Modeling multivariate parametric densities of financial returns (in Russian)

This paper compares several bivariate conditional density parameterizations for stock market returns in terms of in-sample fit and out-of-sample predictive ability for the whole conditional density. We consider Skew-Normal, Skew-Student, Skew-GED and Gram-Charlier densities. We focus on the ability of these density specifications to capture asymmetry and so called 'multivariate tails'. Using a test based on Kullback-Leibler information criterion we conduct pairwise comparisons of estimated conditional density models in sample and out of sample. The models are ranked according to their quality of fit and predictive ability. We discuss the causes behind superiority of this or that density specification.