RIJKSUNIVERSITEIT GRONINGEN
Faculty of Economics
Department of International Economics and Business
Doctoral Thesis
THE EFFECTS OF 11
THOF SEPTEMBER 2001 ON INVESTOR
BEHAVIOUR IN THE AVIATION INDUSTRY
R.H. Scholtens 1199021
August 2007
2
ABSTRACT
3
TABLE OF CONTENTS
4 1. INTRODUCTION
The terrorist attacks of September 11th 2001 (9/11) have changed the daily life for many people. One of the industries that has to cope with many changes due to 9/11, the aviation industry, is also a victim of these events. One of the avenues of academic research into the aviation industry is the response of investors to the news of an accident. In the wake of 9/11, the academic community has researched the effects of these events on the aviation industry. Carter & Simkins (2004) wrote an article that combines both these approaches of research. They investigated the reaction of airline stock prices to the attack immediately after the terrorist attack and during the following period. They find evidence that the hypothesis of rational pricing holds and the evidence suggests that the market differentiated between various air-transport firms. Although this article provides insight in the direct effects of 9/11 on the aviation industry it does not provide insight in the indirect effects. One such indirect effect could be that the attitude of investors towards the aviation industry has changed. This is the central theme of my paper.
According to Carter & Simkins (2004), the “rational pricing hypothesis” is not affected directly after 9/11, but this does not necessarily mean that is not affected at all. In order to investigate whether there is an indirect effect on rational pricing I will study the behaviour of investors before and after 9/11. One particular aspect of investor behaviour that has been studied is the reaction of investors to aviation accidents. I think that a traumatic event like 9/11 has made investors less rational with respect to aviation accidents. Therefore, I will base my study on the comparison of the reaction of investors to aviation accidents before and after 9/11. In my investigation, I want to answer the following research question: Did the
events of 9/11 effect investor reactions to aviation accidents? And if so, how?
With the above presented problem statement there are three elements in the question that need further clarification: 9/11, investor reactions, and aviation accidents. Both a good definition and understanding of these elements is needed to answer the problem statement. I will define each element, explain how it will be measured, and indicate its relevance to the research question.
5 changes when looking from a finance perspective. Carter & Simkins (2004) argue that the tremendous emotional impact of the September 11th attacks makes this event potentially very different from other events that are frequently studied in the finance literature. In order to measure the effects of 9/11 from a finance perspective I will compare investor reactions before and after 9/11. Based on this comparison I will be able to draw a conclusion about the research question.
The term investor reaction also needs further clarification. The reaction of investors to news relating to publicly quoted companies is one of the ways in which investor behaviour becomes apparent. The relation between the investor and the company is of importance. Jensen & Meckling (1976) and Fama & Jensen (1983) identify this relationship as an agency relationship. The agency theory describes the relationship as one where one party, the principal, delegates work to another party, the agent. Since shares are a part of ownership in the company, the willingness to pay (the price at the stock market) for them reflects how investors value the company. In my research, I want to investigate how investors react to news that influences the value of a company. The value of a company is influenced by a substantial loss (Sprecher & Pertt, 1983) and an aviation accident is a good example of a substantial loss. By measuring the willingness to pay of investors, their reaction to the news can be measured, which in turn allows for inferences about investor behaviour. Measuring the willingness or unwillingness of investors to pay for shares of a public company in the wake of a substantial loss allows inferences about investor behaviour.
The final element in my problem statement that needs to be clarified is aviation accidents. An aviation accident can be briefly be defined as any accident relating to or as result of operating an aircraft. With an aviation accident defined, I will now look at how to measure an accident and how it relates to my problem statement. What is important in relation to the aviation accident is that the news of the accident reaches the market. If this is not the case, investors cannot update their valuation of involved company because they do not know the accident. In light of this news criterion, I will only consider accidents with fatalities and check if the news actually has reached the market.
The aim of this study is to examine the effects of 9/11 on investor behaviour to
aviation accidents. This paper is able to make some significant contribution to the current
6 before by my knowledge. Secondly, this is the first time a two samples t-test is used to compare aviation price stock reactions from different periods. Furthermore, this thesis expands the knowledge of the indirect effects of 9/11, more specifically whether investor behaviour has changed due to 9/11. Existing literature has only focused on the effects direct effects of 9/11 of investor behaviour.
7 2. LITERATURE REVIEW AND HYPOTHESES
In order to answer the research question of this thesis properly a thorough analysis of relevant academic literature is necessary. The contribution of the academic literature to answering my research question is twofold, a thorough analysis facilitates the development of theoretical model underlying the research, and previous research and its findings helps in deriving the hypothesis to test. This section will first give an in-depth analysis of the concepts in my research question. It will continue with the derivation of the conceptual model with which I will be working. This chapter will conclude with the derivation of the hypotheses.
2.1 Analysis of concepts
This section proceeds with an in-depth analyses of the three characteristic concepts in my research question: aviation accidents, investor behaviour, and 9/11. For an orderly analysis I will undertake three steps with which I will analyse the concepts
1. Concept definition
2. Theoretical background relating to the concept 3. Review the relevant literature of the concept
The purpose of the concept definition is to meticulous define the concept in question. The analysis of the theory will serve as basis for the derivation of the conceptual model. The literature review analysis will be used as input for constructing the hypothesis.
2.1.1 Aviation accidents
The U.S. National Transportation Safety Board (NTSB) has defined an aviation accident as: “an occurrence associated with the operation of an aircraft which takes place
between the time any person boards the aircraft with the intention of flight and all such persons have disembarked, and in which any person suffers death, or serious injury, or in which the aircraft receives substantial damage” (NTSB, 1999). In addition to this definition
8 The basis of academic theory relating to aviation accidents stems from the work of Davidson et al. (1987). The authors identified three possible ways in which an aviation accident can cause a loss to an airline and its stockholders. Physical damage to the aircraft is the first type of loss to which an airline company is exposed. This kind of loss is covered under an aircraft hull insurance policy. The second type of loss is liability losses suffered by the airline as result of the crash. These losses include death, injury, and damage both off and on the ground as result of the crash. This type of loss is covered under various aircraft liability policies (Ball & Brown, 1968). The third and final type of loss is loss of goodwill, which is not an insurable risk. This means that passenger might not want to use the airline because of perceived safety problems with the airline, which results in loss of passengers.
The physical damage to the aircraft is covered by the hull policy, which is mandatory. This hull policy covers the air and ground risk from accidental damage (Petroni, 2000). Related to the hull policy is the hull war policy, which covers airlines aircrafts against loss or damage from a range of war and war-related risks. An important relation that the author mentions in his article is that of the number of claims (measured by the number of hull losses) which used to influence insurance rates. Due to changes in airline insurance markets the fluctuations in rates and premiums are the result of different views between various underwriters, who use different methods for assessing the price for a risk. This has made the influence of hull losses on rates ambiguous. Since the loss of an aircraft is fully covered under hull policy of an airline and the hull loss has no direct effect on the insurance rate, the loss of the aircraft itself does not cause a loss to an airline.
With respect to liability loss, some regulatory changes have taken place since the article of Davidson et al. (1987). When this article was written the liability loss of airlines arranged in the Warsaw convention, and its amendments. The upper limit was set in 1966 to 75,000$ and remained in effect until 1997. The authors mention a lower limit for liability insurance of $300,000 set by the U.S. Department of Transportation (DOT). On initiative of the International Air Transport Association (IATA) and the DOT an international intercarrier agreement was reached starting in 1997. The intercarrier agreement removes the limit of liability and allows recovery full compensatory damages for physical injury or death in an accident, according to the law of the place of residence of the passenger (Kolczynski, 2001). Currently the intercarrier agreement is signed by the 130 airlines1. The implications of these
1
9 regulatory changes are ambiguous. The liability of an airline in the case of an accident is dependent whether an airline has signed the intercarrier agreement and the home country of the passengers. Furthermore, in the U.S. the DOT sets minimum limits for the airline industry on insurance coverage for passenger liability insurance, which also influences the exposure of an airline to liabilities. The intercarrier agreement does have implications for the theory of Davidson et al. (1987). They work under the assumption that the only the loss of goodwill can lead to a loss, both physical damage and the liability loss is assumed to be fully covered by insurance policies. However, with the intercarrier agreement the signatory airlines are exposed to unlimited liability that can lead to claims that exceed the coverage of the liability insurance. With lower limits set by regulatory bodies, the exposure to risk is still under control of the airline itself. If the liability claims are in excess of the coverage the airline has, the implications for the profitability of the airline can be large.
Based on the above presented potential sources of loss, airlines will be most susceptible to the loss of goodwill. The basic argument behind the loss of goodwill can work both ways. Potential passengers can switch because of perceived safety problems, but the airline can also be perceived to be safer since it will be under closer scrutiny of regulatory bodies. Research (Borenstein & Zimmerman, 1988) finds weak deviations from expected demand due to aviation accidents. This conclusion should be interpreted with care. Another option to consumers instead of flying less, is switching to another airline company. Bosch, Eckard & Singal (1998) find that this indeed happens, when market overlap exists. Furthermore, negative spillover effects (consumers do not switch but fly less, which also affects other demand airlines negatively) are also present in cases of low market overlap.
Table 1: Review of literature on aviation accidents
author key questions Airlines2 observations period model key hypothesis key results
Sprecher & Pertl, 1983
the effect of large losses
on the value of firms NYSE/AMEX listed 27 1969-1978
event study (market model)
large losses have an impact of the firm, reflected in the value of the firm's stock.
prices of stock declined 4% on average as result of the report of a large loss
Barrett et al., 1987
price stock reaction to fatal commercial airline
crashes
NYSE/AMEX listed
airlines, 1fat. 78 1962-1985
event study (market model)
the efficient market hypothesis holds, so all news is directly incorporated
The EMH holds, significant one day reaction to an airlines crash
Chance & Ferris,1987
price stock reaction to air crashes of (not) involved airlines and the aircraft
manufacturer NYSE/AMEX listed domestic airlines, 3 largest aircraft manufacturers, 10 fat. 46 1962-1985 event study (market model)
negative price stock reaction of airline (not) involved and the aircraft
manufacturer
Immediate one-day reaction on the day of the crash with crash airline, no reaction at non-crash airlines or
manufacturers of aircraft involved
Davidson et al., 1987
effects of large losses on airlines NYSE/AMEX listed airlines 57 1965-1984 event study (market & constant mean return model)
large losses in the airline industry have a
negative impact on firm value no long term significant abnormal returns Borenstein
& Zimmerman,
1988
quantify the costs that airlines incur due to
crashes US certified air carriers, NYSE/AMEX listed, 1 fat. 67 1962-1985 event study (market model)
a consumer demand response to the accident which can be measured by a price
stock reaction
unclear if consumer respond to crashes (average firm loss appears to below total social cost)
Mitchell & Maloney
(1989)
examine the brand-name effect of airline crashes
US NYSE/AMEX
listed airlines, 1 fat. 56 1965-1987
event study (market model)
Accidents where the carrier is at fault cause consumers revise their expectations
about the probability of an accident
when the airline is most likely at fault, there is a significant negative stock market reaction to the event
Bosch, Eckard, & Singal (1998) expected adverse consumer response causes a stock price reaction to the accident
US NYSE listed national airlines, 1
fat.
25 1978 (Q4)- 1996 (Q4) event study
a positive price stock reaction for non crash airlines which overlap the crash
airlines market
positive price stock reaction when overlap exists, when there is no overlap a negative spillover effect
exists (negative price stock reaction)
Walker, Thiengtham,
& Lin (2005)
short/long-term stock price performance of the
aviation industry after crashes
US publicly traded airlines & aircraft manufacturers, 1 fat. 138 07-1962 - 12-2003 event study + univariate test + regression analysis
significant price stock reactions to accidents, US territory & cause of
influence, EMH holds
Significant price reaction to accidents, crash location, fatalities, airline size, criminal activity influence
reaction. EMH does not hold
Here I will continue with the review of the literature relevant to aviation accidents. Table 1 (page 10) lists the key literature relevant to aviation accidents in chronological order with the key assumptions, key hypothesis, data & methods, and key findings of the respective authors. The following paragraphs will discuss the method, findings, and conclusions of the relevant authors.
The interest of the academic community in aviation accidents found its bases in the research of Sprecher & Pertt (1983) who researched the effect of large losses firms experienced on the market price of the firm‟s stock. They found negative abnormal returns of 4% on the day of a large loss, which not reversed itself in the following days. Davidson et al. (1987) placed the large loss, as defined by Sprecher & Pertt (1983), in the context of the aviation industry, an aviation accident. Davidson et al. (1987) employed a 57 crashes sample, for which they controlled for confounding crashes (crashes that occurred within a short time period, six months, of an earlier crash at the same airline). They performed a test on the whole sample as well on a subsample with multiple deaths and the most severe crashes. They only found significant negative investor reaction for the most severe crashes of around 2.5%, but this reaction reversed itself in the following five days. The authors came to the conclusion that investor reaction might have been limited due to the fact that the selected airlines carried an adequate liability coverage. Further reasons for the differences in result with Sprecher & Pertt (1983) are explained by industry specificity, firm size, and limited goodwill effects according to investors. The methods, findings, and conclusions that Davidson et al. (1987) appear sound at first sight. However, the results are contradicting those of Barrett et al. (1987) and Chance & Ferris (1987), who both investigated an almost identical sample and have different findings.
12 after the event. Based on this and the fact that there were no significant increases either after the second day the authors concluded there was no over or under reaction, however they do not define either. In my opinion, this conclusion cannot be drawn from these findings. Whether overreaction occurred can be better evaluated when it know what an appropriate reaction to the news is. The fact that no significant decreases occurred after the second day could mean that overreaction persisted and was not corrected. In other finance literature (De Bondt & Thaler, 1987; Fama & French, 1992; Lakonishok, Shleifer, & Vishny, 1994) it is more usual to asses over or underreaction by looking to other measures of valuation, and not past returns. In their article, Barrett et al. (1987) did attempt to include another measure of valuation. They repeated the analysis for a subsample, which had the highest ratio of deaths (in the crash) to the book value of the firm (at the time of crash) as a proxy for the loss. This analysis resulted in a significant Abnormal Return (AR) for the event date and marginal significance for one day after the event. These results provided no evidence that under reaction or overreaction appeared in the initial response period. Although, the authors did attempt to incorporate other measures of valuation, then just using past returns, the findings still do not support their conclusion. First, the article does not describe what are to be considered „high‟ death-to-asset ratios, nor does it say how many crashes this subsample contains. Secondly, the usage of a death-to-asset ratio as a proxy as for the magnitude of the loss is not very sound. As mentioned in theoretical background, minimum limits are set by the DOT for liability coverage of airlines. Since in general airlines have coverage well above these limits the number of deaths of an accident will not influence the size of the loss of the airline and this known by investors3. Without an assessment of how an accident has changed the expected cashflow or risk to the firm (Davidson et al., 1987), nothing can be said about over or under reaction. A more sound judgement about over or underreaction should be based on the ex post information about how the accident influenced the expected cashflow of the airline. Although I do not agree with the conclusion draw based on their findings, the findings itself are in order and can be used to underpin my hypothesis.
3
13 Chance & Ferris (1987) also examined the price stock reaction, for both airlines and aircraft manufacturers, to aviation accidents. They found negative abnormal returns of 1.2% for airlines on the day of the event, no reaction was found for both airlines not involved in the crash as well as for the manufacturer of the aircraft involved in the crash. The negative abnormal return did not reverse itself in the following period that was researched. The research of the authors is sound and straightforward.
In the year 1987, three articles were published on the stock price reaction to aviation accidents (Barrett et al., 1987; Chance & Ferris, 1987; Davidson et al., 1987). The results, the reversal of the initial reaction, from Davidson et al. (1987) contradict those of Chance & Ferris (1987) and Barrett et al. (1987). This is somewhat strange since both the methodology and samples are quite similar. The initial reaction found by Davidson et al. (1987) is low with only 0,7% at an 10% significance level. With such a low initial reaction, the reversal itself could be caused by the normal volatility of the airline stock. Therefore, I will disregard the results of Davidson et al. (1987) when deriving my hypothesis.
14 appropriate to link the quantity effect to the aviation accident then to the stock market reaction to the aviation accident. Finally, the post deregulation sample used by the authors is small and result could idiosyncratic (the authors do mention this in their conclusion).
Mitchell & Maloney (1989) performed what can be considered as a follow up to the study of Borenstein & Zimmerman (1988). They investigated the effect of airline crashes on the brand name with the event study method. In their sample, they distinguished crashes caused by pilot error and were the carrier was not at fault. Brand theory suggests, according to the authors that in the case of pilot error consumers revise their consumption pattern, leading to a lower demand and hence lower goodwill. An insurance rate function showed that insurance costs do increase due to pilot-error crashes and not to crashes for which the carrier is not to blame. According to the authors, this partially explained the stock market decline (38%), and following the theory of brand the rest of the stock market decline was caused by a loss in goodwill of consumers (increased insurance cost reflects a change in the probability of an accident which also will be taken into account by consumers). The authors used the price stock reaction as proxy for the reaction of consumer, under the assumption of market efficiency this is correct. It must be realized here that this assumes that investors can correctly assess the effects of an aviation accident on consumer demand. The authors concluded that the results are straightforward and support the notion that airline crashes cause consumers to reduce their demand for the services provided by negligent carriers, which is the prediction of the theory that brand names are a quality assuring mechanism.
15 have the same shortcoming. In their research the expected adverse consumer reaction is measured by the stock price reaction of investors to the accident, this assumes that investors correctly asses the adverse consumer reaction. Furthermore it assumes that the price stock reaction is only caused by this expected adverse consumer reaction. The conceptual models have shortcomings in both the assumptions of cause as well as the measurement of investor stock price reactions. It seems more appropriate to measure factual effect of the aviation accident on the turnover of the airline then to measure this with valuations of investors.
The most recent study, performed by Walker et al. (2005), is comprehensive and focused on the impact of aviation disasters on the short- and long-term performance of airlines and airplane manufacturers. Their focus was on a sample of U.S. carriers and aircraft manufacturers between July 1962 and December 2003 and included tests of analysis for accident cause, accident location, firm type, and firm size. They found an average price stock drop of 2.8% on the day of the accident, and another 1.08% the following week for airlines, for aircraft manufacturers a cumulative price drop of around 1% was found. The effects remained negative and significant for six month after the crash, but no long-term effect was found. McWilliams & Siegel (1997) rightfully point out in their article that long event windows increase the chance of confounding effects, however Walker et al. (2005) do not make any reference about controlling for confounding effects. A long event window as used by the authors (24 months) cannot be reconciled with the key assumption underlying the event method, that of market efficiency. The usage of long event windows reflects disbelieve that the efficient market holds, if such is the case then the event study method should not be used at all. In addition, the effect of several other characteristics was researched. The following characteristics had significant effects on the reaction: location (accidents within U.S. territory leads to larger declines), fatalities (more than 100 fatalities caused -7% opposed to -3.3% in the case of less than 100 fatalities), airline size, and criminal activity (such accidents have larger declines). Although the long-term results of this study should be disregarded in my opinion, the short-term results are sound.
16 literature here discussed will be used as input for deriving my hypothesis, with the exception of the article of Barrett et al. (1987).
2.1.2 Investor behaviour
To define investor behaviour the first step is to define who are considered investors. This is not as simple as it sounds, investors are heterogeneous group of stock market participants. Research (Barber and Odean (2000,2001); Locke and Mann (2001)) confirms that the response of investors to news is not uniform. Ekholm (2006) was able to research how different types of investors in Finland react to new earnings information by usage of the official register of shareholding. He finds that large investors show behaviour opposite to that of the majority of investors. One of the problems researchers face is that standard stock market data do not allow to test how different investors types react to new earnings information. As a result I will work under the assumption as stated by Fama (1970), that one condition for market efficiency is that all investors „agree on the implications of current information for the current price and distributions of future prices of each security‟. In other words, I will work under the assumption that investors are a homogeneous group of market participants. The behaviour of this group can be observed in the market place via the price stock reaction to news. To fully assess the behaviour this price stock reaction should not only be monitored but also the news that reaches the market. If for instance, the news of an accident the market, and that the management doesn‟t foresee any adverse effects because of adequate insurance coverage reaches, the behaviour of investor will be different then when adequate insurance is not in place.
At the basis of the relationship between investors and a corporation lies the separation of ownership and control, which was recognized by Smith (1838) as early as the 18th century. Berle & Means (1932) confirmed that as countries industrialized and their market developed, ownership and control were separated. This basis has lead to three theories in the academic literature that are relevant when considering the relationship between investors and a corporation: the agency theory, the transaction cost economics theory, and the stakeholder
theory.
17 and the directors of the corporation are the agents. There two sources of potential conflict in the agency theory: the agency problem, and the problem of risk sharing. The agency problem arises when the desires or goals of the principal and agent conflict and it is difficult or expensive for the principal to verify what the agent is actually doing. The problem of risk sharing arises when the two parties have different attitudes toward risk. The problem here is that the principal and the agent may prefer different actions because of the different risk preferences. When an accident occurs information asymmetry can exist between the principal and the agent. Although air crashes might get extensive media coverage, the agent might have more reliable information on the crash, and might better oversee the consequences the crash for the company. The Securities and Exchange Commission (SEC) requires companies listed on U.S. stock exchanges to disclose the material economic consequences. However, in the in between period (such disclosures appear some time after the accident) information asymmetry will exist between the agent and the principal. The agent will have a better insight on how the accident has affected the company. Using this information for an ex post assessment of investor behaviour in the case information asymmetry between the principal and the agent, might not give sound insight if investors reacted more fiercely to news of an accidents since 9/11. Consequently, results of this method should be interpreted with care.
Two streams can be distinguished in the agency theory literature: the principal-agent research, and the positivist stream. The former being the stream concerned with describing the general principal-agent relationship that can be applied on different relationships. The latter is of the most interest to my research since it is almost exclusively concerned with the relationship between owners and manager of large corporations. The positivist stream focuses on situations were the principal and the agent have conflicting goals (agency problem) and finding solutions that limit the agent‟s self-serving behaviour.
18 may be of overriding importance to the airline. This can lead to a focus away from shareholder value. Marcus & Goodman (1991) investigate the conflict between shareholder interest and the interest of victims in case of an accident. The corporate policy that a company can undertake a reaction to an accident may take two directions: an accommodative reaction (a statement in which management accepts responsibility), and a defensive reaction ( a statement in which management insists that the problem does not exist).
The main implication for my research that arise from the above discussed theory is the information asymmetry that can exist between the principal and the agent in case of an accident, and that shareholder interests might not be of paramount concern when an airline considers it reaction to an aviation accident. Because information asymmetry may exist results of ex post assessment of investor behaviour should be interpreted with care. What becomes clear from the stakeholder theory and subsequent research is that the reaction of an airline to an aviation accident does not have to be in the best interest of shareholders.
19
2.1.3 The terrorist attack of September 11th, 2001
The terrorist attacks of September 11th, 2001 were committed by 19 Islamic extremist affiliated with al-Qaeda. The extremist hijacked 4 American commercial airlines (American Airlines Flight 11, American Airlines flight 77, United Airlines flight 93, and United Airlines flight 175) each loaded with about 90.000 litres of jet fuel and used them as rockets. Three of these for hit their intended targets (both WTC towers, and the pentagon) the fourth jet crashed in the southwest of Pennsylvania. The motive behind these terrorist attacks is, according to the United Stated government, a holy war against the United States as declared by Osama bin Laden. Together with hijackers, 2.992 people died as an immediate result of the attacks.
The terrorist attacks also had immediate economic effects. The U.S. financial markets were closed until September 17 due to the damage to the New York City financial district. When the markets reopened some sectors were hit hard with stocks prices falling (insurance, airlines and aviation, and tourism) while other sectors stock (communications, pharmaceuticals, and military/defence) rose. For the U.S. aviation industry the financial consequences were unprecedented (Government Accounting Office, 2001). These led to a government aid package for the U.S. airline industry. The Air Transportation Safety and System Stabilization Act, which provided $15 billion of aid, was signed into law on September 22.
Table 2: Review of literature on the effects of 9/11
author key questions area observations period model key hypothesis key results
Carter & Simkins (2004)
airline price stock reaction to 9/11 US & international airlines, US freight firms 29 September 17-24, 2001 multivariate regression model
significant abnormal returns in response to 9/11
Rational pricing hypothesis holds, significant negative abnormal returns for each of the airlines studied and
smaller negative returns for airfreight firms and international airlines
Chen & Siems (2004)
the effects of terrorism on global capital markets
US & global capital markets 14 for US, 2 for global capital markets 1915-2002 event study (mean adjusted returns model)
significant negative abnormal returns in US/global capital markets after terrorist
and military attacks
U.S. capital markets are more resilient than in the past and recover sooner from terrorist attacks than other
global capital markets.
Drakos (2004)
the effects of terrorism on airline stocks US & international airlines 5 US & 8 international airlines 12/07/2000 to 26/06/2002 adapted event study
9/11 did not affect the systematic/idiosyncratic/total risk of
airline stocks
systematic/idiosyncratic/total of airline stocks has significantly
increased since 9/11
Ito & Lee (2005a) the impact of 9/11 on airline demand US 216 1986-2003 reduced form model of demand
There is an ongoing shift in demand in the U.S. airline industry. And there is transitory shock in the demand in the U.S.
airline industry.
9/11 resulted in both a transitory, negative demand shock of more than 30%
Ito & Lee (2005b)
the impact of 9/11 on airline demand
Australia, Canada, Europe, Japan, and
the US
around 2160 1986-2003
reduced form model of
demand
An ongoing downward shift in the demand for air travel resulting from increased fear, and an initial panic drive fear of flying
directly following 9/11.
All countries/regions in our analysis suffered significant declines (between -15% and 36.5%) in international air travel demand as a result of 9/11, the
Since the effects of 9/11within the aviation industry are widespread I will restrict the in-depth analysis to the effects which bear relevance to my research. Table list a review of the relevant literature. Drakos (2004) used a method that is not formally an event study, but does bear considerable resemblance to it. The author researched the effects of 9/11, to the extent that terrorism was perceived as adverse to demand lowering airline profits and consequently dividends, on airline stocks over the period 12-07-2000 until 26-06-2002. I must be noted here that this research focused on the short term effects of terrorism on the airline industry. For each stock, the overall risk can be decomposed into two components: systematic and
idiosyncratic (Sharpe, 1964; Litner, 1965). The distinction between both being that systematic
risk cannot be diversified away, and idiosyncratic risk can be eliminated by effective diversification (also defined as an asset return‟s co variation with the market portfolio return, known as beta). Drakos found that the systematic risk of airline stock has significantly increased since the terrorist attacks, also volatility has dramatically increased in the post 9/11 period. By decomposing total risk into its constituents (systematic and idiosyncratic), the findings suggested that systematic risk has on average more than doubled (as measured by beta), while the percentage it represents over total risk has shown a considerable increase. In my opinion, the article of Drakos (2004) and its findings are robust. The large increase in the systematic risk found clearly indicates that the attitude of investors towards airline stocks has changed. Together with the finding that idiosyncratic risk of airline stock has increased, the behaviour of investors does appear to have changed after 9/11.
22 airlines. This suggests that the market believed the long-term consequences of the attacks were much more significant for U.S. airlines than for either international carriers or airfreight firms. In addition, they found that investors distinguished between airlines based on the level of their cash reserve, suggesting according to the authors, that the market was concerned about the airlines‟ ability to survive a prolonged downturn in air travel. The authors also had examined the period during which the Air Transportation Safety and System Stabilization Act was passed and signed into law. Their results suggest the market believed that the major airlines benefited while smaller airlines did not, they also find evidence that the market had concerns for airlines that were more heavily involved in international travel. The final result from the research of the authors is that their results are important because they support rational pricing in the U.S. stock markets following the September 11th crisis, particularly given the psychological impact of the attacks. Although the authors did not mention the increased liability risk to which airlines are exposed their conclusions are still sound. This is because the authors linked their findings to how investors reacted to the events of 9/11 and what firm specific characteristics affected this reaction, and not linked it to directly to expected loss consumer demand.
Ito & Lee (2005a) assessed the impact of 9/11 on the U.S. airline demand. The basic methodology used was a reduced form model of demand for domestic air services using monthly time-series data from 1986 until 2003, this method controlled for cyclical, seasonal and other unique events impacting the industry. The authors found that 9/11 led to both an initial demand shock of more than 30% as well an ongoing downward shift in the demand for commercial air service of roughly 7.4%. However, the authors acknowledged some limitations to their results. The post 9/11 results are based on 27 observations (which limits the degrees of freedom of the post 9/11 analysis) and the recovery of 9/11 is still in progress. In my opinion the authors wrote an meticulous article with reliable results.
23 effects that 9/11 had on the aviation industry. This will serve as useful input for deriving my hypothesis.
2.2 Derivation of the method and conceptual model
With the research question and its underlying concepts in mind, an appropriate method needs to be selected in order to answer the research question. I will start of by defining the requirements that my method needs to satisfy. Because of the three characteristic concepts in my research question my methods needs to satisfy the following characteristics:
a) it needs to be able to measure the investor reaction over multiple events
b) it needs to take into account the time span in which the investors update their valuation of the company under consideration
c) it needs to be able to make an comparison between before and after 9/11
To gain insight into investor reactions to aviation accident their behaviour to the news of an accident needs to be determined. In order to get an accurate picture of the behaviour of investors in such a case is to evaluate their reaction of multiple events. The reaction of investors can be measured along three dimensions: the sign of the reaction (positive or negative), the size of the reaction, and the length of the reaction. When these three dimensions are all measured over multiple reactions to aviation accidents, investor behaviour can be accurately characterized. By comparing the characteristics of investor behaviour to aviation accidents before and after 9/11, the research question can be answered.
In light of these characteristics, I have selected the event study as an appropriate method. The event study as it is currently used stems from Fama, Jensen, & Roll (1969). The event study method can measure all three dimensions of investor reaction. The method also allow to measure the investor reaction to multiple events, allowing characterization of investor behaviour, and by implementing a two sample t-test into this method a comparison of the investor behaviour before and after 9/11 can be made.
24 which means that in conceivably can affect the relationship between the dependent and independent variable. The figure also displays the intended method of measuring the different variables. They are the result of the research question and the event study method that is selected to answer this question. The independent variable is measured by the date on which the accident is occurred. On this date a effect is expected in the dependent variable, so on the date of the accident investor reaction is expected and this will be measured by the stock price reaction of the crash airline. The extraneous variable is assumed to influence this relationship. To measure this two sub samples will be constructed with accidents that occurred before and after 9/11. Performing a two sample t-test allows to compare the stock price reaction before and after 9/11.
Figure 1: Conceptual Model
Dependent Variable Independent Variable Extraneous Variable Investor Reaction Aviation Accidents 9/11 Stock Price Date of Accident Pre vs Post Concept Measurement Variable
2.3 Derivation of the hypotheses
25
Table 3: findings of previous studies
authors sample
initial reaction α
cumulative
reaction days α
Chance & Ferris, 1987 all airline crashes (46) -1,17% 0.05 - - -
Barrett et al., 1987 all crashes (57) negative4 0.5 - - -
Borenstein & Zimmerman, 1988 all crashes (67) -0,94% 0.01 -0,97% 2 0.05 Mitchell & Maloney, 1989 pilot error crashes (34) -3,80% 0.01 -2,62% 6 0.01
Bosch et al., 1998 all crashes (22) -1,17% 0.05 -2,67% 3 0.05
Walker et al., 2005 all airline crashes (138) -2,80% 0.001 -3,88% 7 0.001
Table 3 lists all the results of event study research into aviation accidents. The research of Davidson et al. (1987) is excluded because of ambiguous results. Previous research clearly shows that there is a relationship between the behaviour of investors and aviation accidents. If an aviation accident occurs and this news reaches the market the price of the stock of the airline company falls because investor expect lower dividends and hence want to pay less. The behaviour of investors in case of an aviation accident is clear-cut, there is a negative initial price stock reaction. This means that the price paid at the stock market is lower than the day before. The size (measured in percentages) of the reaction found by previous literature varies from 0.94% (Borenstein & Zimmerman, 1988) up to 2.80% (Walker et al., 2005). Mitchell & Maloney (1989) find a large reaction but this due to the specific nature of their sample, it contained only crashes caused by pilot error. Other factors that influence the size of the reaction are the location of the crash, number of fatalities, if it caused by criminal activity (Walker et al., 2005). With my first hypothesis, I want to confirm the results on the initial price stock reaction that earlier studies have found. Since my sample has not been selected based on specific characteristics, like the ones mentioned above, I expect a significant initial price stock reaction that is negative, since all previous research has also founds such a reaction. Based on previous research the range of this initial reaction should be between -0.94% and -2.80%. Since the research of Walker et al. (2005) is the most recent and also employs the largest sample over the longest time period I expect the initial price stock reaction to be closer to -2,80% then to -0.94%. My first hypothesis is:
H1: The airlines initial price stock reaction of investors to the news of an aviation
accident is negative.
26 The cumulative reaction to an aviation accident measures reaction of the price over several trading days starting at the moment that the news of the accident reaches the market. Table 3 lists both the size and the length of this cumulative reaction. With my second hypothesis, I want to confirm the results on the cumulative price stock reaction that earlier studies have found. The findings of previous research are less clear-cut about the cumulative reaction to aviation accidents. Two articles do not find a cumulative reaction and the four other articles do find a negative cumulative reaction. The two articles that did not find a cumulative reaction are the most dated ones. The finding of Walker et al. (2005) might be the result of a wrongful application of the event study method and the results from Mitchell & Maloney (1989) are for a sample with specific characteristics. The two remaining articles both find a significant cumulative reaction with a length of around two/three days and a size between 0.97% and 2.67%. I expect a reaction that is in line with these two articles, therefore, my second hypothesis is,
H2: The airlines cumulative price stock reaction of investors to the news of an accident
is negative.
27 Drakos are a relative short period after 9/11, but I expect that the events of 9/11 had a long-term effect on investor behaviour. If this is the case I expect that this changed investor behaviour also manifests itself in the relationship between investor behaviour and aviation accidents. Since the relationship between investor behaviour can be measured by both an initial reaction and a cumulative reaction, my next hypotheses are,
H3: The airlines initial price stock reaction of investors to the news of an aviation
accident is higher post 9/11 then pre 9/11.
H4: The airlines cumulative price stock reaction of investors to the news of an aviation
28 3. METHODOLOGY
3.1 Introduction
The two conceptual models that I have derived have as implication that I will be using two methods to test the hypothesis. I will use the event study method to test the four hypotheses. For the first two hypotheses, a straightforward application of the event study method is used. For the third and fourth hypothesis an extraneous variable is inserted into conceptual model, this variable will be used as control variable in a two sample t-test.
3.2 The event study method
The event study is a method that allows the measurement of firm-specific events on the price of the stock of the firm. The method works has several assumptions that need to be satisfied in order to ensure reliable results (McWilliams & Siegel, 1997):
1. Markets are efficient.
2. The event under research was unanticipated
3. There are no confounding effects during the event window.
Furthermore, there are critical issues that need to be addressed when using an event study method:
1. Sample size
2. Correcting for outliers
3. The length of the event window and its justification
I will discuss the above-mentioned assumptions and issues and the effect they have on how I will execute my research below.
The main underlying assumption of the event study approach is the Efficient Market Hypothesis, which states: “an efficient financial market as one in which security prices always fully reflects the available information (Schleifer, 2000)”. In effect this mean that as soon as new information relating a company comes available this information in instantly used update expectations about current and future performance. Therefore, when new information about a company, for instance an accident, becomes available the company‟s stock price should respond immediately and reflect the updated valuation of investors.
29 to be efficient. In order to ensure that the reaction of the stock market to the news is measured correctly it is important that news is actually new to the market. It is possible that an event is anticipated or news is leaked before the official announcement. In such cases, the stock market has already started to react to event before it took place. The event that I investigate is an aviation accident. These events are clearly unanticipated and conform to this assumption.
The final assumption underlying the event study method is that there are no confounding effects during the event window. Confounding effects are other events that might have an impact on the share price during the event window. Examples of such events are the declaration of dividends, merger announcements, product announcements, earnings announcements. When such an event occurs during the event window the effect of both the event and the confounding effect are measured. Therefore, it is necessary to check for confounding effects to ensure that only the event is measured. I will control for confounding effects by checking LexisNexis NewsPortal if any events might have taken place during the event period that meet the criterion of a confounding effect.
The first critical issue, sample size, is a concern because the test statistics used in the event study framework are based on normal distribution assumption associated with large sample. Descriptive statistics can confirm if the sample if the normal distribution assumption is met. If this is not the case, an alternative method like a bootstrap method should be used in order to control for the sample size. However, this method can only be applied certain instances of the event study method. Next to this, I will perform a post hoc power of analysis to evaluate if the number of observations has implications for the results of the testing procedure.
Another concern raised by the authors is the correction for outliers. The test statistic used in event studies is sensitive to outliers, especially since in a small sample the impact of one firm‟s observations is much greater than in the large sample case. Therefore, it is important to asses if the results are driven by outliers and if so to adjust the research method to this influence into account. In order to control for outliers I will implement a testing procedure that attributes less weight to observations of firms with a high variance in returns. This testing procedure can be used next to the normal testing procedure for event studies and is less sensitive to distortions from very noisy observations. A comparison of the results from both testing procedure will provide insight if the results are driven by outliers.
30 will reduce the power of the test statistic, which in turn can lead to false inferences about the significance of an event (Brown & Warner, 1980; 1985). Secondly, long event windows increase the chances of confounding effects. The nature of the event should be determinant of the event window used. The longest cumulative reaction that is found within academic research is 6 days (Mitchell & Maloney, 1989), except for in the research of Walker et al. (2005) but their findings are ambiguous. I will use an event window of ten trading days in my research, which is four days more than the longest cumulative reaction, to ensure that window is long enough. I do not expect longer cumulative reaction.
3.2.1 Data sources
This research will rely on secondary data, this is because of the methodology that I selected to answer my research question. I will use three sources to collect data from: Datastream, www.aviation-safety.net, and LexisNexis NewsPortal. Datastream is database containing daily stock price data for companies throughout the world and is frequently employed in finance literature. The second data source, www.aviation-safety.net, is also a source that was used in earlier research (Walker et al., 2005). The authors cross-referenced this database together with two other databases (www.airdisaster.com and www.airdisasters.co.uk) and found no inconsistencies. Another source that is frequently used is the database of the NTSB, however since this databases is not focused on all aviation accidents worldwide it is not appropriate. The LexisNexis NewsPortal is the third source I will use. It is used in order to control for confounding effects that can influence investor behaviour.
3.2.2 Sample selection
The sample that I will be working consists of publicly quoted airline companies that experienced an aviation accident in the period June 1996- July 2007. To construct the sample the following criteria need to be met:
1. The airline needs to be publicly quoted at the time of the accident as well as the estimation period.
31 5. The accident should be the results of the commercial operation of aircraft by
the airline.
The first criterion for inclusion in the sample is straightforward. Since the reaction of investors is researched in this thesis, the airlines that are included in the sample should be publicly quoted. This is because of the measure selected to measure the variable investor behaviour.
The second criterion is product of the market efficiency assumption underlying the event study method. Since in an efficient market security prices always fully reflect the available information, the event that is being studied needs to contain information that is relevant and reaches the market. Using the event study method it is crucial that news actually reaches the market, if this is not the case not the reaction to the event is measured. In order to ensure that the news of the aviation accident actually reaches the market only cases were fatalities occurred are considered. This is a usual constraint employed in this avenue of research (i.e. Barrett et al., 1987; Borenstein & Zimmerman, 1988; Bosch et al., 1998; Dillon et al., 1999; Walker et al., 2005). Although some authors (Bosch et al., 1998) only research cases with at least one on board fatality, in my opinion this distinction should not influence the results. In both cases, the airline will be liable and it does not affect the chance of the news reaching the market in my opinion.
The accident under consideration should not be caused by criminal occurrences is a criterion to ensure that the appropriate investor reaction is measured. Although in the definition of the NTSB of an accident acts of terrorism are also considered an accident, in my opinion these cases should be excluded. These occurrences are the result of unlawful acts by third parties. The measurement of investor reaction to unlawful acts airlines is not the scope of this research. The investor reaction that is under investigation is that to aviation accidents that fall under the normal risks of the aviation industry. Another reason for excluding these cases is that will introduce a bias into my sample. The investor reaction to an accident caused for instance an act of terrorism can result in outliers within the sample that will influence the results.
32 The final criterion is to ensure that only airline accidents are included where the airline is performing its core service, providing transportation to customers. Although other accidents related operating an airline could influence the performance of the company, the goal is to have homogenous group of accidents. This group should therefore exclude cases like pilot training accidents, and accidents with cargo aircraft.
The method of sampling employed, where sample members are selected to conform to specific criteria, is known as purposive judgment sampling (Cooper and Schindler, 2001). One of the problems associated with a non-random sampling procedure is that it can lead to a selection bias. However, the criteria above have identified several biases that could influence the results as to control for these. In table A1 (see appendix) the sample obtained is presented together with data on the date of the event, aircraft and registration, airline, and fatalities (both ground and on board fatalities), if the news has reached the market, and confounding effects. Of the 38 crashes, three crashes cannot be included in the sample. The news of the Singapore airlines crash of 21st of July 2007 did not reach the market, the crashed airplane was a training aircraft probably this is reason that this news did not reach investors. The crash of Korean Air on the 22nd December 1999 is not included because of a confounding effect (other accident at Korean Air) in the estimation window. The final crash that is excluded is the crash of British Airways on the 5th of September 2001, this is due to the confounding effect of 9/11 in the event window. This has resulted in 35 crashes in my sample of which 21 occurred before 9/11 and 14 after.
3.2.3 Variables
33 𝑅𝑡 = 𝑃𝑡
𝑃𝑡−1− 1 [1]
With Pt and Pt-1 being the price of asset at date t and t-1 respectively, and Rt being the simple net return.
The independent variable, which is the aviation accident, as I will use it in this method determines the date on which investor behaviour is measured. The date on which the investor behaviour will be measured depends on when the news of the aviation accident reaches the market. Table A1 (in the appendix) list the dates on which the accidents took place as well as the local time. If the accident took place an hour before closing of the relevant market the event date is defined as the trading day after the accident. In case the accident takes place at a time when the relevant exchange is closed, the event date is the first trading day after accident.
34 All other accident listed in table A1 are included in the sample. Other information listed in the table is the crash airline and its parent company, number of fatalities both on board and on the ground
The extraneous variable that I employ within my conceptual model, which is 9/11, will be my control variable. The control variable will be used in the testing procedure that I will use for testing the third and fourth hypotheses. Working under the hypothesis that the events of 9/11 affect investor behaviour I two sub samples are formed one pre 9/11 crashes and one for post 9/11 crashes. To test if the returns for both sub samples are different a two sample t-test will be used.
3.2.4 Methodology
The event study approach works as following (MacKinlay, 1997). To value an event‟s impact there is a need for a measure of Abnormal Return (AR), this is the actual ex post return of the security over the event window minus the normal return of the firm over the event window. This normal return is the return that would have been earned when the event had not taken place. So,
𝐴𝑅𝑖𝑡 = 𝑅𝑖𝑡 − 𝐸 𝑅𝑖𝑡 𝑋𝑡 [2]
With ARiτ, Riτ, and E(Riτ|Xτ) are the abnormal, actual and normal returns respectively for time period t.
Figure 2: Time line of the event study
-230 0 10
T0 T1 T2 T3
Estimation window Event Window
-30
Figure 2 depicted above shows the time line for my event study. The numbers on the horizontal scale represent the number of trading days from the event date. For both the market
model and the and the constant-mean-return model an estimation window needs to be decided
on. The estimation window, which is from T0 until T1, is the time period over which the two
35 and T3 are listed in table A1). The parameters that stem from either the market model or the constant-mean-return model can then be used to calculate the abnormal returns for the event
window.
The normal return is commonly measured in two ways. The first method is the
constant-mean-return model where the normal return is a constant.
𝑅𝑖𝑡 = 𝜇𝑖 + 𝜉𝑖𝑡 [3]
𝐸 𝜉𝑖𝑡 = 0 𝑣𝑎𝑟(𝜉𝑖𝑡 ) = 𝜎𝜉2𝑖
With μi being the mean return for asset i, Rit is the period-t return for security i and ξit is the time period t disturbance term for security i with an expectation of zero and variance 𝜎𝜉2𝑖. Using μi as input in formula [2] the measure for abnormal return can be calculated,
𝐴𝑅𝑖𝑡 = 𝑅𝑖𝑡 − 𝜇𝑖 [4]
with 𝐴𝑅𝑖𝑡 being the abnormal period-t return for security i, the abnormal returns of interest are over the period T2 until T3, so this results in 30 values for the 35 airline companies in my
sample.
From this step onwards the method is the same of the constant-mean-return model and the market model. Therefore I will first describe how to calculate the abnormal return using the market model. Then I will continue with the following steps for the event study approach which are the same for both the constant-mean-return model and the market model.
The second method is by usage of the market model where the normal return depends on the market return, this relation is tested over period prior to the event. For any security i the market model is
36 where Rit and Rmt are the period-t returns on security i and the market portfolio, respectively, and εit is the zero mean disturbance term. αi, βi and 𝜎𝜖2𝑖 are the parameters of the
market model. Employing the market model, the abnormal return can now be calculated as follows,
𝐴𝑅𝑖𝑡 = 𝑅𝑖𝑡 − 𝛼𝑖 − 𝛽𝑖𝑅𝑚𝑡 [6]
Brown and Warner (1980) have investigated the differences in results from employing both methods and find that: “A simple methodology based on the market model performs well under a wide variety of condition. In some situations, even simpler methods which do not explicitly adjust for market wide factors of risk perform an no worse than the market model.” So the choice for either the constant-mean-return model or the market model can be justified from this research since it is questionable if more complicated models yield better results.
With the abnormal return calculated the next step is aggregation. This can be done along two dimension, time, and securities. Aggregation across securities gives insight in the average response of investors to accident included in the sample. Aggregating across time is done by summing up the abnormal return of that day and all previous days in the event window, this gives insight into the total reaction to the aviation accident up to that day. The first hypothesis requires an aggregation of all securities. For the second and fourth hypothesis, the returns need to be aggregated along both time and securities. To test hypotheses one I need the average abnormal return (the abnormal return aggregated across securities). So,
𝐴𝑅 𝑇2, 𝑇3 = 1 𝑁 𝐴𝑅𝑖(𝑇2, 𝑇3) 𝑁 𝑖=1 7
With 𝐴𝑅 being the average abnormal return over the event window 𝑇2, 𝑇3 , and 𝑁 being the
total of securities aggregated in the group.
37 𝐶𝐴𝑅 𝑇2, 𝑇3 = 𝐴𝑅 (𝑇2, 𝑇3) 𝑇3 𝑠=𝑇2 8
with 𝐶𝐴𝑅 being the average cumulative abnormal return for security over the event window 𝑇2, 𝑇3 .
The third hypothesis requires
3.3.2 Testing procedures
In order to test the first hypothesis, to test whether the average abnormal return is significant (so if the return can be considered abnormal) I will use the t-test. This method assumes that 𝐴𝑅 is independent and identically normally distributed through event time, if this is not the case the t statistic will overstate the true value. With 36 observations a normal distribution can be assumed. The t-test is computed as following,
𝑡 = 𝐴𝑅 𝑡
𝑆( 𝐴𝑅 𝑡) 9
𝐴𝑅 𝑡 is the day t return of the average abnormal return and 𝑆( 𝐴𝑅 𝑡) the standard deviation of the day t average abnormal return. 𝑆( 𝐴𝑅 𝑡) is estimated over the estimation window,
𝑆( 𝐴𝑅 𝑡) = 𝐴𝑅 𝑡− 𝐴 2
𝑡=−30
𝑡=−230
200 [10]
38 The t-statistic can now be calculated by using formula 9. The first hypothesis is that the initial
price stock reaction of investors to news of an aviation accident is negative. Therefore, the
null and alternative hypotheses are,
𝐻10: 𝐴𝑅0 ≥ 0 𝐻1𝑎: 𝐴𝑅0 < 0
In order to accept the alternative hypothesis the t-value should be below -1.96.
The analysis of power, which is the probability that the test will reject a false null hypothesis, also needs to be taken into consideration. Since my research is limited to the number of observed aviation accidents in the specified period, a will use a post hoc power analysis. I will use the observed abnormal return and the sample size to determine the power. The general formula for testing for the power analysis is,
𝑧 = 𝑥 − 𝜇
𝜎 𝑛 [12]
I will report the power of analysis for all hypotheses with the results of the testing procedures In line with Borenstein & Zimmerman (1988), my second significance test is based on the individual t-statistic for each firm. The test calculates the t-statistic for each abnormal return (which follows an asymptotically distribution normal with mean zero and variance equal to the number of observations), these are summed up and divided by the square root of the number of observations, giving the z-statistic. The advantage of this test is that it attributes less weight to high variance observations of firms, and therefore is less sensitive to distortions.
𝑧 = (𝐴𝑅 𝑇𝑁1 1, 𝑇3 𝜎𝐴𝑅 𝑇0,𝑇1 )
𝑁 13
For the alternative hypothesis to be accepted at a 95% confidence level, the z-value should be lower than -1.645.
39 of the day t average abnormal return as calculated in formula 10 is used. The denominator of the test statistic represents the standard deviation of the average cumulative abnormal return.
𝑡 = 𝐴𝑅 𝑡 𝑡𝑐𝑎𝑟 𝑡=0 𝑆2( 𝐴𝑅 𝑡) 𝑡𝑐𝑎𝑟 𝑡=0 14
𝑡𝑐𝑎𝑟 represents the end of the period over which the car is calculated. The second hypothesis, the airlines cumulative price stock reaction of investors to the news of an accident is negative,
Translates into the following hypotheses,
𝐻20: 𝐶𝐴𝑅𝑡 ≥ 0
𝐻2𝑎: 𝐶𝐴𝑅𝑡 < 0
With 𝐶𝐴𝑅𝑡 being the average cumulative abnormal return for day t. In order to accept the
alternative hypothesis a t-statistic of -1.96 or lower is required.
The third hypothesis is based on the comparison of the two means from two populations. Since this is a small sample case with 21 observations pre and 15 observations post 9/11, a sample test about the difference between the means of two populations with independent sample needs to be performed. The drawback of such a test statistic is that it assumes equal variances for both populations, I will use the variance of abnormal return for this test, 𝑠𝑎𝑟2 .The
test statistic for the small-sample case is
𝑡 = 𝑥 𝑝𝑟𝑒 − 𝑥 𝑝𝑟𝑒 − 𝜇𝑝𝑟𝑒 − 𝜇𝑝𝑜𝑠𝑡 𝑠𝑎𝑟2 𝑛1 𝑝𝑟𝑒 − 1 𝑛𝑝𝑜𝑠𝑡 [15]
𝑥 is sample mean of the relevant period, and 𝜇 is the population mean. The third hypothesis,
the airlines initial price stock reaction of investors to the news of an aviation accident is higher post 9/11 then pre 9/11, translates in the following hypotheses,
𝐻30: 𝐴𝑅𝑝𝑟𝑒 ,0− 𝐴𝑅𝑝𝑜𝑠𝑡 ,0 ≥ 0 𝐻3𝑎: 𝐴𝑅𝑝𝑟𝑒 ,0− 𝐴𝑅𝑝𝑜𝑠𝑡 ,0 < 0
