Extracting Information from the Corporate Yield Curve: A Machine Learning Approach

Abstract

We document strong evidence on the cross-sectional predictability of corporate bond returns based on 48 yield predictors that capture the information in the yield curve one to 48 months ahead. In addition to standard regression forecasts, we generate forecasts based on machine learning, which improves the forecast performance especially for junk bonds, and find that short- and long-term predictors are most informative. Return predictability is economically and statistically significant, and is robust to various controls. The pronounced bond anomaly uncovered in this paper joins a host of equity anomalies that challenge rational pricing models.

Keywords: Yield signals; moving averages; cross-sectional predictability; machine learning; corporate bond returns

JEL Classification: G12; G14