Local Projections vs. Vars: Lessons from Thousands of Dgps

We conduct a simulation study of Local Projection (LP) and Vector Autoregression (VAR) estimators structural impulse responses across thousands data generating processes, designed to mimic the properties universe U.S. macroeconomic data. Our analysis considers various identification schemes several variants LP VAR estimators. A clear bias-variance trade-off emerges: have lower bias than but substantially higher variance at intermediate long horizons. Consequently, unless researchers are overwhelmingly concerned with bias, shrinkage via Bayesian VARs or penalized LPs is attractive.