CMStatistics 2022: Start Registration
View Submission - CFE
Title: Copula-based multivariate mixture normal GARCH Authors:  Alexander Georges Gretener - University of Kiel (Germany) [presenting]
Markus Haas - University of Kiel (Germany)
Marc Paolella - University of Zurich (Switzerland)
Abstract: Univariate mixture normal GARCH models have been shown to provide accurate density and risk forecasts for financial returns. Current multivariate extensions of this model class are only applicable to low-dimensional return vectors. To overcome this limitation, a novel model coupling a univariate mixture of normal GARCH specifications for the conditional marginals with a mixture of Gaussian copulas for the dependence structure is proposed, resulting in a highly flexible multivariate return process which is also applicable to high-dimensional portfolios. The properties of the model and estimation issues are discussed. An application to the returns of the Dow Jones Industrial Average stocks shows that the model provides plausible disaggregation of the conditional multivariate distribution and delivers competitive risk forecasts and risk-based portfolio allocations.