B0418
Title: Semiparametric bivariate conditional copula regression with binary and continuous marginals
Authors: Thomas Kneib - University of Goettingen (Germany) [presenting]
Nadja Klein - Georg-August-University Goettingen (Germany)
Giampiero Marra - University College London (United Kingdom)
Rosalba Radice - Cass Business School (United Kingdom)
Abstract: Copulas are a general and versatile tool for constructing multivariate distributions by combining a specific type of dependence structure specified via the copula with arbitrary marginal as long as these marginals are continuous. This allows for the construction of regression models for bivariate responses within the framework of distributional regression where regression predictors are placed on potentially all distributional parameters including the dependence parameter of the copula. We extend this framework by considering bivariate regression models where at least one of the responses is binary and therefore discrete. Based on the latent utility representation of binary regression models, we can still formulate a copula specification and combine the binary response part with a flexible specification of the dependence structure and the second marginal (unless this is also binary). We develop both penalized likelihood and Bayes inference and compare the results in an application on adverse birth outcomes where we combine a binary regression model for the presence/absence of low birth weight with a three parameter dagum specification for gestational age.
Journal