Multiple Measures Realized GARCH Models

Realized volatility has become the most popular empirical measure in fitting and forecasting volatility. However, as the properties of this class of estimators depend on the sampling frequency of intraday returns, a number of alternative realized estimators have been proposed, generating additional uncertainty in the modelling process. Aiming to mitigate the impact of modelling uncertainty in forecasting tail-risk, this paper investigates the benefits of combining information from several realized measures computed at multiple frequencies. In this framework, extensions of the Realized GARCH model based both on feature selection methods and time-varying parameters are proposed. To assess the implications for financial risk management, an application to the prediction of Value-at-Risk and Expected Shortfall for the Standard {\&} Poor's 500 Index is presented. We find that significant forecasting gains result from modelling approaches combining several realized multi-frequency measures.