Quotation Frühwirth-Schnatter, Sylvia. 2003. Strategies for Improving Bayesian Inference Using MCMC. TU München, Zentrum Mathematik, Lehrstuhl für Mathematische Statistik, München, Deutschland, 03.12.




A well-known problem for Bayesian estimation of highly parameterized models is slow convergence of straightforward MCMC schemes. Typical examples are random effects model including random effects that are nearly deterministic or state space models with state vectors that are nearly constant. In the first part of the talk various strategies for improving MCMC for such models are reviewed and discussed, among them grouping and collapsing in the spirit of Liu (1994); and reparametrization in the spirit of Gelfand, Sahu and Carlin (1995) and Meng and Van Dyck (1999). As one of the major causes of poor convergence seems to be the attempt to fit over-parameterized models, parsimonious MCMC is considered as a new strategy of dealing with poorly identified models. Within parsimonious MCMC sampling the unknown model parameters is carried out jointly with finding a parsimonious representation of the underlying model structure. To this aim suitable selection variables are introduced, that are sampled jointly with the parameters. In the second part of the talk, details will be presented for the random effects model, where Bayesian variable selection is introduced into the covariance structure of the latent process. Apart from the computational merits, parsimonious MCMC also leads to a gain in statistical inference. By exploring the posterior draw of a parsimonious MCMC sampler we also learn what parsimonious model structure fits best to the data at hand.


Press 'enter' for creating the tag

Publication's profile

Status of publication Published
Affiliation WU
Type of publication Paper presented at an academic conference or symposium
Language English
Title Strategies for Improving Bayesian Inference Using MCMC
Event TU München, Zentrum Mathematik, Lehrstuhl für Mathematische Statistik
Year 2003
Date 03.12
Country Germany
Location München
URL http://www.stat.uni-muenchen.de/sfb386/oldtalks/fruehwirth2.html


Frühwirth-Schnatter, Sylvia (Details)
Google Scholar: Search