Abstracts > Jolly Emilien

Conditional copulas play a central role in modeling complex dependence structures and constitute a fundamental building block of vine copula constructions. Their estimation, however, becomes particularly challenging when one does not adopt the simplifying assumption, which replaces conditional dependence by an unconditional copula. In this talk, we present a novel estimator of conditional copulas based on finite mixtures of partial copulas. The main idea is to represent the conditional copula as a convex combination of unconditional bivariate copulas, where the mixing weights depend on the conditioning variables. The proposed methodology combines clustering techniques based on local dependence structures, bivariate copula estimation, and adaptive weighting functions. We establish consistency results in the (L^2) norm, both when the true conditional copula belongs to the finite-mixture model and in more general settings.