Abstract:This paper combines asset pricing theory with deep learning for pricing the cross section of corporate bonds. The proposed deep learning model can flexibly introduce the well-established factors and provide us with deep factors that are not subsumed in those existing factors. The deep factors are tradable long-short portfolios based on a large number of bond and equity characteristics and hence are economically more interpretable. We show empirically that our deep learning factor model improves the asset pricing performance on various corporate bond test portfolios over standard factor models and recommends bond investment portfolios that outperform the leading corporate bond strategies.

Interpretable and Arbitrage-Free Deep Learning for Corporate Bond Pricing