Exchange rate expectations and explainable AI

This paper examines the determinants of long-term exchange rate expectations using survey data for advanced and emerging economies. We complement standard regressions with explainable machine learning (XGBoost with SHAP values) to capture nonlinearities and interactions. We find substantial differences across country groups, with inflation forecast errors and uncertainty more important in emerging markets. Monetary policy shocks become relevant once nonlinearities are accounted for. Overall, the results highlight state-dependent expectation formation and the value of explainable machine learning as a complement to standard methods.