Dynamic forecasting of exchange rate spillovers with TVP-VAR and deep learning models

Juan Liu, Huiming Zhu, Zishan Huang, Lingfeng Deng,
Dynamic forecasting of exchange rate spillovers with TVP-VAR and deep learning models,
Finance Research Letters,
Volume 86, Part E,
2025,
108677,
ISSN 1544-6123,
https://doi.org/10.1016/j.frl.2025.108677.
(https://www.sciencedirect.com/science/article/pii/S1544612325019312)
Abstract: This study develops a novel framework integrating a deep-learning forecasters with Time-Varying Parameter VAR to analyze and predict dynamic spillover effects in global currency markets. Based on exchange rate data, we estimate time-varying connectedness and evaluate four deep-learning architectures—DeepAR, TiDE, PatchTST, and TFT—for volatility forecasting, while incorporating Economic Policy Uncertainty (EPU) as an exogenous predictor. Empirical Results reveal pronounced asymmetries: advanced currencies (EUR, GBP, AUD) are net risk exporters, whereas emerging currencies (ARS, TRY, RUB, IDR) are net importers. The inclusion of EPU substantially enhances predictive performance, particularly within TFT models. Findings suggest advanced economies should coordinate policies to limit spillovers, emerging markets need stronger buffers, and institutions should integrate EPU-enhanced forecasts into risk management systems.
Keywords: TVP-VAR; Exchange rate; Deep learning; TT-Transform; LSTM Encoders