CMStatistics 2017: Start Registration
View Submission - CMStatistics
Title: Extended D-vine quantile regression with applications Authors:  Claudia Czado - Technische Universitaet Muenchen (Germany) [presenting]
Daniel Kraus - Technische Universitaet Muenchen (Germany)
Thomas Nagler - Leiden University (Germany)
Abstract: Non-Gaussian dependence patterns between the response and the covariates cannot be captured by ordinary quantile regression. In addition estimated quantile regression lines can cross for different levels, which is non desirable. Therefore we propose to use a tailored D-vine copula model to capture the dependence. It is tailored to select important covariates in a forward selection procedure by maximizing the conditional log likelihood. Extensions allow for non parametric pair copulas and discrete covariates in a parametric and non parametric setup. The approach will be introduced and illustrated in applications.