&P to proxy the market portfolio. Applying this to the normal return model, the abnormal returns are calculated as:
𝐴𝑅 it= 𝑅 it− 𝛼 ̂i− 𝛽 ̂ ∗ 𝑅i 𝑚𝑡 (2) ARit is the abnormal return and 𝑅it the actual return. The final terms make up the normal return that would have been earned in the absence of the event according to the model.
3.4
Cumulative average abnormal returns (CAARs)
Abnormal returns must be aggregated over observations and time to enable analysis. First of all, the average abnormal return (AAR) of all observations, N, is taken for each day of the event window. 𝐴𝐴𝑅𝑡 = 1 𝑁∑ 𝐴𝑅𝑖𝑡 𝑁 𝑖=1 (3)
The final step in calculating the CAARs is to cumulate the AARs over the days of the event window.
𝐶𝐴𝐴𝑅1,𝑡= ∑ 𝐴𝐴𝑅𝑡 𝑇
𝑖=1
16 This process is done separately for firms which were added to the index, and firms which were dropped from the index. The CAARs show the short term reaction of firm stock returns resulting from the announcement of reconstitutions to the social index. The significance of the CAARs can be tested through the use of a two sided t-test in the event that there are no issues regarding the normality of the underlying CARs which make up the CAARs (MacKinlay, 1997). The existence of non-normally distributed CARs can under certain circumstances necessitate an alternative treatment. Brown and Warner (1985) find that in a cross section of securities the abnormal returns converge to normality with higher numbers of securities. They find that samples much in excess of 50 securities meet this criterion, so given that this paper employs samples of 453 firms added to the index and 153 firms dropped from it the issue of normality in the CARs should be unimportant. Nevertheless, the results of the Corrado rank test are presented as robustness tests. The Corrado rank test is a nonparametric test which can be used to test the first three hypotheses in the presence of non-normal CARs (Kolari and Pynnonen, 2011).
3.5
Regression model
A significance test of the null hypothesis that CAARs for the relevant event windows are not significantly different from zero for the two groups of firms is the first step of the analysis as described in the previous section. The next step is to treat the individual firm CARs as the dependent variable in a linear regression model. Recall that the CAAR, which is used for testing the first three hypotheses, is simply the average of the all of the firm CARs. Two distinct regression models are estimated. The first is estimated separately for firm additions and firm deletions, which facilitates a comparison of how the different independent variables included in the model influence the two distinct dependent variables. The second regression model combines the samples and includes a reconstitution dummy, as well as an interaction term between the reconstitution dummy and internationalisation. This interaction term shows the influence of internationalisation on the relationship between the reconstitution dummy and the dependent variable. This second regression model is estimated twice: once with the reconstitution dummy equalling one for additions and zero for deletions, and another time with the reconstitution dummy equalling one for deletions and zero for additions. This is done to ensure that the influence of internationalisation as a moderator of the relationship between CSP and CFP can be distinguished for the instances of good CSP (firm additions) and bad CSP (firm deletions).
17
3.5.1 Internationalisation
Internationalisation is defined here as the fraction of foreign sales earned by each firm as a percentage of their total sales. Sullivan (1994) considers different measures of internationalisation and finds that the ratio of foreign sales to total sales is the most frequently used approach to measuring it. The measure remains sensitive to the possibility that it is driven by factors which do not truly reflect the firm’s international orientation. This point is developed further in robustness tests, where different measures are applied to the analysis.
3.5.2 Control variables
Equations 5 and 6 includes numerous variables whose inclusion helps to mitigate the possibility that the relationship between CSP and CFP is influenced by other factors. Specifically, these are:
Industry: Margolis et al. (2007) state that socially responsible practices can differ per industry,
18 industry’s are assigned a number for the dummy variable. Excluding one facilitates the circumvention of the dummy variable trap, which could otherwise cause multicollinearity within the regression model.
Management quality: Waddock and Graves (1997) coin the term ‘good management theory’ to
describe the idea of good CSP being linked to strong management. The assertion is supported in more recent studies including Garcia-Castro, Ariño, and Canela (2010). Quality of management is measured with a dummy which equals one if the company had Six Sigma and Quality Management Systems in the year that the event took place.
Profitability: According to the slack resource hypothesis, firms with high profitability will
have higher CSP as they are not as constrained in their spending. The argument sees CSP as optional, meaning firms with some slack in their finances are more likely to spend on CSP. Profitability is measured by the ratio of net income to total assets (ROA).
Size: Firm size is commonly used as a control in studies of CSP and CFP. The argument for it
is not dissimilar to that used in contending that international firms are likely place more importance on CSP. Larger firms are likely to have more stakeholders to please, and therefore likely invest more in maintaining relationships and to have a higher CSP. Additionally, larger firms have a stronger reliance on reputations and have more to lose from poor CSP. Wu (2006) finds evidence for a small, positive relationship between firm size and CSR. Size is measured as the natural logarithm of total assets as is common practice.
Time trend: Flammer (2013) documents an increasingly negative response to environmentally
harmful actions by firms as time passes. This is driven by external pressures which determine the norms firms are adapting to, and result in stronger expectation for firms to act responsibly.
Environmental efforts of firms are closely related to CSP. In the social index context, it is one of the key issues making up the scores on which firm reconstitutions are decided. Thus, the effect is relevant to this study despite the fact that the data are distributed over a small time frame (six years). Setting the time of the first observation equal to zero allows for a numerical value ranging from zero to six to be assigned to each observation, which can be incorporated in the regression model.
Market capitalisation: Doh et al. 2010 suggest the additional use of market capitalisation as a
19
Leverage: High leverage can indicate possible future financial distress, which can restrict a
firm’s willingness to spend large sum on initiatives which do not provide immediate benefits (Verwijmeren and Derwall, 2010). Equity holders may fear bankruptcy, and prevent management from spending on such initiatives with the fear that they will suffer losses in the event of financial problems. Leverage is measured as the ratio of debt to equity.
4 Data and descriptive statistics
Given the multifaceted nature of the study, the analysis conducted in this paper uses data from several sources. The initial stages of the analysis were conducted with data I personally obtained from the S&P Environmentally and Socially Responsible Index. It included a list of firms that were added to, or deleted from the index in the period starting 2010 and running up to 2016. For the purpose of the event study this data was supplemented with stock price information from Datastream. Finally, data for the regression and control variables were obtained through the use of Datastream and Orbis.
Having constructed the methodological basis for the analysis of the events, this section introduces the S&P Environmentally and Socially Responsible Index in more detail and explains its use as an indicator of CSP. It then focuses on the sample employed by the study, and explains the removal of certain observations, before describing the descriptive statistics.
4.1
The S&P environmentally & socially responsible index
The index attempts to track S&P firms, as well as firms from other sister S&P indices, which meet sustainable investing criteria, without significantly altering the risk and performance profiles of the underlying index (S&P Dow Jones Indices, 2016). The index specifically excludes companies from certain industries, such as fossil fuel heavy industries and firms involved in the production of tobacco products and weapons. Companies receive Environmental and Social dimension scores (E&S scores), which are compiled by RobecoSAM. RobecoSAM analyse companies along three dimensions, where each dimension consists of a list of criteria (39 criteria are used in total) on which each
company is assessed (RobecoSAM, 2015). The first dimension is economic, and consists of criteria such as codes of conduct and customer relationship management. The second is environmental and consists of criteria like environmental footprint and climate strategy. Finally, the social dimension consists of criteria including human capital development and labour practice indicators.
20 deterioration, means that the announcement of reconstitutions provides a signal to the market
regarding recent changes in the quality of the firms CSP. This fact provides one key explanation for the use of social indices, as their use solves the endogeneity problem described by Margolis et al. (2007), where the studies focus on events in which it is difficult to determine whether the CSP measures precedes the CFP measure.
A second and related advantage is that there is minimal event uncertainty as the announcement date is the day of the event, and the news is published by the index. Relying on media releases on the other hand can introduce great uncertainty, as the story may have been run by other news outlets or the news release could be delayed. The final advantage mentioned here stems from the fact that it
overcomes the difficulty of measuring CSP. Many authors rely on news releases as the basis for the event, and assume these releases infer an equal magnitude of CSP impact. The use of social indices on the other hand relies on the methodology and analysis of a professional body, which bases CSP judgements on homogenous criteria. Another popular methodology is the use of CSP ratings, however these are criticised by Kappou and Oikonomou (2012) as the ratings are only available to subscribers meaning not all investors have access to them, which undermines the use of ratings as the basis for an event study.
4.2
The sample
The original dataset garnered from the S&P Environmentally and Socially Responsible Index consisted of 708 observations, encompassing 444 firms. Of this, 497 of the observations related to firm additions to the index, and 211 to deletions. Upon conducting the event study the dataset had been reduced to 605 observations, encompassing 397 firms. 453 of these observations referred to additions of firms to the index, and 152 to drops. The full sample is listed in Appendix A.
21 Some additions to the index take place outside the normal rebalancing windows for some firm specific reason (S&P Dow Jones Indices, 2016). An example of this is Hewlett Packard Enterprise Co, whose stocks were inactive until the 19.10.2015, and were then subsequently added to the S&P environmental and socially responsible index around two weeks later on 02.11.2015. The inclusion of such an observation would pose two problems. First, the addition of the company to the index is based on the fact that it just started trading, and thus was not eligible for inclusion prior to its date of
addition. The second issue is empirical, as the two weeks of prior data do not provide a large enough estimation window over which to reliably estimate the coefficients alpha and beta. Similarly, firms may be deleted outside the rebalancing window for firm specific reasons (S&P Dow Jones Indices, 2016). As a result all deletions occurring outside the rebalancing window were removed. An example of this is Pepco Holdings Inc., which stopped trading on 24.03.2016 as the result of a merger with Exelon Corporation (Investorplace, 2016). Pepco Holdings Inc. was subsequently deleted from the S&P Environmentally and Socially Responsible Index on the 30.03.2016. The deletion of such firms from the index was likely influenced by the fact they had stopped trading shortly before.
The removal of a total of 103 observations for the various reasons outlined above, helps to increase the likelihood that the firms which are analysed, were added to, or deleted from the index as a result of changes in their social performance. The omission of observations did not result in any significant changes to the makeup of the data with respect to the number of firms in the data as a fraction of total observations. The removals did however slightly reduce the proportion of observations which referred to firm deletions from the index. The reason for this is that the vast majority of index rebalances occurring outside the main rebalancing windows, related to firm deletions from the index.
4.3
Descriptive statistics
Tables 1 lists the descriptive statistics and correlations of the variables included in the
regression model for firm additions in the upper panel, deletions in the middle panel and the combined sample in the lower panel. The mean CAR listed in the table is of course the same as the Cumulative Average Abnormal Return (CAAR), as will be seen in the results section. Hence the mean CAR is referred to as the CAAR from here on.
22 deleted firms. The difference in the time trend variable is largely the result of the early reconstitutions to the index being made up of additions. Firms dropped from the index tend to be more highly levered than firms added, which is in accordance with expectations as high leverage may lead to a reduction in spending on social issues as mentioned previously. Firms added to the index make a greater proportion of their sales in foreign countries than firms deleted from it, while a higher number of these firms have implemented Six Sigma and Quality Management Systems.
In the upper and middle panels, only two of the independent variables are strongly correlated with the CARs corresponding to each observation. The strongest is the internationalisation correlation with CARs for firms deleted from the social index, which is negative. This implies that more
international firms are more heavily punished (more negative CARs) than less international firms, which is in line with hypothesis five. Although the sign of the correlation between the CARs and internationalisation has the expected sign in the sample of added firms, it is not significant. Finally, correlations between the independent variables are all low and therefore not indicative of issues with multicollinearity.
23 Variable Mean S.D. 1 2 3 4 5 6 7 8 9 10 Additions 1. CAR (-1, 0) -0.001 0.016 2. Internationalisation 0.332 0.284 0.036 3. Industry 3.614 2.705 0.003 0.274 *** 4. Management 0.264 0.442 -0.111 * 0.258 *** 0.007 5. Profitability 0.699 0.552 0.046 0.059 0.389 *** 0.148 ** 6. Size 16.640 1.364 0.026 -0.086 -0.361 *** -0.030 -0.343 *** 7. Time trend 1.992 2.222 0.017 -0.074 -0.073 -0.085 -0.134 ** -0.129 * 8. Market capitalisation 16.742 1.772 -0.083 0.044 -0.042 0.044 -0.009 0.356 *** 0.050 9. Leverage 0.407 11.113 0.080 -0.033 -0.098 -0.098 -0.044 -0.037 0.015 -0.042 Deletions 1. CAR (-1, 0) -0.004 0.017 2. Internationalisation 0.267 0.274 -0.151 * 3. Industry 3.687 2.788 -0.084 0.275 *** 4. Management 0.173 0.380 -0.126 0.260 *** -0.024 5. Profitability 0.670 0.627 -0.015 0.046 0.400 *** 0.099 6. Size 16.381 1.222 -0.009 -0.128 -0.320 *** -0.153 * -0.180 ** 7. Time trend 3.713 1.880 0.091 -0.094 0.061 -0.099 -0.193 ** -0.165 * 8. Market capitalisation 16.674 0.876 0.035 0.013 0.010 -0.068 0.038 0.466 *** 0.018 9. Leverage 1.139 3.192 0.041 -0.056 -0.068 -0.033 -0.086 0.050 0.029 -0.031
Additions and deletions
1. CAR (-1, 0) -0.002 0.016 2. Reconstitution dummy 0.621 0.486 0.102 ** 3. Internationalisation 0.308 0.282 -0.022 0.112 ** 4. Reconstitution dummy * 0.206 0.276 0.083 0.585 *** 0.709 *** 5. Industry 3.641 2.733 -0.032 -0.013 0.271 *** 0.165 *** 6. Management 16.542 1.317 -0.104 ** 0.105 ** 0.267 *** 0.234 *** -0.005 7. Profitability 0.688 0.581 0.023 0.024 0.056 0.050 0.392 *** 0.130 *** 8. Size 0.230 0.421 0.023 0.095 ** -0.088 ** -0.001 -0.345 *** -0.059 -0.275 *** 9. Time trend 2.644 2.257 0.001 -0.370 *** -0.115 ** -0.263 *** -0.020 -0.121 ** -0.152 *** -0.165 *** 10. Market capitalisation 16.716 1.496 -0.050 0.022 0.038 0.046 -0.029 0.022 0.003 0.369 *** 0.031 11. Leverage 0.684 8.976 0.062 -0.040 -0.037 -0.049 -0.083 * -0.087 * -0.046 -0.027 0.029 -0.041
Table 1: Descriptive statistics and correlations
Descriptive statistics and correlations are for the analysis of 453 additions to the index, 152 deletions from it and 605 observations for the combined sample.
24
5 Results
The first three hypotheses can be analysed through the use of significance tests on the CAARs of firms in the period surrounding the announcement of their addition to, or deletion from the S&P Environmentally and Socially Responsible Index. The final hypotheses are analysed with the results of the aforementioned regression model.
5.1
CAARs
Table 2 shows the CAARs as a percentage for firm additions to, and deletions from the index for different event windows around the reconstitution announcement. The number of CAARs which were positive and negative is also reported, along with the p values for two tailed t tests. Firms added to the S&P Environmentally and Socially Responsible Index surprisingly had negative CAARs for all windows. Firms which were deleted displayed a similar reaction, although their CAARs were all highly significant (mainly at the 1 per cent level). The CAARs are negative for firms added to the index, but not significantly so, with the exception of the (-1, 3) event window.
The analysis supports the idea of negative returns to firms deleted from the S&P
Environmentally and Socially Responsible Index. However firms added to the index do not appear to benefit from a favourable market reaction, but instead appear to be punished for it. The negative returns to firms added to the index are generally not significant. Whilst the hypothesis that firms experience positive stock market reactions after being added to the index is of course not supported, it is also not possible to conclude that these firms experience negative reactions. Although the lack of a positive reaction is surprising, such a finding is not unheard of as Kappou and Oikonomou (2012) and Krüger (2015) both find that the impact of positive CSP on CFP can be negative, or zero. In the discussion section, reasons for why positive CSP might not result in positive market reactions are discussed.
25
Event time CAAR pos : neg p value CAAR pos : neg p value
(-3, 3) -0.002 212 : 241 0.236 -0.006** 65 : 87 0.04 (-1, 3) -0.003 ** 203 : 250 0.035 -0.008*** 60 : 92 0.001 (-1, 1) -0.002 202 : 251 0.104 -0.006*** 60 : 92 0.002 (-1, 0) -0.001 200 : 253 0.444 -0.004*** 60 : 92 0.007 Deletions Additions
Table 2: CAARs for the time surrounding the index reconstitution
CAARs are for reconstitutions of the S&P Environmentally and Socially Responsible Index. Event time is in days, CAAR is the cumulative average abnormal return as a percentage. The analysis is for 453 additions to the index, and 152 deletions from it. Pos : neg refers to how many CAARs were positive or negative in each event window. * Significant at the 10% level, ** significant at the 5% level and *** significant at the 1% level.
5.2
Regression results
Including the models for firms added to the index, firms deleted from the index and the combined sample, seven regression models are estimated. Table 3 shows the results of separate regressions for each type of reconstitution. For each sample of firms (firms added to the index, and firms deleted), two models are estimated, to help isolate the impact of internationalisation. Three separate regressions are estimated for the combined sample, as shown in table 4 and discussed shortly.
Table 3 provides the results for when the CARs generated by each set of firms are regressed on the corresponding independent variables described previously. As expected the coefficient on internationalisation in the model with deleted firms is negative, to reflect the idea that increased internationalisation is associated with more negative CARs for firms that are deleted from the social index. The internationalisation coefficient of firm added to the socially responsible index also has the expected sign, given that more international firms are likely to be find it more beneficial to be socially responsible. However, the coefficients on internationalisation are not significant in either set of results. Despite this, the R2 increases greatly in each model when internationalisation is added to the
regression (although they are low overall). This means that internationalisation does have an impact on the amount of variation in CAARs that is explained by the model. Similar to the findings in the
analysis of CAARs, the influence of internationalisation is found to be stronger for firms displaying negative CSP, than those displaying positive CSP.
Management quality is significantly negatively related to the CAARs of firms added to the social index. This means that strong CSP is not as strongly rewarded for firms with strong
management. This is perhaps not surprising; as these firms are likely to promote their CSP, meaning that investors are likely to already know of their CSP quality prior to social indices recognising it. The negative coefficients on management quality for firms added, albeit insignificant, also fit this
26 from the index they are likely to be more heavily punished than firms which have less control over their public image. Finally, there is also a significantly negative relationship (at the 10 percent level) between market capitalisation and firm internationalisation for firms added to the index. This suggests that more recognised firms are less rewarded for strong CSP.
In terms of the signs that were expected for the control variables the results are mixed. For instance, the time trend has a positive coefficient for both firms added and deleted. According to prior research one would expect the punishment for negative CSP to increase over time (Flammer, 2013), which is not borne out. It is less clear what should be expected to happen over time with regards the development of the CFP reaction for positive CSP. The sign on the size coefficient does however fit the expectations outlined before, as bigger firms appear to be more heavily punished for bad CSP, and more rewarded for good CSP.
Table 4 shows the results of the models estimated for the combined sample. Model 5 only includes the control variables, model 6 additionally adds the reconstitution dummy and
internationalisation, whilst the final model incorporates the interaction term. Models 6 and 7 are estimated twice to incorporate the influence of the two values the dummy can take. While the P-value’s are of course identical, it serves to illustrate the difference in the sign of the coefficient on the interaction term. The reconstitution dummy is significant in its own right in model 6, which provides further evidence that the type of reconstitution does influence the CAARs. Model 7 then additionally adds the interaction term which is significant at the 10 percent level, but not at the 5. The signs of the coefficients reflect the idea that more international firms are more rewarded for strong CSP
27
Coefficient P-value Coefficient P-value Coefficient P-value Coefficient P-value Coefficient P-value
Reconstitution dummy 0.004 0.017 ** 0.001 0.671 -0.004 0.017 ** -0.001 0.671 Internationalisation 0.001 0.742 -0.006 0.215 0.001 0.742 0.005 0.181 Reconstitution dummy * Internationalisation 0.011 0.061 * -0.011 0.061 * Industry 0.000 0.472 0.000 0.432 0.000 0.438 0.000 0.432 0.000 0.438 Size 0.001 0.430 0.001 0.446 0.001 0.455 0.001 0.446 0.001 0.455 Profitability 0.002 0.221 0.002 0.180 0.002 0.182 0.002 0.180 0.002 0.182 Management quality -0.004 0.044 ** -0.005 0.029 ** -0.005 0.027 ** -0.005 0.029 ** -0.005 0.027 ** Time trend 0.000 0.898 0.000 0.320 0.000 0.324 0.000 0.320 0.000 0.324 Market capitalisation -0.001 0.241 -0.001 0.211 -0.001 0.199 -0.001 0.211 -0.001 0.199 Leverage 0.000 0.312 0.000 0.284 0.000 0.275 0.000 0.284 0.000 0.275 Observations 605 605 605 605 605 R2 0.021 0.036 0.045 0.036 0.045 Model 7 Model 6
Model 5 Model 6 Model 7
Reconstitution dummy = 1 for additions Reconstitution dummy = 1 for deletions
Coefficient P-value Coefficient P-value Coefficient P-value Coefficient P-value
Internationalisation 0.005 0.197 -0.006 0.274 Industry 0.000 0.913 0.000 0.798 -0.001 0.146 -0.001 0.286 Size 0.001 0.161 0.001 0.149 -0.001 0.433 -0.001 0.403 Profitability 0.003 0.154 0.003 0.119 0.001 0.551 0.001 0.666 Management quality -0.004 0.098 * -0.005 0.052 ** -0.006 0.115 -0.005 0.222 Time trend 0.000 0.534 0.000 0.489 0.001 0.356 0.001 0.439 Market capitalisation -0.001 0.100 -0.001 0.085 * 0.001 0.510 0.001 0.471 Leverage 0.000 0.255 0.000 0.266 0.000 0.669 0.000 0.695 Observations 453 453 152 152 R2 0.036 0.043 0.040 0.048 Additions Deletions Model 4 Model 3 Model 2 Model 1
Table 3: regression analysis with CARs as the dependent variable
ARs are regressed on the independent variables. Internationalisation is the ratio of foreign sales to total sales, industry is dummy reflecting the firms line of business, management is a dummy for high management quality, profitability the firms ROA, size the natural logarithm of total assets and time trend measures the distance of the individual observation from the first across time. * Significant at the 10% level, ** significant at the 5% level and *** significant at the 1% level.
Table 4: regression analysis with CARs as the dependent variable
28
5.3
Statistical assumptions
To test whether the errors in the model have constant variance and are therefore
homoscedastic, White’s heteroscedasticity test and the Breusch-Pagan test are used for each of the seven models. Results of the tests are shown in appendix B, and both fail to reject the null hypothesis of homoscedasticity. Appendix C shows the results of the Durbin-Watson (DW) and Breusch-Godfrey tests for autocorrelation. The DW statistic is close to 2 for each model, while the Breusch-Godfrey test provides no significant results, which indicates that autocorrelation is not a problem. The assumption of normality is investigated with the use of Jarque-Bera (JB) test. Appendix D presents plots of the distribution of residuals under each of the seven regression models, along with the JB test statistic. The test shows that the data for all models are not normally distributed, as the JB statistics are significant under each, meaning there is a problem of skewness and excess kurtosis. The sample size should make the non-normality a non-issue. However given that the non-normality is likely to be caused by outliers, as can be seen in the distributions of residuals, the data is Winsorized as a robustness test as discussed in section 5.4.
5.4
Robustness tests
Within the analysis conducted in this paper, several important choices and assumptions are made which could influence the results of the tests. In the initial phases of the event study choices regarding the estimation windows and event windows need to be made. Then in defining a regression model the choice of event window again proves important as it is the dependent variable in the analysis. The independent variables are also defined in accordance with what is considered best practice, but it is possible the measures are not optimal. Finally, the results are interpreted on the assumption that the finding of non-normal data is not important given the large sample size. These tests require the re-estimation of the regression equations. The final three models combining both firms are used for this purpose as the overall conclusions can be judged from these. The issues mentioned here are dealt with now in the order discussed.
29 differences in the CAARS and p values found for deleted firms, although the conclusions are
unchanged as the CAARs are negative and significant under all estimation windows.
A second possible source of variation in results could come from the event window used. Some initial robustness checks for events windows are shown in table 2 of section 5.1 where the CAARs for different event windows are shown. The robustness check is already included in results because it is also important to the interpretation of the findings. The inclusion of additional event windows is an important robustness check as it is feasible that an effect would only show up in certain windows. If this were the case, there would be conflicting results as there is no hard and fast rule on which event window to select for analysis. Table 2 shows that these results are robust to the use of different event windows surrounding the announcement. Next, appendix F shows the regression results for the combined models, when the CARs found for different event windows are used as the
dependent variable. It is immediately obvious, that event windows expanding over a longer time horizon do not capture the effect found in the shorter windows. As was argued earlier, the event uncertainty and anticipatory effects are minimal which means that small event windows are likely to capture the impact of the index reconstitution, but longer ones may not.
The robustness checks conducted so far show that the analysis is robust to different measures of the dependent variable, but it is important to also consider measures of the independent variables. internationalisation is measured as the proportion of foreign sales to international sales within the analysis. The results are shown in appendix G, and the conclusions drawn do change slightly when internationalisation is instead measured as foreign income to total income in the regressions. The reconstitution dummy remains significant when the interaction term is excluded, so the type of event does influence the short term stock returns. However, the interaction term is no longer significant, which casts doubt over the findings which used the initial measure of internationalisation. The other conclusions are unchanged.
30 In the event that CARs are not normally distributed, the t-test used to determine whether CAARs for testing the hypotheses would not be adequate. Whilst not strictly necessary in this instance due to the large sample size, the use of nonparametric tests provides robustness to possible outliers (Doh et al., 2010). The Corrado rank test is a nonparametric test that can be used in such instances (Kolari and Pynnonen, 2011). Results are shown in appendix H. Except for the (-3, 3) event window all CAARs in the deletion window remain highly significant and negative. The added firms all remain negative and insignificant. Clearly, the results are not sensitive to the non-normality of the data. A secondary means to confirm the robustness of results to non-normality in the regression variables is conducted where all variables are Winsorized at the 0.95 level, and the analysis rerun. Appendix I has the results of this analysis. Results are almost identical, with the interaction effect again proving significant in the full model although now at a higher level (5 per cent instead of 10). The
reconstitution dummy is significant on its own in model 6, where no interaction term is used. Thus the non-normal data does, as expected, have minimal influence on the results and conclusions drawn from them.
6 Discussion
This paper has applied the event study methodology to analyse the impact that reconstitutions of the S&P Environmentally and Socially Responsible Index have on stock returns, as measured by the CAARs. The reconstitutions of the index act as a signal to markets regarding the CSP of a given firm. If a firm is added to, or deleted from the index it is the result of a change in their environmental and social dimension scores (E&S scores). An addition to the index implies an improvement in the firms E&S score, while a deletion from the index suggests a deterioration of the score. The use of socially responsible indices as the source of event information related to CSP remains uncommon but there are at least three advantages of using social indices, as discussed in the data section: it solves the
31 Findings on the influence of internationalisation on the CSP-CFP relationship are mixed. Whilst for firms deleted from the S&P Environmentally and Socially Responsible Index the internationalisation coefficient is negative and significant, it is positive but not significant for firms added to the index. The sign of the coefficients is as would be expected, given that poor CSP is likely to be more costly for international firms and strong CSP to be more beneficial. The interaction term used to understand if the moderating influence of internationalisation on the CSP-CFP relationship is significant in model 7, and provides evidence that more international firms are more rewarded for strong CSP (additions), and punished more for weak CSP (deletions). The findings from the regression analysis are robust to the non-normality of the data, but not to longer event windows or alternative measures of internationalisation. The lack of robustness to different measures of internationalisation the more problematic of the two issues, as it was already anticipated that reconstitution effects would be very short term. There is therefore mixed evidence on the role of internationalisation in moderating the relationship between CSP and CFP. Researchers have demanded studies on the role of industry in social index reconstitutions (Nitani, Carriere, and Bleackley, 2015), but no industry differences are found to exist in this paper as the industry dummy is insignificant in all models.
The finding that firms deleted from the S&P Environmentally and Socially Responsible Index have negative CAARs in the period surrounding the announcement is as predicted, based on previous literature and theory. Researchers are increasingly finding that weak CSP is punished by stock markets, and this finding holds in the social index context. The related theory also predicts such an outcome. In line with stakeholder theory, firms that pay little attention to vital interest groups can suffer from a breakdown in key relationships. Possible repercussions include damaged reputation, inability to attract a high quality workforce and a vulnerability to crises due to a lack of goodwill in society.
