Abstract:
|
From 1970 to 2015, there were 7584 terrorist attacks in Latin America and Asia. We investigate modeling these events using dynamic statistical models with a monthly time step. Methodologically, dynamic models are the most straight forward when based entirely on normal probability structures. A potential complication is that the number of terrorist attacks are counts, with a substantial number of zero values when considered on a monthly basis. We consider a traditional additive error dynamic model in which the latent mean process evolves through time following an auto-regressive structure, another latent process follows normal distributions with these means. The actual observation process is then a discretized version of the intermediate latent process. We contrast this model with a model that takes the observation process to follow Poisson distribution directly. The estimation and inference proceed via Markov Chain Monte Carlo methods and the models are assessed based on the combination of the frequency of months with zero attacks and the maximum number of recorded attacks.
|