Title: Realized volatility modelling with measurement errors and nonlinear effects
Authors: Fulvio Corsi - University of Pisa and City University London (Italy) [presenting]
Giuseppe Buccheri - University of Rome Tor Vergata (Italy)
Abstract: Despite its effectiveness, the approximate long-memory HAR model neglects measurement errors and exhibits several evidences of misspecification due to the inherent nonlinearity of the realized volatility dynamics. We propose new extensions of the HAR model apt to address these effects separately with the aim to disentangle them and quantify their contribution in improving volatility forecasts. First, we combine the asymptotic theory of the realized volatility estimator with Kalman filter to account for measurement errors. Secondly, nonlinear effects are captured by introducing time variations in the HAR parameters driven by the score of the predictive likelihood. The two approaches are then combined to simultaneously account for both measurement errors and nonlinearities. The proposed models are simply estimated through standard maximum likelihood methods and are shown, both on simulations and on real data, to provide better out-of-sample volatility forecasts compared to existing approaches.