Machine learning models to measure the importance of Geopolitical Tensions on Trade

We contribute to the international trade literature by extending the analysis of geopolitical tensions into an out-of-sample predictive framework. We use the adaptive nature of the Random Forest to verify whether the rising importance of geopolitics provides a forecasting signal without imposing parametric constraints. We find that it does. Geopolitical distance emerges as a dominant predictor of trade flows, consistently outranking traditional policy variables such as tariffs and RTAs—a result we validate against regularized linear benchmarks. For policymakers, this suggests that forecasting models accounting only for physical distance and traditional cost factors may suffer from systematic bias. • Proposes a random forest model to study trade flows. • Interpret results using classical gravity models.