Forecasting realized variance using asymmetric HAR model with time-varying coefficients

This paper proposes an asymmetric HAR model with time-varying coefficients (TVC-AHAR) for modeling and forecasting realized variance. The TVC-AHAR model includes good and bad volatilities and assumes the associated time-varying coefficients to be driven by a latent Gaussian autoregressive process. The model is easy to estimate and implement by using maximum likelihood based on Kalman filter. Empirical analysis using two stock market indices of China, the Shanghai Stock Exchange Composite Index and Shenzhen Stock Exchange Component Index, shows that our proposed TVC-AHAR model yields more accurate out-of-sample forecasts of realized variance compared with the other models.