Forecasting trade flows under geopolitical uncertainty: A hybrid econometric–LSTM approach

Abstract

This study develops a forecasting framework to improve the prediction of international trade flows under increasing geopolitical uncertainty, trade tensions, and global value chain fragmentation. The framework combines econometric and machine learning techniques. Stochastic convergence algorithms are used to identify convergence clubs among countries trading with China and the United States. Principal Components Analysis (PCA) and LASSO regression reduce dimensionality and select relevant predictors, which are then incorporated into a Long Short-Term Memory (LSTM) network. The proposed model outperforms comparable LSTM benchmark specifications across multiple forecast horizons. Panel Diebold-Mariano tests indicate statistically significant forecasting improvements for one-month-ahead predictions in most cases, particularly relative to models that include exogenous variables but exclude convergence club formation. The investigation demonstrates that incorporating cross-country convergence structures enhances forecast accuracy, improves dataset management and model specification, and provides a scalable framework that preserves the flexibility and nonlinear capabilities of deep learning models.