Quantile regression as the estimation of conditional quantiles finds applications in various fields. Using a parametrically estimated D-vine copula to model the dependency of the response variable and the predictors, we propose a generalization of the continuous D-vine quantile regression that is explicitly tailored to handle both continuous and discrete variables. Thereby, we extend the usage of D-vine quantile regression to situations where discrete variables are present. A D-vine as a subclass of the more general vine copulas provides us with a very flexible class of models that also enables an easy computation of conditional quantiles. Similar to its continuous counterpart, the discrete-continuous D-vine quantile regression remedies some of the major shortfalls of classical quantile regression, e.g. quantile crossing is not possible and the method directly takes into account the effects of dependent or collinear predictors. Based on an extensive simulation study, we show that our method can compete with other established methods for quantile regression. Particularly with a non-linear relationship between the response variable and the predictors, a relatively small sample size and variables taking on only a small number of values, our method shows superior prediction quality. We demonstrate the functionality of the method by performing a bank stress testing, where we compute the conditional quantiles of bank CDS spreads when conditioning on other banks being in distress and additionally conditioning on a discrete market sentiment index.     «

Quantile regression as the estimation of conditional quantiles finds applications in various fields. Using a parametrically estimated D-vine copula to model the dependency of the response variable and the predictors, we propose a generalization of the continuous D-vine quantile regression that is explicitly tailored to handle both continuous and discrete variables. Thereby, we extend the usage of D-vine quantile regression to situations where discrete variables are present. A D-vine as a subclas...     »