European markets' reactions to exogenous shocks: A high frequency data analysis of the 2005 London bombings

Kollias, Christos; Papadamou, Stephanos; Siriopoulos, Costas Article European markets' reactions to exogenous shocks: A high frequency data analysis of the 2005 London bombings International Journal of Financial Studies Provided in Cooperation with: MDPI – Multidisciplinary Digital Publishing Institute, Basel Suggested Citation: Kollias, Christos; Papadamou, Stephanos; Siriopoulos, Costas (2013) : European markets' reactions to exogenous shocks: A high frequency data analysis of the 2005 London bombings, International Journal of Financial Studies, ISSN 2227-7072, MDPI, Basel, Vol. 1, Iss. 4, pp. 154-167, https://doi.org/10.3390/ijfs1040154 This Version is available at: https://hdl.handle.net/10419/103585 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/3.0/ Int. J. Financial Stud. 2013, 1, 154–167; doi:10.3390/ijfs1040154 International Journal of Financial Studies ISSN 2227-7072 www.mdpi.com/journal/ijfs Article European Markets’ Reactions to Exogenous Shocks: A High Frequency Data Analysis of the 2005 London Bombings Christos Kollias 1, Stephanos Papadamou 1,* and Costas Siriopoulos 2 1 Department of Economics, University of Thessaly, Korai 43, Volos 38333, Greece; E-Mail: [email protected] 2 Department of Business Administration, University of Patras, Rio, Patras 26504, Greece; E-Mail: [email protected] * Author to whom correspondence should be addressed; E-Mail: [email protected]; Tel.: +30-24210-74963; Fax: +30-24210-74772. Received: 29 September 2013; in revised form: 7 November 2013 / Accepted: 11 November 2013 / Published: 18 November 2013 Abstract: Terrorist incidents exert a negative, albeit usually short-lived, impact on markets and equity returns. Given the integration of global financial markets, mega-terrorist events also have a high contagion potential with their shock waves being transmitted across countries and markets. This paper investigates the cross-market transmission of the London Stock Exchange’s reaction to the terrorist attacks of 2005. It focuses on how this reaction was transmitted to two other major European stock exchanges: Frankfurt and Paris. To this effect, high frequency intraday data are used and multivariate Genralised Autorgressive Conditional Heteroskedasticity (GARCH) models are employed. This type of data help reveal a more accurate picture of markets’ reaction to exogenous shocks, such as a terrorist attack, and thus allow more reliable inferences. Findings reported herein indicate that the volatility of stock market returns is increased in all cases examined. Keywords: capital markets; contagion; terrorism; multivariate GARCH JEL Code: G14; G21 OPEN ACCESS Int. J. Financial Stud. 2013, 1 155 1. Introduction The high velocity, with which the shock waves from major financial episodes, irrespective of the source that has generated them, travel across markets and countries, has attracted increasing attention in the relevant financial literature. A plethora of studies, a survey of which can be found in Pericoli and Sbracia [1], have examined both on a theoretical as well as empirical level the mechanisms and the channels through which financial shocks that occur in one country are transmitted and affect markets in another or indeed, have a major international impact on global markets and economic sentiment (inter alia: Goetzmann, et al. [2]; Saleem [3]; Meric and Meric [4]; Asimakopoulos et al. [5]; Chiang et al. [6]). In particular, a number of studies have examined the interdependence of equity market volatility using the framework of autoregressive conditional heteroskedasticity (GARCH) time series models (inter alia: Saleem [3]; Hamao et al. [7]; Theodossiou and Lee [8]; Lin et al. [9]; Longin and Solnik [10]). A strand of the aforementioned literature, has focused on how markets react to exogenous events and shocks including natural or anthropogenic catastrophes and accidents, political risk and violent events such as conflict and terrorism while the contagion potential of this reaction has also been the subject of empirical investigation (inter alia: Kaplanski and Levy [11]; Capelle-Blancard and Laguna, [12]; Asteriou and Siriopoulos [13]; Herbst et al. [14]; Blose et al. [15]; Kalra [16]; Bowen et al. [17]). Following mega-terrorist attacks of recent years, such as for instance 9/11 in New York, and the Madrid and London bomb attacks of 2004 and 2005 respectively, the number of studies that examine the impact terrorism exerts on the economy in general and on financial markets in particular, has steadily grown (inter alia: Brounrn and Derwall [18]; Ramiah et al. [19]; Graham and Ramiah [20]; Amelie and Darne [21]; Fernandez [22]; Nikkinen and Vahamaa [23]). As it has been pointed out in a number of previous papers (inter alia: Drakos, [24]; Kollias et al. [25,26]; Chesney et al. [27]), although the threat of a terrorist attack is omnipresent, particularly in countries such as Israel, Spain or the UK that are or have been the venues of systematic terrorist activity with the concomitant casualties and damages, terrorist events when they occur are unexpected. Depending among other things on their seriousness in terms of victims, damages or target(s) attacked, they have the potential to shake and rattle investors and markets. Just as in the case of natural or anthropogenic accidents, terrorist attacks are unanticipated. Hence, market agents cannot hedge against them. Such incidents can also have a high contagion potential as studies that have addressed this question and the channels of the cross-market transmission of terrorist induced shocks have shown (inter alia: Hon et al. [28]; Mun [29]; Drakos [30,31]). Factors that seem to affect the transmission potential of such exogenous shocks from the market of the country that has been targeted by the terrorists to others include the degree of bilateral integration between the stock markets and the degree of integration into the global economic and financial markets (see Drakos [30,31]). Moreover, Kollias et al. [32] indicate that terrorist attacks trigger a flight-to-safety effect (from stock to bond market within a country) primarily in France and Germany and to a smaller degree in Great Britain and Spain. Within this particular thematic focus of this strand of literature, this paper addresses the cross-market transmission of the shock generated by a major European terrorist event. The study focuses on the 7 July 2005 London bomb attacks that, along with the 2004 Madrid bombings, are considered to be the European equivalent of 9/11 albeit on a much smaller scale in terms of the number of victims and destruction to property and infrastructure (Kollias et al. [26]). However, unlike previous studies those Int. J. Financial Stud. 2013, 1 156 rely on daily data to assess the impact of terrorist events on financial markets as well as their contagion effect, this study uses high frequency data to investigate the issue at hand. By opening a window on high frequency data, investors may discover profit opportunities not easily detected in daily data. In total, each series of 10-minute interval frequency contains over 10,000 observations. As it has been argued and shown, intraday data help reveal a more accurate picture of how markets and market agents react and adapt to changes and exogenous shocks (inter alia: Connolly and Wang [33]; Hanousek [34]; Égert, and Kočenda [35]; Markelos et al. [36].). Consequently, the more detailed account and the information contained in high frequency data allow more reliable inferences and conclusions to be drawn vis-à-vis daily data. Moreover, recently a series of papers propose successful methods on high-frequency predictions of trading volumes, market depth, bid-ask spreads and trading costs to optimize order placement and order execution. Among others, Groß-Klußmann and Hautsch [37] successfully apply a long memory autoregressive Poisson model to predict bid-ask spreads. Härdle, Hautsch and Mihoci [38] introduce a semiparametric dynamic factor approach to model high-dimensional order book curves. Hautsch, Schienle and Malec [39] propose a novel approach to model serially dependent positive-valued variables which realize a non-trivial proportion of zero outcomes, a quite common phenomenon on high frequency financial data. The important role of automated news feeds on intraday dynamics is recognized by Groß-Klußmann and Hautsch [40]. Significant reactions in volatility and trading activity after the arrival of news items which are indicated to be relevant are identified. Moreover, Hautsch, Hess and Veredas [41] study the impact of the arrival of macroeconomic news on the informational and noise-driven components in high-frequency quote processes and their conditional variances. One of their major findings is that all volatility components reveal distinct dynamics and are positively influenced by news. Therefore, this paper following this line of literature use a modified BEKK(1,1)-GARCH model on intraday data in order to incorporate any possible effect on stock market volatilities and covariance of the mega-terrorism event occurred in July 2005. This type of modeling suggested can be generalized to incorporate other insecurity shocks when studying two variables of interest. More specifically, it examines how two other major European stock markets—Paris and Frankfurt—were affected by the 7 July 2005 mega-terrorist attack in London. The choice of the markets was very much dictated by data availability constraints and the level of capitalization. In the section that follows we proceed with the presentation of the data and the methodology employed. In section three, the findings are presented and discussed, while section four concludes the paper. 2. Data and Methodology The London terrorist incident, involved a series of coordinated suicide bomb attacks that targeted the city’s public transport system during the morning rush hour of 7 July 2005. The bomb blasts caused 52 fatalities and injured 700 people. They also caused extensive and widespread disruption of the city's transportation system and of the mobile telecommunications infrastructure. As the results reported by Kollias et al. [26] show, the London Stock Exchange (henceforth LSE) suffered significant negative abnormal returns on the day of the attack. Both the general as well as sectoral indices were Int. J. Financial Stud. 2013, 1 157 negatively affected by the event. However, this impact was rather short lived since the market quickly recovered and rebounded. Market volatility was also significantly affected but again, this was of a transitory nature (Kollias et al. [26]). The use of 10-mimute interval data by this study will shed more light on how the market reacted and market agents adjusted to the event as it unfolded between 08:50 and 09:50 given that the bomb blasts took place in different moments in the morning of 7 July 2005. Furthermore, the paper will examine whether or not contagion between financial markets can be established in the case of this major terrorist incident that took place in a city that is one of the most important financial and trading centres of the world. Given LSE’s significance as one of the major financial markets globally with a market capitalisation well over $ 3 trillions 1, one would intuitively expect that the shock waves could have been transmitted to other major European markets. The possible contagion and shock transmission is examined through the use of multivariate GARCH models. Apart from the London stock exchange, our data set consists of data for the German and French stock market returns over the period 21 January 2005to 28 October 20052. We have to mention at this point that all data are expressed in British summer time (BST) in order to take into account time differences between countries of the study and compare the results of our findings. Frankfurt and Paris are the second and third largest markets in Europe in terms of capitalization. Hence, one can intuitively expect that if shock waves were indeed transmitted from London to other European markets, Frankfurt and Paris are the markets with the highest probability of contagion given the degree of integration between these three large European Union economies and markets. The German stock returns are calculated from the DAX-30 index, the UK returns from the FTSE-100 index, and the French returns from the CAC-40 index. Apart from the information rich nature of such frequency data, a further advantage stemming from the use of intraday data over a short time period is that the possibility of structural breaks is much smaller compared to a longer time period needed if daily data was employed. Indeed, the multivariate GARCH model might give inaccurate forecasts if the underlying process which generates asset prices undergoes a structural break. Moreover, intraday data capture all the main features of the data generating process. In our case the bivariate unrestricted BEKK-GARCH (1,1) model, proposed by Engle and Kroner [42], is used to investigate any possible contagion effects between the London, the German and the French stock markets on the day of the terrorist attack in question. What will be examined is whether or not and to what extent this terrorist event affected the volatilities and the correlation of the stock markets in question using intraday data given that these effects may be hidden in a daily frequency (see for instance previous studies on daily data: Fernandez [22]; Ramiah et al. [19]). In order to avoid any severe convergence problems, the bivariate unrestricted version of the general BEKK(p,q)-GARCH model with p = q =1 is used: tiiti R,,     (1) with ),0(~| 1ttt HGED    and Ht = 0 ' 0CC +  ' 11 ' -- tt  + BHB t1 ' - (2) 1 LSE was the bigger in terms of market capitalization in 2005, in Europe followed by the German and French markets (see: www.world-exchanges.org/statistics). 2 Data are collected by http://www.tickdata.com. Int. J. Financial Stud. 2013, 1 158 where Equation (1) gives the expression for the conditional mean; Ri,t, and ει, t are the return vector (i = 1 for the FTSE-100 index and 2 for the other two indices for each estimation), and the residual vector respectively; and μi is the mean of this process. In Equation (2) H is the conditional variance-covariance matrix that depends on its past values and on past values of ει, parameter. C0, is a 2 × 2 matrix, the elements of which are zero above the main diagonal; and A, B are matrices. More analytically: Ht = ' 11 21 22 0 c c c            2221 11 c 0 c c+ ' 2221 1211 α α         a a' 11  tt          2221 1211   + ' 2221 1211           1t H        2221 1211   The main advantage of the BEKK-GARCH model is that it guarantees by construction that the covariance matrices in the system are positive (Engle and Kroner, [42]). Assuming multivariate General Error Distribution (GED), the maximum likelihood methodology is used to jointly estimate the parameters of the mean and the variance equations. More specifically, in an attempt to identify the possible effects the terrorist incident in question had on FTSE and y (where y =DAX or CAC) stock index returns co-movement, we employ the unrestricted BEKK-GARCH (1, 1) model including a dummy variable about terror activity in 7 July 2005 in the construction of variances and covariance matrices. This dummy variable takes the value of one for the 10-minute ticks between 7 July 2005 8:50 and 7 July 2005 10:50 (British summer time, BST) and zero anywhere else3. Therefore, the functional form of our model is the following: tiiti R,,   (3) with ),0(~| 1ttt HGED    and t H = 0 ' 0CC +  ' 11 ' -- tt  + BHB t1 ' -+t D um   (4) where the K is the coefficient matrix for terror index and the operator “•” is the element by element (Hadamard) product. In that case the model may be written in single equation format as follows: t tttttttt Dum hhhch 11 1,22 2 211,1221111,11 2 11 2 1,2 2 211,21,12111 2 1,1 2 11 2 11,11 22      (5)   ttt tttttt Dumhh hcch 121,2222211,1222111221 1,111211 2 1,222211,21,122111221 2 1,112112111,12       (6) tt ttttttt Dumh hhcch 221,22 2 22 1,1222121,11 2 12 2 1,2 2 221,21,12212 2 1,1 2 12 2 22 2 21,22 22       (7) The error terms in each model represent the effect of news in each model on the different indices. In particular, the terms 2 1,2 2 1,1 , tt  represent the deviations from the mean attributed to the unanticipated event in each market. The cross values of the error terms 1,21,1  tt  represent the news in the first and second index in time of period t – 1. By 1,121,221,11 ,,  ttt hhh we describe the conditional variance for the first stock index (in our case FTSE-100) at time t – 1, conditional variance for the second stock index 3 In order to count for the effect induced by the events occurred in 08:50 and 09:50 respectively we have used a dummy taking the value of one from the first bomb explosion lasting one hour and for the second bomb also lasting also one hour. Int. J. Financial Stud. 2013, 1 159 (in our case CAC-40 or DAX-30) at time t – 1, and the conditional covariance between the first and the second index in our model. 3. Findings and Discussion As mentioned above, daily data can often conceal a significant part of underlying dynamics of a time-series especially when it comes to the reaction and eventual adjustment to an unanticipated event such as the terrorist attack in question here. Table 1 provides information using daily data for ±1 days around the event day. As it can easily be seen, the markets in the event day exhibit negative returns (FTSE –1.37%, CAC –1.39% and DAX –1.86%). However, a more detailed scrutiny of the data presented in the last column of Table 1, reveals that the difference between the high and low values during the day is appreciably higher for each market compared to the relevant values of the ±1 days window. This reinforces the argument that high frequency data of the type used here contains a more accurate and credible account of how markets react and adjust to exogenous and unanticipated events; in this case the terrorist attacks of 7 July 2005 in London. Hence, vis-à-vis studies that employ daily observations to examine the impact terrorism exerts on financial markets, the intraday data used in the estimations that follow, have an advantage in that they allow for more reliable inferences to be drawn. Indeed, Figure 1, plotted with the use of high frequency 10-minute interval data, graphically reveals the magnitude of market agents’ reaction to the event in question as it unfolded between 08:50 and 09:50. All three stock markets exhibit a negative reaction to the news as the event started unfolding. It seems that, a considerable amount of selling orders exerted a significant downward pressure in all three cases. The markets begin to recover after 10:50, probably as a result of discounting the short-term economic, political and security repercussions of the incident4. Table 1. Stock Prices ± one day of the event—Daily data frequency. Date Open High Low Close Daily Return High-Low 6 July 2005 5190.10 5237.60 5190.10 5229.60 0.758% 47.50 FTSE-100 Index 7 July 2005 5229.60 5229.60 5022.10 5158.30 –1.373% 207.50 8 July 2005 5158.30 5232.20 5158.30 5232.20 1.422% 73.90 6 July 2005 4607.57 4636.96 4607.57 4615.49 0.257% 29.39 DAX-30 Index 7 July 2005 4595.23 4595.23 4444.94 4530.18 –1.866% 150.29 8 July 2005 4560.43 4597.97 4559.57 4597.97 1.485% 38.40 6 July 2005 4272.64 4292.07 4264.00 4279.95 0.638% 28.07 CAC-40 Index 7 July 2005 4269.56 4269.77 4089.27 4220.62 –1.396% 180.50 8 July 2005 4264.71 4300.31 4252.07 4300.31 1.871% 48.24 4 See for instance Kollias et al. [26] that compare this attack to the one in Madrid in 2004. Int. J. Financial Stud. 2013, 1 160 Figure 1. Stock Prices the days around and during the event. The stock market returns are used to conduct the empirical analysis that follows. Table 2 presents the descriptive statistics for the return series in all three stock markets. In terms of the mean, standard deviation and maximum returns, the three markets present fairly similar characteristics. Skewness and kurtosis measures indicate deviation from normality. The latter is confirmed by the Jarque-Bera test that provides evidence against normally distributed tick-by-tick returns. Therefore preliminary statistical analysis confirms well-known stylized facts of financial markets including significant asymmetry and kurtosis. Hence, the use of GARCH type models as a tool to take into account nonnormal covariations between stock index returns seems to be appropriate. Table 2. Descriptive Statistics of Intraday data. CAC40 Index DAX30 Index FTSE -100 Index Mean 1.16 × 10–5 1.35 × 10–5 8.09 × 10–6 Maximum 0.0128 0.0156 0.0101 Minimum –0.0168 –0.0229 –0.0138 Std. Dev. 0.0009 0.0011 0.0007 Skewness –0.9576 –1.9041 –0.7345 Kurtosis 40.8166 62.1455 40.6416 Jarque-Bera 612336.3 1500214 606051.9 Probability (0.00) *** (0.00) *** (0.00) *** Observations 10250 10250 10250 Notes: The sample contains every ten minutes index returns from 21 January 2005 to 28 October 2005. The total number of usable observations is 10250. The values in parenthesis are the actual probability values. *** indicates statistical significance at 1% level. The estimated results for the unrestricted BEKK-GARCH (1,1) model are presented in Table 3 for both pairs of indices i.e., the FTSE-CAC (column RFTSE-RCAC) and FTSE-DAX (column RFTSE-RDAX) with the concomitant diagnostics. As far as the whole sample is concerned the majority of the estimated parameters are statistically significant, with the only exception being the coefficient c22 for the FTSE-CAC pair and the coefficient k12 for the FTSE-DAX pair. Int. J. Financial Stud. 2013, 1 161 Table 3. BEKK-GARCH estimation results. RFTSE-RCAC RFTSE-RDAX Whole Sample Pre-Bomb period 21/01/2005-05/07/2005 Post-Bomb period 07/07/2005-28/10/2005 Whole Sample Pre-Bomb period 21/01/2005-05/07/2005 Post-Bomb period 07/07/2005-28/10/2005 Coeff Signif. Coeff Signif. Coeff Signif. Coeff Signif. Coeff Signif. Coeff Signif. μ1 2.06E-05 (0.00) *** 2.34E-05 (0.00) *** 1.20E-05 (0.21) 1.43E-05 (0.01) ** 2.20E-05 (0.01) ** 1.17E-05 (0.19) μ2 3.53E-05 (0.00) *** 3.70E-05 (0.00) *** 1.60E-05 (0.23) 2.29E-05 (0.00) *** 3.07E-05 (0.00) *** 1.89E-05 (0.16) c11 1.38E-04 (0.00) *** 1.21E-04 (0.00) *** −2.12E-04 (0.00) *** 1.69E-04 (0.00) *** 3.67E-04 (0.00) *** 1.71E-04 (0.00) *** c21 -2.03E-04 (0.00) *** −1.97E-04 (0.00) *** 1.23E-04 (0.00) *** −1.71E-04 (0.00) *** −1.78E-04 (0.00) *** 2.10E-04 (0.00) *** c22 -9.49E-08 (0.99) 3.40E-07 (0.99) −2.00E-09 (0.99) 9.05E-05 (0.01) ** −3.63E-07 (0.99) −2.86E-04 (0.00) *** α11 0.0708 (0.00) *** −0.0103 (0.68) 0.3132 (0.00) *** 0.0466 (0.00) *** −0.3445 (0.00) *** 0.1215 (0.00) *** α12 -0.1859 (0.00) *** −0.0809 (0.00) *** 0.2333 (0.00) *** −0.3094 (0.00) *** −0.0891 (0.00) *** −0.3445 (0.00) *** α21 0.4181 (0.00) *** 0.5131 (0.00) *** −0.0476 (0.09) * 0.3560 (0.00) *** 0.5186 (0.00) *** 0.2712 (0.00) *** α22 0.6211 (0.00) *** 0.6166 (0.00) *** 0.3092 (0.00) *** 0.6613 (0.00) *** 0.6731 (0.00) *** 0.6776 (0.00) *** β11 0.9832 (0.00) *** 0.9825 (0.00) *** 0.6514 (0.00) *** 0.9622 (0.00) *** 0.0686 (0.53) 0.9693 (0.00) *** β12 0.2914 (0.00) *** 0.2375 (0.00) *** −0.1966 (0.00) *** 0.2865 (0.00) *** −0.1268 (0.02) ** 0.1226 (0.00) *** β21 -0.1112 (0.00) *** −0.1316 (0.00) *** 0.2478 (0.00) *** −0.0851 (0.00) *** 0.5139 (0.00) *** −0.0756 (0.00) *** β22 0.6962 (0.00) *** 0.7085 (0.00) *** 0.9896 (0.00) *** 0.7365 (0.00) *** 0.9177 (0.00) *** 0.8010 (0.00) *** κ11 1.96E-03 (0.00) *** 1.34E-03 (0.01) ** κ12 3.20E-03 (0.00) *** 1.20E-03 (0.24) κ22 1.51E-03 (0.00) *** 1.79E-03 (0.00) *** GED Parameter 0.9268 (0.00) *** 0.9218 (0.00) *** 0.9117 (0.00) *** 0.9459 (0.00) *** 0.9122 (0.00) *** 0.9577 (0.00) *** Observations 10250 6068 4182 10250 6068 4182 Log Likelihood 121740.71 72829.31 49053.41 120011.73 72276.21 48032.47 Notes: ***,**,* indicate statistical significance at 1%,5% and 10% level.