Group Local Projections

This paper considers the estimation of heterogeneous impulse responses in large panels. I introduce an efficient data-driven clustering methodology for grouping heterogeneous responses within the local projection-IV framework. The proposed group local projection (GLP) estimator consistently recovers the latent group structure and the group-specific impulse responses when the panel dimensions increase. Simulation evidence illustrates the reliable finite sample performance of the estimator even under misspecification of the group structure. With the GLP estimator I revisit the debate on the effects of monetary policy shocks on house prices and document significant price appreciation after a contractionary shock in an economically large cluster of MSAs in the US. Importantly, this cluster is ignored by conventional grouping criteria.