Yield spreads as predictors of economic activity: a real-time VAR analysis

We undertake a real-time VAR analysis of the usefulness of the term spread, the junk-bond spread, the ISM's New Orders Index, and broker/dealer equity for predicting growth in non-farm employment. To get around the "apples and oranges" problem described by Koenig, Dolmas and Piger (2003), we augment each VAR we consider with a flexible state-space model of employment revisions. This methodology produces jobs forecasts consistently superior to those obtained using conventional VAR analysis. They are also superior to Federal Reserve Greenbook forecasts and to median forecasts from the Survey of Professional Forecasters. The junk-bond spread is by far the best single predictor of future jobs growth. However, the term spread has some incremental predictive power at medium-to-long horizons. The incremental predictive power of broker/dealer equity, while small, exceeds that of the ISM index at every horizon.