On Trivariate Copulas with Bivariate Linear Spearman Marginal Copulas

Based on the trivariate reduction technique two different trivariate Bernoulli mixtures of univariate uniform distributions and their associated trivariate copulas with bivariate linear Spearman marginal copulas are considered. Mathematical characterizations of these Bernoulli mixture models are obtained. Since Bernoulli mixture trivariate reduction copulas are not compatible with all valid grade correlation coefficients, and there exist linear Spearman compatible non-Bernoulli mixture trivariate copulas, one can ask when there exists at all a trivariate copula with given linear Spearman marginal copulas. Based on a known concordance ordering compatibility criterion, a set of grade correlation inequalities, which must necessarily be satisfied for compatibility, is derived. The existence question for trivariate copulas with compatible linear Spearman marginal copulas is settled in the main result, which states that this set of inequalities is also sufficient for compatibility. The constructive proof makes use of two new classes of trivariate copulas that are obtained from the Bernoulli mixture trivariate copulas through a natural parametric extension. Finally, the obtained classes of trivariate copulas are compared with another class that contains as special case some trivariate copulas with linear Spearman marginal copulas. Since the latter class is incompatible with some type of linear Spearman copulas, the new classes of trivariate copulas build a richer class in this respect. Moreover, in contrast to the mentioned class, which requires in general 11 different elementary copulas in the defining convex linear combination, the new classes require at most five of them, which results in a more parsimonious parametric modelling.