What's the Price of Monotonicity? A Multi-Dataset Benchmark of Monotone-Constrained Gradient Boosting for Credit PD

TL;DR

This study benchmarks the impact of monotonicity constraints on gradient boosting models for credit risk prediction, showing that constraints often incur minimal accuracy loss, especially on large datasets, thus balancing interpretability and performance.

Contribution

It quantifies the 'Price of Monotonicity' across multiple datasets and libraries, providing practical insights into when monotonicity constraints are cost-effective.

Findings

PoM ranges from 0% to 2.9%, often near zero on large datasets.

Constraints are most costly on small datasets with extensive coverage.

Monotonicity constraints can preserve accuracy while enhancing interpretability.

Abstract

Financial institutions face a trade-off between predictive accuracy and interpretability when deploying machine learning models for credit risk. Monotonicity constraints align model behavior with domain knowledge, but their performance cost - the price of monotonicity - is not well quantified. This paper benchmarks monotone-constrained versus unconstrained gradient boosting models for credit probability of default across five public datasets and three libraries. We define the Price of Monotonicity (PoM) as the relative change in standard performance metrics when moving from unconstrained to constrained models, estimated via paired comparisons with bootstrap uncertainty. In our experiments, PoM in AUC ranges from essentially zero to about 2.9 percent: constraints are almost costless on large datasets (typically less than 0.2 percent, often indistinguishable from zero) and most costly on smaller datasets with extensive constraint coverage (around 2-3 percent). Thus, appropriately specified monotonicity constraints can often deliver interpretability with small accuracy losses, particularly in large-scale credit portfolios.