Forecast approach using neural network adaptation to support vector regression grey model and generalized auto-regressive conditional heteroscedasticity

In order to reduce the volatility clustering effect that deteriorate the efficiency and effectiveness of time series prediction and gives rise to large residual errors, a composite method, which is SVRGM/GARCH model with neural network adaptation, is introduced to improve the predictive accuracy of the complex time series, e.g. stocks price index or futures trading index. A support vector regression (SVR) is employed to improve the control and environment parameters of grey model (GM) denoted by SVRGM. Thus, SVR learning functions to highly reduce the overshoot effect when GM is applied to time series prediction. Moreover, a generalized auto-regressive conditional heteroscedasticity (GARCH) is utilized to resolve the problem of volatility clustering in time series so as to best fit the model. Incorporating GARCH model into SVRGM prediction is scheming to effectively and efficiently tackle two crucial problems, the overshoot and volatility clustering effects, simultaneously. This composite system is adapted optimally by back-propagation neural network (BPNN).