Statistically identified structural VAR model with potentially skewed and fat-tailed errors

Abstract

We introduce a structural vector autoregressive model in which the mutually independent errors follow skewed generalized t-distributions, whose flexibility compared with commonly considered Student's t-distributions diminishes the risk of misspecification and strengthens identification. Because of statistical identification due to non-Gaussianity, the plausibility of economic identifying restrictions can be formally assessed. In an empirical application, the data support narrative sign restrictions in identifying the US monetary policy shock. In contrast to some of the previous literature, we find a strong negative response of real activity to contractionary monetary policy after a few months' delay.