Testing identification strength of narrative sign restrictions for SVAR

Coauthors Bulat Gafarov

Abstract

Sign restrictions provide a range of models that most economists agree upon. In the recent years, additional proposals has been made including restrictions in dynamic factor models, narrative sign and ranking restrictions that can provide large and growing number of identifying inequalities resulting in point identification. In this paper, we provide testable necessary and sufficient conditions under which these bounds collapse to single point. The main necessary condition is positive linear dependence of the vectors of sign-restricted IRF coefficients. We provide a simple test for this condition with standard chi-squared critical values that remain valid in the presence of redundant sign-restrictions. Our simulations suggest that the proposed test has high power against the alternative data generating process with partial-identification.

keywords: SVAR, point identification, sign restrictions