Forecasting Realized Volatility Dynamically Based on Adjusted Dynamic Model Averaging (AMDA) Approach: Evidence from China’s Stock Market

Abstract

In this study, we forecast the realized volatility of the CSI 300 index using the heterogeneous autoregressive model for realized volatility (HAR-RV) and its various extensions. Our models take into account the time-varying property of the models’ parameters and the volatility of realized volatility. The adjusted dynamic model averaging (ADMA) approach, is used to combine the forecasts of the individual models. Different from DMA method, the least and second least probability of particular models are excluded from the process of averaging the forecasts across the different models when using ADMA method. Our empirical results suggest that ADMA can generate more accurate forecasts than DMA method and alternative strategies. Models that use time-varying parameters have greater forecasting accuracy than models that use the constant coefficients. Time-varying parameter (TVP) models can generate more accurate forecasts than constant coefficient models in China’s stock market. The robustness test also indicates that the prediction accuracy of these DMA and ADMA models based on different parameters is higher than that of most single models, which further proves the effectiveness of the multi-model realized volatility prediction model based on dynamic averaging method in the prediction effect. These findings indicate the importance of considering parameter change and model specification change in volatility forecasting.