A0228
Title: Forecasting value-at-risk in time of ultra-high-frequency data
Authors: Mawuli Segnon - University of Münster (Germany) [presenting]
Abstract: A factorial hidden Markov duration (FHMD) process is proposed for modelling the dynamics governing the financial price durations. Its statistical properties are derived and the exact maximum likelihood approach is applied to estimate its parameters. The adequacy of the proposed model is first assessed via the density forecast evaluation tools. Second, we derive an FHMD price duration-based realized variance estimator for forecasting daily value-at-risks at 5\% and 1\% confidence levels. In an empirical study using time series on price durations of the ten most traded stocks on the New York Stock Exchange (NYSE), it is found that the FHMD model is dynamically specified correctly and produces more accurate and valid daily value-at-risk forecasts than the standard $GARCH(1,1)$ at all significant levels.