Comparing and combining realized measure and implied volatility for volatility prediction

In this study, we compare realized measure and implied volatility for forecasting of the volatility. Firstly, we employ a family of homogenous loss functions as the evaluation criteria in order to run a fair and complete forecast comparison. Our results show that, predictors based on realized measures are superior to that derived from implied volatility for both within-sample fitting and one-step-ahead forecasting, whereas the superiority of the latter type is presented in multi-step-ahead forecasting. Secondly, as a result of the comparison, a new model average approach with weights that depend on market conditions is developed to combine the information from implied volatility and realized measure for volatility forecasting. And our results show the superiority of the proposed approach in combining realized measures and implied volatility for volatility prediction.