A1273
Title: Moment condition verification for structural VARs in the Bayesian empirical likelihood framework
Authors: Paul Nguyen - University of Melbourne (Australia) [presenting]
Abstract: A method is developed to verify the validity of moment conditions for the identification of structural VARs in the Bayesian Exponentially Tilted Empirical Likelihood (BETEL) framework through the use of slab-and-spike priors. By building upon previous work, we propose a moment condition verification procedure that requires the estimation of only one unified model, allows for a probabilistic interpretation of moment condition validity, and is consistent with the Bayesian model selection guidelines, as used by Chib, Shin and Simoni. To demonstrate the validity of the moment condition verification procedure, we evaluate its performance in simulation by measuring rejection and non-rejection frequencies in a false discovery rate (FDR) controlled environment. We show that across a range of different identification sources, including exclusion restrictions, heteroskedasticity, higher-order moment restrictions, and instrumental variables, the moment condition verification procedure is able to correctly distinguish between valid and invalid moment restrictions. Furthermore, we demonstrate this procedure with an application to a tax policy structural VAR example, evaluating previously used sources of identification such as narrative-based instruments, higher-order moment conditions, and heteroskedasticity.
