Real Time Macro Factors in Bond Risk Premium

Abstract

We examine whether bond returns are predicted by real-time macro variables with possible nonlinear predictive relationship and possible presence of weak factors. We propose a scaled sufficient forecasting method to deal with the issues arising from these two features, and study its asymptotic properties. Applying both existing and our new methods, we find that real-time macro variables have significant in- and out-of-sample forecasting power, generate sizeable economic values, and their predictability is not spanned by the yield curve. We also find that the forecasted bond returns are countercycilcal and the magnitude of predictability is stronger in economic recessions, thereby lending empirical support to well-known macro finance theories.