The impact of Terrorism and National Security Levels on
Stock Markets in Western Europe
Supervisor: dr. E. Karmaziene Student: Wim Mateboer s2058162 Words: 12167
Abstract
The objective of this paper is to analyze terrorism in Western Europe after 11 September 2001 on both national and industry level by event study methodology. Furthermore it is examined if changes in national security levels take away part of the uncertainty and as a result, take away part of the price effect after a terrorist attack. Lastly, these questions are answered by using a non-parametric test, for robustness purposes. Event study methodology does not give significant results, where the non-parametric test shows evidence that
Introduction
This papers explores the impact of terrorist attacks and changes in the terrorist threat level on domestic stock markets and industries. This is relevant since Markowitz (1952) shifted the focus from solely returns to the combination of risks and returns while taking correlations between stocks into consideration. If terrorism indeed impacts stock returns, it is important to have a portfolio that is not over weighted in countries with a high risk on terrorism or industries that are more sensitive to terrorism. In prior research, the effect of terrorism is measured mainly in the volatility of markets and industries. In this paper, the abnormal returns for national markets and industries are measured. Since stock value is determined by all relevant information that is publicly available to the market, not only the terrorist attack itself is relevant in this analysis. Also the changes in national threat level, announced by governments, are relevant for the impact of terrorism on stock markets. Therefore, abnormal returns are measured for the days on which governments announce a change in the national threat level. If there is indeed an effect, part of the information about a terrorist attack is already incorporated in the market at the moment the attack occurs. This is
important, since it might take away a part of the effect when a terrorist attack actually occurs. By using event study methodology, we will try to get a better understanding about the impact of terrorism on stock markets. The main critique on the event study methodology is the assumptions that have to be made regarding the statistical test. The nature of this test is parametric and the data is not always parametric. Therefore the data is tested for
normality and an additional non parametric test, the Wilcoxon signed rank test, is carried out as well. Not only is the impact on national level analyzed, also the impact on individual industries is considered. There are signals in the literature that some industries are more sensitive to terrorism than others. The method applied to industries is the same as for national stock markets. Both a parametric and a non-parametric test are carried out, after the normality of the data is analyzed. The main contribution of this paper is that not only the terrorist attack itself, but also the changes in the national threat level are considered for measuring the impact of terrorism.
Within ten days after a terrorist attack, the impact of terrorism on financial markets is not found on both the national and the industry level by using event study methodology.
Although there are sporadic results(<2% of the observations), there is no structure in those results. Therefore we assume those significant outcomes are a result of exogenous factors. The explanation for finding only significant results in a non-parametric test, where a
parametric test does not show significance, is due to the fact that financial data is often not normally distributed. Especially in both tails of the distribution is a deviation from normality observable. Therefore an abnormal return has to be more extreme in a parametric test, compared to a non-parametric test.
Literature
The attention for research on the impact of terrorist attacks on the stock market increased drastically after Al-Qaeda targeted the Twin Towers in New York on 9/11. These attacks had a huge impact on stock markets. If papers mentioned terrorism in Finance related papers before 2001, it was often combined with war or political instability and uncertainty. One exception is a paper by Abadie and Gardeazabal. (2001) Their paper is published in September 2001, in the same month 9/11 happened. This added significantly to the
relevance of the paper when it was published. The 9/11 attacks, which were solely aimed on the USA, caused negative stock reactions all across the world. (Chen & Siems, 2004) Despite the tragic nature of the incidence, it was an interesting case to study the impact on stock returns for several reasons. It was the first huge act of terrorism in the USA in recent history. It was aimed at the financial center of the world. The availability of financial data in the USA is high. Therefore, most studies that analyze the impact of terrorism on the stock market focus on this attack, and those studies gained consequently more awareness for stock market behavior after terrorist attacks. (Karolyi, 2006)
It is relevant to find out what the impact of terror is on the stock market, since if this impact is not taken into account, investors possibly construct portfolios that are overly sensitive to terrorism. Furthermore, there is the risk that foreign investors will withdraw their
investments after a terrorist attack and move to a country with more stability. If a government understands the impact of terrorism on the stock market better, it can take more appropriate measures and keep investors satisfied. A way to balance portfolios correctly regarding terrorism risk is investing in the defense and security related industries. Berrebi and Klor (2010) found that a terrorist attack has a positive impact on those
industries, while the impact on the other industries was negative or insignificant. By
investing in the defense and security related industries, terrorism risk will be partly hedged away.
Chesney, Karaman and Reshetar analyzed in 2011 77 terrorist attacks across 25 countries over 11 years. They find that the insurance sector and the airline sector are the most vulnerable to terrorism. The banking sector is the least sensitive. They also conclude that a significant reaction is measurable only in the short term. In general, after a few days the market returns to the equilibrium again, indicating that investors do not believe that the attack has an enduring effect on the financial market. A third point they make is that the target of an attack is essential. The targeted industries experience more sensitivity.
market is low or negative. Furthermore, an adequate response by the authorities effects the recovery of the market in a positive way.
Karolyi and Martell (2005) are the first who compare multiple terrorist attacks with distinguished characteristics all across the world. Their focus is on individual firms. One of their expectations was that firms in wealthier countries would suffer more from terrorism than firms in poorer, less educated and less democratic countries, since prior research indicated that terrorism occurs more often in poor, unstable countries. Hence, the increased exposure to terrorism risk should already be more incorporated in the price. Their findings confirm this expectation.
Most researchers find a negative impact of a terrorist attack, but stabilization after a short period of time. Eldor and Melnick (2004) investigated the impact of terrorism on Israel’s stock exchange. They find that since the amount of attacks in Israel increased significantly, (the year 2000) the market performs significantly worse than before the year 2000. This is in contradiction with most other research. A possible reason for this could be that the high frequency of attacks does not allow the market to fully recover. Eldor and Melnick test also the diminishing effect hypothesis. This diminishing effect would mean that investors take already into account that a certain number of attacks will happen and incorporate this in the market price. Eldor & Melnick conclude that there is no evidence for such an effect in Israel. What drives the price shock after terrorist attacks? The main reason is increased uncertainty. A terrorist aims to destabilize and create chaos. This chaos causes higher volatility and makes it more difficult to forecast the market. Johnston & Nedelescu (2005) differentiate between direct and indirect costs of terrorism. Direct costs are the loss of lives, and as a result human capital, damage to infrastructure and medical healthcare for those affected. These costs are in general relatively low compared to indirect costs. Examples of indirect costs are the fall of investor confidence, consumers who tend to spend less when they feel unsafe and higher insurance costs. The government can influence especially these indirect costs by reacting adequately to the situation and regaining control.
government, facilities etc.) and place and time of occurrence.’’ When certain possible targets are better protected, terrorists can decide to move on to another, more vulnerable target. This makes it more difficult, also from an economic point of view, to find correct measures to prevent those situations.
Arin, Ciferri & Spagnolo (2008) analyzed six different markets across the globe for mean returns and volatility. They find that both mean returns and volatility is affected. The volatility effect results in an extra negative impact on the mean returns. Interestingly, the two European countries in their research, the United Kingdom and Spain, are less sensitive to shocks from terrorism than the four other investigated countries outside Europe. A reason for this could be that the level of terrorist risk is lower in these countries.
It is expected that investments in countries with less terrorist risk are performing well compared to countries with more terrorist risk. A paper about this topic is from Abadie & Gardeazabal (2008). Since productive capital is mobile in an open market, they expect that countries with a lower terrorist risk attract more foreign investment. They conclude that terrorist risk is indeed a factor which attract less foreign investment. This result is robust when demographic factors, country specific risk, governance culture and several financial factors are taken into account.
In general, the literature about terrorist attacks is distinguishable in two groups. One group focuses on the firm level. The impact of terrorism on the firm level is often analyzed by event study methodology. The other group focuses on national or industry level. For this group the focus lies often on volatility analysis. An exception is the event that induced the rise of studies regarding the financial impact of terrorism, the attack on the Twin Towers in the USA. This attack, which attracted the most attention by researchers by far, brought a huge shock to the financial markets. It was an extraordinary attack in many aspects. It was totally unexpected. The first large Jihad based attack on American soil. It brought the realization that the involvement in foreign countries is not always without consequences. It was a tragedy for the families of the almost 3000 people that did not survive the attacks. And the fact that the attack was aimed on the Twin Towers in Manhattan, the financial center from the USA and arguably the world, brought a large shock to the financial markets. Therefore 9/11 is an event that is very well suited to be subject to an event study. According to the literature, terrorism should have a negative impact on national markets or at least
literature, is not extensively found in the empirical field. Therefore, this paper will focus on the national markets and industries, with an emphasis on event study methodology, to see if the impact on the market level is not only visible in volatility, but also in negative (C)AR’s. Those negative CAR’s can be expected, since the investment climate deteriorates due to increased uncertainty and the direct costs mentioned by Johnston & Nedelescu. (2005) Investors will signal this and relocate their investments to countries with more certainty. There is a possibility that the focus in the literature about the terrorism effect on industries and markets is not on event study methodology because in general there is no effect observable by using this method. If only volatility is measurable, but no significant negative Abnormal Return, the focus in the literature will be more on volatility, since this research is valuable to draw conclusions from. If that is indeed the case and no statistical significant abnormal returns are found, this paper will analyze if price adjustments are potentially already made when governments change their national terror level. It would make sense that governmental expectations influence investors. When the terror level in a country increases, this could have a negative impact on investors. The actual attack is then not an unexpected event, but the confirmation that the national terror level was not unnecessarily raised.
Data
The financial data will be retrieved from Datastream. The period over which the data is obtained is from October 2000 until November 2017. This will give us the opportunity to derive for each event an estimation window of 1 year and an event window of 10 days. The datatype is the price index (PI) since in the PI only price movements are incorporated while excluding dividends, rights offerings and other factors that influence the stock price are excluded. Subject to investigation are the country indexes from Switzerland, Spain, United Kingdom, Norway, France, Belgium and Germany, since these are the countries in which the events took place. Furthermore the country indexes from the Netherlands is obtained for comparison and to see if there are effects for a country where no terrorist attacks occurred. Also the FTSE industry indexes per country are part of the data, to check for industry effects. The reason for this is that the literature indicates that there is often not a country wide effect measurable, while there is a firm or industry specific impact. Additionally, the MSCI World Index is involved in the analysis as a benchmark for the national markets. The MSCI World Index consist of approximately 14000 securities all across the world and is one of the main used Indexes to replicate the global market. Like the MSCI World index is the
benchmark for national indexes, the national indexes are benchmarks for their respective national industries. Lastly, 3 Month Bond Yields of the involved countries are chosen as substitutes for the risk free rate. In general it can be expected that the payout on these bonds over such a short period is relatively certain.
organization. (Nationaal Coördinator Terrorisme Bestrijding en Veiligheid) The system is implemented in June 2005. All over Europe are different system in place and they are all implemented independent from each other. Therefore, comparisons of price effects from changing from a particular level to another are not possible. However, it is possible to determine the change from one level to another. When in the UK the terror level goes from Severe to Critical, this might be a more serious sign for foreign (and domestic) investors than the terrorist
attack in itself, after which the eventual impact is clear relatively soon after the attack. Still, the most massive attacks, like 9/11, would influence the domestic market or industry, but the ‘regular’ terrorist attacks might be already incorporated in the price at the moment that there is announced that the national terror threat level is adjusted. Norway’s system is slightly different. The PSA, Norway’s intelligence service, publishes an annual report with the current risk assessment. This assessment changed in format over the years and is currently split up in analyses about Islamic, left wing and right wing extremism. Furthermore the threats about dignitaries are assessed. There are six safety categories ranged from very unlikely to very likely.
The Global Terrorism Database will be consulted to filter for the events that meet the abovementioned requirements. In this database more than 170.000 cases worldwide over the period 1970-2016 are documented. The results of this query will be the events in the event study that follows. Since this database is annually updated, the information for 2017 is not available yet. Therefore the terrorist attacks in Manchester, London and Barcelona in 2017 are manually added.
out three times. In this paper is chosen to consider attacks on the same day in the same city as the same event. Furthermore, in all occasions the offenders were assumingly cooperating, so the attacks can be seen as one terroristic act.
Terrorist attacks in Western Europe 2001-2017
Event Date City Country Fatalities Target Type
1
27-09-2001 Zug Switzerland 14 Government
2
11-03-2004 Madrid Spain 191 Transportation
3
07-07-2005 London United Kingdom 41 Transportation 4
22-07-2011 Utoya Norway 69 Private Citizens & Property 5
13-11-2015 Paris France 127 Business, Private Citizens & Property 6
07-01-2015 Paris France 12 Journalists & Media, Police, Private Citizens & Property
7
22-03-2016 Brussels/Zaventem Belgium 35 Airports& Aircraft, Transportation, Private Citizens & Property 8
14-07-2016 Nice France 87 Private Citizens & Property 9
19-12-2016 Berlin Germany 12 Private Citizens & Property 10
22-05-2017 Manchester United Kingdom 23 Private Citizens & Property 11
03-06-2017 London United Kingdom 11 Government, Private Citizens & Property 12
17-08-2017 Barcelona Spain 16 Private Citizens & Property
Methodology
The purpose of this research is to analyze the impact of terrorist attacks in Europe. Within this field, the focus lies on terrorist attacks in Europe with at least 10 casualties that did not survive the attack. Small terrorist attacks in which there is a small number of casualties, ‘only’ a kidnapping case or other terrorist activity without further stretching have, based on prior research an effect at the firm level, but do not have influence on a whole national market or industry.
The following research questions will be addressed in this study:
Research question 1: Do terrorist attacks in Western Europe impact the domestic stock markets and industries according to event study methodology?
Research question 2: Do changes in National Security Levels in Western Europe impact the domestic stock markets and industries according to event study methodology?
Research question 3: Do the results for the above stated questions hold if the domestic markets and industries are analyzed with a non-parametric test?
A widespread method to measure abnormal returns for financial data is the event study methodology. This methodology is in place since the 1930s and developed since then. Fama et al. (1969) introduced the framework for the event studies currently used. In these studies the returns in the period of interest, the event window, are compared to the returns in another period, usually the period prior to the event window. This is the estimation window. By deriving a model based on the estimation window, the returns in the event window can be predicted. The discrepancy between the expected return and the actual return is the abnormal return. This abnormal return is subject to statistical analysis in order to find out if the difference between the expected return and the actual return is significant.
Despite the possibility that the data is not entirely normally distributed, in this paper is chosen for the use of daily data, since this keeps possibly useful information in the data intact, whereas weekly or monthly data can fail to measure very short term effects. In this paper the Estimation Window is determined on one year, which is equal to 261 data points. Estimation Windows of one or two years are most common in similar event studies,
especially since MacKinley published his paper in 1997 where he stated that longer
possibility is that the correlation between a country and the world economy changes. The Beta is not a fixed value over time. Therefore, an estimation window that is too extensive, has a higher chance on a biased Beta. In case the returns change over time or the returns from the benchmark, the Beta is not a proper estimation of the sensitivity. This makes it difficult for investors to predict future returns with the CAPM model or another model involving Beta. Furthermore, if there are differences between the Estimation Window and the Event Window in case of an event study, the expected returns in the Event Window will be biased. Therefore it is important to have an Estimation Window that is sufficiently large to contain useful information and is simultaneously not so large that historical sensitivity to the market is incorporated that is nowadays irrelevant. The Cumulative Abnormal Returns will be tested in different event windows. The event windows that will be used are just the Abnormal Return of day 0, [0] and furthermore the five and ten day event windows [0-4] and [0-9].
To interpret the data a model is required. Throughout the years several models are
developed, all with their strengths and weaknesses. Chosen is for the CAPM model, due to it’s being straightforward and having a strong basis in the literature. The CAPM model has weaknesses that has to be kept in mind. For example, the beta that is assumed to be constant in theory is not constant in practice. Furthermore it is not certain if the market portfolio is indeed representing the whole market. The general consensus is that the CAPM model is not a fully accurate model that is able to predict future prices with high certainty, but in the mean time it gives a useful insight in the way prices develop and is a valuable tool to analyze financial assets and markets. In formula form the CAPM model is formulated as follows:
𝑅𝑅𝑖𝑖 = 𝛼𝛼𝑖𝑖+ 𝛽𝛽𝑖𝑖𝑟𝑟𝑚𝑚𝑚𝑚+ 𝜀𝜀𝑖𝑖𝑚𝑚, 𝜀𝜀 = ~𝑁𝑁(0, 𝜎𝜎𝜀𝜀𝑖𝑖2) (1) In which Ri is the excess return of asset i, 𝛼𝛼𝑖𝑖 is a measure that indicates how the asset
The next step is to derive the abnormal return. This is the discrepancy between the return estimated by the model and the actual return. The estimated return is calculated by
multiplying the β of the national market index with the return of the MSCI World Index for a specific day and add the α to that. The variance for those error terms is derived from a well-known paper by MacKinlay. (1997) He analyzed the Event Study Methodology thoroughly and his paper is since then used and cited thousands of times. Since the CAPM model is estimated not necessarily at its true value, measuring the variance of the error term by simply taking the squared differences between the observations and their mean, divided by N, could result in a biased variance. Therefore the conditional variance of those abnormal returns is defined by:
𝜎𝜎�2(𝑟𝑟
𝑖𝑖𝑚𝑚𝑎𝑎) = 𝜎𝜎�𝜀𝜀𝑖𝑖2 +𝑁𝑁1(1 +(𝑟𝑟𝑚𝑚𝑚𝑚−𝜇𝜇𝑚𝑚)
2
𝜎𝜎�𝑚𝑚2 ) (2)
Another advantage of the conditional variance is that the conditional variance compensates for potential heteroscedasticity as well. 𝜎𝜎�𝜀𝜀𝑖𝑖2 is the variance of the error term in the CAPM model. The second term is a correction for the sampling error. Since we can not be absolutely sure if the model is estimated correctly, this term is added. N is the amount of observations in the estimation window. The more observations there are, the smaller the correction will be. In this research, with 261 observations in the estimation window, the correction term will be very small. 𝜇𝜇𝑚𝑚 and 𝜎𝜎�𝑚𝑚2 are respectively the mean and variance of the market in the estimation window. (MacKinley, 1997)
Most studies that analyze the impact of terrorism use the event study methodology for the impact of terrorism on individual firms and countries in war zones, while more stable countries and industries are analyzed by (E)GARCH models which measures volatility. This paper will analyze the impact of terrorism on the national and industry level by using event study methodology. Starting with event window [0], the statistical significance of the Abnormal Return will be determined by a t-test. The hypothesis that will be tested is:
𝐻𝐻0: 𝐶𝐶𝐶𝐶𝑅𝑅 = 0 𝐻𝐻1: 𝐶𝐶𝐶𝐶𝑅𝑅 ≠ 0
(3)
If the Abnormal Returns are not significantly different from zero, the model is able to predict the return in a sufficient way. However, if the Abnormal Returns differ from zero in a
different from the situation in the estimation window. The construction of the t-test is as follows:
𝑡𝑡 =𝐶𝐶𝐴𝐴𝐴𝐴𝐴𝐴𝑟𝑟𝐴𝐴𝐴𝐴𝐴𝐴 𝑅𝑅𝑅𝑅𝑡𝑡𝑅𝑅𝑟𝑟𝐴𝐴𝜎𝜎 𝑖𝑖
(4) After determining the degrees of freedom a p-value can be derived from the t-value. This will show the statistical significance of the abnormal return. The Cumulative Abnormal Return is the Abnormal Return over consecutive days. This indicates the effect of the event, in this case the terrorist attack, over an extended period. Daily returns that are not
statistically significant on itself might be significant together if the effect is spread out over a longer period. The CAR and the variance of the CAR can be specified by:
𝐶𝐶𝐶𝐶𝑅𝑅𝑖𝑖 = 𝛴𝛴𝐶𝐶𝑅𝑅𝑖𝑖 𝜎𝜎𝐶𝐶𝐶𝐶𝐶𝐶2 𝑖𝑖 = 𝛴𝛴𝜎𝜎�2(𝑟𝑟𝑖𝑖𝑚𝑚𝑎𝑎)
(5) Additionally, the above described procedure will be carried out for the different industries per country per event. As described in the literature section, some industries might be more sensitive to terrorism than others. Furthermore, the target of the attack can impact the industries in a different way as well. For example, a negative impact on Consumer Services, in which for example airlines and railway companies are incorporated, could be partly offset by a positive impact on the Insurance sector, which is part of the financial industry. In such a scenario the significant abnormal returns for two or more industries could cancel each other out.
the occurrence of a threat and the following de-escalation have opposite impact, a possible significant effect can be neutralized. This has to be taken into account if there are no significant results found for the CAR’s.
The issue with the normality of the data should still be addressed. Therefore, an additional robustness test will be carried out. One factor models like the CAPM model and multifactor models who are set up in a similar way are often used to carry out a t-test or another parametric test. However, the important condition of normality to use these tests is often not fulfilled. When data is not normally distributed, the estimated Alpha’s and Beta’s can deviate from their true value. In a comparison between an event study with a T-Test, a non-parametric test and a GARCH-EVT approach, Chesney, Reshetar & Karaman (2011) they recommend the non-parametric test since the GARCH-EVT approach is limited to the event day only and computationally intensive. Furthermore they mention the statistical limitations discussed earlier that are associated with a parametric test.
First Normal Probability Plots will be created in order to see if the data is approximately normally distributed. The CAPM model on which the event study is based, is constructed based on the estimation window. Therefore the Normal Probability Plot contains data only from the estimation window. To construct a Normal Probability Plot, all excess returns for both the MSCI World Index and the national indexes are sorted from small to large. Then, the mean and standard deviation of the data is taken. Based on the mean, the standard deviation and the amount of observations, it is possible to calculate by means of the inverse normal distribution the expected value of the observations if the data is perfectly normally distributed. The expected value is plotted against the Z-Value of this observation. Similarly, the value of the actual observations are plotted against the Z-Value. If the data is normally distributed, these two functions mainly overlap each other. However, if the actual returns appear above the expected value- curve, this indicates that the probability curve is skewed to the right. This means that more observations are above the average excess return than below it. Similarly, an actual value curve below the expected value curve indicates skewness to the left. A third possibility is that the actual value curve is an S-curve through the
In case the data is not normally distributed, the next step is to carry out a non-parametric test in addition to the T-Test. The non-parametric test of our choice will be the Wilcoxon Signed Rank Test. This test is chosen because it is a good addition to the T-Test in order to check for extreme deviations from the expected values. The Wilcoxon Signed Rank Test is designed to test paired data. CAR’s are single values and can not be paired. However, the CAR’s are based on market returns. Since we are comparing national indexes with world indexes and industry indexes with national indexes, the data is clearly paired. This test is preferred over a regular rank test, since the fact that the data is paired improves the power of the signed rank test. (Gibbons & Chakraborti, 2011) Three assumptions should hold in order to use the Wilcoxon Signed Rank Test. (Lowry, 2003) Firstly, the paired values should be randomly and independently drawn. Second, the dependent variable should be
intrinsically continuous, meaning that it is possible to estimate it’s value precisely. Thirdly, the paired values should have properties of an ordinary scale, which makes it possible to rank the observations.
The null hypothesis tested is:
𝐻𝐻0: 𝑅𝑅𝑚𝑚𝑚𝑚 = 𝑅𝑅𝑖𝑖𝑚𝑚 𝐻𝐻1: 𝑅𝑅𝑚𝑚𝑚𝑚≠ 𝑅𝑅𝑖𝑖𝑚𝑚
(6) In which 𝑅𝑅𝑚𝑚𝑚𝑚 is the excess return on the MSCI World Index and 𝑅𝑅𝑖𝑖𝑚𝑚 the excess return on the national market index. In order to test this hypothesis, the Wilcoxon-statistic has to be computed. The definition of this statistic is:
Therefore it is required to calculate the standard deviation. Since the he standard deviation of W is solely dependent on N, the number of observations, the formula for the standard deviation is:
𝜎𝜎𝑊𝑊= �𝑁𝑁(𝑁𝑁 + 1)(2𝑁𝑁 + 1)6
(8) The more observations available, the more reliable the test. Therefore only for the [0-9] Event Window the ranked sign test will be carried out. For N=10, the standard deviation of the test will in this case be 19.6. Combining this standard deviation with the W-Statistics for the different Events will allow us to calculate Z-Values and corresponding P-Values.
Results
Research question 1: Do terrorist attacks in Western Europe impact the domestic stock markets and industries according to event study methodology?
The Abnormal Returns per event are given below. The table is divided in three sections, for three different Event Windows. Above these three sections, some characteristics of the model are described. The Alpha of the model is the part of the return that is not explained by the CAPM model. Ideally, this should be zero, since all return should be explained by the sensitivity of the market, which is the Beta. Although no statistical test is performed on the Alpha’s, they seem to be very close to zero, with values around 0.000 for all events. The Beta, which is the sensitivity of the national market to the World Index, is between 0.5 and 1.2 for all events. These are different national indexes, all compared to the World Index. Event 5 and 6 have a Beta close to each other with 0.608 and 0.618, while event 8 is also reasonable close with 0.677. Those three attacks took all place in France, so it is expected that their Beta is approximately the same. However, especially the difference between the first two events and the last one shows that Beta is indeed not fixed. This becomes even clearer in the case of the UK, where three attacks happened. Two of them were relatively short after each other, Event 10 and 11, in Manchester and London in 2017. Their Beta’s are approximately the same, 0.712 and 0.702 respectively. However, Event 3, which is the bombing in London in 2005, has a Beta of 0.503, which is almost 30% lower than the more recent Beta’s. Although Event Windows that are longer than 10 years are far from common, this is a clear indication that Beta’s indeed change over time.
Those results, that are highly insignificant, might be an explanation for the tendency in existing literature to focus on country or industry level on volatility and on the firm level on abnormal returns. While there is proof in other research that volatility appears in markets and industries after a terrorist attack, the focus was only for individual firms on Abnormal Returns. Only in case of the 9/11 attack the effect was strong enough to find significant results on a national level. A remark that has to be made is that the Event Study
Methodology is only suitable to analyze the short term. In this case the longest Event Window is 10 days. Arin, Ciferri & Spagnolo (2008) find in their research about six countries across the world a negative Beta for national indices between -0.0003 and -0.0026 on the longer term, using a GARCH model. However, this negative impact is not yet noticeable in the short term. Note as well that the sign of the Cumulative Abnormal Returns are not necessarily negative, not even on the first day. On this day the uncertainty is peaking in general and it would not be surprising if that would have been visible in the AR.
Descriptive Statistics and T-Test Abnormal Returns Domestic Markets per Terrorist Attack
Attack Zug Mad Lon Uto Par Par Bru Nic Ber Man Lon Bar
Event 1 2 3 4 5 6 7 8 9 10 11 12 Alpha 0.000 0.001 0.000 0.000 -0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.001 Beta 0.638 0.684 0.503 0.913 1.099 0.925 0.802 1.045 1.136 0.712 0.702 0.668 R² 0.349 0.574 0.373 0.311 0.618 0.608 0.619 0.677 0.605 0.347 0.337 0.214 Std Error 0.011 0.006 0.004 0.009 0.006 0.008 0.008 0.009 0.008 0.006 0.006 0.008 σ²(rait) -0.035 0.044 0.046 0.035 0.049 0.020 -0.005 0.000 0.023 0.058 0.053 0.038 Event Window [0] AR 0.012 -0.008 -0.011 -0.003 -0.009 -0.002 -0.001 0.006 -0.006 0.003 -0.004 -0.002 σ² AR 0.235 0.672 0.613 0.379 1.165 0.246 0.075 0.138 0.012 0.054 0.019 1.342 t-test AR 0.025 -0.010 -0.014 -0.005 -0.008 -0.005 -0.002 0.016 -0.057 0.013 -0.029 -0.002 P-Value 0.980 0.992 0.989 0.996 0.993 0.996 0.998 0.990 0.954 0.990 0.977 0.998 Event Window [0-4] CAR -0.007 -0.041 -0.001 0.004 0.024 -0.005 -0.010 -0.004 -0.003 0.001 -0.007 -0.013 σ² CAR 1.200 0.065 0.452 0.379 2.108 0.604 0.075 0.442 0.098 0.932 0.164 1.671 t-test CAR -0.006 -0.162 -0.001 0.007 0.017 -0.006 -0.037 -0.006 -0.011 0.001 -0.018 -0.010 P-Value 0.995 0.871 0.999 0.994 0.987 0.995 0.971 0.995 0.992 0.999 0.986 0.992 Event Window [0-9] CAR -0.002 -0.052 -0.015 -0.047 0.036 -0.011 -0.002 0.012 -0.004 -0.004 -0.015 -0.021 σ² CAR 1.832 0.899 1.490 0.379 3.909 1.020 0.075 0.493 0.679 1.294 0.074 2.278 t-test CAR -0.001 -0.054 -0.012 -0.076 0.018 -0.011 -0.009 0.018 -0.005 -0.003 -0.055 -0.014 P-Value 0.999 0.957 0.990 0.939 0.986 0.991 0.993 0.986 0.996 0.998 0.956 0.989
The next step is to see if the same conclusion can be drawn if the focus is on the industry level. Since the literature states that some industries are more prone to terrorism than others, this is an aspect that has to be analyzed. In order to keep the table layout
structured, the industries are labeled with letters. The letters correspondent to the following industries.
Letter Industry O&G Oil Gas
BM Basic Materials IND Industrials CG Consumer Goods HC Health Care CS Consumer Services TEL Telecom UTI Utilities FIN Financials TEC Technology Table 3: Industries
The table shows that across all events and across all industries no significant results are found for the day that an act of terrorism took place. All P-Values are above 0.870. For a few national industries is no result obtained, since the data was not available in all periods for all national industries. The results for the Event Windows [0-4] and [0-9] can be found in
P-Values T-Test Abnormal Returns for Event Window [0] per Terrorist Attack
Attack City O&G BM IND CG HC CS TEL UTI FIN TEC Event 1 Zug N/A 0.946 0.875 0.985 0.990 0.908 0.937 1.000 0.986 N/A Event 2 Mad 0.999 0.995 0.992 1.000 0.980 0.992 0.985 0.996 0.989 0.985 Event 3 Lon 0.992 0.991 0.990 0.993 0.993 0.991 0.994 0.990 0.994 1.000 Event 4 Uto 1.000 0.973 1.000 0.984 N/A 1.000 0.979 1.000 0.998 1.000 Event 5 Par 0.982 0.995 0.993 0.998 0.992 0.998 0.988 0.972 0.968 0.965 Event 6 Par 1.000 0.998 0.993 0.995 0.996 0.991 0.919 0.990 0.991 0.996 Event 7 Bru 0.989 0.990 0.979 0.982 0.988 1.000 0.981 N/A 0.997 0.998 Event 8 Nic 0.991 0.992 0.998 1.000 0.951 0.984 0.996 0.984 0.981 0.994 Event 9 Ber N/A 0.994 0.981 0.989 0.993 0.990 0.991 0.975 0.980 0.986 Event
10 Man 0.977 0.914 0.979 0.999 0.983 0.999 0.985 0.997 0.964 0.954 Event
11 Lon 0.977 0.994 0.976 0.986 0.973 0.991 0.972 0.981 0.964 0.984 Event
12 Bar 0.999 0.969 0.995 N/A 0.994 0.953 0.997 0.982 0.992 N/A
Table 4: P-Values per Industry per Event. All industries show insignificant results over all events. For some industries the data is missing (N/A) for individual countries. Therefore not every event could be analyzed for all 10 industries.
Research question 2: Do changes in National Security Levels in Western Europe impact the domestic stock markets and industries according to event study methodology?
Since there is no proof for abnormal returns on the national level and the industry level, the announcements of changes in national threat levels by governments become interesting. The same structure and method as for the actual terrorist attacks is applied and therefore the T-Test for changes in threat level is carried out first on the national level and then on the industry level for France, The United Kingdom and the Netherlands. Eleven events are
distinguished for France which are analyzed. By testing for abnormal returns after the French government announces a change in the national threat level, a T-Test on the national stock market gives no indication for any significance. The abnormal returns on the day of the announcement are between -0.9 and 1.3 percent. Since the standard errors are high compared to the abnormal returns, the P-Values are all relatively close to 1. Furthermore, there is no pattern in the sign of the abnormal return and the direction of the change in national threat level. For example event 5 represents an increase in terror level, which indicates more risk on a terrorist attack, which would in turn have a negative effect on the expected abnormal return for event 5. Meanwhile the abnormal return is positive, which indicates a higher return than expected by the CAPM-model based on the estimation
and can be found in Appendix 2. No P-Values under 0.900 are found and most are in the 0.907-1.000 range. Also those national indexes do not show any pattern.
There is still a possibility that the results for a few events are biased by event date clustering. In 2012 the threat level in France changed twice between 19 March and 24 March, due to a terrorist attack in Toulouse. When the terrorist was arrested on March 24, the terror level decreased again. This could bias especially the CAR results for the Events 8 and 9 and to less extent the AR’s for event 9. However, the outcome for all other events, where the time gap between events is bigger, give reason to believe that event date clustering is not a decisive factor for finding insignificant results.
Descriptive Statistics and T-Test Abnormal Returns France per Change in Threat Level
Event Window [0] Event
1 Event 2 Event 3 Event 4 Event 5 6 Event Event 7 Event 8 Event 9 Event 10 Event 11
Terror Level Change + + - + + + + + - - +
Alpha 0.000 0.000 0.000 0.000 0.000 0.000 0.000 -0.001 -0.001 0.000 -0.001 Beta 0.695 0.707 0.707 0.957 1.257 1.239 1.139 1.308 1.306 0.983 1.099 R² 0.422 0.416 0.432 0.550 0.552 0.526 0.498 0.561 0.561 0.453 0.618 Std Error 0.005 0.005 0.005 0.016 0.010 0.010 0.010 0.012 0.012 0.007 0.006 σ²(rait) 0.073 0.064 0.078 -0.009 0.011 0.009 0.012 0.004 0.000 0.032 0.002 AR 0.001 -0.001 0.011 -0.002 0.013 0.001 0.002 -0.001 0.000 0.000 -0.009 σ² AR 0.168 0.021 1.117 0.128 0.510 0.050 0.022 0.049 0.110 0.148 0.267 t-test AR 0.002 -0.008 0.011 -0.005 0.019 0.004 0.015 -0.006 0.000 -0.001 -0.017 P-Value 0.999 0.994 0.991 0.996 0.985 0.997 0.988 0.995 1.000 0.999 0.986 Event Window [0-4] CAR -0.011 0.000 0.009 0.003 0.015 -0.007 -0.013 -0.007 -0.007 0.001 0.024 σ² CAR 1.122 0.765 1.539 0.062 0.340 0.421 0.184 0.370 0.131 0.622 2.037 t-test CAR -0.010 0.000 0.007 0.011 0.025 -0.010 -0.030 -0.011 -0.020 0.001 0.017 P-Value 0.992 1.000 0.994 0.991 0.980 0.992 0.976 0.991 0.984 0.999 0.987 Event Window [0-9] CAR -0.033 -0.002 0.004 0.012 -0.009 -0.028 -0.032 -0.014 -0.037 -0.003 0.036 σ² CAR 3.468 1.538 0.849 0.069 0.145 0.033 0.306 0.501 0.500 0.660 3.391 t-test CAR -0.018 -0.002 0.005 0.045 -0.023 -0.156 -0.057 -0.020 -0.052 -0.003 0.019 P-Value 0.986 0.999 0.996 0.964 0.982 0.876 0.954 0.984 0.959 0.997 0.985
Table 5: Descriptive Statistics and T-Test for the France Markets’ Abnormal Returns per Change in Threat Level. A ‘+’ indicates an increase in Threat Level, i.e. a more dangerous situation, a ‘-‘indicates a decreasing Threat Level.
industries for the United Kingdom and the Netherlands can be found in in Appendix 3 and give similar results. The high P-Values show that there is no reason to assume an effect after a change in the threat level. Investors do not withdraw their investment when the threat level increases, nor invest extra in certain industries when the threat level decreases. Table 7 shows the P-Values when the event window is extended to respectively 5 and 10 days. They show a similar situation as the analysis on the event day itself. A change in the behavior of investors is also not observable after respectively one and two weeks following change in threat level. This indicates that there is also no effect after investors evaluate the
information the following days after they assessed the actual implications of the change in the threat level.
T-Test Abnormal Returns France per Industry for Changes in Threat Level
[0] O&G BM IND CG HC CS TEL UTI FIN TEC
Event 1 0.999 0.999 0.988 0.993 1.000 0.995 0.991 0.987 0.998 0.985 Event 2 0.991 0.994 0.997 0.990 0.991 0.998 0.986 0.990 0.992 0.985 Event 3 0.951 0.998 0.998 0.976 0.992 0.999 0.992 0.994 0.995 0.994 Event 4 0.988 0.932 0.388 0.993 0.965 0.982 0.999 0.957 0.972 0.974 Event 5 0.993 0.994 0.995 0.992 1.000 0.996 0.994 0.989 0.999 0.944 Event 6 0.988 0.987 0.984 0.997 0.995 0.982 0.987 0.982 0.962 0.996 Event 7 0.985 1.000 0.984 0.985 0.987 0.995 0.995 0.990 0.998 0.989 Event 8 0.996 0.998 0.992 0.989 0.990 0.991 0.971 0.948 0.737 0.957 Event 9 0.993 0.997 0.994 0.992 0.999 0.995 0.977 0.996 0.946 0.993 Event 10 0.987 0.988 0.994 0.995 0.976 0.989 0.997 0.984 0.972 0.994 Event 11 0.982 0.995 0.993 0.998 0.992 0.998 0.988 0.972 0.968 0.965
Table 6: P-Values per Industry per Event for Event Window [0]. A changing Threat level has no significant effect on Industry Returns for France, for both increases and decreases of the Threat Level.
T-Test Abnormal Returns France per Industry for Changes in Threat Level
[0-4] ([0-9]) O&G BM IND CG HC CS TEL UTI FIN TEC
Event 1 0.983(0.990) 0.996(0.994) 0.997(0.997) 0.962(0.991) 0.998(0.999) 0.998(0.996) 0.945(0.992) 0.982(0.987) 0.972(0.971) 0.976(0.963) Event 2 0.972(0.998) 0.996(0.997) 0.997(0.986) 0.993(0.997) 0.998(0.999) 0.999(0.972) 0.998(0.976) 0.982(0.990) 0.991(0.995) 0.990(0.986) Event 3 0.995(0.983) 0.986(0.983) 0.999(0.995) 0.984(0.995) 0.988(0.996) 1.000(0.982) 0.995(0.964) 0.985(0.996) 0.993(0.995) 0.971(0.986) Event 4 0.932(0.915) 0.994(0.920) 0.951(0.932) 0.940(0.932) 0.910(0.926) 0.943(0.940) 0.928(0.957) 0.898(0.874) 0.889(0.876) 0.993(0.993) Event 5 0.981(0.961) 0.989(0.968) 0.985(0.975) 0.956(0.997) 0.996(0.972) 0.902(0.981) 0.997(0.963) 0.980(0.957) 0.993(0.993) 0.929(0.953) Event 6 0.979(0.971) 0.986(0.985) 0.986(0.971) 0.998(0.993) 0.999(0.983) 0.945(0.970) 0.957(0.975) 0.913(0.968) 0.994(0.999) 0.939(0.949) Event 7 0.972(0.967) 0.991(0.966) 0.992(0.998) 0.974(0.969) 0.966(0.957) 0.981(0.992) 0.994(0.997) 0.986(0.978) 0.933(0.994) 0.997(0.970) Event 8 0.983(0.960) 0.986(0.940) 0.995(0.994) 0.984(0.998) 0.993(0.982) 0.991(0.984) 0.928(0.962) 0.988(0.974) 0.964(0.998) 0.928(0.556) Event 9 0.962(0.982) 0.969(0.954) 0.961(0.988) 0.971(0.846) 0.977(0.995) 0.984(0.996) 0.977(0.973) 0.965(0.998) 0.961(0.956) 0.936(0.968) Event 10 0.977(0.976) 0.984(1.000) 0.991(0.992) 0.998(0.969) 0.988(0.995) 0.988(0.981) 0.901(0.939) 0.982(0.970) 0.970(0.976) 0.999(0.998) Event 11 0.981(0.988) 0.995(0.996) 0.995(1.000) 0.989(0.989) 0.998(0.991) 0.993(0.994) 0.983(0.995) 0.987(0.975) 0.970(0.984) 0.984(0.980) Table 7: P-Values per Industry per Event for Event Window [0-4] and [0-9]. The results for Event Window [0-9] are between
Robustness
In this section the found results will be validated. The focus in this Robustness Test is on the distribution of the data, since the condition that the data should be normally distributed is often violated in event studies. The first obtained result is that the data is in general normally distributed, except for both positive and negative outliers in some occasions. The skewness for the national markets and the international Index is given in Table 8.
Skewness
Nat. Market MSCI
Event 1 -0.529 0.183 Event 2 -0.042 0.157 Event 3 -0.285 -0.571 Event 4 0.088 -0.357 Event 5 -0.228 0.005 Event 6 -0.216 -0.597 Event 7 -0.080 -0.467 Event 8 -0.502 -0.565 Event 9 -0.644 -0.100 Event 10 -1.000 -0.127 Event 11 -1.013 -0.158 Event 12 0.378 0.203
Table 8: Skewness for the National Market Indexes and the international benchmark for all events.
Skewness between -2 and 2 is considered acceptable.
curve. This indicates that the more extreme abnormal returns that occur in the market in the Estimation Window are indeed extreme values.
By way of illustration, two Normal Probability Curves are depictured here. Event 7, which is the bombings in Brussels on 22 March 2016, shows a very close relationship between the expected Excess Return values and the actual Excess Return values of the national market in the Estimation Window. The difference between both curves is only visible around the lowest values, apart from an almost unnoticeable bubble close to the maximum value. In contrast to the Belgian market, the UK market shows between June 2016 and June 2017, prior to the attack on London bridge a clear reversed S-shape. A reversed S-shape indicates a greater variance than under a normal distribution, since the values at both ends are more extreme than under normality. All other Normal Probability Plots show a situation
somewhere between those two examples. Overall, the data is not perfectly normally distributed. The values in the middle follow the normal distribution. However, both the extreme positive as the extreme negative excess returns deviate from a normal distribution. Therefore, in order to add robustness to the prior results, a non-parametric test is carried out as counterpart for the performed t-tests.
Figure 1: Normal Probability Plot Excess Returns for the national Market Index Event 7. The blue line gives the expected values for a normal distribution, based on the mean and the standard deviation of the estimation window. The red line gives the actual values of the excess returns.
-0.06 -0.05 -0.04 -0.03 -0.02 -0.010 0.01 0.02 0.03 0.04 0.05 -4 -2 0 2 4 Return Z-Value
Normal Probability Plot Event 7
Figure 2: Normal Probability Plot Excess Returns for the national Market Index Event 11. The blue line gives the expected values for a normal distribution, based on the mean and the standard deviation of the estimation window. The red line gives the actual values of the excess returns.
Research question 3: Do the results for the above stated questions hold if the domestic markets and industries are analyzed with a non-parametric test?
Since there is no significance found in the T-tests, there is a possibility that the
non-normality of the data is leading to results that are biased. Therefore a non-parametric test, the Wilcoxon Signed Rank Test is carried out. This test is carried out to test all hypotheses that are tested with a T-Test as well. Starting with the Abnormal Returns for the National Indexes in Table 9, it can be observed that also for the ranked sign test no significant results are found. However, 10 out of the 12 Events exhibit a negative W-Score and especially Event 2 and Event 4, the attacks in Madrid 2004 and Norway 2011 show P-Values that are fairly close to the 10 percent level. Although this is still insignificant, the P-Values are much lower than for a parametric test.
Table 9: Wilcoxon Signed Rank Test for Domestic Markets in the Event Window [0-9]
Similar to the national level, the Wilcoxon signed rank test is also performed at the industry level for each event. The letters A-J correspondent again with the industries from table 3. Again, most industries do not show significant results. Incidentally, the data for a certain industry is not available, but this does not prevent us from getting a good image of the
-0.05 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 -4 -2 0 2 4 Return Z-Value
Normal Probability Plot Event 11
Expected Country Actual Country
Wilcoxon Signed Rank Test France for Terrorist Attacks
Attack Zug Mad Lon Uto Par Par Bru Nic Ber Man Lon Bar
situation. Over all the data, there are 9 industries that show abnormal returns that are significant at the 10 percent level. At the 5 percent level only 1 of them is left and there is no statistical significant result at a 1 percent level. Across the events, Event 3, Event 8 and Event 10 show statistical significance for multiple industries. Those events are respectively the London bombing in July 2005, the Nice attack in July 2016 and the Manchester attack in May 2017. This means that 2 out of the 3 attacks in the United Kingdom show significant results on the industry level for a non-parametric test. Furthermore, the Oil & Gas industry shows significance for 3 out of 10 and the Utility industry for 2 out of 11 available results, while all other industries are significant at best once. This confirms that the impact of terrorism on the oil price is existing, although this effect is only temporary. (Kollias, Kyrtsou &
Papadamou, 2013) It is interesting to notice that all significant results occur in France and the United Kingdom. These countries have the most developed financial markets with the highest degree of liquidity. Information is incorporated quick and efficient.
P-Values Wilcoxon Signed Rank Test National Industries for Terrorist Attacks
O&G BM IND CG HC CS TEL UTI FIN TEC
Zug N/A 0.314 0.314 0.551 0.510 0.591 0.279 0.351 0.470 N/A
Madrid 0.800 0.591 0.951 0.852 0.551 0.551 0.116 0.246 0.551 0.591 London 0.054* 0.186 0.137 0.551 0.017** 0.160 0.279 0.054* 0.390 0.314 Utoya 0.668 0.591 0.969 0.668 N/A 0.969 0.985 0.969 0.116 0.969 Paris 0.470 0.551 0.510 0.894 0.738 0.351 0.770 0.137 0.351 0.097* Paris 0.770 0.668 0.991 0.081* 0.738 0.215 0.770 0.591 0.470 0.770 Brussels 0.649 0.370 0.721 0.983 0.571 0.863 0.370 N/A 0.449 0.863 Nice 0.081** 0.770 0.551 0.961 0.510 0.351 0.066* 0.097* 0.630 0.874 Berlin N/A 0.785 0.785 0.449 0.934 0.610 0.903 0.934 0.148 0.721 Manchester 0.097* 0.066* 0.279 0.470 0.215 0.279 0.470 0.186 0.215 0.551 London 0.852 0.215 0.551 0.470 0.314 0.246 0.429 0.630 0.704 0.551 Barcelona 0.551 0.969 0.985 0.630 N/A 0.551 0.470 0.116 0.738 0.314 Table 10: Wilcoxon Signed Rank Test per Industry in the Event Window [0-9] Significance at a 5 percent level is marked with **, significance at a 10 percent level is marked with *.
events and only at a 10 percent level. Event 2 gives a P-value of 0.081, while event 10 gives a P-Value of 0.097. The CAR for both significant events in the event window [0-9] is negative, which is expected, since in both instances, the government of the United Kingdom changed the threat level from severe to critical.
The Wilcoxon Signed Rank test on the industry levels shows a similar pattern. The French industries do not respond significantly to a change in threat level. In the Netherlands, there is one significant result for the most current change in threat level at a 95 percent
confidence interval for the Health Care industry. The United Kingdom, where the threat level has changed 13 times in the past 10 years, shows only one statistical significant result as well. The Industrials sector had an abnormal return with a corresponding P-Value of 0.035 at Event 5. It is interesting to observe that not a single industry in the UK shows significant abnormal returns for Event 2 and Event 10, while the market index gives a significant
abnormal return. This indicates that the small insignificant abnormal returns of the separate industries can have a significant effect on the national stock market when accumulated.
Wilcoxon Signed Rank Test France for National Security Level
Event
1 Event 2 Event 3 Event 4 Event 5 6 Event Event 7 Event 8 Event 9 Event 10 Event 11
Ter. Level Change + + - + + + + + - - + Nat. Ret. -0.038 0.018 0.009 -0.017 -0.010 0.000 -0.017 -0.043 -0.042 0.014 0.082 MSCI Ret. -0.015 0.024 0.003 -0.028 0.002 0.026 0.010 -0.018 0.001 0.016 0.049 AR -0.024 -0.006 0.007 0.011 -0.012 -0.026 -0.028 -0.025 -0.043 -0.001 0.034 W-Stat -21 -5 5 -1 -11 -9 -5 -23 -23 3 25 P-Value 0.137 0.390 0.591 0.470 0.279 0.314 0.390 0.116 0.116 0.551 0.894
Table 12: Wilcoxon Signed Rank Test per Industry in the Event Window [0-9] All P-Values are insignificant.
Conclusion
The effect of terrorism on stock markets is limited. Using a parametric test, there is no evidence that terrorism affect stock returns in the short term on both the domestic level as the sector level. Concerning the changes in national threat level, there is no reason to assume that a changing threat level is observable in the stock prices. The T-Test gives highly insignificant results, similar to the T-Tests for the actual terrorist attacks.
The normal probability plots show deviations from the normal distribution. Those deviations are found mainly for both the lowest and the highest returns. The Wilcoxon Signed Rank Test, which allows data to be not normally distributed, gives significant results on the industry level. Especially the Oil & Gas industry is sensitive to terrorism. However, also Consumer Goods, Health Care, Telecom, Utilities and Technology show incidentally significance. Furthermore, all industries that show significant abnormal returns after a terrorist attack have their origin in France or the United Kingdom. These are the countries with the most developed markets in which the information is adopted with most efficiency. For other countries this information might be adopted less complete or outside the 10 day event window.
Therefore we can conclude that markets in Western European are only influenced by terrorist attacks within a 10 day event window if they are highly developed and sufficiently liquid. For other markets the industry effect might be still present, although this is not measured within 10 days after the attack.
Changes in the national threat level do not impact the return of stock market. Investors do not respond to neither an increasing threat level as a decreasing threat level. On the industry level there are incidentally significant results for an industry when the threat level changes, however, there is no structure in these sporadic significant results. Therefore we attribute these incidental cases to unrelated factors. This means that the information that a certain threat level contains is not reliable enough for investors to base their investment decisions on. Since the threat level does not impact the market, governments can increase their threat level if there is reason to do so, without being concerned about negative economic
consequences. This is important, since raising the security level is sometimes necessary for the government to take measures which are under safer circumstances not allowed by law. Although there is no evidence for the impact of terrorism in most Western European
Literature
Abadie, A., & Gardeazabal, J. (2001). The economic costs of conflict: a case-control study for the Basque country (No. w8478). National Bureau of Economic Research.
Abadie, A., & Gardeazabal, J. (2008). Terrorism and the world economy. European Economic Review, 52(1), 1-27.
Arin, K. P., Ciferri, D., & Spagnolo, N. (2008). The price of terror: The effects of terrorism on stock market returns and volatility. Economics Letters, 101(3), 164-167.
Berrebi, C., & Klor, E. F. (2010). The impact of terrorism on the defence industry. Economica, 77(307), 518-543.
Binder, J. (1998). The event study methodology since 1969. Review of quantitative Finance and Accounting, 11(2), 111-137.
Chen, A. H., & Siems, T. F. (2004). The effects of terrorism on global capital markets. European Journal of Political Economy, 20(2), 349-366.
Chesney, M., Reshetar, G., & Karaman, M. (2011). The impact of terrorism on financial markets: An empirical study. Journal of Banking & Finance, 35(2), 253-267.
Eldor, R., & Melnick, R. (2004). Financial markets and terrorism. European Journal of Political Economy, 20(2), 367-386.
Fama, E. F., Fisher, L., Jensen, M. C., & Roll, R. (1969). The adjustment of stock prices to new information. International Economic Review, 10(1), 1-21.
George, D., & Mallery, P. (2010). SPSS for Windows step by step. A simple study guide and reference (10. Baskı).
Gibbons, J. D., & Chakraborti, S. (2011). Nonparametric statistical inference. In International encyclopedia of statistical science (pp. 977-979). Springer Berlin Heidelberg.
Karolyi, G. A. (2006). The consequences of terrorism for financial markets: what do we know?.
Kollias, C., Kyrtsou, C., & Papadamou, S. (2013). The effects of terrorism and war on the oil price–stock index relationship. Energy Economics, 40, 743-752.
Lowry, R. (2013). The Wilcoxon signed-rank test. Concepts & Applications of Inferential Statistics.
Appendix 1
P-Values T-Test Abnormal Returns for Event Window [0-4] per Terrorist Attack
[0-4] City O&G BM IND CG HC CS TEL UTI FIN TEC Event 1 Zug N/A 0.95
0 0.890 0.835 0.984 0.858 0.915 1.000 0.979 N/A Event 2 Ma d 1.000 0.997 0.993 1.000 0.959 0.987 0.982 0.986 0.970 0.989 Event 3 Lon 0.97 5 0.985 0.982 0.980 0.982 0.989 0.991 0.981 0.989 0.994 Event 4 Uto 0.99 6 0.954 1.000 0.962 N/A 1.000 0.955 1.000 0.998 1.000 Event 5 Par 0.98 1 0.995 0.995 0.989 0.998 0.993 0.983 0.987 0.970 0.984 Event 6 Par 0.96 9 0.948 0.985 0.964 0.992 0.991 0.978 0.979 0.989 0.997 Event 7 Bru 0.98 3 0.999 0.993 0.948 0.992 0.949 0.988 N/A 0.995 0.966 Event 8 Nic 0.96 4 0.986 0.993 0.997 0.992 0.965 0.934 0.981 0.981 0.974 Event 9 Ber N/A 0.99
1 0.965 0.986 0.980 0.987 0.988 0.959 0.996 0.994 Event 10 Ma n 0.963 0.933 0.965 1.000 0.972 0.960 0.982 0.999 0.941 0.928 Event 11 Lon 0.97 5 0.963 0.994 0.989 0.988 0.983 0.976 0.998 0.957 0.957 Event 12 Bar 0.98 9 0.945 0.997 N/A 0.987 0.995 0.995 0.991 0.995 N/A
Table 14: P-Values per Industry per Event. All industries show insignificant results over all events. For some industries the data is missing (N/A) for individual countries. Therefore not every event could be analyzed for all 10 industries.
P-Values T-Test Abnormal Returns for Event Window [0-9] per Terrorist Attack
[0-9] City O&G BM IND CG HC CS TEL UTI FIN TEC Event 1 Zug N/A 0.97
8 0.776 0.972 0.989 0.926 0.844 1.000 0.979 N/A Event 2 Ma d 0.989 0.993 0.999 1.000 0.959 0.997 0.996 0.991 0.964 0.994 Event 3 Lon 0.97 5 0.983 0.981 0.948 0.982 0.989 0.999 0.978 0.995 0.995 Event 4 Uto 0.98 7 0.977 1.000 0.995 N/A 1.000 0.810 1.000 0.997 1.000 Event 5 Par 0.98 8 0.996 1.000 0.989 0.991 0.994 0.995 0.975 0.984 0.980 Event 6 Par 0.96 6 0.939 0.982 0.956 0.990 0.956 0.995 0.987 0.984 0.997 Event 7 Bru 0.98 2 0.988 0.962 0.951 0.978 0.994 0.994 N/A 0.994 0.956 Event 8 Nic 0.87 8 0.985 0.999 0.980 0.983 0.997 0.905 0.967 0.996 0.977 Event 9 Ber N/A 1.00
Appendix2
Descriptive Statistics and T-Test Abnormal Returns UK per Change in Threat Level
Descriptive Statistics and T-Test Abnormal Returns The Netherlands per Change in Threat Level Terror Level Change - + - + Event 1 2 3 4 Alpha 0.000 0.000 0.001 0.000 Beta 1.009 1.032 1.023 0.989 R² 0.559 0.471 0.544 0.405 Std Error 0.006 0.009 0.015 0.008 Event Window [0] AR -0.014 -0.001 0.002 -0.007 σ² AR 0.895 0.261 0.016 0.090 t-test AR -0.015 -0.001 0.019 -0.022 P-Value 0.988 0.999 0.985 0.983 Event Window [0-4] CAR -0.009 0.016 -0.003 0.009 σ² CAR 0.903 0.584 0.148 0.879 t-test CAR -0.010 0.020 -0.008 0.010 P-Value 0.992 0.984 0.993 0.992 Event Window [0-9] CAR -0.012 0.026 0.004 0.015 σ² CAR 1.836 0.717 0.436 1.810 t-test CAR -0.009 0.030 0.007 0.011 P-Value 0.993 0.976 0.995 0.991
Appendix 3
P-Values T-Test Abnormal Returns UK per Industry for Changes in Threat Level
[0] O&G BM IND CG HC CS TEL UTI FIN TEC
Event 1 0.840 0.993 0.993 0.998 0.993 0.999 0.992 0.997 0.998 0.994 Event 2 0.995 0.966 0.990 0.994 0.986 0.994 0.973 0.987 0.998 0.994 Event 3 0.991 0.978 0.999 0.993 0.995 0.989 0.934 0.995 0.995 0.987 Event 4 0.989 0.966 0.990 0.990 0.987 0.990 0.980 0.991 0.990 0.995 Event 5 0.988 0.944 0.997 0.999 1.000 0.997 0.992 0.975 0.998 0.978 Event 6 1.000 0.998 1.000 0.887 0.981 0.985 0.965 0.990 0.978 0.972 Event 7 0.984 0.956 0.985 0.985 0.981 0.983 0.992 0.980 0.982 0.975 Event 8 0.967 0.963 0.997 0.981 0.943 0.998 0.984 0.983 0.994 0.961 Event 9 1.000 0.990 0.995 0.994 0.988 0.988 0.984 0.997 0.994 0.990 Event 10 0.972 0.979 0.987 0.994 0.989 0.988 0.964 0.981 0.966 0.995 Event 11 0.988 0.985 0.988 0.991 0.990 0.986 0.983 0.988 0.982 0.985 Event 12 0.986 0.982 0.984 0.983 0.994 0.983 0.957 0.984 0.993 0.976 Event 13 0.989 0.988 0.995 0.989 0.984 0.982 0.991 0.988 0.998 1.000
Table 17: P-Values per Industry per Event for Event Window [0]. A changing Threat level has no significant effect on Industry Returns for the UK, for both increases and decreases of the Threat Level.
P-Values T-Test Abnormal Returns UK per Industry for Changes in Threat Level
[0-4]([0-9]) O&G BM IND CG HC CS TEL UTI FIN TEC
Event 1 0.982(0.994) 0.928(0.957) 0.996(0.992) 0.984(0.977) 0.991(0.987) 0.979(0.983) 0.952(0.998) 0.982(0.978) 0.908(0.978) 0.995(0.986) Event 2 0.973(0.980) 0.986(0.979) 0.993(0.994) 0.986(0.986) 0.990(0.997) 0.992(0.992) 0.953(0.963) 0.996(0.981) 0.998(0.987) 0.977(0.976) Event 3 0.957(0.966) 0.958(0.978) 0.994(0.990) 0.981(0.999) 0.971(0.996) 0.993(0.986) 0.959(0.989) 0.976(0.974) 0.984(0.987) 0.982(0.989) Event 4 0.994(0.999) 0.974(0.983) 0.984(0.979) 0.980(0.969) 0.979(0.981) 0.972(0.999) 0.955(0.947) 0.974(0.970) 0.991(0.993) 0.999(0.999) Event 5 0.988(0.993) 0.983(0.992) 0.988(0.980) 0.993(0.991) 0.985(0.981) 0.994(0.992) 0.973(0.965) 0.984(0.979) 0.996(0.990) 0.988(0.987) Event 6 0.964(0.896) 0.974(0.943) 0.900(0.966) 0.987(0.975) 0.980(0.989) 0.933(0.981) 0.965(0.928) 0.611(0.848) 0.991(0.953) 0.992(0.939) Event 7 0.989(0.974) 0.997(0.977) 0.994(0.998) 0.990(0.991) 0.976(0.965) 0.990(0.990) 0.984(0.923) 0.992(0.976) 0.901(0.928) 0.970(0.927) Event 8 0.941(0.970) 0.963(0.983) 0.992(0.997) 0.887(0.984) 0.975(0.993) 0.992(0.990) 0.990(0.934) 0.988(0.985) 0.995(0.928) 0.979(0.939) Event 9 0.980(0.985) 0.989(0.984) 0.997(0.995) 0.987(0.986) 0.996(0.999) 0.997(0.992) 0.999(0.981) 0.994(0.989) 0.99(0.958)3 0.998(0.984) Event 10 0.971(0.973) 0.942(0.938) 0.993(0.985) 0.994(0.987) 0.981(0.988) 0.967(0.997) 0.981(0.958) 0.991(0.976) 0.926(0.992) 0.999(0.985) Event 11 0.979(0.992) 0.969(0.975) 0.982(0.967) 0.989(0.993) 0.979(0.974) 0.978(0.992) 0.998(0.978) 0.973(0.975) 0.976(0.950) 0.969(0.957) Event 12 0.973(0.968) 0.982(0.919) 0.984(0.982) 0.992(0.998) 0.999(0.982) 0.988(0.988) 0.978(0.973) 0.985(0.993) 0.985(0.987) 0.986(0.987) Event 13 0.982(0.977) 0.989(0.931) 0.995(0.989) 0.980(0.974) 0.998(0.992) 1.000(0.993) 0.954(0.998) 0.986(0.982) 0.997(0.994) 0.996(0.971) Table 18: P-Values per Industry per Event for Event Window [0-4] and [0-9]. The results for Event Window [0-9] are
P-Values T-Test Abnormal Returns The Netherlands per Industry for Changes in Threat Level
[0] O&G BM IND CG HC CS TEL UTI FIN TEC
Event 1 1.000 0.991 0.957 0.996 1.000 0.997 0.970 N/A 0.996 0.984 Event 2 1.000 0.998 1.000 0.993 1.000 0.991 0.965 N/A 0.996 0.982 Event 3 0.971 0.962 0.996 0.991 1.000 0.983 0.984 N/A 0.869 0.981 Event 4 0.968 0.981 0.984 0.988 1.000 0.992 0.962 N/A 0.970 0.987
Table 19: P-Values per Industry per Event for Event Window [0]. A changing Threat level has no significant effect on Industry Returns for the Netherlands, for both increases and decreases of the Threat Level.
P-Values T-Test Abnormal Returns The Netherlands per Industry for Changes in Threat Level [0-4]([0-9]) [0-4]
([0-9]) O&G BM IND CG HC CS TEL UTI FIN TEC
Event 1 1.000(1.000) 0.991(0.938) 0.957(0.982) 0.987(0.943) 1.000(1.000) 0.992(0.992) 0.974(0.971) N/A 0.987(0.999) 0.962(0.964)
Event 2 1.000(1.000) 0.932(0.932) 0.976(0.966) 1.000(0.994) 1.000(1.000) 0.956(0.949) 0.995(0.947) N/A 0.982(0.998) 0.992(0.945)
Event 3 0.982(0.979) 0.975(0.971) 0.953(0.956) 0.785(0.988) 1.000(1.000) 0.977(0.984) 0.961(0.941) N/A 0.992(0.972) 0.951(0.950)
Event 4 0.975(0.923) 0.942(0.959) 0.999(0.993) 0.994(0.999) 1.000(1.000) 0.999(0.984) 0.926(0.868) N/A 0.980(0.979) 0.741(0.870) Table 20: P-Values per Industry per Event for Event Window [0-4] and [0-9]. The results for Event Window [0-9] are
