On Priors for Impulse Responses in Bayesian Structural VAR Models

We proposed clear, methodologically sound framework for analyzing SVAR with priors on impulse responses. We showed it poses no difficulties in deriving the posterior which even in case of unidentified SVAR with flat prior on impulse functions (under the appropriate requirement tying number of observations, lags and variables) is necessarily proper. Accordingly, useful factorization of the posterior was given and efficient method for sampling from the posterior was outlined.