Wavelet-Driven LSTM Modelling for Exchange Rate Forecasting in BRICS Economies

This study evaluates the predictive performance of Wavelet-LSTM models in forecasting exchange rates for BRICS currencies (INR, CNY, RUB, BLR, and ZAR). The data period studied is from March 2013 to February 2024. The analysis reveals significant variability in model performance, with the Chinese Yuan (CNY) exhibiting robust predictive accuracy, as indicated by low errors and high R² values on both training and test datasets. Conversely, the Indian Rupee (INR) and South African Rand (ZAR) show poor test set performance, highlighting overfitting and limited generalization capabilities. Moderate performance for the Russian Ruble (RUB) and Brazilian Real (BLR) suggests potential for improvement through fine-tuning. These findings underscore the need for strategies such as regularization, cross-validation, and enhanced data preprocessing to address overfitting and improve generalization. A tailored, currency-specific modelling approach is recommended to account for diverse exchange rate dynamics. Future research should explore advanced techniques, including ensemble learning, to enhance model robustness and applicability across varying economic contexts. JEL Classification - C45, F31, C53, O57