Bayesian Inference for Switching Mean Models with ARMA errors

Abstract

Bayesian inference is considered for switching mean models with the ARMA errors. We use noninformative improper priors or uniform priors. The fractional Bayes factor of O`Hagan (1995) is used as a Bayesian tool for detecting the existence of a single change or multiple changes and the regular Bayes factor is used for identifying the orders of the ARMA error. Once the model is fully identified, the Gibbs sampler with the Metropolis-Hastings subchains is constructed to estimate parameters. Finally, we perform a simulation study to support theoretical results.