Collaborative Research: Factor-Augmented Vector Autoregressions [FAVARs] and the Analysis of Monetary Policy

点击翻译按钮获取中文摘要

A considerable literature has developed that employs vector autoregression (VAR) methods to attempt to identify the effects of monetary policy innovations on various macroeconomic variables. These methods generally deliver empirically plausible assessments of the dynamic responses of variables such as output and prices to policy shocks, and they have been widely used both for assessing the fit of structural models and in policy applications. However, the various VAR-based approaches have certainly not escaped criticism. Several of the criticisms of the VAR approach to monetary policy identification center around the relatively small amount of information used by low-dimensional VARs. The sparse information sets used in typical VAR analyses create at least two potential problems. First, to the extent that central banks and the private sector have information not reflected in the VAR system, the measurement of policy innovations is likely to be contaminated. A standard illustration of this potential problem is the perverse response of prices to monetary policy shocks in some VARs. It is argued that this price puzzle results from imperfectly controlling for information that the central bank may have about future inflation. A second problem arising from the use of sparse information sets in VAR analyses is that impulse responses can be observed only for variables included in the VAR, which generally constitute only a small fraction of the variables that we care about. This project develops an econometric approach that addresses both of these issues while retaining the benefits of small-dimension VAR analyses. Specifically, it combines the standard VAR analyses with factor analysis. Recent research in dynamic factor models suggests that the information from large numbers of time series can be usefully summarized by a small number of indexes, or factors. The investigators add estimated factors to otherwise standard VARs, obtaining factor-augmented VARs (or FAVARs). FAVARs can be estimated by two-step methods or by maximum likelihood methods that account for uncertainty in the factor estimation in second-stage VAR analysis. Preliminary work shows that FAVARs can help solve both problems alluded to above: First, FAVAR analyses of monetary policy yield plausible and tightly estimated impulse response functions; in particular, the price puzzle is greatly ameliorated. Second, FAVARs allow estimates of the responses of a wide variety of macro variables to policy shocks within a single unified approach. This project pursues a number of extensions to the basic analysis, both econometric and substantive. Econometric extensions include the development of empirical weighting schemes to provide for more precise measurement of the underlying factors in the economy. Substantive extensions include developing methods for real-time measurement of latent variables such as the output gap; factor-based analysis of data revisions and their forecast ability; and characterization of the effects of monetary policy on stock prices.