Long term stock price effect of industrial disasters

UNIVERSITEIT VAN AMSTERDAM

Long term stock price effect of

industrial disasters

Student: Jelle Krikke Student number: 10073345

Thesis supervisor: Timotej Homar Research field: Asset pricing Version: June 26th_{, 2015}

1 TABLE OF CONTENTS

1. Introduction ... 2

2 Literature review ... 3

2.1 Background on disaster data and restrictions ... 3

2.2 Emperical studies ... 4

2.3 Factors that explain the impact of accidents ... 5

3. Hypothesis ... 6

4. Data and Methodology ... 6

4.1 Data limitations ... 6 4.2 Disaster selection ... 7 4.3 Model ... 9 5. Results ...10 5.1 Analysis ...11 5.2 Discussion ...16 6. Conclusion ...17 7. Bibliography ...19

2

1. Introduction

In the modern industrialized world we live in today, we are very much dependent on fossil fuels and other chemical power sources that provide us with the energy we use every day. The fact that the modern society depends on these sources more than ever before obviously creates a huge market for all kinds of products we so desperately demand. As we know a lot of these production processes are not without risk. Every couple of years we hear about another industrial disaster, which causes terrible

environmental pollution, incredible danger and sometimes even has deaths as a result. The incredible industrial development of the last century also brought more and more ‘man-made’ accidents. In the 20th _{century the EM-DAT, which is the global} emergency disaster database, recorded 268 industrial accidents in the 30 OECD

countries. OECD stands for organization for economic co-operation and development and they recorded these records from their 30 members, which mostly are the well-developed countries in the world. Before 1970 there were on average 7.3 disasters per decade in these countries, after the 1970 this number grew to an average of 72.3 per decade (Coleman, 2006).

Because there are a lot of interests at stake these disasters cost a lot of money for the stakeholders. Right now the yearly insured cost for man-made disasters is about 8 billion dollars (Coleman, 2006). These are cost that are visible. It is a lot harder to determine what the consequences followed by the disaster are for the responsible

company in the sense of stock price fluctuations. It is to be expected that the days after a severe disaster the stock price of related companies will decline. Cappelle-Blancard and Laguna (2010) wrote about this in their paper ‘how does the stock market react to

chemical disasters’ and concluded that petrochemical firms on average experience a drop of 1.3% market value over the two days following the disaster. The long-term stock price fluctuations can be caused by a lot of factors other than the disaster. To get an idea of the effect the disaster did have in the long-run several disasters will be examined and stock prices will be analyzed at different points in time. In this paper research will be done on what the long-term effects of such a disaster will be on the stock price of the responsible company.

The relevance of this study is that although it is a small study of ten firms which stock returns will be analyzed, it does create a sample of firms that are very comparable industry wise. This makes it possible to be very specific about this branch and the impact a disaster could have on their financial situation. Furthermore this paper focusses on a

3

point in time where a lot of disasters and their consequences have not been analyzed yet. Hence this could give a different results on market reaction to industrial accidents than previous studies. Concluding this research will, as mentioned earlier, will as focal point look at the long term stock price effect caused by the concerned catastrophe, this is something that has not been studied as deeply as the short term effects yet.

2 Literature review

2.1 Background on disaster data and restrictions

Thinking about industrial disasters most people will immediately think of big disasters like the Bhopal disaster in India at December 3, 1984 (Jasanoff,1988) or the Exxon Valdez oil tanker that stranded on a reef in the Prince William Sound at March 24, 1989 (Herbs, Marshall and Wingender, 1996). These are indeed two of the biggest industrial disasters of all time, but there have been so many others over the last decades which have not been in the news as broadly as these two. One reason for the fact that not all disasters discussed in the media is the magnitude of the different disaster, although that is not always a clear measure. In the database of the EM-DAT can be seen that in that same period from 1984 to 1989 there were 113 more disasters worldwide, most of them smaller but some in the same range as the ones mentioned above. Another explanation for the difference in media attention could be the power of the company responsible for the disaster, but also the approach of the media in different countries with different cultures, both arguments cannot be scientifically backed up and therefore will be left for further research.

To get a good view on what to examine it is important to look at what exactly is explained as a disaster. Turner and Pidgeon (1997) in their research on man-made disasters stated that there is no universal accepted definition of a disaster. Parker and Handmer (2013) did a study on hazard management and emergency planning and they found that an appropriate interpretation of a disaster would be the following:

… an unusual natural or man‐made event, including an event caused by failure of technological systems, which temporarily overwhelms the response capacity of human communities, groups of individuals or natural environments and which causes massive damage, economic loss, disruption, injury, and/or loss of life. This definition encompasses medical accidents and disasters such as those which affect of whooping cough vaccine, Opren and HIV/AIDS haemophiniac cases

4

To be more specific about the definition of an industrial disaster further literature has been inspected which gave some clearer insides. An industrial disaster can be defined as ‘a destructive event which, relative to the resources available, causes many casualties, usually occurring within a short period of time’ (Rutherford and de Boer, 1983). In the EM-DAT database at least one of the following conditions has to be met in order be called a disaster and be included in the database:

1. 10 or more people reported killed 2. 100 people reported affected 3. A call for international assistance 4. Declaration of state of emergency

Having the definition of an industrial disaster clarified it is time to look more closely into the frequency of industrial disaster over the last fifteen years. The EM-DAT international disaster database as mentioned has been a useful tool in doing so, for instance we could assemble that during this timeframe there have been 708 industrial disasters of which 517 took place in Asia. Unfortunately it is not a publicly available database. Accordingly data with respect to names of companies responsible for disasters was not available for this analysis. The way of gathering and analyzing the data will be discussed later on.

2.2 Emperical studies

Literature available on the financial effect of industrial disasters focusses mostly on the short term effects. For instance the paper of Cappelle-Blancard and Laguna (2010), as mentioned above analyzed the short term stock returns. They presented some interesting facts about the frequency of industrial disasters in petrochemical facilities. This is one of the most relevant industries when looking into industrial disasters and according to the US Environmental Protection Agency, 8% of the all petrochemical facilities reported at least one accident over the period between 1994 and 2000. This establishes ones again that there are a lot of industrial accidents that are not publicly discussed in the media. Keeping in mind the definition we collected from the literature earlier and the media aspect it could be thought that sizeable companies would be more likely to cause industrial disasters that also will be broad under the attention of the public. Unfortunately no literature has been written on this and during the course of

5

writing this paper I had no available access the detailed EM-DAT database to examine this question.

Studies by Klassen, Curtis and Mclaughlin (1996) and Rao (1996) both

considered a relatively small sample, but found a significant drop in stock price followed by a disaster. Jones and Rubin (1999) on the other hand found no significant decline in the stock market in there study of 14 oil and power disasters that happened between 1970 and 1992. Salinger (1992) did an event study annualizing among others the Bhopal disaster of 1984 finding a CAR of -31.5% at a 1% significance level in December 1984, a CAR of -22% at a 5% significance level in January 1985 and a CAR of 8.9% in

December 1985 for Union Carbide suggesting no long term stock price effect either. How big and powerful a firm is within an industry obviously influences the magnitude of the effect of the disaster in the stock market. The Exxon Valdez oil spill mentioned earlier is a good example to show this. Exxon Mobil is, and was back then, one of the biggest oil companies in the world who caused one of the biggest industrial tragedies of all time. Surprisingly there was no significant effect on their stock prices volatility (Herbst, Marshall and Wingender, 1996). The first month after the disaster they did incur a firm decline in the stock price, but after that recovered fast. What was

noticeable during the period right after the spill was that other oil tycoons like Texaco and Arco, who did not have a stake in the Prudhoe Bay, where the spill happened, did draw a significant decline in volatility (Herbst, Marshall and Wingender, 1996). In 1992 Exxon was named the most profitable U.S. Corporation in 1991 (Herbst, Marshall and Wingender, 1996). This example of Exxon Mobil would imply that we will not find a significant effect on the stock prices in the long term. Nevertheless every catastrophe is different and the subject companies react to it in contrasting ways.

2.3 Factors that explain the impact of accidents

One aspect that seems to be an important factor in the market reaction to a chemical disaster is the environmental disclosure of a company. According to

Blacconiere and Patten (1994) firms with extensive environmental disclosure experience a less negative reaction in stock price fluctuations after a disaster. This is an indication that investors value information that firms disclose about their operations even if this knowledge does not seem to benefit the firm in the first place. Besides that, a positive market reaction is expected for firms that reduce the emissions of toxic chemicals (Konar

6

and Cohen, 2001). During an accident with toxic release and or fatalities the market reaction in the long run seem to be significantly different from accidents without toxic release or fatalities. Cappelle-Blancard and Laguna (2010) found this evidence despite the fact their research mainly focused on the short term abnormal returns. Since 1989 the United States Environmental Protection Agency publishes a yearly toxic release

inventory (TRI). There have been a few studies on the relation between the stock market and TRI. Hamilton (1995) states that investors in firms that report TRI pollution figures experience a significant abnormal return after the release of the report. This could also imply that during this study a significant drop in stock price after a disaster is more likely to occur than during the Exxon Valdez accident mentioned earlier, when this TRI report was not published yet. Another study moreover implies that although there is a negative impact on the stock price, the generated toxic release of the relevant firm does not reduce significantly (Madhu Khanna, Wilma Rose H. Quimio and Dora Bojilova, 1998).

3. Hypothesis

The research question that will be tested during this study is: What is the long-term effect of an industrial disaster on the stock price of the company responsible for this disaster? That results in the following hypothesis to be tested:

H0: There is no significant long-term effect on the stock price of the responsible company after an industrial disaster

H1: There is a significant long-term effect on the stock price of the responsible company after an industrial disaster

This will be tested using the stock price data of ten different firm responsible for ten different disaster of the last fifteen years. The approach that is taken during this research will be further explained in the following paragraphs, explaining the methodology.

4. Data and Methodology

4.1 Data limitations

The data selection for the analysis to be performed is based on data from companies selected by several criteria. First of all, a sample was needed of industrial firms that had all been involved in at least one industrial disaster over the last fifteen years. In previous

7

paragraphs several definitions of a disaster has been mentioned concluding with a clear explanation of what is understood by an industrial disaster. In this research some extra restrictions are added to this definition. This is done so that proper research can be done on the effect of a disaster on stock price fluctuations. First condition in order to do so is to do research on companies with enough available stock price information. The

databases that were referred to investigating the data were Datastream, from which we collected the most returns of the several companies and CRSP database was examined to gather the rest of the stock return data. In order to be able to use the three factor model, which will be discussed later on, the Fama and French three factor model data has been used too. The companies and type of data that has been collected will also be discussed later on, but first the other general restrictions to the data will be further explained. Starting the investigation there has been looked at a lot of industrial disasters over the last couple of decades to get an idea of what industries have been concerned with the highest rate of industrial disasters over the years. No exact numbers were available, but what stood out were the numerous oil industry accidents over the years. This led to the aim to search for industrial disasters, caused by listed companies preferably active in the oil industry. As last, to specify the subject a little bit more and be able to draw

conclusions that indeed have an effect on the modern economic climate, the timeframe is set at 2000 to 2015. This recent timetable of fifteen years was chosen because there has not been a lot of research on the subject in this period yet, which would make it more interesting to compare the results to the studies that has been done over different time periods.

4.2 Disaster selection

Starting the research a lot of desk research has been done in order to find disaster data that could be useful for the data analysis. The EM-DAT database among others has been examined to find the most appealing disasters of the last decades. Since no company names were publicly available at EM-DAT the companies have been determined through searching the internet by disaster date. The 2010 BP deep water horizon oil spill served as a benchmark in selecting disasters considering different aspects of the event in terms of size, total deaths, total effected and damage in dollars. After identifying the companies that were responsible for the most suitable disasters the next selection restrictions were considered. First of all the home country of the company was determined, then the ownership structure was explored to the point where it was clear if the association was

8

listed one of its countries stock exchange markets. This is where a lot of potential interesting research material was lost. A lot of disasters that happened in the business area analyzed appeared to be caused by government owned companies, non-listed companies or bankrupt companies with not enough stock return data available.

Subsequently completing the list of companies considered to be appropriate, the search for applicable equity information began. Datastream and CRSP were both analyzed and fortunately for most selected companies the equity details were available. The rest of the essential information was obtained by data library of Kenneth R. French, which is publicly available. Almost every month Kenneth French and Eugene Fama update this library with new calculations for the factor variables used in the economic models they designed over the years, one of them is the three factor model which will be adopted in this premise.

This disaster selection resulted in a series of ten disasters over the last fifteen years. The first catastrophe that was chosen could not have been left out because of its gigantic impact. The deep water horizon oil spill in the Gulf of Mexico is the second largest oil spill in the world (Mason, 2012). British Petroleum took full responsibility for the accident where eleven people were killed and an amount about 4.9 million barrels of oil were spilled into the sea (Mason, 2012). As mentioned above this accident was taken as a benchmark for the rest of the disasters selected although it was impossible to find events of the same magnitude as this one.

The companies that were most suitable in this research were: Exxon Mobil, Repsol, Total, Shell, PetroChina, Chevron, Petrobras, Statoil, American Airlines. Nine out of ten companies selected are oil companies of which eight belong to the fifteen biggest oil firms in the world in terms of market capitalization (Ross, 2012), all of them were responsible for at least one major accident over the last fifteen years. The other corporation is a valuable variable in answering the research question because it incurred a disaster of comparable encounter in terms of firm and public harm. American Airline flight 587 crashed on November 12th_{, 2001 right after takeoff. There were two hundred} and sixty people on board, they all died, there were also five people on the ground that did not survive the accident (Raju, Glaessgen, Mason, Krishnamurthy and Davila, 2007). Moreover both firms are publicly traded firms in the United States. This makes it possible to analyze the daily stock returns of these companies instead of the monthly return. For publicly traded companies outside the United States this is not possible

9

because Fama and French only generate monthly return factors for exchange markets outside the U.S. Consequently the returns that are obtained from the Datastream and CRSP databases are the monthly returns from five year before the disaster and five years after the disaster plus a constant for the non-U.S. companies being: Repsol, Total, Shell, Petrobas, Statoil, BP and PetroChina. Additionaly daily returns are collected for the four remaining firms that are listed on a United States stock exchange market.

4.3 Model

The Capital Asset Pricing Model (CAPM) of Sharpe (1964) and Lintner (1965) is a famous asset pricing model which tries to explain the relation between risk and return based on simplified assumptions about investors behavior (Womack and Zhang, 2003). Years after that well known researchers in the field Eugene Fama and Kenneth French developed the Three factor model (1993) used in this paper. It could be stated as an expanded CAPM where Fama and French discovered apart from the market risk two other factors helped significantly explain realized returns on publicly traded stocks (Womack and Zhang, 2003). These components are the size and value of the firm, hence these are controlled for in the three factor model. The attitude towards the model among economists is not always laudatory although the main tendency towards it is that it is has at least some extra explanatory power compared to the CAPM. Gaunt (2004) is one of these followers who tested the model on the Australian stock returns, also Al-Mwalla and Karasneh (2011) who also compared their results with those of the CAPM concluded that it explains more. Faff (2004) who also analyzed the Australian stock market is not as convinced of the potential of the model, but also does not claim the contrary.

In this paper several dummy variables will be added to the equation also in order to find a potential relationship between the return on stock and the applicable disaster. For the first set of regressions generated, a dummy variable was used that took value 0 for returns corresponding with a date before the disaster and value 1 for return

corresponding with a date after the disaster. The second set of regressions was done using the model with year dummies added. The third regression is added to see if there at least is an effect on the short-term if the long-term effect cannot be found. The two

10 Regression 1: 𝑅_𝑖,𝑡− 𝑅_𝑓,𝑡 = 𝛼 + 𝛽₁(𝑅_𝑚,𝑡 − 𝑅_𝑓,𝑡) + 𝛽₂𝑆𝑀𝐵𝑡 + 𝛽₃𝐻𝑀𝐿_𝑡+ 𝛽₄𝐷𝑑𝑖𝑠 + 𝜀_𝑖,𝑡 Regression 2: 𝑅𝑖, 𝑡 − 𝑅𝑓, 𝑡 = 𝛼 + 𝛽₁(𝑅𝑚, 𝑡 − 𝑅𝑓, 𝑡) + 𝛽₂𝑆𝑀𝐵𝑡 + 𝛽₃𝐻𝑀𝐿𝑡 + 𝛽₄𝐷𝑌𝑒𝑎𝑟(−4) + 𝛽5𝐷𝑌𝑒𝑎𝑟(−3) + 𝛽6𝐷𝑌𝑒𝑎𝑟(−2) + 𝛽7𝐷𝑌𝑒𝑎𝑟(−1) + 𝛽8𝐷𝑌𝑒𝑎𝑟(0) + 𝛽9𝐷𝑌𝑒𝑎𝑟(1) + 𝛽10𝐷𝑌𝑒𝑎𝑟(2) + 𝛽11𝐷𝑌𝑒𝑎𝑟(3) + 𝛽12𝐷𝑌𝑒𝑎𝑟(4) + 𝛽₁₃𝐷𝑌𝑒𝑎𝑟(5) + 𝜀𝑖, 𝑡 Regression 3: 𝑅_𝑖,𝑡 − 𝑅_𝑓,𝑡 = 𝛼 + 𝛽₁(𝑅_𝑚,𝑡− 𝑅_𝑓,𝑡) + 𝛽₂𝑆𝑀𝐵𝑡 + 𝛽₃𝐻𝑀𝐿_𝑡+ 𝛽₄𝐷𝑑𝑖𝑠𝑚𝑜𝑛𝑡ℎ + 𝜀_𝑖,𝑡 Meaning the following:

- 𝑡  point in time

- 𝑖  company

- 𝑅_𝑖,𝑡− 𝑅_𝑓,𝑡  the stock return minus the risk free rate. This is known as the risk premium

- α  This is the constant in this multiple regression model

- (𝑅_𝑚,𝑡 − 𝑅_𝑓,𝑡)  The market risk premium with 𝛽1 as its coefficient - 𝑆𝑀𝐵𝑡  Small Minus Big is the size premium with 𝛽2 as its coefficient - 𝐻𝑀𝐿𝑡  High Minus Low is the value premium with 𝛽3 as its coefficient - 𝐷𝑑𝑖𝑠  Dummy for disaster (as explained earlier) with 𝛽4 as its coefficient - 𝐷𝑌𝑒𝑎𝑟(𝑗)  Dummy for years before, during and after the disaster with

coefficient (𝛽₄, 𝛽₁₄), with Year0 is disaster year.

- 𝐷𝑑𝑖𝑠𝑚𝑜𝑛𝑡ℎ Dummy for the month that the disaster happened

In the second set of regressions another set of dummy variables was generated in order to make a distinction between every year.

5. Results

In this part the results of the regressions that are done using the method explained above will be presented and analyzed.

11

5.1 Analysis

The first set of regressions that were performed consisted of ten separate regressions carried out in STATA. Regressions were done on the monthly return data of every firm, controlling for the monthly Fama and French factors and the Disaster dummy to make a distinction between before and after the disaster. Results are shown below:

Table 1, regression 1. Effect of disaster dummy on ab-normal returns

BP Exxon Mobil Repsol Total Statoil

Constant -.2136679 (.7269888) -.1003963 (.6477089) -.230244 (.842492) 1.051287 (.7805131) -.457676 (.6360307) Market premium .6305027*** (.1130637) .6026793*** (.124438) .6786275*** (.1447274) .4998058*** (.1454054) .5242966*** (.1240823) SMB -.5766409** (.2787904) -.0285694 (.1269472) .0573302 (.2502902) .0156884 (.2441962) .2097925 (.295857) HML .2878248 (.2953294) .4566404 (.1580838) .1936406 (.2276489) -.0418138 (.220807) -.0432789 (.2649344) Disaster dummy -.7802531 (1.07681) .8120514 (.9398805) -.4232634 (1.244565) -.9858754 (1.205483) .1679194 (1.327238) 𝑹𝟐 _{0.3583 0.1920 0.1670 0.1153 0.2919} Obs. 120 120 120 120 76

Table 1 (continued) regression 1: Effect of disaster dummy on ab-normal returns Shell Chevron American Airlines PetroChina Petrobras Constant -.6443698 (.623399) .3911684 (.5891924) -4.957662** (2.409588) 3.823758** (1.620784) -.4004826 (1.550792) Market premium .5580033*** (.094105) 1.004685*** (.1193573) 3.02281*** (.434719) .950812*** (.2216319) .9369784*** (.2812648) SMB -.3136337 (.2384018) -.5218509** (.2369671) 2.170995** (.5802663) .0662092 (.3987033) -.0653534 (.4597683) HML -.2996658 (.2685258) .0365181 (.2130003) 2.170995*** (.5802663) -.887077** (.3663269) .3544314 (.443539) Disaster dummy .6759136 (1.041411) -2.071351* (1.052307) 1.959609 (3.330353) -3.53734* (2.122481) .5253031 (2.367226) 𝑹𝟐 _{0.2661 0.4685 0.3164 0.2487 0.0967} Obs. 132 99 120 94 120

Significance indication: *statistically significant at a 10% level , ** statistically significant at 5% level and ***statistically significant at a 1% level

12

For most companies that we analyzed not a lot of significant alpha’s were found. Most of the companies only generate a significant coefficient for the market risk premium. It could have been expected that at least this risk premium would be significant since there almost always is a relation between the market movement and the movement of separate companies within that market, especially in this industry with cyclical companies.

Looking at the results of the other control variables, of which two are part of the Fama & French Three Factor Model and one indicator variable that was added later, we see that the SMB factor only for three companies has a significant influence on the stock price, according to this model. For the HML variable it is even worse, only to companies find significant results. Last variable that was tested in this model also is the most important one. The Disaster Dummy is the one variable that could show whether there is a

significant dissimilarity in stock return before and after the disasters. As can be seen in the table above for only two of the companies that attended there has been found a 10% significant relationship between the disaster and the stock price of the company over the next five years compared to the five years before the accident. These are not very

convincing numbers, therefor they will be discussed in the discussion later in the paper. To make get a clearer distinction between the difference in return per year, ten more regressions were performed by STATA to see if there would be a significant between the disaster and the return, at least in the short run. This time indicator variables were used for every year. The aim was to measure for ten years, five years before and five years after the disaster, but sometimes only eight or nine years were measured depending on the situation. In this case also monthly data was used in order to get results of the full sample. Results are shown on the next pages.

13 Table 2, regression 2: effect of year dummies on ab-normal returns.

BP Exxon Mobil Repsol Total Statoil

Constant .9250557 (1.71054) .5909347 (1.546646) -.2141692 (1.958867) .9690865 (1.831053) -.5988752 (1.496717) Risk premium .6847565*** (.122892) .6148105*** (.1310303) .6496906*** (.1533345) .5128464*** (.1522858) .5355583*** (.1293831) SMB -.4540661 (.2993973) .0314312 (.1407402) .001586 (.2648629) -.0181491 (.2500822) .3136883 (.3122097) HML .2584793 (.3079028) .5365057 (.179851) .3438675 (.2509645) -.0183073 (.2393817) .0171967 (.2702784) Year(-5) -3.550703 (2.407633) -.1072146 (2.155354) 2.694475 (2.799878) 1.671039 (2.589673) -.426183 (1.987729) Year(-4) -1.455613 (2.433219) -.9360017 (2.262454) -2.481993 (2.819069) -3.443783 (2.587353) 2.210631 (2.07528) Year(-3) .5549058 (2.5013) -1.76033 (2.252622) .0709718 (2.837179) 1.575237 (2.611149) -1.195668 (1.97951) Year(-2) -1.515522 (2.427626) -1.081831 (2.211049) 0.5679123 (2.682629) .7482794 (2.653351) -.2729537 (1.977034) Year(-1) -1.678341 (2.031134) -.9593533 (2.216826) 1.2470103 (2.763920) -.071976 (2.659777) -.0180403 (2.030918) Year(0) -3.102884 (2.423733) .0413568 (2.180819) -2.020061 (2.821856) -1.807038 (2.631241) 1.660131 (2.55043) Year(1) -.3862618 (2.479941) .0413568 (2.180819) -.5546312 (2.816636) -1.904874 (2.630957) 0.979091 (1.86354) Year(2) -2.777133 (2.414284) -.4423894 (2.166202) .5377032 (2.777145) -1.221095 (2.592935) 1.086573 (2.13456) Year(3) -1.204465 (2.406025) .5376501 (2.18067) 1.442162 (2.762683) .7986225 (2.582766) Year(4) -1.949128 (2.42593) 1.213624 (2.139224) -1.495026 (2.776397) 𝑹𝟐 _{0.3851 0.2068 0.2079 0.1689 0.3309} Observations 120 120 120 120 76

14 Table 2 (continued), regression 2: effect of year dummies on ab-normal returns.

Shell Chevron American

Airlines PetroChina Petrobras Constant -1.385796 (1.751603) 1.060929 (1.565614) -3.419944 (3.530567) 3.699779 (6.444357) 2.264329 (4.070913) Risk premium .5446849*** (.1509835) 1.100081*** (.1266206) 2.98855*** (.6603157) .859748*** (.2882586) 1.031367*** (.3312099) SMB -.3254403 (.299959) -.579648 (.2816611) 1.103764* (.6174909) -.0197172 (.429715) -.0371157 (.511148) HML -.2231931 (.2891498) -.0283154 (.2566116) 2.325884*** (.5106147) -.9022164 (.4723989) .4233642 (.5717267) Year(-4) .1296465 (2.357493) 1.27181 (2.345143) -4.440694 (6.615968) -1.446813 (6.715642) .3711043 (4.707655) Year(-3) .8019297 (2.113169) -2.856811 (2.522399) 3.778226 (4.1532) -.9746496 (6.493394) -1.365192 (4.791831) Year(-2) 2.731829 (2.560013) -.6286201 (2.03978) -4.314289 (7.817689) 3.03562 (6.949083) -8.242968 (5.45359) Year(-1) .2949459 (2.085532) -.2572851 (1.808092) -3.031799 (4.623648) -1.26371 (6.57463) -2.607142 (4.990343) Year(0) .9663861 (2.35641) -2.32772 (2.016173) -3.905397 (7.423156) -5.452454 (6.335956) -4.616315 (5.759407) Year(1) -.0647427 (3.029251) -2.396939 (1.745391) -.4480349 (10.16076) -1.051513 (6.510381) -.9038115 (6.138072) Year(2) 1.090212 (2.308102) -3.339539 (2.069431) -2.818779 (5.159714) -1.221451 (6.506196) 1.439621 (6.604325) Year(3) 2.606448 (2.19055) -3.509946 (2.681699) 6.739824 (6.32139) -3.918259 (7.481425) -3.141103 (4.89106) Year(4) 2.341431 (2.204159) .6615458 (4.587075) -12.13584 (7.245565) -5.189126 (5.512783) Year(5) -.2007129 (1.987242) 19.08651*** (3.226684) 7.027763 (4.547198) 𝑹𝟐 _{0.2905 0.4984 0.3479 0.2968 0.01487} Observations 132 99 120 94 120

Significance indication: *statistically significant at a 10% level , ** statistically significant at 5% level and ***statistically significant at a 1% level

The result show that for every company the market risk premium is a significant control variable which, as stated before, could be expected. Furthermore we do not see a lot of significant results which would presume that the control variables do not have an effect on the stock return of any company and also there is no disaster effect measured. Noticeable is that the 𝑅2 _{in all the regressions performed are very low. This tells us that} the explanatory power of this model is very small.

15

As mentioned above the daily stock return of the firm listed in the U.S. were meant to be analyzed too. The results gathered from this analysis were not significant at all and therefore is not taken into further consideration for the rest of the research.

So far not a lot of significant results are found. To see at least some effect of the disaster can be measured the following regression was done. In this regression only a dummy was added for the month in which the disaster happened.

Table 3, regression 3: effect of dummy disaster month on ab-normal returns

BP Exxon mobil Repsol Total Statoil

Constant -.5266661 (.5740502) .2714666 .482424 -.4861331 .6753669 .785463 .7129874 -.3300025 .535745 Market premium .6160547*** (.1857534) .6233784*** .1507006 .6685907*** .1709725 .4189627** .194407 .4958114*** .1207267 SMB -.5242998 (.3179002) -.0273044 .1318444 .0166368 .3050338 -.1149248 .2593742 .1935229 .307967 HML .3266481 (.2943671) .4716225** .2013278 .2034473 .2446398 -.0792221 .2041272 -.0255645 .2293878 Dummy for disaster month -5.566669*** (.9561635) .9795902 .8170512 9.17686*** .9689445 -7.53551** 3.287885 -3.94363*** 1.418781 𝑹𝟐 _{0.3603 0.1870 0.1798 0.1195 0.2982} Obs. 120 120 120 120 76

Significance indication: *statistically significant at a 10% level , ** statistically significant at 5% level and ***statistically significant at a 1% level

Table 3, regression 3 (continued): effect of dummy disaster month on ab-normal returns

Shell Chevron American

Airlines PetroChina Petrobras Constant -.3561463 .5099943 -.2395161 .5216684 -3.953104** 1.815669 1.445179 1.184126 .0710837** 1.198455 Market premium .5755914*** .1456528 .9735243*** .1406015 3.056007*** .6343057 .8558097*** .2799298 .885934 .3443508 SMB -.3239909 .3063307 -.4878127 .267334 1.066032 .5643939 .1341421 .3287896 .1230175 .4644134 HML -.3503485 .2946552 .0337261 .2522335 2.219147*** .5136272 -.6293201 .4141862 .2987584 .4460275 Dummy for disaster month -6.138133*** .7615029 -.4340831 .551385 -5.277032 4.89638 35.62478*** 1.733542 -19.9341*** 4.57409 𝑹𝟐 _{0.2703 0.4467 0.3148 0.3269 0.1160} Obs. 133 99 120 94 120

Significance indication: *statistically significant at a 10% level , ** statistically significant at 5% level and ***statistically significant at a 1% level

16

As shown in the table above this time some disaster effect was measured. This time seven out of ten dummy coefficients are significant on a 5% or 1% level. From this can be concluded that there at least is an effect of the disaster in the short-run.

5.2 Discussion

Analyzing the results of the regressions that were done in order to answer the research question, some notable comments have to made. As explained in the previous

paragraphs and shown in the tables above the results that were found related to the long-run are not significant. This does not necessarily mean that no scientific conclusion could be drawn from these results, but it a critical note should be given.

The biggest shortcoming of this research is the amount of data that has been used. The sample of ten companies can be considered reasonably small, but especially the fact that only monthly data has been used could be a reason for the insignificant results. Stock prices are volatile and change daily because of a lot of different factors. Using the monthly returns makes that part of the daily effects are neutralized because of averaging. Analyzing the daily stock prices of these companies could be valuable. In this research there were some difficulties with matching the Three Factor Model factors from the Fama and French databases, with the daily stock returns and therefore no useful results were found. This will be left for further research.

If looking at the tables above there are, apart from the results itself, some other aspects that could be noticed. In the first place not all firms have been analyzed for the same amount of years. The aim was to analyze five years previous to the disaster and five years after the disaster, but not in all cases this amount of data was available. A couple of times the reason for this was that the concerned disaster happened after the year 2010. This obviously made it impossible to measure five year after the disaster. Accidents with this shortcoming are still chosen to be analyzed, even while the aim the paper is to catch the long term effect of the disaster, because they still were considered to be relevant. In most of these cases the years missing would only be one or two which is surmountable in a ten year time frame. Second this research tried to pick a sample within a relatively small time frame, with disasters that were very similar to each other. This made it a very specific sample with not a lot of choice in different sample candidates.

17

6. Conclusion

The objective of this paper was to analyze the stock returns of several companies that incurred an industrial disaster in the last fifteen year, to see if there is a significant relationship between an industrial disaster and long-term stock price fluctuations. In order to do so a sample of ten companies was selected, aiming for companies from the same industry that experienced similar disasters. Most of the companies were oil companies belonging to the biggest in the world.

Ten year monthly stock return data was selected and the factors data for the Three Factor Model was collected. Before analyzing the data existing literature was reviewed. The theory on this subject mostly considered the short-term return after a disaster, accordingly there seems to be a significant decline in stock price in the short-run. In the long-run whatsoever not a lot of effect has been found, although some studies seem to find some effect. Hence the expectation for this research was not necessarily to find a significant drop, but to discover whether there would be some effect. Research was done by regressing the Three Factor Model Factors together with several dummy variables for year specification and before and after the disaster on the stock return data. The results came out surprisingly. No disaster effect was found on the stock price over the long-run, but this could have been expected from the theory. To see if there can be at least found an effect the short-term effects were measured as well. This resulted in mostly significant results which tells us that there is an effect in the short-run, just like expected from the literature.

To improve the study more firms should have been included and where possible daily data should be analyzed to get more reliable results, also more comprehensive models could be a valuable addition to the study. In this paper there was not enough time to further get into this and therefor this is left for further research.

Concluding, the results of this research suggest that there is no significant long-term effect of an industrial disaster on the stock return of the responsible company, however the amount of data used for this study is limited. Together with the sample size, the data analyzed was not a detailed enough representation of the real fluctuations in the stock price. The limited time period that has been looked at should in further research be expanded to get a good representation of the disaster population. In this research a lot of time has been spent finding suitable disasters. Through EM-DAT database only the disaster date could be gathered. After that searching the internet for that disaster date did

18

not always give a result. This tells us that there are a lot of disasters that happen over the world where no information is published about. It was hard to find enough suitable accidents that happened over the last fifteen years. Further research, preferably on a longer time period, more disasters and daily data is needed to be able to answer the question whether long-term stock prices are influenced by industrial disasters.

19

7. Bibliography

Al-Mwalla, M., & Karasneh, M. (2011). Fama and french three factor model: Evidence from emerging market. European Journal of Economics, Finance and Administrative Sciences, 41, 132-140.

Blacconiere, W. G., & Patten, D. M. (1994). Environmental disclosures, regulatory costs, and changes in firm value. Journal of Accounting and Economics, 18(3), 357-377. Capelle-Blancard, G., & Laguna, M. (2010). How does the stock market respond to

chemical disasters? Journal of Environmental Economics and Management, 59(2), 192-205.

Coleman, L. (2006). Frequency of man‐made disasters in the 20th century. Journal of Contingencies and Crisis Management, 14(1), 3-11.

Connor, G. (1995). The three types of factor models: A comparison of their explanatory power. Financial Analysts Journal, 51(3), 42-46.

Gaunt, C. (2004). Size and book to market effects and the fama french three factor asset pricing model: Evidence from the australian stockmarket. Accounting & Finance, 44(1), 27-44.

Hamilton, J. T. (1995). Pollution as news: Media and stock market reactions to the toxics release inventory data. Journal of Environmental Economics and Management, 28(1), 98-113.

Herbst, A. F., Marshall, J. F., & Wingender, J. (1996). An analysis of the stock market's response to the exxon valdez disaster. Global Finance Journal, 7(1), 101-114.

Jasanoff, S. (1988). The bhopal disaster and the right to know. Social Science & Medicine, 27(10), 1113-1123.

Jones, K., & Rubin, P. H. (1999). Effects of harmful environmental events on reputations of firms. Available at SSRN 158849,

Khanna, M., Quimio, W. R. H., & Bojilova, D. (1998). Toxics release information: A policy tool for environmental protection. Journal of Environmental Economics and Management, 36(3), 243-266.

Klassen, R. D., & McLaughlin, C. P. (1996). The impact of environmental management on firm performance. Management Science, 42(8), 1199-1214.

Konar, S., & Cohen, M. A. (2001). Does the market value environmental performance? Review of Economics and Statistics, 83(2), 281-289.

Mason, O. U. (2012). New insights into microbial responses to oil spills from the deepwater horizon incident. SIMS News, 16, 60-65, 2011,

20

Parker, D., & Handmer, J. (2013). Hazard management and emergency planning: Perspectives in britain Routledge.

Raju, I., Glaessgen, E., Mason, B., Krishnamurthy, T., & Davila, C. (2007). Structural analysis of the right rear lug of american airlines flight 587. Computer Modeling in Engineering and Sciences, 22(1), 1.

Rao, S. M. (1996). The effect of published reports of environmental pollution on stock prices. Journal of Financial and Strategic Decisions, 9(1), 25-32.

Ross, M. (2012). The oil curse: How petroleum wealth shapes the development of nations Princeton University Press.

Rutherford, W., & De Boer, J. (1983). The definition and classification of disasters. Injury, 15(1), 10-12.

Salinger, M. (1992). Value event studies. The Review of Economics and Statistics, , 671-677. Salinger, M. (1992). Value event studies. The Review of Economics and Statistics, , 671-677. Shaluf, I. M., Ahmadun, F., & Mat Said, A. (2003). A review of disaster and crisis.

Disaster Prevention and Management: An International Journal, 12(1), 24-32. Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under

conditions of risk*. The Journal of Finance, 19(3), 425-442.

Vorderbrueggen, J. B. (2011). Imperial sugar refinery combustible dust explosion investigation. Process Safety Progress, 30(1), 66-81.

Womack, K. L., & Zhang, Y. (2003). Understanding risk and return, the CAPM, and the fama-french three-factor model. Tuck Case, (03-111)