Attribution Methods in Asset Pricing: Do They Account for Risk?
Dangxing Chen, Yuan Gao
TL;DR
This paper develops domain-driven axioms for attributing predictions in asset pricing and systematically evaluates two axiomatic explainers—Baseline Shapley (BShap) and Integrated Gradients (IG)—in light of these axioms. It shows that BShap preserves several first- and second-order risk axioms while IG preserves a complementary set, with DIM failing for IG in some regimes and CG not guaranteed for BShap, highlighting when each method is appropriate. Analytical arguments and an empirical 2008-option-pricing study demonstrate how attribution reflects risk patterns under varying market conditions and caution against out-of-training-domain extrapolations. The results offer guidance on using attribution methods in finance and point to future work on developing axioms-aware, time-series-aware explanations for risk-sensitive tasks such as portfolio attribution.
Abstract
Over the past few decades, machine learning models have been extremely successful. As a result of axiomatic attribution methods, feature contributions have been explained more clearly and rigorously. There are, however, few studies that have examined domain knowledge in conjunction with the axioms. In this study, we examine asset pricing in finance, a field closely related to risk management. Consequently, when applying machine learning models, we must ensure that the attribution methods reflect the underlying risks accurately. In this work, we present and study several axioms derived from asset pricing domain knowledge. It is shown that while Shapley value and Integrated Gradients preserve most axioms, neither can satisfy all axioms. Using extensive analytical and empirical examples, we demonstrate how attribution methods can reflect risks and when they should not be used.