1 The effect of air crashes on plane manufacturers value.
Abstract.
The effect of air crashes on the airline industry is widely discussed in previous literature. The effect on the manufacturer of the plane is a less commonly discussed topic and this paper will examine this. There is literature available that discusses the effect of air crashes on the manufacturer of the plane; however, this literature is outdated and was based on small samples. This paper uses two models to test the hypothesis that plane manufacturers are negatively affected in their value because of air crashes. Those models are the capital asset pricing model and the market model. Also there will be tested if there is a difference in effect for crashes before the year 2000 and after the year 2000. This study finds that air crashes have no significant effect on the returns of aircraft manufacturers when using the entire sample. The average abnormal returns are not significant for both models. Moreover, when comparing the subsamples with the two different models the test results do not show any difference. The value of plane manufacturers is not negatively affected by plane crashes, regardless of when they happened.
Bachelor thesis for E&BE, BE
Jochem Loonstra, S2543427
Rijksuniversiteit Groningen
2 1. Introduction
Ever since the beginning of aviation in human history there have been accidents. Accidents involving aircrafts often create horrible scenes because they frequently lead to loss of human life. A loss of human life is not something that cannot be accounted for or compensated.
Airline companies suffer, as described by Davidson, Chandy, and Cross(1987), in at least three ways. Firstly, the damage to the aircraft itself, mostly beyond repair. Secondly, the compensation for physical injury or death to the passenger. Lastly (and perhaps the most important one) is the loss of goodwill, often resulting in lower demands for the industry or switching to different airlines. That the airline suffers major losses due to an aviation disaster is a commonly known fact. Whether the manufacturers of the involved planes also suffer from a decrease in value will be tested in this paper.
Aviation is an industry that is sensitive to political and economic changes. Therefore, a change in the value of companies involved in air crashes is something that can be expected.
The hypothesis that the value of the involved airlines decrease after air crashes has also been shown by research. Chance and Ferris(1987) proved that there is a decrease in stock value of the carrier right after a crash. On average, the manufacturer of the airplane has less harm of a crash, this is because there are less substitutes for them. A consumer can easily switch between airlines, but plane manufacturers are less sensitive to switching as there are less options, and often there are contracts involved. Commercial aviation has only been around since 1914 and people were more sceptical towards flying in the early days. The aviation industry has developed a lot in the past decades and has become more widely available. Scepticism has decreased and rules for safety have increased. Therefore, we will also test if investors reacted stronger to aviation disaster before the year 2000 than after the year 2000.
In this paper the effect of air crashes on the returns of plane manufacturers will be tested,
by the use of event study methodology. The hypothesis which will be tested, is that the
value of plane manufacturers is negatively affected because of plane crashes. The second
3 hypothesis that will be tested is the one that crashes that occurred before the year 2000 have more negative effect than crashes after the year 2000.
2.1 Literature review
Literature on the effect of air crashes on the value of companies involved is easily available.
Three papers are very relevant to my paper because of their similarity in topic and use of event study methodology. These papers are are Chance and Ferris(1987), Davidson, Chandy, and Cross(1987) and Walker, Thiengtham, and Yi Lin(2005).
Chance and Ferris(1987) examined the effect of air crashes on the value of the two companies that are most involved in an air crash; the manufacturer of the plane and the carrier. Their samples consist of crashes that occurred in the period 1962-1987, a period in which aviation was still developing on a large scale. Their main findings is that the stock market immediately reacts to an air crash by a significant negative abnormal returns of 1.2%
on average. This effect does not continue after the day of the crash suggesting that the stock market only reacts on the economic impact on one trading day. They do not find significant effects on the value of the plane manufacturer after a crash. As an explanation for this they argue that investors apparently see aviation disaster as a carrier specific event, that has no financial relevance for the manufacturer.
Davidson, Chandy, and Cross(1987) also examined how the value of airlines is affected by air crashes. They also use event study methodology and have a sample of 57 crashes. They also examine whether the most severe crashes in their sample have a more significant effect than the less severe ones. They find that, when using all crashes in their sample, the value of the airliners decreases by 0.785% on average. They explicitly mention that the negative effect is reversed within 5 trading days after the crash.
Walker et al.(2005) also examine the impact of plane crashes on the value airline and
manufacturer. They also test the effect of different causes and severity of the crash. They
find that investors react to a crash by a decrease in value of 2.8% and 0.8% respectively for
the carrier and the manufacturer. Moreover, they find that crashes which are caused by acts
of terrorism have a more significant effect and this is the same for crashes with a higher
4 number of fatalities. This counts both for manufacturers and carriers, so they do find a negative effect for manufacturers, in contradiction to the findings of Chance and Ferris (1987).
The effect of plane crashes on the carrier has been tested many times, and all literature agrees on the fact that it has a negative effect. When it comes to manufacturers of the planes, the literature is contradictory. Chance and Ferris(1987) use old data and have a small sample, they find no effect. Walker et al.(2009) on the other hand use a large sample that is up to date; however, they have included 9/11, which is more than a firm-specific event and they do find an effect. By using a large sample and leaving out 9/11 this paper hopes to give a decisive answer to the question whether plane manufacturers are also significantly affected by plane crashes. Thereby, the test to see if there is a more negative effect for crashes before than after the year 2000 will be conducted.
2.2 Hypothesis.
This paper will test two hypotheses.
Hypothesis 1: the value of plane manufacturers is negatively affected by air crashes.
Hypothesis 2: crashes that occurred before the year 2000 have more negative impact on the manufacturers value than crashes that occurred after the year 2000.
3. Data and methodology 3.1 Sample
The sample used for this study consists of 120 events, that occurred between 1963 and 2015. For this study it was necessary to have crashes from a wide range of time to make sure the effects for the entire sample are not time related. The first event used in this study was in 1963, one year after the Boeing went public for the first time. The sample was collected by working down a list of most severe crashes, provided by the aviation
network(https://aviation-safety.net/), an online database, that is updated every week and contains descriptions from over 15,800 aviation accidents and incidents since 1921.
The sample contains three different plane manufacturers, Boeing, Airbus and Lockheed.
Appendix 1 shows the event number, date, airline, manufacturer and number of fatalities of
the selected crashes. The subsample of crashes that happened before 2000 contains 82
5 events and the one subsample of crashes after 2000 has 38 events in total. The average number of fatalities was 131 and the average date of occurrence was 28-4-1992. The data on historical stock prices of the plane manufacturers was obtained from Yahoo Finance.
3.2 Methodology
This study uses event study methodology using two different models. These models are the constant mean return model and the market model.
According to MacKinlay (1997) the constant mean return model is the simplest model and uses an estimation window with the stocks on returns to obtain an expected return for the event window. Although it is a very simple model Brown and Warner(1980,1985) find that it often creates results similar to more complex models.
The market model works in a manner that it relates the return of any given security to the performance of the market portfolio. The models linear specification comes forth out of the assumed joint normality of the returns of the security and the market. In case the day of the crash was not a trading day, the first following trading day was used as the event day.
Every event study, no matter what model, uses abnormal returns to measure the economic impact of an event on the company’s value.
The return function used in this study is:
R
i,t= LN(P
i,t) – LN(P
i,t-1) (1)
In this formula R is the return on day t, given as a function of the adjusted closing price of stock i, on time t and t-1. The adjusted closing price is the stock price adjusted for dividends and stock splits. This function makes use of natural logarithm because it reduces problems of non-normality.
The formula for the abnormal returns looks as follows:
AR
i,t= R
i,t– E(R
i,t) (2)
In this formula R is the actual return of security i on time t and E(R
i,t) is the expected return
of security i on time t.
6 This study uses two different methods to obtain expected returns, the mean return of an estimation period and the market model which runs a regression using the market returns as the independent variable, and the actual returns as the explanatory variables.
Expected return in the constant mean return model is the mean return of an estimation period. This study uses a estimation period of 200 days.
The expected return function looks like this:
E(R
i,t) = (1/200) * ∑
200𝑡,𝑖=1𝑅𝑖, 𝑡 + e (3)
In this formula E(Ri,t) is the expected return from stock i, on time t, R
i,tis the actual return of security i on time t, and e is an error term. The 200 represents the 200 days of the
estimation window which ends 11 days before the event date.
The market model, on the other hand, uses a market portfolio to derive expected returns.
This study uses the S&P500 returns as a benchmark for the returns of the company. Hereby a regression was run using the S&P500 returns as the independent variable and the returns of the companies used in this study as the explanatory variable. The expected returns formula of the market model looks as follow:
E(R
i,t) = α + β R
i,t+ e (4)
In this formula E(Ri, t) is the expected return of security i, on time t. Furthermore α is the intercept obtained from the regression, β is the slope and e the error term. R is the return of the benchmark which is used, in this case the S&P500, on time t.
To test the average abnormal returns per day on normal distribution a Jarquebera test was used with the following formula:
𝐽𝑎𝑟𝑞𝑢𝑒𝑏𝑒𝑟𝑎 =
𝑁−𝑘6
+ (𝑆
2+
14
(𝐶 − 3)
2) (5)
In this test N is the sample size, k the number of regressors, S the skewness of the
distribution and C the kurtosis. The Jarquebera test tests the null-hypothesis of normal
distribution.
7 To observe whether the market reacts to a plane crash we are going to test if the average abnormal return of a day is different from zero. The average abnormal return on time t is given by this formula:
AAR
t= 1/N * ∑
120𝑖=1ARi, t (6)
Here, N is the number of observations used and AR
i,tis the abnormal return of security i on time t.
There is a good possibility that the stock market does not react significantly on the day itself but over the window of several days. Therefore this study also tests if the cumulative
average abnormal returns deviate from zero. The formula of the Cumulative average abnormal returns is given by:
CAAR
(t1 , t2)= ∑ 𝐴𝐴𝑅𝑡
𝑡1𝑡2(7)
In this formula the CAAR is given as the sum of the average abnormal returns in the period that starts at t1 and ends at t2.
Now, to test if the average abnormal returns for a specific day deviates from zero we need to calculate the variance of the average abnormal returns. The formula for this variance is given by:
VAR(AAR
t) = ∑
−11𝑡=−210((𝐴𝐴𝑅𝑡 − 𝑀𝐴𝐴𝑅)
2/(𝑥 − 1)) (8)
In this formula AAR
tis the average abnormal return on day t, MAAR is the mean of the average abnormal returns per day and x the number of days used in the calculations.
Assuming normal distribution a t-test can be used to test whether the average abnormal returns deviate from zero.
t = AAR
t/ (VAR(AAR
t))
^0.5(9)
We expect the abnormal returns to be negative, so a one-sided test is used. The degrees of
freedom for this test is N-1, so 119 here. The null-hypothesis of AAR
tbeing equal to zero can
be rejected if the t-value exceeds the critical values, in this case 1.984 for 95% confidence
level and 2.626 for 99% confidence level.
8 Now, for testing if the cumulative average abnormal returns deviates from zero we first need to obtain the variance of the CAAR. The formula for the variance of the CAAR is:
VAR(CAAR(t
1,t
2)) = ∑
𝑡1𝑡=𝑡2𝑉𝐴𝑅 (AAR
t) (10)
In this formula the variance of the cumulative average abnormal returns from t
1to t
2is obtained by adding up the variances of the average abnormal returns from t
1to t
2.
To test if the CAAR(t
1,t
2) is different from zero, assuming normal distribution, we use the following formula:
t = CAAR(t
1,t
2) / VAR(CAAR(t
1,t
2))
^0.5(11)
All formulas to test the statistical significance so far have assumed normal distribution.
Also, we test the hypothesis that crashes that happened before 2000 have a more negative effect on the manufacturers value than crashes that happened after 2000. We test the difference in means using the following formula:
t = MAAR
a– MAAR
b/ ((VAR^0.5(AAR
t) * √
1𝑁1
+
1𝑁2
) (12)
Here MAAR
ais the mean of the average abnormal returns per day of all crashes after 2000, MAAR
bis the mean of all average abnormal returns for crashes after 2000 and N
1and N
2are the number of observations used to compute the average. The Average abnormal returns here are for all crashes and all days in the event window. The test uses df = N
1+ N
2– 2 degrees of freedom. The VAR^0.5(AAR
t) term is the standard deviation of the entire sample for the average abnormal returns.
We also used a non-parametric Corado-test to examine if the average abnormal returns on the event-day deviates from 0. The Corado test looks as follows:
Θ
3= (1 / N) *
∑ (𝐾𝑖,0−𝐿2+1 2 ) 𝑁
𝑖=1
𝑆(𝑘)
(13)
In this test N is the total number of events, K
i,0is the rank event i on day 0, L2 is the number
of days used in the test and S(k) is the standard deviation given by the following formula:
9 S(k) = √𝐿21 ∑
𝑡2𝑡=𝑡1+1(
1
𝑁
∑ (𝐾𝑖, 𝑡 −
𝐿2+12
𝑁𝑖=1
))^2 (14)
Again, L
2is the number of days used in the test, K
i,tis the rank of the return from stock i on time t and N is the total number of events.
The Mann-Whitney u-test is a rank-test that can be used to compare both subsamples used.
First, a U value is computed for both subsamples using this formula:
U
i= R
i– (N
i(N
i+ 1)/2) (15)
In this formula R
iis the sum of ranks of subsample i, and N
iis the number of observations in the subsample, in this case 31 days.
For larger samples, U is normally distributed and the test statistic is:
Z = (U - ɱ
U) / Ϭ
u(16)
Here ɱ
uis given by:
ɱ
u= (N
1N
2)/2 (17) And Ϭ
uis given by:
Ϭ
u= √
𝑁1𝑁2(𝑁1+𝑁2+1)12