Modeling and forecasting return volatility: A new reduced form approach using nonparametric variation measures

This paper motivates a reduced form discrete-time series approach that models realized volatility by using its separated components, continuous variation and variation due to jumps. For this purpose, I combine Engle and Russell’s (1998) autoregressive conditional duration (ACD) model applied to the continuous and jump size variation with Hamilton and Jordà’s (2002) autoregressive conditional hazard (ACH) model applied to jump durations. Further, the paper develops and discusses a methodology to evaluate density and probability function forecasts of this model class. The successful model fit and the favorable point and density forecast results show that the approach proposed in this paper qualifies as a useful forecast model for daily return variation.