A1392
Title: Identification and inference in triangular models with flexible copula dependence
Authors: Robinson Dettoni - Universidad de Santiago de Chile (Chile) [presenting]
Abstract: Recursive triangular models are widely used for causal inference with endogenous treatments, but identification becomes challenging when outcomes are nonlinear and latent dependence is non-Gaussian. Identification and inference results are developed for triangular models with binary or continuous endogenous treatments, binary and duration outcomes, and flexible copula-based dependence structures. Particular attention is given to two-parameter copulas, including BB1 and BB7 families, which allow overall latent association and tail dependence to vary separately. For binary-treatment models, identification is studied through the mapping from structural and copula parameters to observable joint probabilities under exclusion and support conditions. For continuous endogenous treatments, the framework combines monotone triangular structure with latent treatment-rank variation induced by excluded instruments. Local identification of the copula dependence block is established through generic rank conditions. Estimation is based on maximum likelihood and complementary Bayesian methods, and finite-sample behavior is evaluated through Monte Carlo experiments under weak instruments and copula misspecification. An empirical application using UK Biobank data studies the causal effect of adiposity on incident hypertension and type 2 diabetes using genetic instruments.
