On Forecasting Conflict in the Sudan: 2009–2012

The Center uses science and technology to reduce armed conflict, sustain families and communities during conflict, and assist states to rapidly recover from conflict. Research has established a fundamental link between food insecurity and conflict. Food prices, particularly wheat prices, appear to be an important focus in much of the literature. A study by Hendrix and Brinkman (2013) reviewed the correlation between conflict events and poverty or food scarcity. This paper considers the forecasting merit of a particular fit relationship: wheat prices and conflict in the Sudan over post-2000 data. The study uses univariate and multivariate models to forecast monthly conflict events in the Sudan from 2009 to 2012. The models were based on a specification from a machine learning algorithm fit to 2000-2008 monthly data. The model that includes previous month's wheat price performs better than a similar model that does not include past wheat prices. Both models did not perform well in forecasting conflict in a neighborhood of the 2012 'Heglig crisis'. The study demonstrates that wheat prices are a mover of conflict numbers, but also shows time periods where wheat prices are clearly not the cause of conflict events. The results highlight the need to develop models capable of offering credible forecasts of extreme events. The study's probabilities do pass a statistical test of credibility, but suffer from overconfidence at particular critical periods. The literature on forecasting conflict is relatively recent, with three threads characterizing these models: expected utility-game theoretic models, logistic regression, and Bayesian vector autoregressions. The study uses a structural Bayesian vector autoregression model to offer probabilistic forecasts of conflict numbers. The model conditions on past wheat prices and is compared to a model that does not condition on previous wheat prices. The study provides evidence that grain prices in the Sudan Granger-cause conflict numbers in the Sudan based on monthly data. Using a model similar in its dynamic structure to that of Brandt et al. (2011) and Chen et al. (2014), the study shows that wheat price explains a considerable proportion of the (within sample) forecast error variance of conflict numbers in the Sudan using monthly data observed over 2001-2012. The data used in the study are monthly wheat prices, conflict numbers, and fatalities from conflict numbers from the Sudan for the period 2001-2012. The nominal price data per ton of wheat reflect the monthly wholesale prices in Khartoum port. The study's findings have implications for the development of models capable of offering credible forecasts of extreme events, particularly in the context of conflict-affected countries.