CEBA talks 2020-2022
Invited talks Past talks

Uniform Priors for IRFs

Authors: Jonas Arias, Juan Rubio-Ramirez, Daniel Waggoner
Abstract: The usefulness of a popular approach for conducting Bayesian inference based on structural vector autoregressions identified with sign and zero restrictions has been called into question. First, we show that although in general the methods impose informative marginal prior distributions over individual impulse response functions (IRFs), the problem is not as severe as the critics seem to imply. If one does not condition on the reduced-form parameters, the marginal prior distributions over the IRFs do not drive the marginal posterior distributions over the IRFs. Second, we show that when the focus is on joint inference about IRFs or on any other objects of interest parameterization, the quest for “noninformative” priors is not destined to fail. In particular, we propose variants of the criticized methods to conduct joint inference about IRFs using a joint prior distribution for IRFs or about any other objects of interest parameterization using a joint prior distribution for the objects of interest. 

Link to work
Presentation slides