Artificial intelligence and exchange rate forecasting: assessing predictive accuracy and macroeconomic sensitivity

Abstract

This study evaluates the effectiveness of Machine Learning (ML) models in forecasting the EUR/USD exchange rate from January 2014 to December 2024, focusing on the relationship between forecast errors and key macroeconomic indicators, including interest rates, inflation, unemployment, and GDP growth. The forecasting framework integrates three widely used architectures—Multilayer Perceptron (MLP), Random Forest (RF), and Long Short-Term Memory (LSTM) networks—applied to monthly exchange rate and macroeconomic data drawn from Yahoo Finance, the World Bank, and the International Monetary Fund. The findings indicate that the LSTM model outperforms both MLP and RF in predictive accuracy, achieving an R-squared value of 0.9234. While all models demonstrate strong short-term forecasting performance, macroeconomic variables have limited explanatory power regarding forecast error, with only Eurozone interest rates showing weak statistical significance. These results suggest that ML models can effectively model exchange rate dynamics even when macroeconomic indicators provide limited statistical relevance. The study contributes to the literature on AI in financial forecasting by highlighting the comparative strengths of deep learning and ensemble methods and identifying persistent challenges in integrating economic fundamentals, underscoring the value of hybrid and interpretable AI frameworks that bridge macroeconomic theory and data-driven learning for improved financial forecasting.