Transitions into and out of food insecurity: A probabilistic approach with panel data evidence from 15 countries

Recent advances in food insecurity classification have made analytical approaches to predict and inform response to food crises possible. This paper develops a predictive, statistical framework to identify drivers of food insecurity risk with simulation capabilities for scenario analyses, risk assessment and forecasting purposes. It utilizes a panel vector-autoregression to model food insecurity distributions of 15 countries between October 2009 and February 2019. Least absolute shrinkage and selection operator (LASSO) methods are employed to identify the most important agronomic, weather, conflict and economic variables. The paper finds that food insecurity dynamics are asymmetric and past-dependent, with low insecurity states more likely to transition to high insecurity states than vice versa. Conflict variables are more relevant for highly critical states, while agronomic and weather variables are more important for less critical states. Food prices are predictive for all cases. A Bayesian extension is introduced to incorporate expert opinions through the use of priors, which can lead to significant improvements in model performance.