Can terrorist threats be forecasted in a systematic way? Which variables help to do so in the most accurate way? The present study examines the relative importance of features when building...Show moreCan terrorist threats be forecasted in a systematic way? Which variables help to do so in the most accurate way? The present study examines the relative importance of features when building forecasting models on terrorist threats. To do so, it draws on both academic literature and publications by counterterrorist practitioners. This study addresses three key gaps in existing research. Specifically, it allows for comparing the utility of different theoretical models to each other, it puts an explicit focus on machine learning-based forecasting with out-of-sample performance metrics, and it explicitly aims to incorporate knowledge from the practitioner sector, which is understandably less open about their work than the academic community but has still produced several insightful publications on the topic of forecasting terrorist threats. The outcomes of the analysis do not confirm the expectation that variables of interest to both academics and practitioners would have the highest predictive power. Rather, it is the population of a country that scores highest, followed by GDP, data on weapon flows into the country, and religious fragmentation in models with no features based on lagged versions of the outcome variable. In models including such variables, the lag of the terrorist attack occurrence consistently scores second highest, and these models consistently out-perform their counterparts missing these variables. The results obtained in this paper are arguably of most use to academic research, in that they add onto a so far relatively limited body of work on out-of-sample forecasting and provide insight into the relative predictive power of existing theoretical models. Practitioners may be more interested in the methodological approach taken in this piece, which can be of use to them when evaluating the priority list of warning indicators to take into consideration when assessing the severity of terrorist threats.Show less
