Implementing Bayesian Vector Autoregressions

Abstract

This paper discusses how the Bayesian approach can be used to construct a type of multivariate forecasting model known as a Bayesian vector autoregression (BVAR). In doing so, we mainly explain Doan, Litterman, and Sims (1984) propositions on how to estimate a BVAR based on a certain family of prior probability distributions, indexed by a fairly small set of hyperparameters. There is also a discussion on how to specify a BVAR and set up a BVAR database. A 4-variable model is used to illustrate the BVAR approach.