The Predictive Power of the Yield Spread Under the Veil of Time

Abstract This paper studies the sources of cyclical information delivered by the term spread for output growth predictability in the U.S. I use a wavelet-based time-frequency decomposition to decompose the predictive power of the yield spread across time scales, both in-sample and out-of-sample, over various forecast horizons. Spreads between interest rates on 10-year and 3-month Treasuries have a predictive ability for output growth that changes largely over different time scales. I find evidence of a negative correlation between the spread and future GDP growth for fluctuations with a frequency of 4 to 8 years per cycle. A linear combination among filtered yield spreads shows a sizable improvement in forecasting out-of-sample. The time-frequency decomposition is also used to propose an interpretation for the breakdown of in-sample predictability documented by Dotsey (1998) that arises after 1985. JEL classification numbers: C19, E43, E27. Keywords: Multiresolution analysis, Term structure, Predictability.