Exchange rate forecasting with macroeconomic data: Evidence from a novel comprehensive ensemble approach

Abstract

The foreign exchange market operates as a high-dimensional, dynamic, and complex system influenced by multiple factors and their interrelations. In this paper, we propose a comprehensive ensemble framework for exchange rate forecasting that effectively captures the intricate fluctuation patterns inherent in exchange rate data. Our framework integrates economic theories, technical indicators, and other relevant factors to enhance predictive accuracy. To achieve this, we first decompose the exchange rate time series using ensemble empirical mode decomposition with adaptive noise (CEEMDAN). The resulting components are then segmented into high- and low-frequency groups using the Wilcoxon rank test. Based on macroeconomic fundamentals and technical indicators, we select predictive variables to be included in the model. Next, we conduct comparative experiments to verify the role of export and import (EI) data in exchange rate forecasting. We employ the time convolutional network (TCN) model to predict four important exchange rate time series. The empirical results—validated across forecasting horizons of 1, 3, and 6 months—consistently demonstrate that the proposed method outperforms benchmark models, offering a more accurate and reliable framework for exchange rate predictions. These findings underscore the robustness and predictive power of our approach, confirming its effectiveness in anticipating fluctuations in exchange rates over different time scales. The results highlight the strong correlation between exchange rates, macroeconomic conditions, and investment transactions. Moreover, the comparative experiments reveal that the inclusion of EI data significantly improves the prediction accuracy of the model, emphasizing the importance of this factor in exchange rate forecasting.