SVR with Nonlinear Conditional Heteroscedasticity

Abstract

A support vector regression (SVR) is very useful on modeling the predictor for forecasting complex time series. However, SVR cannot avoid volatility clustering effect and thus worst its prediction accuracy. NGARCH(p,q) is utilized for dealing with the problem of volatility clustering or fat-tail effect to best fit the modeling. Therefore, SVR with nonlinear conditional heteroscedasticity is introduced herein in order for dealing with volatility clustering phenomenon while it is applied to forecasting complex financial indexes, e.g. stock price indexes or future trading indexes. This proposed method can get the satisfactory results because of improving its generalization.