This paper presents evidence of predictability in the time-varying shape of the U.S. term structure of interest rates using a robust recursive modelling approach based on a Bayesian mixture of multi-factor models. We find that variables such as default spread, equity volatility, short-term and forward rates, among others, can be used
to predict changes in the slope of the yield curve, and also, albeit to a lesser extent, changes in the curvature of the yield curve. By using systematic trading strategies based on butterfly swaps, we also find that this evidence of predictability in the shape of the yield curve is economically significant in addition to being statistically significant.