COMPARATIVE FORECASTING PERFORMANCE OF DAILY EXCHANGE RATE VOLATILITY ACROSS DEVELOPED AND EMERGING ECONOMIES

Abstract

This study comparatively examined the forecasting performance of machine learning and traditional volatility models in predicting daily exchange rate volatility across selected economies from 01 January 2015 to 08 May 2026. The study employed Random Forest (RF), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Ensemble learning (ENS), XGBoost (XGB), GJR-GARCH, and EGARCH models to forecast exchange rate volatility for Australia, Brazil, Canada, China, India, Japan, Switzerland, the United Kingdom, and the United States. Preliminary diagnostics revealed substantial volatility clustering, serial correlation, asymmetry, and non-normality across all exchange rate series. The Augmented Dickey–Fuller and Phillips–Perron tests indicated mixed stationarity properties, while Ljung–Box statistics confirmed strong autocorrelation effects. Forecasting performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results showed that Random Forest achieved the best overall forecasting performance with the lowest average RMSE (0.0057), MAE (0.0042), and MAPE (11506.7780). LSTM, GRU, and ENS models also demonstrated strong predictive performance with average RMSE and MAE values of 0.0058 and 0.0043 respectively. In contrast, traditional econometric models performed poorly, particularly EGARCH, which recorded the highest average RMSE (2.2130) and MAE (1.2519). The findings indicate that machine learning models outperform traditional GARCH-type models in capturing nonlinear dynamics, volatility clustering, and structural instability in foreign exchange markets. The study concludes that AI-based forecasting frameworks offer superior predictive efficiency and should be increasingly integrated into financial risk management, exchange rate policy analysis, and macroeconomic forecasting systems.