Predicting Individual Corporate Bond Returns

Abstract

We use machine learning methods to find substantial evidence of return predictability and investment gains for public and private individual corporate bonds from 1976 to 2020. The return forecast-implied long-short and market-timing strategies deliver significant risk-adjusted returns net of the transaction cost. We find random forest outperforms all other methods because the ensemble of nonlinear trees helps reduce overfitting. Moreover, given the long history of our bond sample, we can evaluate macro predictors and find they contain more useful information than bond characteristics for the out-of-sample prediction. Finally, predictability differs between private and publicly-listed companies, with investment gains larger for private company bonds, which the literature has overlooked.

Keywords: Bond Characteristics; Machine Learning; Macro Predictors; Return Predictability; Private Bonds

JEL Classification: C55, C58, G0, G1, G17.