Principal components at work: The empirical analysis of monetary policy with large datasets

Two competing methods have been recently developed to estimate large-scale dynamic factor models based, respectively, on static and dynamic principal components. In this paper we use two large datasets of macroeconomic variables for the US and for the Euro area to evaluate in practice the relative performance of the two approaches to factor model estimation. The comparison is based both on the relative goodness of fit of the models, and on the usefulness of the factors when used in the estimation of forward looking Taylor rules, and as additional regressors in monetary VARs. It turns out that dynamic principal components provide a more parsimonious summary of the information, but the overall performance of the two methods is very similar, in particular when a common information set is adopted. Moreover, the information extracted from the large datasets turns out to be quite useful for the empirical analysis of monetary policy.