The puzzling effects of monetary policy in VARs: Invalid identification or missing information?

Abstract

Standard vector autoregressions (VARs) often find puzzling effects of monetary policy shocks. Is this due to an invalid (recursive) identification scheme, or because the underlying small-scale VAR neglects important information? I employ factor methods and external instruments to answer this question and provide evidence that the root cause is missing information. In particular, while a recursively identified dynamic factor model yields conventional monetary policy effects across the board, a small-scale VAR identified via external instruments does not. Importantly, the discrepancy between both models largely disappears once the information set of the VAR is augmented via factors. This finding is comforting news for the recent monetary literature. Two leading empirical advances with different underlying assumptions—namely external instruments (applied to a factor-augmented VAR) and dynamic factor models (identified recursively)—find very similar effects of monetary policy shocks, cross-verifying each other.