Dynamic partial (co)variance forecasting model

Abstract

In this study, we propose a dynamic partial (co)variance forecasting model (DPCFM) by introducing a dynamic model averaging (DMA) approach into a partial (co)variance forecasting model. The dynamic partial (co)variance forecasting model considers the time-varying property of the model's parameters and optimal threshold combinations used to construct partial (co)variance. Our empirical results suggest that in both variance and covariance cases, the dynamic partial variance forecasting model can generate more accurate forecasts than an individual partial (co)variance forecasting model in both the statistical and economic sense. The superiority of the dynamic partial (co)variance forecasting model is robust to various forecast horizons.