Traditional bond value factors profit from both mispricings and risk. A machine learning–based factor earns 79% from repricing, outperforming others after costs. It better controls risk, offering a more accurate approach to “true” value investing.
Interested in having your article published in the Financial Analysts Journal? Find out how.
Abstract
Value investing in the corporate bond market aims to identify mispricings by determining to what extent a bond’s credit spread compensates for its risk. By decomposing returns into a risk-taking and a repricing component, we show that existing value factors earn not only from capturing mispricings but also substantially from taking more risk. To better control for risk, we construct a value factor based on an ensemble of machine learning methods. We find that it earns less from risk-taking and more from repricing and is thus closer to a “true” value factor. It also delivers the highest returns after costs.
KEY HIGHLIGHTS
- We argue that a “true” value factor should earn most of its return from capturing mispricings and not from taking more risk.
- We find that existing value factors from the literature earn substantially from taking more risk.
- We introduce a machine learning–based value factor, whose performance is driven less by risk and more by repricing and is thus closer to a “true” value factor.