On the Heterogeneity Bias of Pooled Estimators in Stationary VAR Specifications

This paper studies asymptotically the bias of the fixed effect (FE) estimator induced by cross-section heterogeneity in the slope parameters of stationary vector autoregressions (VARs). The paper also compares the FE, the mean group estimator (MG), and a simple instrumental variable alternative (IV) in Monte Carlo simulations. The main results are: (i) asymptotically, the heterogeneity bias of the FE may be more or less severe in VAR specifications than in standard dynamic panel data specifications; (ii) in Monte Carlo simulations, slope heterogeneity must be relatively high to be a source of concern for pooled estimators; (iii) when this happens, the panel must be longer than a typical macro dataset for the MG to be a viable solution.