Forecasting Rebel Violence = How Diffusion Patterns of Violence Improve the Prediction of Future Conflicts

다국어 초록 (Multilingual Abstract)

Ensuring the safety of civilians and security personnel in conflict zones remain one of the most significant challenges to the United Nations and troop contributors. This study addresses this issue by predicting future locations of violent event based on empirical investigation of rebel activities. Making such predictions includes a three-step process. First, diffusion patterns of rebel activities are analysed using the contagion model of near-repeat victimization. This method enables to assess how the risk of violent events at a given location is transmitted to neighbouring locations within a limited time. Next, an autoregressive distributed lag model is specified based on diffusion patterns of violence identified in the first stage. This model is then applied to monthly, disaggregated data on African civil wars over the period 1997-2012. Finally, estimation results are used to predict the location of conflict in the following month at the level of 55×55km grids. The generated predictions pose both promises and challenges of conflict prediction, with the potential to provide security personnel with timely and policy-relevant information on future conflict.