Enhancing Time Series Forecasting in Foreign Exchange Markets with a Hybrid Model Based on Histogram-Valued Data

This study explores exchange rate forecasting using a hybrid model with equal weight, alongside traditional models like ARIMA(p, d, q), ETS(A, N, N), TBATS, and the NNAR(p, k) model. We evaluate and compare their performance using RMSE, MAE, and MAPE measures on data spanning from January 2018 to July 2023. Our findings highlight the hybrid model with equal weight as an effective choice for USD/JPY and USD/CAD exchange rate prediction, outperforming other models in terms of lower RMSE, MAE, and MAPE values. In contrast, the ARIMA(p, d, q) model excels in forecasting EUR/USD and USD/CHF, aligning closely with true values. The hybrid model, while not always the best, consistently offers competitive performance, providing a versatile tool for Forex rate prediction. Future research can extend this hybrid model to forecast other financial instruments, including stock indices, digital assets, or commodity markets, offering valuable insights for investors and policymakers.