The Value of Corporate Social Responsibility in Emerging Markets An Event Study using the Dow Jones Sustainability Index

(1)

The Value of Corporate Social Responsibility in Emerging Markets

An Event Study using the Dow Jones Sustainability Index

By G. J. te Velde

University of Groningen Faculty of Economics and Business MSc. International Business and Management

First supervisor: Dr. John Q. Dong Second supervisor: Dr. André A. J. van Hoorn

(2)

2

ABSTRACT

(3)

3

1. INTRODUCTION

In recent years, some companies have suffered from Corporate Social Responsibility (CSR) misbehavior. British Petroleum, also known as BP, for example, saw how one of its offshore drilling rigs, the Deepwater Horizon, exploded and sunk to the bottom of the ocean on 20 April 2010. BP‟s reputation got heavily damaged and its share price almost halved between April 2010 and June 2010 (BBC 2010a, The Telegraph 2010). Another well-known example is the Tokyo Electric Power Company, also known as TEPCO, and its nuclear power plant Fukushima. Triggered by an earthquake on 11 March 2011, The International Atomic Energy Agency (IAEA) classified the disaster on the International Nuclear Event Scale (INES) as a „Major Accident‟ (INES 7), which only happened before with Chernobyl (Steinhauser, Brandl & Johnson 2014). The share price development for both TEPCO and BP starting 8 days before each event and ending 52 days after the event can be found in Figure 1. Here, the share prices are indexed using the 8th day before the event as the base date.

Both examples show that the markets punish firms that have to deal with environmental disasters, which was also confirmed by Flammer (2012). She further states that the negative market reaction towards environmental disasters has increased over time.

Although these examples only show disasters in developed countries, emerging markets also have to deal with similar problems. In 2008, for example, the Sanlu Group announced that its milk powder was poisoned after several babies died. Eventually, the firm filed for bankruptcy and the Chinese government imprisoned several people (BBC 2010b). In 2007, Zhang Shuhong, director at Lee Der Industrial, committed suicide after Mattel, an American manufacturer of toys, accused his firm of being responsible for producing toys painted with

0,00 20,00 40,00 60,00 80,00 100,00 120,00 -8 -6 -4 -2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 Tepco BP

Figure 1: Share price development BP & TEPCO

(6)

6 toxic lead (Barboza 2007). Both examples show that the consequences of being an irresponsible firm from an emerging market can be severe, too.

These examples show that CSR misbehavior can have a tremendous impact on the share price of a firm, and even jeopardize its existence, its consumers, its managers, and employees. They clarify that a company cannot allow itself to behave irresponsible as the consequences can be severe.

Fortunately, many companies have recognized the importance of CSR and are actively communicating their initiatives – McDonalds supports the Ronald McDonald House Charities, Microsoft supports the Bill & Melinda Gates Foundation, and Coca Cola supports Replenish Africa Initiative (RAIN). Flammer (2012) argues that the public increasingly expects firms to behave responsible, but that as the expectations increase the rewards for well-behavior decrease, while miswell-behavior is punished more severe.

Furthermore, Flammer (2012) states that “the external pressure to behave responsibly […] has increased dramatically over the past decades”, which eventually also means that firms have set up more CSR initiatives to satisfy these demands. Because of this development, it becomes increasingly difficult for investors interested in CSR to find out which companies perform best. For firms, however, it also becomes more challenging to find out which firm has the best CSR strategies, which can act as a benchmark. In the end, CSR is also just another way to diversify the firm from competitors and outperform them in another field. This shows that for both investors and firms an overview is needed of the firms that perform best in terms of CSR.

(7)

7 There are already quite some researchers that have investigated how the stock markets react on the membership of such a sustainability index, but with inconclusive results. Some researchers conclude that neither the addition to nor the deletion from a sustainability index affects the stock price over the long term (Curran, Moran 2007, Karlsson, Chakarova 2007, Cheung 2011, Siegmund, Witt 2012). Others argue that only an addition affects the share price (Robinson, Kleffner & Bertels 2011, Nakai, Yamaguchi & Takeuchi 2013) or that only a deletion has an effect (Consolandi et al. 2009, Tillmann 2012).

These studies, however, have concentrated on European (Consolandi et al. 2009, Curran, Moran 2007, Siegmund, Witt 2012), North American (Cheung 2011, Robinson, Kleffner & Bertels 2011) or Japanese (Nakai, Yamaguchi & Takeuchi 2013) firms only. Other researchers have focused on multiple regions, like the ones mentioned above, but excluded emerging markets (Karlsson, Chakarova 2007, Tillmann 2012). No research before has included emerging markets when investigating the effect sustainability index membership has on the share price of a firm.

There are, however, some researchers that focus on CSR initiatives in specific emerging markets. Muller and Kolk (2009), for example, focus on Mexican CSR activities, Aras, Aybars, and Kutlu (2010) on Turkish firms, Cheung, Tang, Ahn, and Zhang (2010) discuss CSR performance in Asian markets, Lourenço and Branco (2013) use a sample of Brazilian firms, and Arya and Zhang (2009) focus their study on CSR activities in South Africa. These studies, however, have not focused on the effect sustainability indices may have on the share price of a firm.

Therefore, this paper makes an important contribution to existing literature by focusing on emerging market firms when investigating the effect DJSI membership may have on the share price of a firm. This means that the event study methodology used in the articles mentioned first will be performed on a sample of firms introduced in the other articles. It is interesting to research this topic because it is generally assumed that emerging markets differ from developed markets. This research will shed some light on this specific topic

(8)

8 only 31 events occurred between 1999 and 2012 and 56 in 2013. Therefore, an event study at an earlier point in time would probably not have been very useful as the sample size was too small.

This limited availability of events is also the reason why this study solely focuses on the additions to and not the deletions from the DJSI. The final sample, which only includes deletions that meet the criteria specified in part 3.2.3 (page 21), consists of 18 deletions. With such a small sample size, it is very difficult, and probably not very useful, to reach any conclusions.

The results of this research show that there is no significant effect. This means that the addition of an emerging market firm to the DJSI does not significantly increase the abnormal returns at the event date or in the event window. There are, however, differences across emerging markets when forming regional clusters or groups based on the level of market efficiency.

Although there are differences between emerging and developing countries, there are also differences between different emerging markets. That is why this research also clusters countries based no their location to form regional clusters. These results show that Latin American markets, for example, react different from Asian markets.

Market efficiency is also an important concept in the event study methodology. If the markets are not efficient, it is unlikely that a specific market reaction will be found. Therefore, countries are also clustered based on their level of market efficiency. The results based on these sub-samples show that high efficient markets react more positively to a DJSI addition than medium or low efficient markets do.

(9)

9 market efficiency. A thorough discussion of the results, including the implications for managers, limitations and future research will follow in part 6 (page 38). This paper will end with some concluding remarks in part 7 (page 41).

2. THEORETICAL FRAMEWORK

In this part, first, the existing literature in the field of CSR in emerging markets will be introduced. After that, literature on DJSI membership will be discussed and the DJSI methodology itself will be presented. Lastly, this part closes with the research question and related hypotheses.

2.1. Literature Review

Several researchers investigate how CSR initiatives and financial performance are related in emerging countries. Aras, Aybars, and Kutlu (2010), for example, test the effect of CSR on the profitability of a Turkish firm. They concluded, however, that there is no significant link between these variables.

Muller and Kolk (2009) focus on CSR initiatives in Mexico and also recognize that research in the field of CSR has mainly concentrated on European or North American firms. Still, they conclude that Mexican firms have put into place several CSR activities, which are not very different from „Western‟ CSR initiatives.

Cheung et al. (2010) study CSR in Asian emerging countries and conclude that between 2001 and 2004 CSR performance has increased, except in Malaysia. Next to that, they found out that if an Asian firm improves its CSR practice, the market will reward this behavior. They admit, however, that Asian emerging markets lack behind Europe and the US in terms of CSR because of their family or concentrated ownership.

Lourenço and Branco (2013) confirm that ownership concentration matters when dealing with CSR in emerging markets. From their Brazilian sample the results show that leading CSR companies have a more dispersed ownership structure. This study also confirms that size matters, as leading CSR firms are larger than firms that care less about CSR.

(10)

10 announcements in South Africa have a positive effect on the share price after institutional reforms are put in place.

These different studies on several emerging markets show that firms from these countries are actively involved in CSR activities. The results, however, are inconclusive as some studies report that the market rewards social responsible behavior, although others do not find any significant benefit.

These studies on emerging markets, however, have not focused on the concept of sustainability indices. Neither the DJSI nor the sustainability index of the Financial Times Stock Exchange (FTSE4Good) have been used to find out how the markets respond if firms from an emerging market are added to or deleted from a sustainability index.

Research on sustainability indices, however, has only included developed market firms. The results of these studies, however, are inconclusive. In the following some important papers will be presented that have studied the financial impact of sustainability indices.

Curran and Moran (2007) study the effect of an addition to or deletion from the FTSE4Good UK Index in 2001 and 2002. They conclude that the share price of a firm from the UK does not increase when being added to the index; neither does the share price decrease when it is deleted from this index.

Consolandi, Jaiswal-Dale, Poggiani, and Vercelli (2009) focus on additions to and deletions from the Dow Jones Stoxx 600 index, which lists sustainable, European firms. Analyzing the period between 2001 and 2006, they find that the share price already increases before the announcement was made and conclude that it might relate to information leakages. This effect, however, is only temporal. The negative effect of a deletion is not temporal and is much stronger than the positive effect of an addition. Still, they argue that the limited time perspective might be a reason for this non-temporal effect.

Siegmund and Witt (2012) also focus on European firms, but use the additions to and deletions from the FTSE4Good index between 2006 and 2011. Their results show that an addition has no positive effect on the share price and a deletion only temporally decreases the share price of a firm.

(11)

11 no effect on the share price. There is, however, a positive temporal effect around the day the addition becomes effective and a temporal negative effect when the deletion becomes effective.

Robinson, Kleffner, and Bertels (2011) had a closer look on if and how the share price of US firms changes when these firms are added to or deleted from the DJSI. They conclude that being added to the DJSI results in a sustained rise of the share price, whereas being deleted from the DJSI had only a temporal negative effect on the share price.

Nakai, Yamaguchi, and Takeuchi (2013) focus their research on additions to and deletions from the Morningstar Socially Responsible Investment (SRI) index between 2003 and 2010. They study Japanese firms and conclude that additions to this index lead to an increase in share price, but that deletions from this index have no significant effect.

Whereas the researchers introduced above focus on one specific country or region, others use a broader approach. Karlsson and Chakarova (2007), for example, focus on the Nordic countries (Sweden, Denmark, Norway, and Finland), France, Germany, Japan, UK, and USA. They use the additions to and deletions from the DJSI World between 2002 and 2007. Although their results show that an addition or deletion does not change the share price significantly, this differs across regions and time.

Tillmann (2012) also focuses on firms from several European countries (Switzerland, Germany, Spain, France, the UK, Italy, and The Netherlands), Australia, Canada, Japan, and the US. He uses the additions to and deletions from both the DJSI and the FTSE4Good Index from 2002 to 2008. The outcome of this study is that the addition to an index does not result in an increase in the share price. There is, however, a significant negative effect if a firm is deleted from the FTSE4Good Index.

(12)

12

2.2. Research Question and Hypotheses

As can been seen in the literature review, no research has been done yet to investigate if the share price of a firm from an emerging market is influenced by the addition or deletion from a sustainability index. Therefore, this paper is based on the following research question:

Is there a positive abnormal return if a firm from an emerging market is added to the Dow Jones Sustainability Index?

The reason why this paper only focuses on the additions to the DJSI is because there is not yet sufficient data available about deletions. Between 2002 and 2013 only 48 firms were deleted from the DJSI compared to 159 additions. Out of these 48 deletions, only 12 deletions are useful after the selection criteria, see part 3.2.2 (page 19), are employed.

Now the research question has been defined, the next step is to create hypotheses that will be tested towards the end of this study.

In general, it is argued that the event window should be as short as possible so that it becomes unlikely that other events distort the results (Karlsson, Chakarova 2007, Curran, Moran 2007). Therefore, the first hypothesis is:

Hypothesis 1: The share price of a firm from an emerging market increases on the event date.

The first hypothesis can also be expressed in statistical terms. To measure an effect at a specific date, the average abnormal return (AAR) is used, which will be explained in detail in part 3.2.4 (page 22). This means that a null, , and an alternative hypothesis, , are created:

and .

MacKinlay (1997) states that it is also important to capture the effects that occur before and after the event. This means that the event window should consist of multiple days surrounding the event date. Therefore, the following hypothesis is tested:

(13)

13 The second hypothesis can also be expressed in statistical terms, where is the null hypothesis and is the alternative hypothesis. The average abnormal return over a specific event window is called average cumulative abnormal return (ACAR), which will also be explained in detail in part 3.2.4 (page 22). Based on this, the second hypothesis can be expressed as:

and .

Detailed information about the precise event date and the event windows that are used in this paper can be found in part 3.2.1 (page 17).

2.3. DJSI Procedure

Here, more information will be given about the procedure of how the DJSI decides which firms are added and which firms are deleted. For more details, please look at the Emerging Markets Index Guide (RobecoSAM, S&P Dow Jones Indices 2013).

It starts with the annual Corporate Sustainability Assessment (CSA), which is used to estimate the sustainability of different firms. The questions of this assessment relate to economic, social, and environmental issues. The economic dimension is related to, for example, corporate governance, stakeholder engagement, and research and development. The social dimension asks for more information about, for example, labor practices, health and safety, and talent attraction. The environmental dimension focuses on, for example, the environmental footprint, climate strategy, and biodiversity. About half of these questions are used for every firm, although the other half is industry specific (RobecoSAM, S&P Dow Jones Indices 2013).

(14)

14 As the assessment is likely to be biased because companies can give their answers themselves, a Media and Stakeholder Analysis (MSA) is performed. This analysis evaluates the response of a firm if there is news about some CSR misbehavior. Although the CSA is only performed on an annual basis, the MSA is continuously. The results of this MSA are eventually combined with the scores of the CSA so that a final Total Sustainability Score appears (RobecoSAM, S&P Dow Jones Indices 2013).

After the Total Sustainability Score has been calculated for each firm, it is looked at the scores within a specific industry. The highest score within a specific industry should at least be 40 per cent of the highest score of all the industries together, otherwise all firms within that industry will be excluded from the DJSI Emerging Markets. Next to that, all the other firms within a specific industry should have a Total Sustainability Score of at least 50 per cent of the industry leader, which means that if the leader scores 90, all the other firms that are interested in being listed at the DJSI Emerging markets should have a score of at least 45.

Eventually, only the best 10 per cent emerging market firms of each industry will be added to the DJSI Emerging Markets. Additionally, the top 15 per cent of all firms from an emerging market will also be added to the DJSI Emerging markets, irrespective of their industry (RobecoSAM, S&P Dow Jones Indices 2013).

3. METHODOLOGY

(15)

15 According to MacKinlay (1997), an event study consists of five different steps. The first task is to describe the event, the corresponding event window, and the estimation period. Second, the selection criteria are determined as not all events should be included in the study. Next, the estimation procedure is started where the expected normal returns are calculated. Fourth, we use the actual and expected return to calculate the abnormal returns. As a fifth, and last task, we need to test if the results are statistically significant. Later, starting in part 3.2.1 (page 17), each of these steps will be discussed extensively.

3.1. Assumptions

As with each methodology, also the event study methodology is based on certain assumptions. McWilliams and Siegel (1997) argue that there are three important assumptions that must be met in an event study. These assumptions are based on the market efficiency, information leakages, and other extraordinary events and will be discussed now.

First, it is important that the market efficiency is high. According to Fama (1970), an efficient market is one where all available information is reflected in the security prices. McWilliams and Siegel (1997) clarify that only in an efficient market an event can have a significant impact on the stock price. Mensi (2012) focuses on market efficiency in emerging markets and argues that there are some differences between individual emerging markets. He concluded that Argentina has the highest market efficiency and Tunis the lowest of the sample. From the sample of countries in this study, South Africa scores best, while Colombia is the least efficient country in this sample.

Based on the importance of this assumption and the finding of Mensi (2012) that market efficiency differs across countries, it is important to keep this into account when testing for significance. Therefore, the market efficiency data provided by Mensi (2012) will be used to form three groups based on the market efficiency score of an individual emerging country. Then, for each group the average abnormal return on the event date and the average cumulative abnormal return for the event window will be calculated to see if and can be rejected.

(16)

16 in the post-event window, this indicates that the market is not very efficient and needs some time to adjust to the new information. Therefore, a post-event window will also be tested, as will become clearer in part 3.2.1 (page 17).

The second assumption is that the new information should not have been leaked before the event date. Kothari and Warner (2006) argue that if information has been leaked before the official announcement, some or all abnormal returns are expected to show up before the event date. In line with Siegmund and Witt (2012), a pre-event window will be tested to assure that the chance of information leakages is not ignored in this study. If the pre-event window has significant positive abnormal returns, it is likely that a leakage of information has occurred.

Third, during the event window there should be no other events that may influence the abnormal returns. If this assumption is not met, and other events have occurred during the event window that may have had an impact on the share price, the results of the event study will be distorted. Therefore, an assessment is performed to discover any extraordinary events that may have had an impact on the share price and thereby the abnormal returns. If such an extraordinary event is found, the specific event, which event window has been assessed, is excluded from this study.

As a parametric test is used in this paper, it must also be taken into account that this test assumes that the abnormal returns are normally distributed. According to Serra (2004), parametric tests are not useful when the normality assumption is violated. Brown and Warner (1985), however, conclude that non-normality does not impact event studies significantly. They argue that with an increasing sample size the abnormal returns get closer to a normal distribution. Still, parametric tests are even useful if only 5 securities are used within a sample. As nonparametric tests, however, are free of this normality assumption (MacKinlay 1997), such test will be performed if the normality assumption is violated.

3.2. Procedure

(17)

17

3.2.1. Event Definition

First of all, the event needs to be defined before we are able to set the event window. According to Henderson (1990) the event date is not the date when the event actually happens, but the date the date on which the market first received the new information. For an event study, it is important that the event is correctly defined; otherwise no useful conclusion can be drawn from the data.

In this study, the objective is to find out if the addition of a firm from an emerging market to the DJSI increases the value of that specific firm. This means that the event date should equal the date on which the market receives this new information for the first time. The DJSI announces the additions to and deletions from its indices via a press release, usually on the third Friday of September (RobecoSAM, S&P Dow Jones Indices 2013).

The sample of this study, however, consists of many firms all over the world, which means that differences in time zones are an important factor to look at. Stocks markets in Asia, for example, are already closed, although the markets in America are not even open. Chakarova & Karlsson (2007) argue that because the headquarter of the DJSI is in Zürich, Switzerland, the event date for European and American stocks equals the announcement date. For Japanese stocks, however, they argue that the event date is one day after the announcement date as the Japanese stock market is already closed when the announcement is made.

Although this argumentation of Chakarova & Karlsson (2007) sounds reasonable, it is not sure that this reflects the truthful circumstances. Therefore, the DJSI has been contacted to get more information about the exact time of the publication of the annual reviews. The DJSI responded that it publishes the results after the United States stock markets are closed. This means that for all firms in the sample, including European and American securities, the event date is one day after the official announcement date.

Event Window

(18)

18

Author Event Window

ED: Event Date; EfD: Effectiveness Date Chakarova & Karlsson (2007: 15) ED-1:ED

ED:ED+1 Curran & Moran (2007: 532)

ED-4:+6 ED-4:ED+5 ED-1:ED+8 Consolandi et al. (2009: 189) Pre-event: ED-10:ED-1 event: ED Post-event: ED+1:EfD-1 Effective: EfD Post-effective: EfD+1:EfD+10 Cheung (2011: 148) ED-15:EfD+60

Robinson, Kleffner & Bertels (2011: 498)

Pre-event: ED-60:ED-1 event: ED:EfD-1

Post-effective: EfD+1:EfD+60 Siegmund & Witt (2012: 21)

Pre-event: ED-20:ED-1 event: ED:EfD-1

Post-effective: EfD:EfD+60

Tillmann (2012: 23) Around ED: -1:ED, ED:+1, -1:+1, ED:+3, -1:+3 Nakai, Yamaguchi & Takeuchi (2013: 74) ED-1:ED+1

Table 1: Event Window Overview

For this study, the main event window will be very short; ranging from the event date until one day after the event date (ED:ED+1). This event window (ED:ED+1) is also in line with the suggestion made by MacKinlay (1997), who states that at least the event date (ED) and the day after should be included in the event window. For robustness purposes, two other event windows will be tested, namely (ED-1:ED) and (ED-1:ED+1). In line with the first hypothesis, introduced in part 2.2 (page 12), one test will be performed only considering the event date (ED).

In part 2.2 (page 12), which introduced the hypotheses, it was mentioned that a longer event window might be beneficial as it helps to capture price effects related to the event date. Therefore, another event window will be tested, which starts 10 days before the event date and ends 10 days after the event date (ED-10:ED+10).

(19)

19

Estimation Period

To be able to calculate the expected normal returns at a later stage, an estimation period is needed. Henderson (1990) admits that the estimation period is set before the event window and MacKinlay (1997) additionally states that the estimation period and event window should not overlap as otherwise the expected normal return might be influenced by the event. Seiler (2004) proposes that when using daily data, the estimation period should not be longer than 250 days (about 1 year). In Table 2 an overview can be found of several estimation periods used by previous papers using event studies about sustainability indices.

Authors Estimation period

Chakarova & Karlsson (2007: 15) Daily trading data of 5 months Curran & Moran (2007: 531) 300 trading days

Consolandi et al. (2009: 190) Daily trading data of 52 weeks

Cheung (2011: 148) 235 trading days

Tillmann (2012: 22) 130 trading days

Nakai, Yamaguchi & Takeuchi (2013: 74) 150 trading days Table 2: Overview of Estimation Periods

Table 2 shows that the estimation periods vary widely, which has also been recognized by Tillmann (2012), who states that there is no agreement on the length of the estimation period. Based on these findings, the estimation period is set on 210 trading days, starting 220 days before the event date and ending on the 11th day before the event date.

For scientific papers, it is also important to test the robustness of the results. For an event study, one possibility to test the robustness is to change the estimation period. Tillmann (2012) uses an estimation period of 130 trading days and an extreme short robustness estimation period of 43 trading days. As the estimation period of this study is considerably larger as it was set on 210 trading days, the robustness estimation period should also be larger. Therefore, an estimation period of 90 trading days will be used, starting 100 days before the event date and ending on the 11th day before the event date.

3.2.2. Selection Criteria

(20)

20 First, each security, of course, must be from an emerging country1 and listed on the DJSI. This paper focuses on the DJSI as it is the oldest index and was the best known and second most credible sustainability index in 2013 (Sadowski, Whan & Guenther 2014).

Second, the DJSI should have clearly confirmed the addition of a specific security to its indices. New securities should have been emphasized in each annual index with the word „added‟. Securities that are new, but are not emphasized, were not included. Only this way it can be guaranteed that the security was added on a specific date, which is crucial for an event study.

The third criterion is that securities are only included if they are added for the first time. Some securities are listed on multiple indices, like DJSI Emerging Markets and DJSI Asia Pacific. If a security was already listed on the DJSI, it is unlikely that there will be abnormal returns if it is listed for a second time, but on another DJSI.

Fourth, some securities were added to multiple indices at the same time. Using, however, multiple events for one company at one specific date is not adding any value to this research. Therefore, it only one event per company for each specific date is used.

The fifth criterion relates to the information about the event date. For an event study it is important that the event window can be set as narrow as possible. Therefore, it is important that clear information is available about the event date. If not, the event is excluded from this study.

The sixth relates to extraordinary events. First, as stated in part 3.1 (page 15), no extraordinary events should take place in the event window that may have an effect on the abnormal returns too, like environmental disasters or dividend announcements. If this is the case, such an event is excluded from this study. Second, to be able to estimate the parameters of the regression well, it is important that the estimation period of one year does not overlap with the event window of another year. Otherwise, the effect of one year may affect the abnormal returns of another year if the firm was just added or deleted the year before.

1

(21)

21 Seventh, as the purpose of this study is to find out how the markets respond to additions to the DJSI, only firms that are listed on the stock exchange and are actually traded are useful. Private companies or firms that are not frequently traded are excluded. Brown & Warner (1985) argue that securities should have at least 30 returns in the estimation period and that the last 20 days should not have any missing returns2. Maynes & Rumsey (1993) argue that, however, most event studies have limited data available and that not every researcher can use such strict criteria as Brown & Warner (1985) did. Curran & Moran (2007), for example, used an event window of 11 days where the daily returns for 4 days were actually missing. Therefore, instead of no missing data in the last 20 days, this study also includes firms that have 3 days without data3. The other criterion, more than 30 returns in the estimation period was kept the same. For robustness purposes, the criterion set by Siegmund & Witt (2012) is also used, which requires that there are no missing data during the event window, adjusted for holidays.

3.2.3. Estimation Procedure

After having defined the selection criteria for this event study, it is time start the estimation procedure. The aim of this procedure is to estimate the expected normal return, which is needed to calculate the abnormal return. MacKinlay (1997) defines the expected normal return as the normal return being expected without the event actually taking place.

According to Henderson (1990) and MacKinlay (1997) there are several methods to calculate the expected normal return. McWilliams and Siegel (1997) point out that the market model, a statistical method, is the most common approach to calculate the expected normal return. Basically, the market model approximates the expected normal return by estimating the regression of the return of a security on the return of the market during the estimation period (Henderson 1990). After this has been done, the parameters of the regression are used to calculate the expected normal return during the event window.

2

The 20 days of Brown & Warner (1985) are part of both the event window and estimation period. The event window reaches from -5 to +5 days (11 days in total), which means that the last 9 days of the estimation period should also have daily return data.

3

(22)

22 Before the regression parameters can be estimated, however, the returns are needed for both the stocks and their related market indices. Henderson (1990) suggests using continuously compounded returns, which is a common approach for event studies. The continuously compounded returns of security i on day t, , can be calculated using this formula:

(

) (1)

where is the price of security i at day t (Curran, Moran 2007). This formula is not only used to calculate the return of stocks, but also to calculate the return of the market index of each stock.

Having explained how to calculate the returns, it is now time to calculate the expected normal returns. Using the market model specified by MacKinlay (1997), the expected normal return of security i at time t, , can be calculated by using the following formula:

(2)

where is the intercept of the regression, is the slope of the regression, and is the disturbance term with a zero mean. Here, and are estimated by regressing the returns of a specific stock on the returns of the market index of that stock during the estimation period. These parameters are then used to calculate the expected normal return, and eventually the abnormal return, during the event window, which will be shown in the next step.

3.2.4. Abnormal Return

The abnormal return of security i at time t, , is calculated by subtracting the expected normal return, , from the actual return, :

(3)

where is the actual return of stock i at day t and is the expected normal return of security i at day t (Curran, Moran 2007). As MacKinlay (1997) suggests, the abnormal return is calculated for the event window by using the parameters and of the estimation period:

(23)

23 Now, the abnormal return for a security at a specific day has been calculated. To be able to test the hypotheses, however, the average abnormal return and the average cumulative abnormal return need to be calculated.

The average abnormal return at day t, , is the sum of the abnormal returns of all securities at a specific day, divided by the amount of events (MacKinlay 1997). This process is also called aggregation across securities and is shown in the next formula:

∑

(5)

where is the amount of events. Using this formula, the average abnormal return of all securities can be calculated for each single day in the event window.

To test the significance of the results, the standard deviation of the average abnormal return is needed, which can be calculated using the following formula:

∑

(6)

“where is the residual variance from the market model regression for security i” (Siegmund, Witt 2012), that equals:

(7)

As, however, the event window is longer than 1 day, the average cumulative abnormal returns needs to be calculated. MacKinlay (1997) explains that the average cumulative abnormal return between day t1 and day t2, , is just the sum of the average

abnormal returns over the event window, which is also called aggregation across time:

∑

(8)

where its variance is:

( ) ∑

(24)

24 It is shown now how the average abnormal return, the average cumulative abnormal return, and their variances can be calculated that are needed for significance testing, which is the next step.

3.2.5. Significance Testing

Event studies in the field of sustainability indices have especially focused on parametric tests (Curran, Moran 2007, Consolandi et al. 2009, Siegmund, Witt 2012, Tillmann 2012, Nakai, Yamaguchi & Takeuchi 2013). As mentioned before in part 3.1 (page 15), however, a parametric test is only useful if the abnormal returns are normally distributed. If this is not the case, a nonparametric test should be applied. Corrado (1989) was about the first to introduce nonparametric tests for event studies, called the rank test. Therefore, both the parametric and nonparametric tests will be shown below.

Parametric Test

As the hypotheses introduced before are based on both the average abnormal return and the average cumulative abnormal return, one must also use two different test statistics. As a reminder, the first hypothesis, , states that the average abnormal return at the event date is higher than 0. According to Siegmund & Witt (2012) the test statistic of the null hypothesis of the first hypothesis, , equals:

√ (10)

The second hypothesis, , states that the average cumulative abnormal return during the event window should be higher than 0. MacKinlay (1997) defines the test statistic of the null hypothesis for the second hypothesis, :

√ (11)

Nonparametric Test

(25)

25 except that the abnormal returns are not only calculated for the event window, but also for the estimation period. In this case, the amount of abnormal returns for each of the stocks is defined as . The parameters and of the regression are still based on the estimation period.

After the abnormal returns are calculated for the time frame , they are ranked according to the value of the abnormal return where the lowest value is ranked 1st and the highest value is ranked . “The rank of the abnormal return of security i for event time period t” (MacKinlay 1997: 32) is defined as .

After having ranked the abnormal returns, the test statistic introduced by Corrado (1989) can be used to test the null hypothesis of the first hypothesis, :

∑ ( ₎

(12)

where is the standard deviation:

√ ∑ ( ∑ ( ) ) (13)

Using this test statistic it is possible to find out if the abnormal return at a given day, presumably the event day, is significantly different without assuming that the abnormal returns are normally distributed.

4. DATA

In this part, more information will be given on the data used for this study4. First, an overview will be given on the DJSI data. After that, the corresponding returns for both securities and markets for each event will be collected. More information will also be provided on the regional clusters and the market efficiency groups.

4

(26)

26

4.1. DJSI Data

As the purpose of this study is to find out how the markets react towards the addition of a firm from an emerging market to the DJSI, data concerning DJSI membership must first be gathered. As Table 3 shows, the annual data of six different indices was received.

Index Years

Asia Pacific 2009 – 2013 Emerging Markets 2013 Europe (former STOXX) 2001 – 2013

Korea 2011 – 2013

North America 2005 – 2013

World 1999 – 2013

Table 3: DJSI Overview

As this study solely focuses on firms from emerging markets, see the first selection criterion in part 3.2.2 (page 19), not all indices were used. Only the Asia Pacific, Emerging Markets, and World Index were used as firms listed in one of these three indices are based in one of the 20 emerging markets mentioned before.

(27)

27 Year Announcement Date Additions Deletions

1999 No information available 4 0 2000 No information available 6 1 2001 No information available 8 3 2002 4 September 2 5 2003 4 September 1 1 2004 2 September 2 2 2005 7 September 2 3 2006 6 September 5 1 2007 6 September 1 1 2008 4 September 6 2 2009 3 September 5 4 2010 9 September 3 0 2010 4 November5 1 0 2011 8 September 9 1 2012 13 September 5 3 2013 21 February6 69 0 2013 12 September 28 19 Total: 157* 46*

*2 additions and 2 deletions are excluded from this table because no exact announcement date could be found

Table 4: Announcement Dates and Amount of Additions and Deletions

In Figure 2 it can been seen where all the firms responsible for the 159 additions and 48 deletions are located. Overall, this original sample shows that Eastern Europe is not very well represented, but that the events are almost evenly distributed across Africa, Asia, and Latin America. In Appendix A, Table 15 (page 49), a more detailed overview can be found on how many events are located in each specific country.

Not all of these 159 additions and 48 deletions are useful as not all of the selection criteria mentioned in part 3.2.2 (page 19) have been applied yet, except for the first one. Applying the other criteria, except the last criterion as it requires the abnormal return that

5

On October 8, 2010, SAM announced that there were some problems with its software so that the results announced in September 2010 may not be correct (SAM 2010). On November 4, 2010, it announced the adjustments to the DJSI, which resulted in one additional firm from an emerging market that was added to the DJSI (Kaminski 2010).

6

On February 21, 2013, the DJSI Emerging Markets was published for the first time. Latin America 34% Asia 40% Africa 25% Eastern Europe 1%

(28)

28 will be calculated later, the sample decreases significantly. The sample of events for which the abnormal return will be calculated consists of 95 additions and 18 deletions, which means that 64 additions and 30 deletions did not satisfy these five criteria (2nd to 6th criterion).

4.2. Return Data

At this point, a selection must be made to choose among the different market indices. Overall, as Philips and Kinniry (2012) introduce, there are four global indices, which are the Morgan Stanley Capital International (MSCI), FTSE, Standard & Poor‟s (S&P) and Dow Jones. The return differences between these indices, however, are negligible so that eventually the decision on which index to use depends on preference, accessibility, and the price (Philips, Kinniry 2012).

For this paper, both the MSCI and FTSE index have been used. Both the S&P and Dow Jones index were not used because there were some data availability issues: there was no market data available for some events, which makes it impossible to calculate the abnormal return. Additionally, the Dow Jones index had no data at all for Colombia.

Although both the MSCI and FTSE index are used, the focus of this paper lies on the MSCI index. The reason behind this decision relates to the security free-float requirements. Basically, free- float describes the percentage of stocks that is traded on the stock markets that is available to public investors. The reason why low free-floats are not preferred is because these securities are at higher risk as its price can significantly increase due to low supply and high demand, which results in an artificially high price (Zeiler 2012). The MSCI index requires that the free-float of a security is higher than 50%, whereas the FTSE only requires a free float higher than 15%.

The FTSE is used for robustness purposes as it has slightly more constituents than the MSCI for 9 out of 11 countries. The more constituents a market index consists of, the less impact a single security will have, which improves the representativeness of the index. The amount of constituents for both MSCI and FTSE for each country can be found in Appendix A, Table 16 (page 50).

(29)

29 window. Here, the Datastream database provided by Thomson Reuters was useful to collect these returns. For each security and its corresponding market index price data is gathered.

The main Datastream variable used for this study is the Total Return Index (RI). The main advantage of using this index is that it represents a theoretical growth where dividends are re-invested (Thomson Reuters ). Next to this variable, one test will be performed using the Price Index (PI) instead of RI. The price index does not include the theoretical growth, but is adjusted for capital changes. Tests are performed using both variables to avoid any potential data bias.

Using formula (1), introduced in part 3.2.3 (page 21), the continuously compounded returns are calculated. Then, estimation procedure is started and the expected normal return is calculated. Eventually, the abnormal return for all the 95 additions is calculated.

Now the abnormal return has been calculated, the seventh, and also last criterion, can be applied. This criterion is related to the trading frequency of a security and requires each security to have more than 30 returns in the estimation period and allows only 3 missing returns in the event window. The final sample used for the statistical tests in this study consists therefore out of 87 additions and 12 deletions; 8 additions and 6 deletions do not satisfy this last criterion and are excluded from this research. As can been seen in Figure 3, however, Europe is still weakly represented in the addition sample, but the rest of the additions are almost evenly distributed across Asia, Latin America, and Asia. In Appendix A, Table 15 (page 49) an overview can be found that shows in which countries the events are located.

Because there are only 12 events left at this stage in which a firm from an emerging market has been deleted from the DJSI, it has been decided at this step to not include the deletions in this research. Table 15 (page 49) in Appendix A also shows that half of these events are located in Africa and only 2 in Asia, which is a completely different distribution compared to the additions. Therefore, reaching a conclusion based on such a small sample will not be very useful. Latin America 35% Asia 38% Africa 25% Eastern Europe 2%

(30)

30 As highlighted in part 3.1 (page 15), it is suggested to test for normality before a parametric test is performed. Overall, it seems that the abnormal returns on the event date of the main test7 are only slightly normally distributed. The histogram in Figure 4 shows that the distribution is skewed to the right, which is also confirmed by Table 5. In a normal distribution, both skewness and kurtosis approximate 0. The positive value for skewness confirms that this distribution is skewed to the right. The positive value for kurtosis indicates that there are fewer cases in the tails compared to a normal distribution. Garson (2012) suggests dividing the skewness and kurtosis by their standard errors. When normality is critical, these values should be between -1 and +1. If normality is less critical, these values can also be in the range between -2 and +2 or, only for kurtosis, even -3 and +3. Table 5 shows that for both skewness and kurtosis the most stringent thresholds are not met. Comparing these values with less stringent criteria shows that the abnormal returns are assumed to be normally distributed.

Skewness .299 Kurtosis 1.065

Standard Error (SE) of Skewness .258 Standard Error (SE) of Kurtosis .511

Skewness / SE of Skewness 1.1589 Kurtosis / SE of Kurtosis 2.0841 Table 5: Skewness and Kurtosis

As the abnormal returns have been calculated now, the average abnormal return can be calculated for the event date to test the first hypothesis. The average cumulative abnormal returns can also be calculated now for each event window to test the second hypothesis. The

7

As a reminder, the main test (Test 1) is based on an estimation period of 220 days, the MSCI index, the Return Index, and a maximum of 3 non-return days (adjusted for holidays).

(31)

31 results of these tests will be shown in the next part of this paper. First, however, more details will be given on regional clustering and market efficiency.

4.3. Regional Clustering

As introduced before, it is also important to look at country specific differences. As the amount of events for each country, however, is rather low, it has been decided to group the countries based on their regions. Therefore, a Latin American, Asian, and African sample is created. An Eastern Europe sample will not be tested as it consists of only 2 events. In comparison, the Latin American sample includes 30 events, the Asian 33 events, and the African sample 22 events.

4.4. Market Efficiency

Next to the three regional samples, countries will also be clustered based on their market efficiency. Here, the data provided by Mensi (2012) is used, who investigates the efficient market hypothesis in 26 emerging countries, including the 13 emerging countries used in this study. This study uses the Shannon entropy, which is a measure in the field of information theory that measures the randomness of a variable8. In Table 6 an overview is shown on the results of this study. The second column presents the Shannon entropy median value. This value indicates that South Africa is the most efficient market and that Colombia is the least efficient market.

With these scores in mind, three groups are formed: a high efficient market (1), a medium efficient market (2), and a low efficient market (3).Each country was assigned to a group based on its rank and the promise that all the events are evenly distributed among these three groups. As, however, there are 22 South African events and 17 Brazilian events, the first group is slightly bigger than the other two groups.

8

(32)

32 Country Median Group Events

South Africa 0.951668 1 39 Brazil 0.951345 Hungary 0.951087 2 22 Taiwan 0.949777 Turkey 0.947732 China 0.947087 Philippines 0.945155 Thailand 0.944713 Mexico 0.944629 India 0.942713 3 26 Malaysia 0.941609 Chile 0.933869 Colombia 0.926141 Source: Mensi (2012: 60)

Table 6: Market Efficiency Ranking

For these three different groups, based on the market efficiency score, tests will be performed to test if and can be rejected for a specific subsample.

5. RESULTS

As the abnormal returns have been calculated for the sample, consisting of 87 additions, the average abnormal return at the event date and the average cumulative abnormal returns for the event windows can be calculated. As we are performing a one-tail test, the critical z-score for is 1.282, for is 1.645, and for is 1.960.

(33)

33

5.1. Parametric Test

As can been seen in Table 7, none of the results is statistically significant using the criteria mentioned below the table. This means that neither nor can be rejected.

Sample Size: 87

Event Date AAR Z-score P-value

ED .00037 .23451 .59271 Cannot reject

Event Window ACAR Z-score P-value

(ED:ED+1) -.00109 -.51216 .30427 Cannot reject (ED-1:ED) .00008 .03598 .51435 Cannot reject (ED-1:ED+1) -.00138 -.53417 .29661 Cannot reject (ED-10:ED+10) -.00497 -.64631 .25904 Cannot reject (ED-10:ED-1) -.00006 -.01209 .49518 Cannot reject (ED+1:ED+10) -.00528 -.94053 .17347 Cannot reject

Significant at: *10% level **5% level ***1% level

These results are based on an estimation period of 220 days, the MSCI index the Return Index, and a maximum of 3 non-return days (adjusted for holidays).

Table 7: Overview of Statistical Results

To test whether both null hypotheses really cannot be rejected, four other tests are performed, but based on different assumptions. In the second test, keeping everything else equal, the FTSE index is used instead of the MSCI index. The sample size, however, decreases from 87 to 85 securities because using the FTSE index results in 2 more stocks having more than 3 non-return days. Table 8, however, clarifies that using the FTSE index instead of the MSCI index does not change the results significantly and that we cannot reject the null hypotheses.

Sample Size: 85

ED .00027 .00161 .16615 Cannot reject

(ED:ED+1) -.00095 -.43644 .33126 Cannot reject (ED-1:ED) .00020 .09184 .53659 Cannot reject (ED-1:ED+1) -.00101 -.38278 .35094 Cannot reject (ED-10:ED+10) -.00461 -.59170 .27702 Cannot reject (ED-10:ED-1) .00031 .06069 .52420 Cannot reject (ED+1:ED+10) -.00519 -.91607 .17981 Cannot reject

These results are based on an estimation period of 220 days, the FTSE index the Return Index, and a maximum of 3 non-return days (adjusted for holidays).

(34)

34 A third test is performed where the estimation period is not 220 days, like in test 1 and 2, but now using an estimation period of 90 days, like mentioned in part 3.2.1 (page 17). Table 9, however, shows that with a shorter estimation period the results do not change significantly and the null hypotheses cannot be rejected.

Sample Size: 87

ED .00048 .31118 .62217 Cannot reject

(ED:ED+1) -.00093 -.43562 .33156 Cannot reject (ED-1:ED) -.00013 -.06194 .47531 Cannot reject (ED-1:ED+1) -.00154 -.59107 .27724 Cannot reject (ED-10:ED+10) -.00354 -.45890 .32315 Cannot reject (ED-10:ED-1) -.00010 -.02023 .49193 Cannot reject (ED+1:ED+10) -.00392 -.69505 .24351 Cannot reject

Table 9: Overview of Statistical Results (90 days)

In the fourth test, the abnormal returns are not based on the Total Return Index, but on the Price Index. In Table 10 the results of this test can be seen, but, again, the results are not significant.

Sample Size: 87

ED .00038 .24144 .59539 Cannot reject

(ED:ED+1) -.00162 -.76490 .22216 Cannot reject (ED-1:ED) -.00001 -.00287 .49886 Cannot reject (ED-1:ED+1) -.00201 -.77821 .21822 Cannot reject (ED-10:ED+10) -.00565 -.73175 .23216 Cannot reject (ED-10:ED-1) -.00001 -.00115 .49954 Cannot reject (ED+1:ED+10) -.00602 -1.06658 .14308 Cannot reject

These results are based on an estimation period of 220 days, the MSCI index the Price Index, and a maximum of 3 non-return days (adjusted for holidays).

Table 10: Overview of Statistical Results (PI)

(35)

35 50 securities, as the other 37 securities had more than 0 non-return days. But also here the null hypotheses cannot be rejected, as can been seen in Table 11.

Sample Size: 50

ED .00036 .17067 .56776 Cannot reject

(ED:ED+1) -.00069 -.24642 .40268 Cannot reject (ED-1:ED) .00033 .11995 .54774 Cannot reject (ED-1:ED+1) -.00072 .00334 -.21423 Cannot reject (ED-10:ED+10) -.00042 -.12942 .44851 Cannot reject (ED-10:ED-1) -.00079 -.11520 .45414 Cannot reject (ED+1:ED+10) -.00098 -.11918 .45257 Cannot reject

These results are based on an estimation period of 220 days, the MSCI index the Price Index, and a maximum of 3 non-return days (adjusted for holidays).

Table 11: Overview of Statistical Results (0 non-return days)

In Figure 5, however, the average abnormal returns used for the five different tests are shown. It becomes clear that especially Test 5 (0 non-return days) differs from the other four tests, but also that, overall, the tests yield similar results. It can be seen that the 9th, 7th, and 4th days before the event date the average abnormal return is positive. The z-values its significance of the average abnormal return for each day in the event window for each single

-0,005 -0,004 -0,003 -0,002 -0,001 0,000 0,001 0,002 0,003 0,004 0,005 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10

(36)

36 test can be found in Appendix B, Table 18 (page 52). However, the result on the 7th day before the event date is also statistically significant at the 10 percent level in one test (Test 2) and in 3 tests (Test 1, 3, and 4) at the 5 percent level. The 4th day is only significant at the 10 percent level in Test 2 and 5. Additionally, the exceptionally high score in Test 5 at day 3 is also statistically significant at the 10 percent level.

5.2. Rank Test

Now it is clear that a nonparametric test might be useful to perform as the abnormal returns are not normally distributed. The test introduced by Corrado (1989) looks on the impact the event has on the abnormal return of the event date. In the second column in Table 12 it can been seen that the sum of all ranks at the event date is -213. The lower this value, the higher the abnormal return on that day compared to other days in the estimation and event window. The table shows, however, that this result is not significant, which means that also a nonparametric test cannot reject .

Sample Size: 87

Event Date SUM Z-score P-value

ED -213 -.30386 0.35958 Cannot reject

Table 12: Results Rank Test

5.3. Regional Clustering

(37)

37

Latin America Asia Africa

Sample size 30 33 22

Event Date z-value z-value t-value

ED .02409 -.23391 .23265

Event Window z-value z-value t-value (DF=20)

(ED:ED+1) .20012 -1.55591 .16942 (ED-1:ED) .40084 -.67633 -.16822 (ED-1:ED+1) .49159 -1.72333 -.14149 (ED-10:ED+10) 1.362085* -2.58342 .28192 (ED-10:ED-1) 1.308135* -1.45294 .08103 (ED-1:ED+10) .683785 -2.15710 .24860

Table 13: Region Comparison

5.4. Market Efficiency

The countries in this study are not only clustered based on their location, but also based on their market efficiency level. For each subsample introduced before, the average abnormal return on the event day is calculated, as well as the average cumulative return for the different event windows. Unfortunately, none of the values is significantly higher than 0, which means that both and cannot be rejected.

(38)

38 (1) High Efficient (2) Medium Efficient (3) Low Efficient

Sample size 39 22 26

Event Date z-value t-value t-value

ED .87125 -.42787 -.63112

Event Window z-value t-value t-value

(ED:ED+1) .74620 -.88622 -1.65766 (ED-1:ED) .50167 -.09564 -.89077 (ED-1:ED+1) .50623 -.56270 -1.71856 (ED-10:ED+10) 1.05041 -1.84542 -1.12141 (ED-10:ED-1) .76133 -.77260 -.41893 (ED-1:ED+10) .50831 -1.70418 -1.03691

6. DISCUSSION

The aim of this study is to find out if DJSI membership has a positive effect on the share price of a firm from an emerging market. The event study methodology is used to investigate this effect. The statistical results of this study, however, reveal that it cannot be confirmed that the addition of a firm from an emerging market leads to significant positive abnormal returns. It can be seen, however, that in each of the five parametric tests, the average abnormal return on the event date is positive. Still, in the main event window (ED:ED+1), the average cumulative abnormal returns are negative and not significant for each of the five tests performed.

The rank test, based on the sample used in the first test, also confirms that there is no significant positive abnormal return at the event date. This means that both the nonparametric test and the parametric test come to the same conclusion and that the slightly not-normal distribution of the abnormal returns does not distort the results significantly.

(39)

39 When countries are clustered into specific regions, interesting results can be found. It seems that Asian markets seem to value DJSI membership less than Latin American markets; Africa is in the middle. This conclusion is based on the fact that in Latin America there are only positive abnormal returns for both the event date and the different event windows. In Asia, however, there are only negative abnormal returns. It is important to note, however, that Cheung et al. (2010) conclude that there are differences in how Asian markets value CSR activities, which means that not all Asian markets need to be negative about DJSI membership.

In Latin America, the results show also significant positive abnormal returns in (ED-10:ED+10) and (ED-10:ED-1). This abnormal return in the pre-event window indicates that some information may have leaked to the market before the official announcement was made. This result is in line with Consolandi et al. (2009), who also found a positive abnormal return in the pre-event window and conclude that the markets found out about the results before the event date.

When clustering countries based on their market efficiency score calculated by Mensi (Mensi 2012), there are also some differences. The results show that the most efficient markets have positive abnormal returns, although not significant, at the event date and in the different event windows. The efficient and least efficient markets, however, only have negative abnormal returns at the event date and in the event windows. This still indicates, however, that market efficiency plays a role and cannot be ignored when testing the market reaction on DJSI membership.

6.1. Managerial Implications

For managers of emerging market firms this research has some implications. In general, it seems that managers should not be willing to be listed at the DJSI because they expect it will have a positive effect on the share price. The markets do not respond significantly positive towards the addition so the firm value will not rise if a firm is added to the DJSI.

(40)

40 managers should think about other possible investment projects than CSR activities that may have a more positive impact on the share price. Next to that, it should also be questioned whether it is useful for firms from Asia or Africa to participate in the DJSI assessment as it consumes a lot of time and therefore money.

The study also shows that some managers should first do something about market efficiency. The underlying reason for this is that the high efficient countries have higher positive abnormal returns than medium and low efficient countries. Therefore, improving market efficiency may result in an improved market reaction when a firm is added to the DJSI.

6.2. Limitations

There are some limitations that must be pointed out in this study. First, it is important to acknowledge that this study only focuses on the short term effect of DJSI membership. It is not known if and how the markets respond outside of this event window. A long term study focusing on DJSI members and non-DJSI members from emerging markets is needed before conclusions over the long term can be reached

Another limitation of this paper is the limited amount of constituents for the MSCI and FTSE index for emerging markets. The MSCI and FTSE Hungary, for example, consist of both respectively 3 and 4 constituents. Appendix A, Table 16 (page 50) shows an overview of the amount of constituents per market, including Japan and North America. If a market index has only few constituents, it might be the case that these constituents do not completely reflect the true market circumstances. Therefore, the results must be interpreted with caution.

Third, the aim of this paper was to show how emerging markets respond to DJSI membership. This research has been performed irrespective of any specific firm characteristics, like R&D investment and industry. It is, however, important to take into account that there may be some significant differences between specific categories of firms, just like region and market efficiency matter.

6.3. Future Research

(41)

41 paper. Therefore, it would be interesting to find out how, for example, R&D investment and the industry, among others, matter.

Furthermore, this study has only focused on the positive abnormal return at the event date or during the event window. In Figure 5 (page 35), which shows the abnormal returns for each of the five parametric tests in the event window (ED-10:ED+10), however, it can been seen that these tests also show some highly negative abnormal returns before, but also after, the event. Looking at the corresponding Z-scores in Appendix B, Table 18 (page 52) it can be seen that some of these negative abnormal returns are significantly different from 0. For future research it would therefore be interesting to find out if and how these negative returns are related to the event, but especially how it comes that the abnormal returns can be very high at day 7th day before the event and then be even negative at the 6th day before the event.

Next to that, as the results show that Latin America, Asia, and Africa differ in their market reaction towards DJSI membership, it would be interesting to know why these differences occur. Now, it can only be concluded that they react differently, without knowing what the underlying causes are. Is this difference related to the value attached to CSR? Are Latin American markets more aware of the DJSI and is it a market efficiency issue?

Lastly, this study has only focused on the additions to the DJSI, but what about the deletions? So far, no one knows how markets respond if a firm from an emerging market is deleted from the DJSI. In developed countries the findings were inconclusive so it is very difficult to have any expectations. This study, however, should only be carried out when there are enough events and the sample is large enough, which may take a while.

The Value of Corporate Social Responsibility in Emerging Markets An Event Study using the Dow Jones Sustainability Index