A Gibbs sampler for structural vector autoregressions

摘要

Structural VAR modeling has played an important role in empirical macroeconomics. The importance sampler used in the existing literature, however, can be prohibitively inefficient for obtaining accurate finite-sample inferences. In this paper we develop a Gibbs sampler for Bayesian inferences of structural VARs that restrict the covariance matrix of reduced-form residuals. Our method is computationally efficient in comparison to the existing method and can be readily applied. We show, by examples, that inferences based on the importance sampler can seriously distort economic interpretations.