B0885
Title: Bayesian analysis of jointly heavy-tailed data
Authors: Karla Vianey Palacios Ramirez - University of Edinburgh (United Kingdom) [presenting]
Abstract: A novel flexible Bayesian nonparametric model is introduced for a jointly heavy-tailed response. The proposed model induces a prior on the space of multivariate distributions with heavy-tailed marginal distributions. The proposal resorts to an infinite mixture of multivariate Erlang distributions, with a specific parameterization, to learn about the body and joint tails of a jointly heavy-tailed distribution. A predictor dependent version of the model is devised to learn about the effect of covariates on a jointly heavy-tailed response. Sufficient conditions to induce a multivariate heavy-tailed distribution are studied, and posterior inference following a Bayesian nonparametric approach is developed. The simulation study suggests that the proposed method performs well in a variety of scenarios. We showcase the application of the proposed methods in a case study in neuroscience.
