Machine Learning and the Cross-Section of Emerging Market Corporate Bond Returns

Abstract

This paper conducts a comparative assessment of machine learning models for predicting corporate bond behavior in a high transaction cost emerging market setting. Our research demonstrates that considering nonlinearities and interactions yields superior out-of-sample returns compared to linear factor models. Additionally, it unveils the potential for significant net returns while addressing challenges like high transaction costs and practical limitations. These limitations include constraints on short selling and deviations from benchmarks in key risk characteristics faced by fixed-income investors in these markets. Furthermore, our analysis reveals unanimous agreement across all methodologies on a relatively small set of dominant predictive signals, with the most potent predictors consistently linked to low-risk, macro, and momentum factors. In summary, we present evidence that machine learning techniques enhance the effectiveness of financial factors in delivering excess returns in Emerging Market corporate bonds.

Keywords: Machine Learning, Return Prediction, Cross-Section of Corporate Bond Returns, Emerging Markets, Random Forest, Gradient Boosting, Practical Limitation, Transaction Cost

JEL Classification: C14, C38, C45, C52, C58, G11, G12, G13, G14, G15