Blended Identification in Structural VARs

We propose a blended approach which combines identification via heteroskedasticity with the widely used methods of sign restrictions, narrative restrictions, and external instruments. Since heteroskedasticity in the reduced form can be exploited to point identify a set of orthogonal shocks, its use results in a sharp reduction of the potentially large identified sets stemming from the typical approaches. Conversely, the identifying information in the form of sign and narrative restrictions or external instruments can prove necessary when the conditions for point identification through heteroskedasticity are not met and offers a natural solution to the labeling problem inherent in purely statistical identification strategies. As a result, we argue that blending these methods together resolves their respective key issues and leverages their advantages, which allows to sharpen identification. We illustrate the blending approach in an artificial data experiment first, and then apply it to several examples taken from recent and influential literature. Specifically, we consider labour market shocks, oil market shocks, monetary and fiscal policy shocks, and find that their effects can be rather different from what previously obtained with simpler identification strategies.