Score-Driven Models for Realized Volatility

Abstract

This paper sets up a statistical framework for modeling realised volatility (RV ) using a Dynamic Conditional Score (DCS) model. It first shows how a preliminary analysis of RV, based on fitting a linear Gaussian model to its logarithm, confirms the presence of long memory effects and suggests a two component dynamic specification. It also indicates a weekly pattern in the data and an analysis of squared residuals suggests the presence of heteroscedasticity. Furthermore working with a Gaussian model in logarithms facilitates a comparison with the popular Heterogeneous Autoregression (HAR), which is a simple way of accounting for long memory in RV. Fitting the two component specification with leverage and a day of the week effect is then carried out directly on RV with a Generalised Beta of the second kind (GB2) conditional distribution. Estimating logRV with an Exponential Generalised Beta of the second kind (EGB2) distribution gives the same result. The EGB2 model is then fitted with heteroscedasticity and its forecasting performance compared with that of HAR. There is a small gain from using the DCS model. However, its main attraction is that it gives a comprehensive description of the properties of the data and yields multi-step forecasts of the conditional distribution of RV.