Reconciling VAR-based Forecasts with Survey ForecastsDecember 21, 2018
Expectations about future economic conditions play a central role in macroeconomic theory. Vector autoregression (VAR) models are often used to both generate forecasts to proxy these expectations and to model the dynamics of survey forecasts. However, jointly analyzing realized data and survey forecasts in a VAR remains a challenge. The primary issue that arises when embedding realized data alongside survey forecast in a VAR is the simultaneous existence of two different expectations of the same variable: the VAR-based forecast and the survey forecast. This paper proposes a Bayesian prior over the VAR parameters which allows the econometrician to impose the desired degree of consistency between these two forecasts at low computational costs. Our approach leverages the existence of multiple forecasts to aid in structural VAR identification and enhance VAR forecasts. We illustrate the usefulness of our approach in several applications including the identification of forward guidance shocks, the degree to which households’ inflation expectations are anchored, and the evolution of inflation tail-risks during and after the Great Recession.
- Doh, Taeyoung, and A. Lee Smith. “Reconciling VAR-based Forecasts with Survey Forecasts.” Federal Reserve Bank of Kansas City, Research Working Paper no. 18-13, December. Available at https://doi.org/10.18651/RWP2018-13