A0679
Title: Dynamic factor model for realized covariance matrices
Authors: Ilya Archakov - York University (Canada)
Roxana Halbleib - University of Freiburg (Germany)
Jasper Rennspies - University of Freiburg (Germany) [presenting]
Abstract: A dynamic factor model is developed for realized covariance matrices. The log-transformation is used, as proposed in a prior study, to decompose realized covariance matrices into realized volatilities and transformed realized correlations. The panel of the resulting series is modelled by aggregating AR(1)-factors to capture persistence in a parsimonious way. A standard Kalman filter is used together with maximum likelihood to extract the latent factors and estimate the model parameters. The Kalman Filter setting is extended to allow for GARCH-type effects in the factors. The aggregation approach allows for an interpretation of the persistence in the factors that is data-driven. An empirical analysis of 29 liquid stocks on the NYSE is performed. Descriptive statistics of the 406 series of transformed realized correlations are reported. An in-sample analysis is performed, in which we inspect the autocorrelation in the residuals, and an out-of-sample analysis is performed for various forecast horizons.
