Genocide Forecasting: Past Accuracy and New Forecasts to 2020

Abstract

We assess the accuracy of genocide forecasts made by the Atrocity Forecasting Project (AFP) for 2011–15, and present new forecasts for 2016–20. Using data from the United Nations, Genocide Watch and the Political Instability Task Force, we evaluate AFP accuracy. We compare AFP accuracy with that of forecasts from the Genocide Prevention Advisory Network. It is relatively rare in most areas of social science that researchers produce (and make public) future forecasts. It is rarer still to evaluate their accuracy once the future has arrived. AFP five-year forecasts are potentially important for genocide and politicide prevention, and have gained attention from policy makers and news media, but a systematic assessment of their accuracy has not been undertaken previously. Our evaluation of past forecast accuracy, with true-positive rates from thirty-three to fifty per cent, true-negative rates around ninety per cent, and area under the curve (AUC) statistics from .81 to .96, gives an indication of how much confidence should be placed in the 2016–20 forecasts.