Walter Enders & Todd Sandler
Abstract
This paper applies a threshold autoregression (TAR) model to a casualties time series to show that the autoregressive nature of such events depends on the level of terrorism at the time of a shock. Following a shock, persistence of heightened attacks characterizes low-terrorism regimes, but not high-terrorism regimes. Similar findings are associated with incidents with deaths, bombings with deaths, and hostage taking. In contrast, the assassinations series indicates some persistence even in the high-terrorism state, while the threats/hoaxes series displays persistence in only the high-terrorism state. For all series studied, the TAR model outperforms a standard autoregressive representation. A forecasting method is engineered based on the TAR estimates, and nicely tracks resource-using events.