This paper applies the maximum likelihood based EM algorithm to a large-dimensional factor analysis of US monetary policy. Specifically, economy-wide effects of shocks to the US federal funds rate are estimated in a structural dynamic factor model in which 100+ US macroeconomic and financial time series are driven by the joint dynamics of the federal funds rate and a few correlated dynamic factors. This paper contains a number of methodological contributions to the existing literature on data-rich monetary policy analysis. Firstly, the identification scheme allows for correlated factor dynamics as opposed to the orthogonal factors resulting from the popular principal component approach to structural factor models. Correlated factors are economically more sensible and important for a richer monetary policy transmission mechanism. Secondly, I consider both static factor loadings as well as dynamic factor loadings, which is important for the response heterogeneity. Thirdly, the monetary policy rate is not a latent factor representation but measured without error and interacts dynamically with the factors in the estimation. Finally, the dynamic factor model is estimated by the one-step maximum likelihood based EM algorithm as an alternative to Bayesian methods and two-step principal component methods. Based on a large panel from 1959:01 to 2012:06 I estimate a number of model specifications and find that the dynamic responses of a monetary policy shock are theoretically more plausible for sufficiently rich factor models compared to the response implied by standard SVAR models. For instance, I do not observe the price puzzle in the dynamic factor model implying that after a contractionary shock prices fall.
Main Research Area:
First Vienna Workshop on High Dimensional Time Series in Macroeconomics and Finance, 2013