Assessing identifying restrictions in SVAR models

This paper proposes a Bayesian approach to assess if the data support candidate set-identifying restrictions for Vector Autoregressive models. The researcher is uncertain about the validity of some sign restrictions that she is contemplating to use. She therefore expresses her uncertainty with a prior distribution that covers the parameter space both where the restrictions are satisfied and where they are not satisfied. I show that the data determine whether the probability mass in favour of the restrictions increases or not from prior to posterior. Using two applications, I find support for the restrictions used by Baumeister & Hamilton (2015a) in their two-equation model of labor demand and supply, and I find support for the true data generating process in a simulation exercise on the New Keynesian model.