New approach to financial time series forecasting - Quantum minimization regularizing BWGC and NGARCH composite model

Abstract

A hybrid BPNN-weighted GREY-C3LSP prediction (BWGC) is used for resolving the overshooting phenomenon significantly; however, it may lose the localization once volatility clustering occurs. Thus, we propose a compensation to deal with the time-varying variance in the residual errors, that is, incorporating a non-linear generalized autoregressive conditional heteroscedasticity (NGARCH) into BWGC, and quantum minimization (QM) is employed to regularize the smoothing coefficients for both BWGC and NGARCH to effectively improve model's robustness as well as to highly balance the generalization and the localization.