Volatility transmission patterns and terrorist attacks

Abstract

The objective of this study is to analyse volatility transmission between the US and Eurozone stock markets considering the financial market responses to the September 11, March 11 and July 7 terrorist attacks. In order to do this, we use a multivariate GARCH model and take into account the asymmetric volatility phenomenon, the non-synchronous trading problem and the turmoil periods themselves. Moreover, a graphical analysis of the Asymmetric Volatility Impulse-Response Functions (AVIRF) is introduced, which takes into consideration the financial market responses to the terrorist attacks. Results suggest that there is bidirectional and asymmetric volatility transmission and show the different impacts that terrorist attacks had on both markets.

Keywords:

• Volatility modelling
• International finance
• International asset pricing
• GARCH models
• Multivariate volatility
• Risk management

Acknowledgements

Financial support from the Spanish Ministry of Education and Science (contract No. SEJ2006-15401-C04-04/ECON), Generalitat Valenciana (contract No. GV/2007/082), Instituto Valenciano de Investigaciones Económicas (IVIE), and Cátedra en Finanzas Internacionales–Santander is gratefully acknowledged. The authors also thank seminar participants at the XIV Foro de Finanzas in Castelló and the 2006 EFMA Annual Meeting in Madrid, several anonymous referees and the editor of this journal for their helpful comments.

Notes

†See also Allen and Gale (2000) for theoretical models on financial contagion.

‡See Koutmous and Booth (1995), Karolyi (1995), Karolyi and Stulz (1996), Darbar and Deb (1997), Ramchand and Susmel (1998), Brooks and Henry (2000), Longin and Solnik (2001), Martens and Poon (2001) and Bera and Kim (2002), inter alia.

†A special issue of the Economic Policy Review of the Federal Reserve Bank of New York (2002, Volume 8, Number 2) analyses general economic consequences of September 11. A special issue of the Journal of Risk and Uncertainty (2003, Volume 26, Numbers 2/3) deals with the risks of terrorism with a special focus on September 11. A special issue of the European Journal of Political Economy (2004, Volume 20, Issue 2) deals with the economic consequences of terror.

‡The use of local, regional and global shocks is similar to the taxonomy for crisis transmission proposed by Dungey and Martin (2007). These authors propose a model that captures a range of common factors, including global shocks, country and market shocks, and idiosyncratic shocks.

†The Dow Jones EURO STOXX 50 Index, Europe's leading blue-chip index for the Eurozone (that is, the European Monetary Union), provides a blue-chip representation of supersector leaders in the Eurozone. The index covers 50 stocks from 12 Eurozone countries: Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, and Spain.

‡As the New York Stock Exchange (NYSE) was closed during September 11 and the three following days, the S&P500 index value recorded the previous day is used for those dates.

§Cointegration between both indexes was also rejected at the usual significance levels when using closing values of both indexes instead of prices recorded at 15:00 GMT.

¶The authors thank one of the referees for this comment.

†The specification in equations (6) to (9) for testing crises contagion can be seen as a particular case of the general framework proposed by Dungey et al. (2005). In our model, the ‘contagious transmission channel’ (in the words of Dungey et al. (2005)) is measured through the significance of w in the market receiving the crisis contagion.

†This could be due to the fact that prices are recorded at 15:00 GMT, when European markets are about to close and the US market has just started trading.

†The standard deviation of the responses depends on the estimated coefficients and their covariance matrix (see equation (8) of Lin (1997)). The exploding confidence bands in figures 5(a) and (c) are due to the large values of the estimated coefficients in matrix L (see table 4), even though GARCH coefficients satisfy stationarity conditions. This instability can be avoided in several ways, for instance by increasing the number of days that this dummy takes values different from zero. Nevertheless, when this is done, no coefficient in L remains statistically significant.