INE Crude Oil Futures Volatility Prediction Based on Deep Learning HAR-RV-CJ-GRU Hybrid Model

Abstract

Aiming at the characteristics of high noise, nonlinearity, and strong volatility of high-frequency data in the crude oil futures market, and to overcome the limitations of traditional financial econometric models in capturing nonlinear features and the overfitting issues in deep learning models, this paper proposes a new hybrid model: Firstly, this study preprocesses the intra-day 5-minute logarithmic return data of INE crude oil futures using wavelet threshold denoising (WTD). Subsequently, a hybrid model combining traditional financial time series models and deep learning algorithms, namely the HAR-RV-CJ-GRU model, is constructed to predict the daily realized volatility series. The prediction performance is compared with that of the single models (HAR-RV, HAR-RV-CJ, GRU) and the HAR-RV-GRU hybrid model, both before and after WTD. The results show that: (1) The high-frequency data of crude oil futures exhibit high noise, nonlinearity, and strong volatility. Wavelet threshold denoising effectively removes noise from the micro-high-frequency data, thereby enhancing the accuracy of volatility prediction. (2) Traditional financial econometric models primarily capture the linear features of financial time series, while deep learning algorithms can effectively capture nonlinear features. The proposed HAR-RV-CJ-GRU hybrid model leverages the strengths of both types of models, significantly reducing the MAE, MSE, and RMSE values and improving prediction accuracy. This demonstrates the superiority of the hybrid model in predicting the volatility of INE crude oil futures.