The predictive power of the term spread revisited : a change in the sign of the predictive relationship

We qualify some of the traditionally accepted results on the predictive power of the term spread over output. We show that in the case of short-term spreads, the direction of the predictive power may be the opposite to that usually found in the empirical literature, which has mostly rested on the use of long spreads, and in theoretical results, that have tended to neglect the consideration of multiperiod dynamics. An analysis of data for Germany and the United States confirms that short-term spreads have low predictive power and sometimes in the opposite direction to the traditional argument. Some suggestions for empirical work are derived from the analysis.