Enhancing Exchange Rate Forecasting with Explainable Deep Learning Models

TL;DR

This study tackles RMB/USD exchange-rate forecasting in a volatile, post-2015 environment by leveraging deep learning to capture nonlinear dynamics in a richly featureful multivariate time series. Among nine architectures, transformer-based models—especially TSMixer, FEDformer, and iTransformer—delivered the strongest predictive performance, outperforming LSTM, MLP, and TCN as measured by $MAE$ and $MSE$ across multiple horizons. A ridge-regression feature selection process identified a compact set of 10 influential features, including HS300, CPI differentials, M2, and U.S.-China trade indicators, highlighting the role of fundamental data. Grad-CAM visualizations provided explainability, revealing which features and time points drive predictions and increasing trust in model-driven financial forecasting. Together, the results illustrate the potential of explainable transformer-based deep learning for improving exchange-rate risk assessment and decision-making.

Abstract

Accurate exchange rate prediction is fundamental to financial stability and international trade, positioning it as a critical focus in economic and financial research. Traditional forecasting models often falter when addressing the inherent complexities and non-linearities of exchange rate data. This study explores the application of advanced deep learning models, including LSTM, CNN, and transformer-based architectures, to enhance the predictive accuracy of the RMB/USD exchange rate. Utilizing 40 features across 6 categories, the analysis identifies TSMixer as the most effective model for this task. A rigorous feature selection process emphasizes the inclusion of key economic indicators, such as China-U.S. trade volumes and exchange rates of other major currencies like the euro-RMB and yen-dollar pairs. The integration of grad-CAM visualization techniques further enhances model interpretability, allowing for clearer identification of the most influential features and bolstering the credibility of the predictions. These findings underscore the pivotal role of fundamental economic data in exchange rate forecasting and highlight the substantial potential of machine learning models to deliver more accurate and reliable predictions, thereby serving as a valuable tool for financial analysis and decision-making.

Figures (2)

• Figure 1: Comparison of True Values vs. Predicted Outputs from six models: TSMixer, FEDformer, LSTM, TimesNet, MLP, TCN
• Figure 2: GradCAM visualization indicates feature contribution for forecasting.