CMStatistics 2022: Start Registration
View Submission - CFE
Title: High-frequency volatility measurement and forecasting: Parametric vs non-parametric approaches Authors:  Bjoern Schulte-Tillmann - University of Münster (Germany) [presenting]
Mawuli Segnon - University of Münster (Germany)
Timo Wiedemann - University of Muenster (Germany)
Abstract: Recently, the plethora of existing (non-parametric) realized variance (RV) estimators was extended by a parametric estimator based on price durations that are modeled by an autoregressive conditional duration (ACD) specification. Encouraged by its superior performance, we study the impact of different specifications on estimation accuracy and forecasting performance. To this end, we introduce the factorial hidden Markov duration (FHMD) process to model the dynamics governing financial durations and derive its statistical properties. Utilizing high-frequency data of ten actively traded stocks, we consider further parametric price duration-based RV estimators and the most common non-parametric approaches. We find that the duration-based approaches achieve substantial accuracy gains in an application to forecasting (i) the integrated variance and (ii) the Value-at-Risk.