Title: Robust shrinkage for set-identified SVARs
Authors: Alessio Volpicella - Queen Mary University of London (United Kingdom) [presenting]
Abstract: Set-identified SVARs, which relax exclusion restrictions and rely on weaker assumptions such as sign restrictions, are increasingly common. However, a known drawback is that the inference is rarely informative. It is shown that robust restrictions on the Forecast Error Variance (FEV) decomposition may dramatically shrink the inference. Specifically, these restrictions are consistent with the implications of a variety of different DSGE models, with both real and nominal frictions, and with sufficiently wide ranges for their parameters. First, in a bivariate and trivariate setting, restrictions on the FEV decomposition are proven to be more informative than traditional sign restrictions. Second, sufficient conditions are provided to guarantee that the identified set is non-empty and convex. Finally, two applications are provided: using models of monetary policy and technology shocks, restrictions on the FEV decomposition tend to be highly informative, greatly shrink and even change the inference of models originally identified via traditional sign restrictions. Remarkably, shrinkage in inference is robust to the recent concerns over the unintended consequences of rotation matrix prior.