Topic: New Approaches to Forecasting Civil Violence and Geopolitical Crises

Presenter: Mahda Soltani

Session Details: Predicting the onset of geopolitical crises is challenging due to two key issues: the difficulty of distinguishing new crises from ongoing ones and the lack of high-frequency covariates capturing rapid political changes. While much research has focused on armed conflict, forecasting the onset (as distinct from the continuation) of these conflicts remains particularly difficult. To address these challenges, we: First, leverage granular, locally-sourced data from the Machine Learning for Peace (MLP) project—which offers monthly civic space measures from 2012 to 2023—to predict conflict onset. Second, beyond armed conflict, we extend our analysis to predict State Department travel advisories, a broader measure of national risk that includes political violence, natural disasters, and social unrest. Third, we introduce methodological innovations, such as transforming panel data into lagged training data to capture the time-dependent nature of political events and employing temporal cross-validation tailored for panel data. Our models demonstrate strong predictive performance and provide interpretable results on the role of key predictors, such as civic unrest, government interventions, and electoral stability in forecasting crises. By addressing the broader challenges of onset prediction, this research contributes both to academic literature and offers practical tools for policymakers to better anticipate and respond to a range of geopolitical crises.

This session will be a seminar and participants are not required to read a paper beforehand.

For more information, reach out to pdri-devlab@sas.upenn.edu