Forecasting the realized range-based volatility using dynamic model averaging approach

- We apply DMA approach as combined models with time-varying weights to forecast RRV.
- HAR-RRV-type models, combined models with constant weights and DMA are compared.
- The models are evaluated by various methods for different high-frequency data.
- Our results show DMA approach has strong forecasting ability for RRV.

In this study, we forecast the realized range-based volatility (RRV) using the heterogeneous autoregressive realized range-based volatility (HAR-RRV) model and its various extensions, which are called HAR-RRV-type models. We first consider the time-varying property of those models' parameters using the dynamic model averaging (DMA) approach and evaluate the forecasting performance of three types: individual HAR-RRV-type models, combined models with constant weights, and combined models with time-varying weights. Our out-of-sample empirical results show that combined models with time-varying weights can not only generate more accurate forecasts, but also beat individual models and combined models with constant weights.