Bayesian Methods for Dynamic Multivariate Models
Christopher A. Sims and Tao Zha
Federal Reserve Bank of Atlanta
Working Paper 96-13
If multivariate dynamic models are to be used to guide decision-making, it is important that it be possible to provide probability assessments of their results. Bayesian VAR models in the existing literature have not commonly (in fact, not at all as far as we know) been presented with error bands around forecasts or policy projections based on the posterior distribution. In this paper we show that it is possible to introduce prior information in both reduced form and structural VAR models without introducing substantial new computational burdens. With our approach, identified VAR analysis of large systems (e.g., 20-variable models) becomes possible.
JEL classification: C11, C53