Improving forecasts of the federal funds rate in a policy model

Vector autoregression (VAR) models are widely used for policy analysis. Some authors caution, however, that the forecast errors of the federal funds rate from such a VAR are large compared to those from the federal funds futures market. From these findings, it is argued that the inaccurate federal funds rate forecasts from VARs limit their usefulness as a tool for guiding policy decisions. In this paper, we demonstrate that the poor forecast performance is largely eliminated if a Bayesian estimation technique is used instead of OLS. In particular, using two different data sets we show that the forecasts from the Bayesian VAR dominate the forecasts from OLS VAR models—even after imposing various exact exclusion restrictions on lags and levels of the data.