Corrigendum: Local Projection Inference is Simpler and More Robust Than You Think∗

It has recently come to our attention that the high-level Assumption 3 on p. 1805 of Montiel Olea and Plagborg-Møller (2021) (henceforth “MOPM”) is more restrictive than intended. As stated, the assumption allows for any VAR(1) process, both stationary and nonstationary, as well as VAR(p) processes with roots bounded away from the nonstationary part of the parameter space. However, several nonstationary VAR(p) models with p > 1 are ruled out. In this note, we therefore propose a modification of Assumption 3 that can be verified for a wide range of VAR(p) models whose autoregressive parameters are contained in the parameter space defined on p. 1804 in MOPM. Our modified assumption is similar to Assumption 3 in the recent paper by Xu (2022), who applies the appropriate “Dickey-Fuller” transformation to the regressors. If our modified Assumption 3 replaces the one in MOPM, all theoretical conclusions in our paper go through as originally stated. All econometric procedures, simulation results, efficiency calculations, and verbal discussions in our original paper are unaffected by the modification of Assumption 3.