Semiparametric model selection for copulas
MetadataShow full item record
This thesis will consider the performance of the cross-validation copula information criterion, xv-CIC, in the realm of finite samples. The theory leading to the xv-CIC will be outlined, and an analysis will be conducted on an assorted collection of bivariate one-parameter copula models. The restriction to the bivariate case is not a grave one, since more complex d-variate samples can be broken down into a study of conditioned bivariate samples, by the methodology of regular vine-copulas, the pair copula construction and stepwise-semiparametric estimation of parameters. As a by-product of our analysis, we can give an advice with regard to the selection of model selection method in the semiparametric realm.