Bayesian Nonparametric Vector Autoregressive Models via a logit stick-breaking prior

Date:

Joint work with Alessandra Guglielmi and Fernando Quintana

Abstract

We propose a semiparametric VAR model with exogenous covariates for nonstationary multidimensional longitudinal data, where a dependent stick-breaking prior is assumed for the autoregressive component through logit stick-breaking. We develop an efficient Gibbs sampling algorithm for MCMC posterior simulation, leveraging recent results on logit stick-breaking priors. We also illustrate the approach through simulations and medical data.

I was awarded a travel grant to present this research project.

See my poster!