Early ViEWS: A prototype for a political Violence Early-Warning System

Abstract:

In this paper we present prototype predictions of intra-state violence from the Violence Early Warning System (ViEWS), a new open-source, forecasting project. The goals for the system are to provide the academic and humanitarian communities with useful and accurate forecasts of the timing and location of different forms of political conflict as well as allow public access to the tools and data that created them. We target forecasting violent events at the PRIO-GRID level with lead times of one month. While we are building the project to be global in scope, we begin with a focus on intra-state violence in Africa. Our approach brings together work on the underlying structural risks of violence, which has identified slowly moving indicators such as GDP per capita, with research on quickly emerging triggers observable in news reports. The core of the project is an ensemble of theoretically informed static and dynamic non-linear models, each of which are weighted and averaged into our overall forecast. We evaluate the performance of our sub-models and overall ensemble prediction using proper scoring rules.

Authors:

Colaresi, Michael, Håvard Hegre and Jonas Nordkvelle