Posterior mean and variance approximation for regression and time series problems

TL;DR

This paper introduces a methodology for approximating the first two moments of the posterior distribution in Bayesian regression and time series models, especially with unknown covariance components, using second-order conditional independence.

Contribution

It develops partially specified models based on second-order conditional independence for efficient posterior moment approximation in complex Bayesian models.

Findings

Effective approximation of posterior moments demonstrated on simulated data.

Application to US investment and inventory data shows practical utility.

Models handle multivariate t, inverted t, and Wishart error distributions.

Abstract

This paper develops a methodology for approximating the posterior first two moments of the posterior distribution in Bayesian inference. Partially specified probability models, which are defined only by specifying means and variances, are constructed based upon second-order conditional independence, in order to facilitate posterior updating and prediction of required distributional quantities. Such models are formulated particularly for multivariate regression and time series analysis with unknown observational variance-covariance components. The similarities and differences of these models with the Bayes linear approach are established. Several subclasses of important models, including regression and time series models with errors following multivariate $t$, inverted multivariate $t$ and Wishart distributions, are discussed in detail. Two numerical examples consisting of simulated data and of US investment and change in inventory data illustrate the proposed methodology.