Terrorist attacks’ impact on the
financial sector: an event study
By:Leonie Jansen
S4058887
l.jansen.12@student.rug.nl
University of Groningen
Faculty of Economics and Business Pre Msc. Finance
Supervisor: J. Offerein Date: May 29, 2020
Abstract
This paper focuses on the impact of 62 terrorist attacks on the stock prices of 66 financial companies. Each of these companies are included in the S&P 500. The study was conducted using the event study methodology, which involved a time period of 41 days. To conduct the results, several statistical tests were performed. The results of this study show that terrorist attacks have a positive effect on the stock prices of financial firms and that financial firms do not have more negative returns with a larger instead of a small terrorist attack.
1. Introduction
The largest terrorist attack is probably known by almost everyone. This refers to the attack on September 11, 2001 when four airplanes were hijacked by supporters of the terrorist
organization Al Qaeda. Two of those airliners flew directly into the Twin Towers in New York. The third plane flew into the Pentagon and the fourth plane crashed near Pennsylvania. This attack on September 11, 2001 has claimed around 3000 lives. After this attack, the stock market was shut down for four days. According to the article written by Hon, Strauss and Yong (2004), shares of companies within the financial sector plummeted after the stock market reopened.
According to Abadie and Gardeazaba (2007), terrorism is expected to reduce the return on an investment. They also mention that terrorism creates a high degree of uncertainty. Investors will be reluctant because of this uncertainty and the expectation of lower returns. A survey within the study found that investors see terrorism as one of the most important factors in deciding to invest in a company. Abadie and Gardeazaba (2007) mainly emphasize the
foreign investors. The more attacks that take place in a certain country, the less investors from abroad will be inclined to invest in the companies within that country. If these international investors have a wide range of companies that they can invest in, they will not choose to invest in a country where relatively many terrorist attacks take place. Since many financial companies depend on investors, it is expected that financial companies will suffer from terrorism.
The first terrorist attack (recorded by the Global Terrorism Database) in the United States took place in 1970. After this first attack, another 2925 followed. The last attack in the United States took place on December 28, 2019 and is therefore very recent. There has been a
The second part addresses the literature, where the expectations are explained on the basis of theoretical underpinning and empirical findings. The section ends with the research question converted into hypotheses for testing. Section three discusses the data and methodology, where the execution of the study is explained. The performed tests and results of the research are discussed in the fourth part. The research ends with the conclusion in section five.
2. Literature review
A well-known quote by Lucas (1977) is: “In cases of uncertainty, economic reasoning will be of no value.” His theory indicates that when a company or country is in an uncertain situation, investors will no longer rely on economic reasoning. Keeping this in mind, Daniel, Hirshleifer and Subrahmanyam (1998) state there are two groups of investors; investors who are well informed and investors who are not well informed. The theory states that the well-informed investors are risk neutral and the uninformed investors are risk averse. This corresponds with the views of Lucas (1977), since uninformed investors are in an uncertain position and are therefore risk averse. As a result, economic reasoning will have less value. Additionally, Harrison and Stein (1997) follow the theory that investors are divided into a group that overreacts and a group that is prone to underreaction. They believe that the market remains the same in most cases, as the two are counterbalanced.
Furthermore, the reaction of the market after an attack has been examined by several studies. The study by Chen and Siems (2004) states that the US capital market suffers a couple of days of even weeks after an attack. Barrett et al. (1987) do not support this statement. Their study has shown that the negative reaction only applies to the day after the event and does not pass on to the following days. Next to the negative effects of these attacks, Chen and Siems (2004) state that the market has become more resistant and repairs itself better after such an event. This finding is contradicted by Eldor and Melnick (2004). They indicate that there is no evidence that the markets are less affected after events like 9/11, which is supported by
Barrett et al. (1987). The study by Eldor and Melnick (2004) does mention that the markets have become more efficient in recording information about terrorist attacks. Keeping in mind the theoretical response and the studies previously conducted, the following hypothesis has been determined:
Previous studies have shown that there is a difference in the reaction after a large and small attack. According to Aslam and Kang (2013), a large attack has more impact than a smaller attack. The research shows that while there is little difference in the direct costs to the economy with a large and smaller attack, there is a big difference in the impact on the stock prices. This difference in direct and secondary effects is consistent with the article written by Johnston and Nedelescu (2006). This article shows that the impact on share prices is mainly caused by secondary effects. Johnston and Nedelescu (2006) argue that terrorist attacks damage the security of consumers, causing them to spend less and save more. Other indirect effects mentioned in the article are higher insurance premiums and costs due to new security measures. Financial markets are therefore faced with major uncertainties. Based on the theoretical and empirical evidence, the following hypothesis emerged:
3. Data and methodology
3.1.
Data
In this paper, two sources were used to collect the data. This data includes the committed attacks and the stock prices of the selected companies.
All attacks are imported from the Global Terrorism Database. The attacks that took place between the years 1995 and 2018 are included in this research. Within these years, a total of 784 attacks were committed. These attacks vary in number of people killed or injured, motive and places where the attack happened. Of these 784 attacks, 62 attacks have claimed 3 or more victims. These 62 attacks are included in this research. Furthermore, this research looks at the 66 companies within the S&P 500's financial sector. The stock prices are obtained from the Yahoo! Finance database. The table below shows a summary of the collected data.
TABLE 1 SUMMARY DATA
VALUE
Number of companies 66
Number of total attacks 62
Number of years involved in the research 14 Number of small attacks (casualties between 3
and 10)
37 Number of large attacks (more than 10 casualties) 25 Average number of casualties per attack 254 Average number of fatalities per attack 57 Average number of attacks per year 2,6 Average of the AAR within event window
(-20,20)
0,029% Median of the AAR within the event window
(-20,20)
0,021%
3.2. Methodology
the market equal. According to the theory, a market is efficient if prices fully reflect the available and relevant information. Since new information is also being released when a terrorist attack takes place, this theory can be used well for this research.
The event study methodology is largely based on the efficient market hypothesis. The article written by MacKinlay (1997) discusses the event study methodology and has been consulted for the execution of this study. The methodology examines the returns on stock prices. The normal returns are compared with the returns after new information is announced or an unexpected event has occurred. Based on this information, we can determine whether an abnormal return has indeed emerged after the occurrence of an event.
In this study, an estimation window of 180 days was chosen. This means that returns 180 days prior to the event were taken into account. In this way, a clear and objective picture of the normal returns will be obtained. In addition to the returns from the companies, we also looked at returns 180 days prior to the event of the market prices. To obtain this information, the index of the S&P 500 has been consulted. This index was chosen because it includes the largest US financial companies. Furthermore, an event window of 41 days has been applied. This event window examines the development of returns in the 20 days before and 20 days after the event. In this way, the abnormal returns for each event in combination with each company becomes transparent, to which the average abnormal returns per event can be calculated. An illustration of the estimation and event window is shown in figure 1.
Estimation window Event window
-180- -20 0 20
Figure 1: Estimation and event window
To determine whether there is indeed an abnormal return as the result of an event, the
test this factor, the data is divided into two subsamples and the Wilcoxon Signed Rank Sum Test (1945) is performed. With this test, the difference in returns can be determined by looking at the difference in CAARs. After this, the differences are ranked and summed up for the positive and negative values. The test then compares the test statistic with the critical value. A critical value of 5 percent for a one-tail test has been chosen.
In order to calculate the CAAR, the risk-adjusted market model is used. The first thing to calculate is the abnormal return for each event in combination with each of the companies on each of the relevant days. The following formula is associated with the abnormal returns:
𝐴𝑅!" = 𝑅!"− (𝛼' + 𝛽# + 𝑅# $"). (1)
Where
𝑅!" represents the return on stock 𝑖 at time t, 𝛼'#the intercept and 𝛽+#the slope. 𝑅$"
gives the return on the market portfolio at time t. The estimator 𝛼'#is obtained by calculating
the intercept of the firm’s return and the market return within the estimation window. Regarding the estimator 𝛽+#, the slope of the firm’s return and the market return within the
estimation window was conducted. After determining all abnormal returns within an event (so for all companies in combination with this event), the average abnormal returns for one event can be calculated. This includes the formula: 𝐴𝐴𝑅" = % & ∑ 𝐴𝑅!" & !'% (2)
where N stands for the total companies. This gives the AAR per event on a certain day within the event window. After this, the average of all AARs of the events on a given day can be calculated. This clarifies what the AAR for the entire event window is.
𝐴𝐴𝑅"
IIIIIII = %
() ∑ 𝐴𝐴𝑅
*
!'% " (3)
Where 62 stands for the total events. To determine the total impact of the events within the event window, we look at the cumulative of the average abnormal returns:
To test the significance of the cumulative average abnormal returns, we first should determine if these returns are normally distributed. This can be tested by using the Jarque-Bera test.
𝐽𝐵 = ()( (𝑆 + %0(𝐾 − 3))) (5)
Within this formula, 62 stands for the number of observations, K for the kurtosis of the AARs and S for the skewness of the AARs. Although the Jarque-Bera test shows that the data is not normally distributed, according to Brown and Warner (1980) it is still useful to test the data by means of a t-test, because they have found that parametric test results correspond to non-parametric test results. The following formula has been used:
𝜃 = 4 (1223)1223 (6)
𝜃 represents the value of the t-test and 𝜎 stands for the standard deviation. Since it emerged that the results are not distributed normally, the Sign Test by Cowan (1992) is used. This allows us to determine whether the number of positive abnormal returns within the event window is higher than the number of positive abnormal returns without abnormal
performance. The formula associated with the Cowan Sign Test is:
𝑝̂ = &% ∑* 𝑆
!'% (7)
Where 𝑝̂ stands for the percentage of positive returns within the event window. N stands for the number of abnormal returns within the estimation window and S represents the dummy variable. This dummy variable equals zero when the return is positive and equals one when the return is negative. To determine the Cowan Sign Test statistics, we can use the formula:
𝑍𝑔 = [*78(%,78)]6,*78!/# (8)
4. Results
This section discusses the results of the study. The CAAR was calculated for two components; the CAAR of the impact regarding all events as a whole and the difference between the impact of small and large attacks. A number of statistical tests and analyzes have been carried out to answer the hypothesis.
TABLE 2
EVENT DAY AAR CAAR
Table two shows the AARs and CAARs for the overall sample. As can be seen in the table, the ratio of positive and negative values within the AARs is very divided. However, the CAAR on days 20, 10 and 5 are positive. In addition, there is also a positive CAAR on the day of the event and the day after.
TABLE 3 CAAR SUBSAMPLES
CASUALTIES BETWEEN 3 AND 10 MORE THAN 10 CASAULTIES
EVENT DAYS CAAR CAAR
-20, 20 1,215% 0,789%
-10, 10 1,008% 0,466%
-1, 5 0,336% 0,168%
All of the CAARs regarding the subsamples in table three are above zero. The difference in these two subsamples has been measured using the Wilcoxon Signed Rank Sum Test. The calculation and some of the outcomes can be found in appendix three. The test showed that the test statistic (83) is lower than the critical value (152), which can be seen in table 3. This means that there is indeed a difference between the two samples.
TABLE 4
WILCOXON SIGNED RANK SUM TEST - SUMMARY
CRITICAL VALUE (𝛼 = 0,05, 𝑜𝑛𝑒 𝑡𝑎𝑖𝑙) POSITIVE SUM NEGATIVE SUM TEST STATISTIC TL TU 778 -83 83 152 313
TABLE 5 JB AND T-TEST (-20, 20) (-10, 10) (-1, 5) St. deviation CAAR 0,0032 0,0028 0,0014 T-test 3,2082 2,8085 1,9432 JB 36,8795 38,4647 52,8934 Skewness 0,40319 0,41331 -0,3547 Kurtosis -0,5585 -0,6381 -1,6791
The JB-test showed a value of 36,8795, which indicates that the data is not normally distributed. The value of 36,8795 refers to the 20-day event window. This and the other values of the skewness and kurtosis can be found in table five. Besides this, the t-test has been calculated using the CAAR and the standard deviation of the CAAR. These results show that all the tests are significant at a one-percentage level. However, the data is not normally distributed, so no conclusion can be drawn from the t-test.
Since the data is not normally distributed, the Cowan Sign Test has been performed. The table below shows that the Z-value of this test is -3,1688714. The corresponding P-value concerns 0,0011962 (0,1%). The outcome of the P-value is below 1 percent and is therefore highly significant. The hypothesis within this study states that with a CAAR that is bigger than or equal to zero, there are no negative abnormal returns. Since none of the CAARs are below zero, we can accept H0. There are no negative abnormal returns.
TABLE 6 COWAN SIGN TEST
N (number of events) 62
-Degrees of freedom 61
Total returns within event window 41
Positive returns 24
Negative returns 17
Percentage positive returns 59%
Z-value -3,1688714
The results do not match with what would be expected according to the literature. For example, Chen and Siems (2004) state that negative returns arise after the occurrence of an attack, which is not the case in this study. This study shows that positive returns occur after an attack happens. This does not really make sense at first sight, as you would expect investors and consumers to become more reluctant after an attack. Based on this research, this does not seem to be the case. A possible reason could be that consumers indeed spend less, forcing banks to raise the interest rates. The rise in interest rates may cause financial companies to take advantage of this, resulting in positive abnormal returns.
Conclusion
The results showed that each of the CAARs have a positive value. This applies to the 20, 10 and 5 days after the event as well as the day the event occurs and the day after the event occurred. Since the Jarque-Bera test showed that the data is not normally distributed, the Cowan Sign test was performed. This resulted in a P-value of 0.1%, which means a high significance. In addition to the Jarque-Bera test and the Cowan Sign test, a t-test was
performed on the overall sample. This test showed that the values for each event window are significant for a one-percent level. These results were included in the study, but since the data is not normally distributed, no conclusion can be drawn from this t-test. Looking at the
results, it can be concluded that there are no negative abnormal returns present. The hypothesis states that with a CAAR greater than or equal to zero, there are no negative
returns. This is indeed the case in this study and therefore the first null hypothesis is accepted.
All CAARs are positive. This means that a terrorist attack would result in a positive return. This is not entirely in line with the expectations, as previous studies by Chen and Siems (2004) and Barret et al. (1987) have shown that such attacks lead to negative returns. Barret et al (1987) indicate that these negative returns only occur the day after the attack, but this is not the case in this study either. According to Lucas (1977), investors and consumers in an
uncertain position will reason less from an economic point of view. The media can provide a distorted picture of the situation, so that consumers are not fully informed and feel like they are in an uncertain position. A reason for the positive returns may be that banks are forced to adjust interest rates upwards, because consumers spend less and companies are therefore affected. Financial companies can benefit from this, which results in a positive return.
Another reason for this may be that we live in an open market economy. As a result, investors know what the risks are and what their influence can be. This transparency makes it possible that investors, as Lucas (1977) states, are risk neutral. This may explain that the attacks do not lead to negative returns, but it does not explain the positive returns. Furthermore, the research by Aslam and Kang (2013) showed a difference between the impact of a large and a small attack. This is in line with the results of this study, as it has indeed shown that stock prices respond to the fact that a large attack has taken place. The results show that the positive returns in the case of a large attack are lower than in the case of a small attack.
The limitations of this investigation are, firstly, that only terrorist attacks with three or more victims were considered. Because the study did not look at attacks with less than three victims, the results are limited. The reason these attacks were chosen, is because the expectation at the start of the research was that attacks with no or very few victims would have a small impact. In retrospect, it may also be interesting to determine whether there is a difference in impact with such attacks. Another limitation of the research concerns the methodology. This study only used the event study methodology, while studies such as those by Chen and Siems (2004) used other methods, for instance the excess returns approach. This could have had a positive effect on the reliability of the results of this study.
References
Abadie, A., Gardeazaba, J., 2007. Terrorism and the world economy. European Economic Review.
Aslam, F. and Kang, H., 2013. How Different Terrorist Attacks Affect Stock Markets. Defence and Peace Economics 26(6): 634-648
Barrett, W., Heuson, A., Kolb, R., and Schropp, G., 1987. The adjustment of stock prices to completely unanticipated events. The Financial Review 22(4): 345-354.
Brown, S.J., Warner, J.B., 1980. Measuring security price performance. Journal of Financial Economics 8, 205-258.
Chen, A., Siems, T., 2004. The effects of terrorism on global capital markets. European Journal of Political Economy 20, 349–366
Cowan, R., 1992. Non-Parametric Event Study Tests. Review of Quantitative Finance and Accounting 2, 343-358
Daniel, K., Hirshleifer, D., Subrahmanyam, A., 1998. Investor Psychology and Security Market Under- and Overreactions. The Journal of Finance.
Eldor, R., Melnick, R., 2004. Financial markets and terrorism. European Journal of Political Economy 20: 367-386
Fama, E. F., 1970. The behavior of stock-market prices. The Journal of Business.
Harrison, H., Stein, J., 1997. An Unified Theory of Underreaction, Momentum Trading and Overreaction in Asset Markets. National Bureau of Economic Research.
Hon, M.T., Strauss J., Yong S.K., 2004. Contagion in financial markets after September 11: myth or reality? The Journal of Financial Research, 95-114.
Johnston, R., Nedelescu, O., 2005. The Impact of Terrorism on Financial Markets. Working Paper, International Monetary Fund, Washington.
Lucas, R.E., 1970. Understanding Business Cycles. In Stabilization of the Domestic and International Economy. North-Holland, Amsterdam, pp. 7-29.
MacKinlay, A. G., 1997. Event Studies in Economics and Finance. Journal of Economic Literature, 13-39.
Noether, G. E., 1992. Introduction to Wilcoxon (1945) Individual Comparisons by Ranking Methods. Springer-Verlag. New York, pp 191-195
Websites
Global Terrorism Database:
https://www.start.umd.edu/gtd/search/Results.aspx?search=&sa.x=54&sa.y=3
Yahoo! Finance Database:
Appendix 1: List with terrorist attacks
DATE FATALITIES INJURED TOTAL
Appendix 2: Subsamples – Hypothesis two
CASUALTIES BETWEEN 3 AND 10 MORE THAN 10 CASUALTIES
EVENT DAY AAR CAAR AAR CAAR
Appendix 3: Wilcoxon Signed Rank Sum Test
SAMPLE 1 SAMPLE 2 DIFFERENCE POSITIVE/ NEGATIVE
