Bayesian Estimation And Subset Selection Of SLTBL Models

Markov chain Monte Carlo (MCMC) technique is used to make Bayesian inference for seperable lower triangular bilinear models, a more general model class in time series analysis. First, we derive all of the conditional posterior distributions, and obtain the estimators of model parameters by virtue of Gibbs sampler. Since it is difficult to sample directly from the conditional posterior of the index in the models, a special Metropolis-Hastings step is designed. Furthermore, we apply the reversible jump Markov chain Monte Carlo approach, a generalized MCMC algorithm, to choose the subset models randomly. The proposed methods are demonstrated by simulated and real examples.