Métodos Bayesianos para seleção de modelos de mistura de distribuições normais e t de Student assimétricas

Abstract

In this work, we consider mixture models whose components of the mixture are modeled by the skew normal and skew t distributions. For the estimation of these skew mixtures models, we used a Bayesian approach, via Markov Chain Monte Carlo methods (MCMC), since this approach allows the development of a joint estimation procedure of the mixture components number and the associated parameters of the mixture components. We also use the Reversible jump method and propose the use of the Data-driven Reversible jump method for modelling the mixture of skew normal and skew t distributions, both to adjust and select the number of components of the mixture. We compare the performances of these two methods (Reversible jump and Data-driven Reversible jump) for selecting the best model through simulations. The Data-driven Reversible jump method was more accurate in pointing out the best model in the simulation studies carried out.