Factor-augmented HAR model improves realized volatility forecasting

Abstract

This paper proposes a factor-augmented heterogeneous autoregressive (FAHAR) model for realized volatility. This model incorporates volatility information from other stock markets into several f actors, hence it is expected to improve forecasting. We also consider nonlinear modeling of the FAHAR based on the LSTM network in deep neural networks. Our empirical analysis shows that factor augmentation indeed improves forecasting for all the stock indices considered, implying the co-movement of world stock markets in the 2010s.