New Approach for Non-Periodic Short-Term Forecasting:GARCH(p,q) Smoothing Hybrid Grey-CLMS with BPNN Weighting

Abstract

A new approach, Garch smoothing hybrid GREY-CLMS prediction with BPNN weighting, is introduced herein for the applications of the non-periodic short-term forecasting. The Grey-CLMS hybrid prediction with BPNN weighting for the applications of the non-periodic short-term forecasting in order to overcome the crucial problem, overshooting and undershooting predicted outputs, by the way of compensation between grey and cumulative 3 points least squared linear predictions to yield the pretty satisfactory results. However, some predicted values have been shown not precisely enough as the observations are really far away from the both grey and cumulative 3 points least squared linear prediction outputs. Therefore, this paper proposes a new approach, GARCH(p,q) smoothing hybrid GREY-CLMS prediction with BPNN weighting, in which ARMAX/GARCH composite model has been incorporated into the hybrid GREY-CLMS prediction, and keep employing back-propagation neural net to train/simulate their weights so as to highly improve the prediction accuracy due to the smoothness enhanced. The proposed method is tested successfully in the empirical examples on the topics of international stock price indexes.