Title: Metropolis Hastings based estimation of generalized partition of unity copulas
Authors: Andreas Masuhr - University of Munster (Germany) [presenting]
Abstract: The recently emerged family of Generalized Partition of Unity Copulas (GPUC) offer a new way for nonparametric modeling of dependencies by using a very general mixture approach. As a special case, GPUC also nest the versatile Bernstein copula, but can also allow for copulas that possess (upper) tail dependence. First, a prior distribution on the parameters of GPUC is established via importance sampling from the space of eligible parameter matrices. Subsequently, two estimation approaches based on the Metropolis-Hastings (MH) algorithm are proposed: a random walk MH and a random blocking random walk MH that makes use of the restrictions on the parameter space. Finally, simulation studies are carried out indicating the superiority of the proposed random blocking algorithm.