A regional investigation of the interrelationships between domestic and transnational terrorism: a time series analysis

Abstract

It is generally deemed that domestic and transnational terrorism respond to different types of events. This study updates Enders, Sandler, and Gaibulloev’s previous analysis to include data through the fourth quarter of 2010 and provides analysis of terrorism at the regional level. Vector autoregressions are used to show that previous findings are accurate on the whole, but that there are important differences between regions. Notably, the Granger-causality for the world depends on whether Iraq and Afghanistan are included in the sample, and impulse response functions highlight the persistent effect a shock to transnational terrorism can have on domestic terrorism.

Keywords:

• Granger-causality
• Domestic versus transnational terrorism
• Regional terrorism

JEL Codes:

• C32
• F51
• F52
• F68

Notes

¹ Note that the regional weights are likely to be different from the aggregate weights reported in ESG. However, since both the domestic and transnational series are multiplied by the same scalar, the vector autoregressive results are not likely to be due to the weights used in the analysis. We confirmed this conjecture by experimenting with various hypothetical weights.

² Results using the negative binomial were included in an earlier version of this paper that is available from the authors on request.

³ Gries, Krieger, and Meierrieks (2011) also use a VAR in order to perform causality tests concerning the relationship between terrorism and the economic performance of seven western European countries.

⁴ Note that the table reports results using robust (i.e. heteroskedastic consistent) standard errors. Results for using the Poisson distribution are reported as a diagnostic check of the OLS results. Granger-causality results using the negative binomial distribution are available from the authors.

⁵ If all coefficients of A₁₂(L) = 0, lagged values of transnational terrorism do not appear in the equation for domestic terrorism, and if all coefficients of A₂₁(L) = 0 lags of domestic terrorism do not appear in the equation for transnational terrorism.

⁶ Results are also available from the authors for estimation with a negative binomial distribution.

⁷ Impulse response functions for the reverse ordering are available in an unpublished referee’s appendix.

⁸ As discussed in Enders (2010), the BEKK specification imposes a quadratic form on the GARCH coefficients so that the estimated variances are necessarily positive. Given that (1) and (2) each have as many as four lags of each variable, the multivariate GARCH(1, 1) system can have a total of 29 coefficients. In order to save space, the estimated coefficients are not reported here; they are available from the authors upon request.

⁹ Specifically, the smoothed values in period t (S_t) are constructed as S_t = S_t−1 + w(ρ_t – S_t−1), where w is the smoothing weight.