Integrating Oil Price Shocks and China–US Geopolitical Risk: A GRU-BiLSTM-Transformer Dynamic Fusion Forecasting Framework for USD/CNY Volatility

Abstract

Forecasting USD/CNY exchange rate volatility is of substantial practical significance for the management of cross-border capital flows and the formulation of monetary policies. In recent years, multiple external shocks arising from China–US geopolitical tensions and sharp fluctuations in the international crude oil market have become increasingly intertwined. Traditional forecasting models are often unable to capture such sudden risk signals in a timely manner, while a single model is also insufficiently adaptable to multi-scale volatility structures. To address these challenges, this paper develops a deep-learning-based dynamic forecasting framework that integrates a China–US geopolitical risk feature system (CUGRI) with international oil price shock signals. Specifically, CUGRI is constructed from news texts published by 15 authoritative Chinese and US media outlets, combining a finance-specific language model with a geopolitical-domain sentiment lexicon to build a five-dimensional daily risk quantification feature system. At the modeling level, this paper proposes a GRU-BiLSTM-Transformer dynamic fusion model, which adaptively assigns fusion weights according to the recent forecasting performance of each sub-model. All empirical analyses are implemented under the Python programming environment with the PyTorch deep learning framework. Using nearly ten years of daily data, this paper conducts out-of-sample forecasting tests. The empirical results show that the proposed dynamic fusion model achieves an out-of-sample R2 of 0.342, while reducing RMSE and MAE by 4.01% and 3.72%, respectively, relative to the best-performing baseline model. CUGRI exhibits significant incremental predictive value, with forecasting gains substantially higher during periods of elevated geopolitical risk than under normal conditions. Oil price shocks provide complementary predictive information, and the dynamic weighting mechanism further improves the accuracy of combined forecasts. In practice, the findings are relevant to cross-border firms and financial institutions when tracking changes in geopolitical and energy-market risks and adjusting foreign-exchange hedging strategies. For monetary and regulatory authorities, the model helps identify periods when external shocks may amplify USD/CNY volatility and supports exchange-rate risk surveillance and macro-financial stability analysis.