GRU-Enhanced Volatility Prediction Using FIGARCH and PLM-GARCH: A Comparative Hybrid Model

This paper proposes a novel hybrid approach to financial volatility forecasting by integrating Gated Recurrent Unit (GRU) neural networks with two advanced GARCH-type models: the Fractionally Integrated GARCH (FIGARCH) and the Periodic Long-Memory GARCH (PLM-GARCH). The goal is to enhance predictive accuracy by capturing both long-memory dynamics and periodic volatility structures in financial time series, while leveraging the nonlinear sequence learning capabilities of GRUs. Using daily closing prices of the Tehran Energy Index from ۲۰۱۶ to ۲۰۲۳ (۱۳۹۶ observations), we construct two hybrid models_x۰۰۱۶_۰_x۰۰۱۹_۸GRU-FIGARCH and GRU-PLM-GARCH_x۰۰۱۶_۰_x۰۰۱۹_۹ and evaluate their performance against traditional and standalone models. The results indicate that GRU-based hybrids significantly outperform their counterparts, with GRU-PLM-GARCH achieving the best performance across all metrics (RMSE, MAE, MAPE, R_x۰۰۱۶_۹). This highlights the importance of combining statistical memory models with deep learning techniques for robust and adaptive volatility prediction in complex financial environments.

کلیدواژه ها:

Volatility Forecasting ، GRU ، FIGARCH ، PLM-GARCH ، Hybrid Models ، Periodic Long Memory