This article presents an augmented deep factor model that generates latent factors for cross-sectional asset pricing. The conventional security sorting on firm characteristics for constructing long–short factor portfolio weights is nonlinear modeling, while factors are treated as inputs in linear models. We provide a structural deep-learning framework to generalize the complete mechanism for fitting cross-sectional returns by firm characteristics through generating risk factors (hidden layers). Our model has an economic-guided objective function that minimizes aggregated realized pricing errors. Empirical results on high-dimensional characteristics demonstrate robust asset pricing performance and strong investment improvements by identifying important raw characteristic sources.

Using machine learning and many predictors, we find strong bond return predictability, with an out-of-sample R-squared of 4.48% and an annualized Sharpe ratio of 3.27. ML models identify important predictors for aggregate predictors (bond market returns, TERM and HML factors, GDP growth) and bond characteristics (downside risk, short-term reversal, return skewness, and credit spreads). Predictability varies over time, being stronger during periods of high investor risk aversion, slow economic growth, and strong cross-sectional factor explanatory power. Our results highlight the benefits of leveraging both cross-sectional and time-series predictors to forecast corporate bond returns while considering public and private bonds.