Proxy Vector Autoregressions in a Data-rich Environment

Abstract

I propose a Bayesian approach to identify vector autoregressive (VAR) models via proxies in a data-rich environment. The setup augments a small-scale VAR model with latent factors. It allows to trace out the responses of disaggregated series in a unified model while controlling for broad economic conditions. The posterior sampler accounts for the estimation uncertainty in these latent factors as well as the measurement precision of the proxy. In a first application to monetary policy, I extract factors from a wide range of real and financial series and find that the effects of monetary policy shocks vary along the yield curve. In a second application to oil market shocks I add disaggregated US series to a standard model of the global oil market. I find that negative news about future oil supply have adverse effects on the US economy.