Presenter: Alejandro Antonio Jara Vallejos, Visiting Professor at BYU 

Description: A rich literature justifies the use of mixture models for making inferences about probability distributions. From a Bayesian nonparametric perspective, the full support of the prior, posterior consistency, and the availability of simple and efficient methods for posterior computation have largely motivated their applications. Unfortunately, commonly employed models do not allow for the use of available prior information about specific aspects of a multivariate distribution; a modeler is unlikely to have prior knowledge about all aspects but may have prior information about specific characteristics, such as the marginal distributions. A marginals-copula representation of a joint distribution can provide a great deal of flexibility in modeling multivariate distributions, allowing for the specification of models for the marginal distributions separately from the dependence structure (copula). However, its use requires the existence of prior models for copula functions with appealing properties. We introduce a new class of priors on continuous copula functions: the Bernstein-Yett-Uniform priors. We investigate the prior and posterior properties of the model and propose alternative Markov chain Monte Carlo algorithms for exploring the posterior distribution. The methodology is illustrated using simulated and real data. Joint work with Nicolás Kuschinski and Richard Warr.

Hosted by: Professor Allen Kei

Zoom link: https://ucsc.zoom.us/j/92903950903?pwd=5QQvJ8uOspL3xhoTDh4GUcGVatWyS0.1