The Financial Market Effects of International Aviation Disasters

The spread of misinformation with regards to aviation disasters continues to be a point of concern for aviation companies. Much of this information usually surrounds speculation based on the cause and responsibility attributed to the incident, implicitly possessing the potential to generate significant financial market price volatility. In this paper, we investigate a number of stylised facts relating to the effects of airline disasters on aviation stocks, while considering contagion effects, information flows and the sources of price discovery within the broad sector. Results indicate a substantially elevated levels of share price volatility in the aftermath of aviation disasters, while cumulative abnormal returns present sharp under performance of the analysed companies relative to international exchanges. When considering an EGARCH analysis, we observe that share price volatility appears to be significantly influenced by the scale of the disaster in terms of the fatalities generated. Significant contagion effects upon the broad aviation index along with substantial changes in traditional price discovery channels are also identified. The role that the spread of information on social media, whether it be correct or of malicious origins, cannot be eliminated as an explanatory factor of these changing dynamics over time and region.

In this paper, we investigate as to whether a number of stylised facts relating to the effects of airline disasters on aviation stocks and contagion effects within the broad sector hold. First, we analyse as to whether there exist time-varying and geographical differences in the response mechanisms of investors to aviation disasters, or indeed, has there been a variation of response that could be considered to be correlated to the level of injury and fatality caused by each individual incident. Further, we generate further novelty through the inclusion of analysis that focused on the interlinkages between the incident companies and the broad aviation sector. Finally, we investigate changes in flow of information and price discovery that could be considered to be abnormal when considering pre-disaster averages between the stock price of incident companies and broad aviation indices that represent sectoral returns.
We first clearly identify that substantially elevated levels of share price volatility, however, there is evidence to suggest that this volatility has somewhat decreased in the periods 2005 through 2019. Such effects are found to be substantial when considering geographical differentials, with both North American and South American companies exhibiting the largest negative effects. When analysing cumulative abnormal returns (CARs), we observe that there exists sharp under-performance of the companies relative to international exchanges throughout each of the analysed time periods with the exception of the period between 2000 and 2004. When considering an EGARCH analysis to investigate specific volatility effects, we observe a sharp increase in unconditional volatility in the ten-day period after the aviation incident, indicative of strong short-term effects. However, while the shock to unconditional volatility appears to be immediate, there is evidence to suggest that it dissipates and returns to pre-aviation incident levels within sixty days after the event. Further, our results indicate that there exists a clear positive relationship between the two variables, indicating that the estimated EGARCH-volatility appears to be significantly influenced by the scale of the disaster in terms of the fatalities generated. When considering the contagion effects of volatility and the flow of information and price discovery between the incident company and the broad aviation sector, we find that there were substantial decreased in dynamic conditional correlations during incidents that occurred in the periods 1995-99, 2000-04 and 2005-09 respectively. However,evidence of such pronounced effects do not appear to occur in the periods between 2010-14 and 2015-19. For every case analysed we find that there were substantial decreases in information flow identified between the interactions between both the airline and the aviation sector. This indicates that there is a permanent disruption to the flow of information and price discovery channels that would have traditional have existed in the period prior to the aviation disaster.
The rest of this paper is as follows. Section 2 presents a thorough review of the literature relating to the interlinkages between aviation disasters and the varying effects on financial markets and sectoral interactions.
Section 3 presents a concise overview of the data used in this research along with the various methodologies employed to capture firm-level volatility, both intra-sectoral and geographic volatility transmission, and indeed contagion effects by type of aviation incident. Section 4 presents a concise overview of the results presented, while Section 5 concludes.

Previous Literature
Although much research focuses specifically on the aviation sector in terms of structure and performance, little has focused specifically on the presence of sectoral interactions between rival and geographically-similar aviation companies. Research of such nature is of substantial value to to the broad sectoral correlations that exist, enabling theoretically-plausible avenues through which large aviation incidents could generate significant sector-wide risk. While this research sets out to investigate the existence of such channels as identified by stock market dynamics, it builds on a number of existing research areas. Chance and Ferris 3 Electronic copy available at: https://ssrn.com/abstract=3550845 [1987] found that such incidents were, in the mid-1980s ring-fenced from the broad sector at large, however, much research today has begun to focus on the dissemination of news through multiple technological channels, leading to quite strong theoretical foundations to signal that such results might not exist almost three decades later. Ho et al. [2013] found that there do exist abnormal returns in competitor companies during major sectoral incidents, however, their share price is found to otherwise increase should the incident be considered minor. This research does not extend its scope beyond abnormal pricing, however, the sentiment of the results further indicate that the work of Chance and Ferris [1987] might not represent the industry today. Kaplanski and Levy [2010] analysed the influence of aviation disasters on stock prices while considering the role of sentiment. Should a market loss of more than US$60 million, the authors identify evidence of substantial negative stock market responses, increased perceived risk and implied volatility, despite no evidence of an increase in actual volatility. Such a result is, of course, not limited to the aviation sector.
Carpentier and Suret [2015] found that such losses exist across a broad number of major accidents, however, they do not persist. Ho et al. [2013] found that airlines that suffer a crash experience deeper negative abnormal returns as the degree of fatality increases, but also, the stock prices of the rival airlines also suffer in large-scale disasters but benefit from the disasters when the fatality is minor. Hung and Liu [2005] use the beta value, an indicator of systematic risk, to estimate the costs of equity and the evaluation of a stock's reasonable price, to find that airline betas are volatile over time and that crashes also impact them in addition to their stock return and volatility.
Outside of just stock market performance in isolation, there might exist channels through which corporate effects can also be transferred to airlines companies. While legal liability, broad reputational damage and the loss of equipment can be found to explain a substantial amount of the target company's losses, guided by the work of Walker et al. [2014] found that there are a number of regulatory factors that extend far beyond insurance premiums and reputational damage. Dillon et al. [1999] identified evidence to suggest that in the aftermath of a single aviation incident, shareholders appear to update their estimates of the probability of a future incident, acting as an explanation for the substantial cumulative abnormal returns that exist. Exogenous events might also have such negative outcomes. For example, Corbet et al. [2019] found that traffic flows fall quite sharply despite significant fare reductions as a result of terrorist incidents in Europe. Such terrorism impacts were also found to be both significant and substantial when considering the persistence of their effects at both sectoral and national levels (Kolaric and Schiereck [2016] where an six-year old article based on the bankruptcy of United Airline's parent company was mistakenly identified as a new bankruptcy filing, causing a 76% fall in the company's share price, but after the case was identified as an error, the stock remained over 11% below opening prices, as the authors identify that contagion effects would dominated competitive effects. Luo [2007] used longitudinal real-world data set that matches consumer negative voice (complaint records) in the airline industry with firm stock prices, this article finds that higher levels of current consumer negative voice harm firms' future idiosyncratic stock returns.
Since energy commodities play an important role for the airlines industry, their price fluctuations can also create problems for aviation stocks in financial markets. Treanor et al. [2014] found that airlines that increase their hedging activity due to higher fuel price exposure are not receive a premium in their valuation when compared to those airlines employing more stable hedging policies. However, Berghofer and Lucey [2014] found that there exists less significant negative exposure coefficients among US carriers. Yun and Yoon [2019] found that there is a return and volatility spillover effect between crude oil price and the stock prices of airlines and that the stock prices of smaller airlines of South Korea and China are relatively more sensitive to the change in oil price. Kristjanpoller and Concha [2016] found a strong positive influence of fuel price fluctuation and airline stock returns using GARCH-family methodologies.

Data
We begin our analysis developing a concise list of aviation disasters that can then be utilised in a thorough and robust methodological investigation through an analysis of parent companies who trade on stock markets and their subsequent performance. Broad speculation based on the cause of such an airline disaster can manifest through many forms, but direct financial punishment due to investor perceptions can present a number of damaging side-effects for the broad aviation sector. To develop such a dataset, we develop a number of strict rules in an attempt to standardise the process across major international financial markets. The first implemented rule is that the specified company must be a publicly traded company with an available stock ticker between the period June 1, 1995 and May 31, 2019. This specific time period is identified due to the relative absence of concise financial market in the period before. Our selected stock price data is taken from Thomson Reuters Eikon. The second news selection rule is based on the source of the aviation disaster data. We develop on a combined search of LexisNexis, Bloomberg and Thomson Reuters Eikon, search for the keywords relating to aviation disasters. For added robustness of our developed dataset, we leverage upon that of the National Transportation Safety Board (available at: https://www.ntsb.gov), the International Civil Aviation Organization, ICAO (available at: https://www.icao.int) and the Aviation Safety Reporting System (available at: https://asrs.arc.nasa.gov).
Insert Table 1 about here To obtain a viable observation, a single result must be present across each of the selected search engines and the source was denoted as an international news agency, a mainstream domestic news agency or the company making the announcement itself. Forums, social media and bespoke news websites were omitted from the search. Finally, the selected observation is based solely on the confirmed news announcements being made on the same day across all of the selected sources. If a confirmed article or news release had a varying date of release, it was omitted due to this associated ambiguity. All observations found to be made on either a Saturday or Sunday are denoted as active on the following Monday morning. All times are adjusted to GMT, with the official end of day closing price treated as the listed observation for each comparable company when analysing associated contagion effects. In Table 1 we observe the relevant summary statistics for the included aviation companies that experienced severe aviation disasters throughout the time period analysed.

Insert Figures 1 & 2 about here
In Figure 1, we observe the cumulative number of incidents and fatalities that occurred on a quarterly basis between Q2 1995 and Q2 2019. There is evidence of a decline in both estimates, with peaks experienced throughout the period between Q1 1996 and Q4 1996. In total, there are 610 incidents included in our analysis, of which there were 12,692 fatalities. The worst incidents include that of the crash of a Saudi Arabian Airlines Boeing 747-100 in New Delhi in November 1996, which led to the death of 312 people.
Further, in the 11 September terrorist attacks of 2001, the Airbus A300-600 that crashed in New York leading to the deaths of 262 persons is also included among the worst incidents included in the database.
The second and fourth most severe incidents included unfortunately involve the same company, Malaysia Airlines. In 2014, an idiosyncratic succession of accidents, unparalleled in aviation history unfolded within a five month time frame as two widebody aircraft operated by Malaysia Airlines crashed under inconceivable circumstances. The first tragedy to strike Malaysia Airlines was the loss of flight MH370, which disappeared while flying from Kuala Lumpur International Airport to Beijing Capital International Airport on 8 March 2014 with the loss of 227 passengers and 12 crew. Inmarsat satellites identified two potential trajectories, that MH370 could have taken, but after much analysis the investigators speculated with a high degree of probability that the perilous aircraft was navigated along a southern trajectory, leading it deep into the southern Indian Ocean. After one of the most expensive searches in aviation history, the aircraft has not been found, however several pieces of debris washed ashore in the western Indian Ocean during 2015 and 2016 which were confirmed to be from the airliner. In the aftermath of the loss of MH370, Malaysia Airlines were then subjected to their second significant loss as MH17, a scheduled flight from Amsterdam to Kuala Lumpur that was shot down on 17 July 2014 while flying over eastern Ukraine, where an armed conflict broke out in April 2014. Flights over the conflict zone were allowed as there was no indication of risk for civil aeroplanes at cruising altitude and Malaysia Airlines, as almost all airlines, assumed that the airspace is safe. A Buk 9M38-series surface-to-air missile with a 9N314M warhead was found to have downed the aircraft. An explosive decompression resulted in the disintegration of the aircraft while in-flight and all 283 passengers together with 15 crew members perished resulting in a wreckage area of 50 square km on the ground. The four most severe incidents in this sample account for 1,111 fatalities in the sample. In Figure   2 we identify the geographic dispersion of first, incidents as denoted by the geographic dispersion of the parent companies of the airlines that have experienced the aviation disaster. The second panel displays the geographic dispersion of the analysed incidents within this research. It is of interest to note that a large number of African nations and countries such as Bangladesh, India, Nepal and Colombia have experienced a broad number of aviation disasters in their respective nations, however, there are few incidents recorded in our dataset that include airlines from these regions that are publicly traded. It is also of interest to note the substantial issues that surround some South American airlines over time, such as substantial debt issues and the impounding of some physical assets (Holden [1986]; Akpoghomeh [1999]).

Methodology
To further the development of our understanding of disasters within the aviation sector, we set out to first specifically analyse the immediate pricing and volatility effects on the stock prices of the company that owned the plane that has been lost. To add methodological robustness to our selected analysis, we have utilised a number of GARCH-family methodologies 1 while we further attempt to mitigate international factors through the inclusion of the Dow Jones Industrial Average (DJIA), West Texas Intermediate oil prices (WTI) and the SPDR S&P Transportation ETF (XTN) which is found to best represent the performance of aviation companies during the period analysed. We further consider as to whether such volatility effect has changed in the period since the broad growth. In a secondary analysis, we then investigate the relationship between each company and the broad measure of the aviation index through the use of a DCC-GARCH analysis.
It has been widely considered that some specific accidents have been large enough to generate substantial reverberations throughout the entire sector due to the presence of a number of technical and regulatory mishaps (Krieger and Chen [2015]; Ho et al. [2013]; Nethercutt and Pruitt [1997]). Such dynamics could be attributed to the market expectations that future regulatory changes could be forthcoming and could even perhaps be restrictive to the future profitability of the sector. Finally, to validate and add further robustness to the presented results, we investigate the sources of price discovery in the relationship between the broad sectoral indices and the companies that have experienced such substantial loss and reputational damage in an attempt to further analyse investor behaviour. We define a distressed airline company as that which has experienced a substantial tragedy in the form of the loss of a plane. Overall, there are a number of specific questions that we then set out to analyse. The above hypotheses thereby set out to analyse multiple facets of financial market distress in the aftermath of such tragedy. The results of this paper are of interest to the broad aviation sector, traders, regulators and policy-makers alike. While it is not surprising to observe that distressed aviation companies in such situations exhibit substantial and significant negative effects, it is very much of interest to further our understanding of the sources of such risk and indeed, as to whether it is contagious upon the entire aviation sector. Should there be evidence identified of a substantial link, this would develop potential fears about the broad financial safety of the sector and a strong potential for issues such as moral hazard and asymmetric information to develop as weaker, less regulatory compliant companies could potentially influence the growth, development and financial viability of companies who are behaving in a regulatory-compliant manner.
To begin our analysis, we first utilise a multivariate EGARCH(p,q) methodology to identify scale of the change in volatility in the period after the identified aviation incidents. At this stage, a number of goodness-of-fit testing procedures identified the EGARCH(1,1) model as the best selected to identify specific volatility changes in the companies' returns, thus we exercise our analysis using this model. 2 The EGARCH specification developed on that of the GARCH specification proposed by Bollerslev [1986] and was designed to include lagged conditional variance terms as autoregressive terms. We specifically develop on an EGARCH methodology to analyse the volatility effects within the aviation industry due to aviation disasters. We first let r t = [r 1,t , ..., r n,t ] be the vector of financial time series returns and ε t = [ε 1,t , ..., ε n,t ] be the vector of return residuals obtained after some filtration. Let h i,t be the corresponding conditional volatilities obtained from a univariate EGARCH process. We assume that is the conditional expectation on ε t , ε t−1 , .... Then the asset conditional covariance matrix H t can be written as is the asset conditional correlation matrix and the diagonal matrix of the asset conditional variances is given by D t = diag(h 1,t , ..., h n,t ). We express the variance equation of our EGARCH model as follows: which states that the value of the variance scaling parameter h t now depends both on the past value of the shocks, which are captured by the lagged square residual terms, and on past values of itself, which are captured by the lagged h t terms. Specification tests found that the EGARCH(1,1) model served as the best fitting to estimate volatility effects through the use of dummy variables that are used to denote both the time-of-the-day and also periods of substantial traditional market volatility. 3 It is also necessary to mitigate international effects which can be completed through the inclusion of the returns of traditional financial products in the mean equation of the EGARCH(1,1) methodology. The volatility sourced in shocks that are incorporated in the returns of traditional financial markets are therefore considered in the volatility estimation of the selected structure. In summary, the estimated model has the following form: R t−n represents the lagged value of stock returns, n days before R t is observed. DJIA t represents the interaction between the distressed aviation company and the Dow Jones Industrial Average (DJIA), while W T I t represents the interaction with West Texas Intermediate oil prices (WTI), which is a market found to be very closely associated with the aviation sector (Kristjanpoller and Concha [2016]; Yun and Yoon [2019]), but also a very strong signal of multiple forms of economic strife (Chuang et al. [2008]), therefore acting as a strong barometer of international effects within our selected methodological structure. Finally, AvET F t represents that of the SPDR S&P Transportation ETF (XTN) which is found to best represent 3 The optimal model is chosen according to three information criteria, namely the Akaike (AIC), Bayesian (BIC) and Hannan-Quinn(HQ), all of which consider both how good the fitting of the model is and the number of parameters in the model, rewarding a better fitting and penalising an increased number of parameters for given data sets. The selected model is the one with the minimum criteria values. We also used a variety of dummy-lengths in Equation (3), denoted as Dt in the variance equation, but the twenty-day period after each selected event was denoted as the most stable specification across our selected methodologies. Results of all these specification tests are available from the authors on request.
the performance of aviation companies during the period analysed. Finally, D t is included in the variance equation to provide a coefficient relating to the volatility response to the thirty-day period after which the aviation tragedy has occurred.
In the next stage of our analysis, we investigate Hypothesis H 4 , which specifically tests as to whether there has been a substantial change in dynamic correlations between the distressed company that has experienced the aviation disaster and the selected aviation indices in the period thereafter. To complete such an analysis, we test for the presence of such comovements in aviation markets and then specifically investigate their responses thereafter using a DCC-GARCH methodology. Engle [2002] models the right hand side of Eq. (1) rather than H t directly and proposes the dynamic correlation structure are non-negative scalars satisfying a + b < 1. The parameters of the DCC model are estimated by using the quasi-maximum likelihood method with respect to the log-likelihood function, and according to the state two-step procedure.
When specifying the form of the conditional correlation matrix R t , two requirements have to be considered.
The first is that the covariance matrix H t has to be positive and the second is that all the elements in the conditional correlation matrix R t have to be equal or less than unity. The DCC model is estimated by using a two-step approach to maximise the log-likelihood function. As proposed by Engle [2002], the DCC-GARCH model is designed to allow for a two-stage estimation of the conditional variance matrix h t .
In the first stage, univariate GARCH (1,1) volatility models are fitted for each of the stock return residuals and estimates of √ h it are obtained. In the second stage, stock return residuals are transformed by their estimated standard deviations from the first stage as z it = it √ hit . Finally, the standardised residual z it is used to estimate the correlation parameters. If we let θ denote the parameters in D t and ϑ, the parameters in R t , then the log-likelihood is: The first part of the log likelihood function is volatility, which is the sum of the individual GARCH likelihoods. The log-likelihood function can be maximised in the first stage over the parameters D t . Given the estimated parameters in the first stage, the correlation component of the likelihood function in the second stage is maximised to estimate the correlation coefficients. Finally, we examine the DCC-GARCH model's change in behaviour before and after each airline disaster, measuring the specific relationship between the associated airline and the broad aviation sector as measured by the SPDR S&P Transportation ETF (XTN).
In the final stage of our analysis, and to provide additional methodological robustness, we analyse the changing behaviour of price discovery in the periods after such aviation disasters.
where ∆p i,t is the change in the log price (p i,t ) of the asset traded in market i at time t. The next stage is to obtain the component shares from the normalised orthogonal vector of error correction coefficients, therefore: Given the covariance matrix of the reduced form VECM error terms 4 where: we calculate the IS using: Recent studies show that IS and CS are sensitive to the relative level of noise in each market, they measure a combination of leadership in impounding new information and the relative level of noise in the price series from each market. The measures tend to overstate the price discovery contribution of the less noisy market. An appropriate combination of IS and CS cancels out dependence on noise, Yan and Zivot Our work on the information share, component share and information leadership share of price discovery sets out to address the final stated hypothesis as to whether there exist substantial changes in information flows between the distressed aviation companies and broad aviation indices in the period after such disasters.

Results
The first analysed hypothesis investigated as to whether the price response of airlines in the aftermath of aviation disasters has varied substantially. To begin such an analysis, we focus on a number of different characteristics surrounding the behaviour of the share prices of companies that have experienced such aviation disasters. In Figure 3 we observe the share price return volatility on a daily level of these companies over time between 1995 and 2019. A one-year sample, both six-months before and after the incident is presented.
We clearly identify that substantially elevated levels of share price volatility on average in the periods incorporating 1995 through 2019, however, there is evidence to suggest that this volatility has somewhat decreased in the period of time thereafter. Throughout each period, there is evidence presented of sharp negative price movements in the days following the aviation disaster, however, as is particularly evident in the periods 2005 through 2019, this is immediately followed by substantial increases in the two-month period after the incident. In Figure 4 we observe the results of a similar analysis that has been separated by the continent in which the parent company of the airline was located at the time of the incident. For each region analysed, there is substantial evidence of immediate negative responses in each jurisdiction with the exception of South America which is portrayed as quite a volatility market throughout the period analysed. It is Asia and North America that present the most substantial decreases in share price in the days immediately after the date on which the incident occurred. On average, North American airlines experience quite a substantial increase in share price within ten days of the incident.

Insert Figures 3 through 6 about here
In Figure 5 we analyse the cumulative abnormal returns (CARs) over time and in Figure 6 we observe the same analysis as separated by the region in which the disaster has occurred. The selection of six month windows before and after each event is made solely for presentation purposes. Focusing on the performance of the CARs over time, we observe that there exist sharp under-performance of the companies throughout each of the analysed time periods with the exception of the period between 2000 and 2004. While considering the relative diversification that aviation stocks presented throughout the multiple international financial crises to portfolio investors, the same stocks were very much exposed to shocks in the market for oil (Kristjanpoller and Concha [2016]; Yun and Yoon [2019]). There is further evidence that CARs do not appear to behave in a similar manner depending on the market in which the airline stock is traded. South American airlines presented evidence of substantial under-performance when compared to international averages in the period before airline crashes leading to fatality. This identified under-performance continued in the period thereafter.
African, European and North American airlines were identified to largely out-perform broad markets by more than 10%, but in the period thereafter, under-performed the same indices by between 6% and 10%. When considering this information, we can validate Hypothesis 1, as we clearly identify time-varying and regional differentials between airline share price performance due to aviation disasters.
Insert Table 2 through 4 about here In the next phase of our analysis, we investigate the changing financial market volatility effects of each individual case analysed. In Table 2 Table 3 we observe the number of companies that experienced a sustained increase in their stock returns' unconditional volatility in the period after each incident. To this effect, we observe as to whether the estimated volatility increases, as measured by both the logarithm of daily returns and excess logarithm of daily returns were significantly different to zero in the periods representing ten, twenty, forty and sixty days after the incident. The results are separated between log-returns with a higher variance and excess log-returns with higher variance. Within this context, in both samples analysed, we observe that there are a large number of companies that experience a sharp increase in unconditional volatility in the ten-day period both before and after the aviation incident, indicative of strong short-term effects. However, such effects are found to dissipate in the following windows of investigation, based on twenty-day, forty-day and sixty-day windows respectively. Such results indicate that while the shock to unconditional volatility appears to be immediate, there is evidence to suggest that it dissipates and returns to pre-aviation incident levels within sixty days after the event. While focusing on the time-variations of the estimated results Table 4 presents the significant estimates of short-term (denoted to be the ten-day period after the airline disaster) price volatility. We observe that companies that have experienced more recent aviation disasters possess an increased correlation with the DJIA and a decreased correlation over time with other aviation companies as measured by the SPDR S&P Transportation ETF.
Within the models presented in Table 4, we observe the sharp, strongly significant increases in short-term volatility in the period after the aviation disaster as measured by D t .
Insert Figure 7 about here The variants dummy variable duration of the estimated EGARCH methodology allow for the investigation of Hypotheses H 2 and H 3 , which specifically analyse as to whether the price volatility response of the airline disasters appears to depend on the scale of the disaster as measured by the number of fatalities and as to whether such volatility effects have changed over time. Figure 7 presents the results of these analyses. While utilising the volatility estimates across all cases included in this analysis, we use a scatter-plot to identify the relationship between these dummy variable estimates and that of number of fatalities per incident.
The results indicate a clear positive relationship between the two variables, indicating that the estimated EGARCH-volatility appears to be significantly influenced by the scale of the disaster in terms of the fatalities generated. These two specific results enable acceptance of both Hypotheses 2 and 3 as there is clear evidence of a relationship between volatility and incident-severity, while further, such volatility effects are found to changed over time.

Insert Figures 8 and 9 about here
In Figure 8, we specifically analyse as to whether the investigated dynamic correlations have changed over time. To complete such an analysis we have broken the analysis into five year windows of analysis and presented the average dynamic conditional correlation for the sixty day period both before and after each incident. We observe that there were substantial decreased in correlation in the periods after incidents in the period 1995-99, 2000-04 and 2005-09 respectively. However, evidence of such pronounced effects do not appear to occur in the periods between 2010-14 and 2015-19. While in Figure 9 we observe the average dynamic correlation by region in the sixty day periods both before and after the investigated aviation disaster. While crashes in Africa are found to have little of substantial change in the period after, there are substantial changes in dynamic conditional correlation observed for incidents that have occurred in Asia, Europe, and both North and South America respectively. The most substantial effects were observed in South America where the dynamic correlation is found to fall from 0.193 on the date of the incident to fall to 0.048 ten days after the incident. Such results indicate that there are geographical differences to be considered where volatility appears to differ based on the region in which the incident occurs.
Insert Table 5 about here In Table 5, we observe the results of the dynamic conditional correlation analysis between the incident company and the selected aviation indices in the aftermath of airline disasters. Such an analysis is used to specifically test Hypothesis H 4 . Two panels are presented in Table 5, where the first presents all dummy coefficients that are found to result in a positive result for the dynamic conditional correlation, while the lower panel presents only significant results. For brevity, only positive results are presented, while negligible results are omitted for presentation purposes only. Each dummy window is presented, based on 1-day through 60day frequencies and an additional analysis which includes the entire sample period after the aviation disaster.
While there is a relatively stable number of incidents that present significantly positive volatility through to the broad aviation industry as represented through the twenty-five through twenty-seven incidents that are found to present positive dummy coefficients. For the entire period thereafter, forty-two companies representing 62.7% of the sample are found to be positive while analysing the entire period after the event.
However, while concentrating on significant results only, there is an interesting observation based on the time frame in which the dynamic correlations are found to be significantly positive. Only one event is found to have generated a significantly positive conditional correlation through to the broad aviation indices, however, the effects of this event are found to dissipate before twenty days after the event. While forty events are found to generate significant positive volatility transfer, it appears that such effects become more substantial over time. This could be attributed to improved information being released to the market as to the exact nature and causes of the incident, which could generate profound industrial changes dependent on the nature of the scenarios. Due to the identification of such significant positive correlations, we can state acceptance of Hypothesis H 4 , namely that individual shocks possess the ability to influence the volatility of the entire aviation index.
Insert Table 6 about here The final hypothesis investigates as to whether there exists a substantial change in information flows between the distressed aviation company and broad aviation indices after an airline disaster. In Table 6 we observe the most substantial changes in the flow of information as measured by the information share, component share and information leadership share of price discovery in the periods both before and after the identified significant incidents. In all cases, there were substantial decreases in information flow identified between the interactions between both the airline and the aviation sector. This indicates that there is a sharp reduction in the flow of information that can be measured by the proportion of the variance in the common efficient price innovations that is explained by innovations in that price series. The permanent component is interpreted as the common efficient price, the temporary component reflects deviations from the efficient price caused by trading fractions. In all cases, there is evidence of a sharp decoupling of information from broad aviation indices upon that of the companies included in the analysis. These results present substantial evidence of a significant decoupling between the airline and other similar companies within the same sector.
This adds further support to the differing interactions between market participants and the manner in which information flows between markets, validating the final, fifth hypothesis that there exists substantial decoupling of broad sectoral indices with the airlines that have experienced substantial tragedy.

Concluding Comments
In this paper, we investigate as to whether a number of stylised facts relating to the effects of airline disasters on aviation stocks and contagion effects within the broad sector hold. Using a number of exceptionally detailed databases, we investigate as to whether there exist time-varying and geographical differences in the response mechanisms of investors to aviation disasters, or indeed, has there been a variation of response that could be considered to be correlated to the level of injury and fatality caused by each individual incident.
Further, we generate further novelty through the inclusion of analysis that focused on the interlinkages between the incident companies and the broad aviation sector, and indeed the flow of information and price discovery that could be considered to be abnormal when considering pre-disaster averages.
We first clearly identify that substantially elevated levels of share price volatility on average in the periods incorporating 1995 through 2019, however, there is evidence to suggest that this volatility has somewhat There is further evidence that CARs do not appear to behave in a similar manner depending on the market in which the airline stock is traded. South American airlines presented evidence of substantial under-performance when compared to international averages in the period before airline crashes leading to fatality, whereas, African, European and North American airlines were identified to largely outperform broad markets by more than 10%, but in the period thereafter, under-performed the same indices by between 6% and 10%. When considering our EGARCH analysis, we observe that there are a large number of companies that experience a sharp increase in unconditional volatility in the ten-day period both before and after the aviation incident, indicative of strong short-term effects. However, such effects are found to dissipate in the following windows of investigation, based on twenty-day, forty-day and sixty-day windows respectively. Such results indicate that while the shock to unconditional volatility appears to be immediate, there is evidence to suggest that it dissipates and returns to pre-aviation incident levels within sixty days after the event. Further, our results indicate that there exists a clear positive relationship between the two variables, indicating that the estimated EGARCH-volatility appears to be significantly influenced by the scale of the disaster in terms of the fatalities generated. Within this context, our research finds that there exist regional and time-varying effects both in terms of volatility and share price response. Further, such share price volatility is found to be directly responsive to the severity of each incident. This can perhaps be explained by the presence of improved information flows through financial markets in the later years analysed within the sample period.
Further, when considering the contagion effects of volatility and the flow of information and price discovery between the incident company and the broad aviation sector, we find that there were substantial decreased in dynamic conditional correlations during incidents that occurred in the periods 1995-99, 2000-04 and 2005-09 respectively. However,evidence of such pronounced effects do not appear to occur in the periods between 2010-14 and 2015-19. Further, crashes in Africa are found to have little of substantial change in the period after, there are substantial changes in dynamic conditional correlation observed for incidents that have occurred in Asia, Europe, and both North and South America respectively. Further, for every case analysed we find that there were substantial decreases in information flow identified between the interactions between both the airline and the aviation sector. This indicates that there is a permanent disruption to the flow of information and price discovery channels that would have traditional have existed in the period prior to the aviation disaster. That is, while companies attempt to return to any form of perceived normality in the period after an aviation disaster, investors appear to treat such companies in a different capacity to industrial peers on a permanent basis. It would be quite interesting for future research to investigate as to whether similar effects area found to permeate throughout the entire supply chain relating to goods an services that are provided to the aviation industry.
Our research has a series of relevant policy implications. Market sensitivity to such sudden, catastrophic shocks such as those related to aviation disasters can of course be considered to be exceptionally negative with regards to expected investor response. However, there has been considerable reservations about the implicit role that social media and the spread of misinformation, or malicious information can have in the aftermath of such events. Much of the spread of such information usually surrounds speculation based on the cause and responsibility attributed to the incident. We feel that analysis and policy review surrounding high-frequency financial market data paired with that of social media data would be considered a worthy direction of future research. While our research presents evidence that financial markets are guided by better quality information through the development of social media, it should also be considered that the presence of such improved efficiency could in fact manifest in side-effects such as an ability to profit from the spread of false information, not only generating further undue distress on the companies and families involved in such tragedy, but also hindering the efforts of rescue teams while further inspiring other market participants who do not fear current regulatory and policing efforts. Electronic copy available at: https://ssrn.com/abstract=3550845    Note: The above figure presents the estimated share price volatility response by investigated region, while presenting the average results for sixty days both before and after each incident for presentation purposes.

24
Electronic copy available at: https://ssrn.com/abstract=3550845   Note: We establish the above list noting that each company must be publicly traded with an available stock ticker between the period June 1, 1995 and May 31,2019. This specific time period is identified due to the relative absence of concise financial market in the period before. Stock price data is taken from Thomson Reuters Eikon.

28
Electronic copy available at: https://ssrn.com/abstract=3550845 Note: This table presents the estimation results of the mean and conditional variance equations; i.e., Rt = a0 + b1Rt−1 + b2DJIA + b3W T I + b4AV I + εt; and ln(h 2 t ) = ω + αεt−1 + γ(|εt−1|−E(|εt−1|)) + β ln(h 2 t−1 ) + Dt respectively. Rt−1 represents the lagged value of the observed company returns. The term ht is the conditional volatility estimated by the EGARCH process and Dt is a dummy term to provide a coefficient relating to the observed changes in the conditional volatility in the subsequent period following each event for each of our investigated companies. Only the results for the companies with a significant positive Dt term is presented. For brevity, only the significant results for entire dummy period are presented. The values in the parentheses are standard errors. ***, ** and * denote significant at the 1%, 5% and 10% level respectively.  Note: This table presents the estimation results of the mean and conditional variance equations; i.e., Rt = a0 + b1Rt−1 + b2DJIA + b3W T I + b4AV I + εt; and ln(h 2 t ) = ω + αεt−1 + γ(|εt−1|−E(|εt−1|)) + β ln(h 2 t−1 ) + Dt respectively. Rt−1 represents the lagged value of the observed company returns. The term ht is the conditional volatility estimated by the EGARCH process and Dt is a dummy term to provide a coefficient relating to the observed changes in the conditional volatility in the subsequent 10 days (2 weeks) following each event for each of our investigated companies. Only the results for the companies with a significant positive Dt term is presented. For brevity, only significant and results above zero are presented. The values in the parentheses are standard errors. ***, ** and * denote significant at the 1%, 5% and 10% level respectively.

31
Electronic copy available at: https://ssrn.com/abstract=3550845 Note: For brevity, only significant and results above zero are presented. Panel A (Panel B) presents the statistical results on the (significant) positive dummy coefficients estimated in the following regression ρ t i,avi = α + Dt + εt. ρ t denotes the dynamic conditional correlations, i stands for the selected company's returns, avi is the returns of the benchmark index where .... Dt is a dummy variable that takes the value one for a certain period of time after company announcements. Values in this table show the number of companies that experience a change in their stock returns' correlation between the above mentioned indices after their announcements. The column headers show the number of days that we analyse the correlation change after the announcements. The values in the parentheses are the percentage of companies within the sub-groups experiencing a change in correlations. Note: The above panel represents the estimated coefficients of price discovery. For brevity, only significant and results above zero are presented. IS represents the information share, IS-r represents the reverse information share criterion, CS represents the component share of information while ILS represent the information leadership share of information. The four right-hand columns represent the estimated changes in price discovery in the periods both before and after the announcements of name changing processes.