A new approach to integrating expectations into VAR models

Abstract

Listen

Translate article

Expectations play a central role in macroeconomics. Expectations are empirically measured from surveys or financial markets and are frequently analyzed in Vector autoregressive (VAR) models alongside realized data of the same variable. However, this leads to two different expectations for the same variable: the VAR-based forecast and the external forecast. This paper proposes a Bayesian prior over the VAR parameters which allows for varying degrees of consistency between these two forecasts. As we demonstrate in two applications, our approach can sharpen structural VAR identification of forward guidance shocks and enhances VAR forecasts of inflation tail risks.