Artificial Intelligence in International Finance: An Empirical Comparison of Traditional and Machine-Learning Models for EUR/USD Forecasting
0 viewsAbstract
This study investigates the use of artificial intelligence and machine learning in international-finance exchange-rate forecasting. The research compares a traditional ARIMA model with Random Forest and multilayer perceptron (MLP) neural-network models for forecasting the monthly EUR/USD exchange rate. The dataset contains 192 monthly observations from January 2010 to December 2025 obtained from the Federal Reserve Board's foreign-exchange series distributed through FRED. The sample is divided chronologically into a training period covering January 2010 to September 2022 and an out-of-sample test period covering October 2022 to December 2025. Machine-learning models use the previous twelve monthly exchange-rate observations as predictors, while ARIMA is selected using AIC from a small grid of candidate specifications. Forecast performance is evaluated using MAE, RMSE, MAPE, and directional accuracy. The results show that the Random Forest model achieved an RMSE of 0.0231 and MAPE of 1.76%, while the MLP achieved an RMSE of 0.0229 and MAPE of 1.75%. These errors were lower than those of the ARIMA benchmark, which produced an RMSE of 0.1156 and MAPE of 9.71%. The findings indicate that nonlinear machine-learning models can provide useful forecasting information for EUR/USD in the examined sample, while also highlighting the importance of out-of-sample validation and the limitations of relying on historical patterns alone.
Keywords
ARIMA, Artificial Intelligence, Eur/usd, Exchange-Rate Forecasting, Machine Learning, Neural Networks, Random Forest, Time-Series Analysis