A synergy of econometrics and computational methods (GARCH-RNFS) for volatility forecasting

This paper presents the application of a rough-set based neuro-fuzzy system (RNFS) in volatility forecasting by synergizing the information extraction of popular generalized auto-regressive conditional heteroscedasticity (GARCH) models with the human like interpretable RNFS. Additional intraday volatility indicators such as realized power variation (RPV) are proposed to further boost volatility forecasts in the hybrid model. Experiments are performed on real-life data (Citigroup and J.P Morgan price series) to compare the volatility forecast and interpretability of the hybrid model against the commonly used GARCH, Exponential GARCH (EGARCH) and Glosten-Jagannathan-Runkle GARCH (GJR-GARCH) models and other soft computing methods. The results show that the accuracy of the proposed hybrid system can match or outperform the GARCH models and other soft computing methods. It also yields improved interpretability in terms of number of if-then fuzzy rules compared against other soft computing methods.