Phichhang Ou and Hengshan Wang
GARCH, FIGARCH, relevance vector machines, nonparametricvolatility
In this study, we propose a nonparametric model of financial volatility based on the relevance vector machine (RVM). The RVM algorithm is fitted twice to the conditional mean and then to the conditional variance of stock returns so that risk and return are related. Simulated data based on GARCH, EGARCH and GJR, and real data of Dow Jones industrial average (DJI) are analysed to validate our proposed model. Parametric GARCH-type models, including double long memory model ARFIMA-FIGARCH, are also applied to compare with the RVM model. The experimental results find that the proposed model yields significant improvement in volatility forecasts from both simulated and real data.
Important Links:
Go Back
