With d(d - 1)=2 bivariate copulas and d one-dimensional distribution functions, a simplified R-vine copula model construction uniquely specifies a d-dimensional distribution function. The wider known theory of continuous R-vine copula models can easily be adapted to cover discrete components as well. This approach is convenient for quantile regression, because if the response variable only appears in leaf notes of the corresponding trees, its conditional quantile function given the covariates can directly be computed. Note that the regression problem becomes a classification problem for discrete response variables. In 2019, Chang/Joe introduced the method copreg, which links the response variable to the trees of an existing R-vine structure on the covariables based on conditional partial correlations. A response focused approach by Kraus/Czado from 2017, called vinereg, builds up a D-vine by iteratively adding the covariate yielding the highest increase in conditional log likelihood. These two methods are compared (i) in a simulation study with datasets generated by simplified R-vine copula models, (ii) in a second simulation study with several different datasets, all not representable by a simplified R-vine copula model, (iii) to two competitor regression methods and two competitor classification methods and (iv) on four datasets from chemistry, economics, biology and health.     «

With d(d - 1)=2 bivariate copulas and d one-dimensional distribution functions, a simplified R-vine copula model construction uniquely specifies a d-dimensional distribution function. The wider known theory of continuous R-vine copula models can easily be adapted to cover discrete components as well. This approach is convenient for quantile regression, because if the response variable only appears in leaf notes of the corresponding trees, its conditional quantile function given the covariates ca...     »