Measuring Monetary Policy: A Bayesian FAVAR Approach with Agnostic Identification

This paper proposes to estimate the dynamic effects of monetary policy in a data rich environment by combining a \textit{Bayesian factor augmented vector autoregression} (BFAVAR) with the agnostic identification method introduced by Uhlig (2005), imposing sign restrictions on the shape of impulse response functions of a small set of variables according to conventional wisdom. We find that our factor generalization of the sign restriction approach although "weak" with respect to the structure and restrictions imposed, combined with the BFAVAR delivers significant results that appear to be reasonable and robust for a broad set of variables. The Results are associated with less uncertainty than the recursive identification scheme. From the results one can conclude that the combination of a rich panel of disaggregated macroeconomic data and the identification scheme is crucial for a successful and robust identification of the impact of monetary policy. Empirically we find for postwar US data some transitory real effects on the financial sector, housing and negative effects on the labor market. Furthermore we find significant short run negative effects on output but modest in size.