Conditional Copula Models with Multiple Covariates

• Activity Number: 584
• Type: Invited
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Elif Fidan Acar*+
• Companies: University of Manitoba
• Keywords: Additive models ; Conditional dependence ; Covariate adjustment ; Local likelihood

Abstract

Conditional copula models provide a flexible framework to study covariate effects on dependence structures. A number of nonparametric estimation techniques have been recently proposed for these models in the case of a single covariate. These approaches, however, are not directly extendible to, or become impractical in, settings with multiple covariates.

This talk will present a nonparametric modelling strategy that can accommodate multiple covariates in conditional copula models. We consider a semiparametric conditional copula model where the copula function belongs to a parametric copula family and the copula parameter varies smoothly with covariates. To alleviate the curse of dimensionality, we use an additive formulation of the copula parameter and estimate smooth component functions associated with each covariate via a local likelihood backfitting algorithm. The finite sample performance of the proposed approach will be demonstrated using simulated and real data. The talk will also address general identifiability restrictions and computational challenges.