The predictive power of Nelson-Siegel factor loadings for the real economy

Abstract We generalize the arbitrage-free Nelson Siegel (AFNS) model to allow λ t to vary over time. We find that the time-varying λ t , which determines the relative factor loadings, typically reaches its local peak before starting to decline right before a recession. Through conducting extensive in-sample and out-of-sample forecast exercises, we show that the information in the time-varying λ t factor has strong predictive power for business cycles and real economic activity. In particular, λ t contains additional useful information beyond those in conventional yield curve predictors, such as the yield spread. We argue and also document empirical evidence that the information in λ t is related to the market perception of the economic risk and uncertainty.