Beyond Long Memory: A Hybrid Periodic GARCH–SVR Framework for Nonlinear Volatility and Risk Forecasting in Financial Markets

Accurate volatility forecasting in financial markets requires capturing long-range dependence, periodicity, and nonlinearities. Traditional GARCH-type models often fail to address these simultaneously. We propose a hybrid PLM-GARCH–SVR model that integrates periodic long-memory GARCH with Support Vector Regression to better model volatility dynamics. Using financial index data, we compare its performance with GARCH, FIGARCH, and machine learning models through loss functions, Diebold-Mariano tests, and Value-at-Risk. Results show that the hybrid model outperforms existing approaches, highlighting the importance of combining periodic memory structures with machine learning for improved volatility and risk forecasting.

کلیدواژه ها:

Volatility Forecasting ، Periodic Long Memory ، PLM-GARCH ، Support Vector Regression ، Hybrid Models ، Value-at-Risk