Improving Federal-Funds Rate Forecasts in VAR Models Used for Policy Analysis

Abstract

Federal-funds rate-forecast errors from vector autoregressive (VAR) models used for monetary policy analysis and fitted by ordinary least squares (OLS) are large relative to those from the futures market. Using three different structural VAR models, we show that forecasts based on a shrinkage estimator dominate the OLS-based forecasts—even after restricting the lag length and/or imposing exact unit-root restrictions—and are broadly comparable to the futures-market forecasts. Our results refute the perception that VAR models forecast the funds rate poorly in general and suggest that using stochastic prior restrictions can provide an effective way of improving forecast accuracy without sacrificing structural interpretation.