Credit Scoring and Underwriting with Quantum

Workshop Description

Credit scoring models have converged. Logistic regression and gradient-boosted decision trees dominate consumer lending, and incremental improvements in Gini coefficient or KS statistic are increasingly difficult to achieve with classical methods alone. Quantum machine learning offers a different approach: quantum kernel methods can map borrower features into exponentially high-dimensional Hilbert spaces that are classically intractable to represent, potentially capturing non-linear relationships between credit variables that XGBoost and LightGBM miss. The question is whether this theoretical expressiveness translates into practically better PD and LGD predictions on real lending data, or whether NISQ hardware noise eliminates the advantage before it materialises.

This workshop answers that question empirically. It covers quantum kernel estimation and quantum support vector machines applied to consumer and commercial credit datasets, provides benchmark-specific performance comparisons between quantum and classical approaches, and addresses the regulatory dimension that makes credit scoring uniquely challenging for quantum adoption. ECOA and Regulation B require adverse action notices that explain why a borrower was declined. The EU AI Act classifies credit scoring as high-risk AI. Both demand model explainability that quantum models do not naturally provide. Participants leave with a working understanding of which credit scoring sub-problems show near-term quantum advantage, an honest assessment of current NISQ hardware limitations for their portfolio sizes, and a regulatory compliance framework for deploying quantum credit models under SR 11-7 governance.

What participants cover

• Quantum kernel estimation: how quantum feature maps create Hilbert space representations that classical kernels cannot efficiently compute
• PD classification with quantum SVMs: benchmark-specific performance comparisons on consumer and commercial lending datasets
• LGD estimation with variational quantum classifiers: where quantum approaches add value and where classical methods remain sufficient
• Fair lending compliance: ECOA adverse action requirements, EU AI Act high-risk obligations, and explainability for quantum models
• NISQ hardware reality check: qubit counts, gate fidelity, and circuit depth limits for production credit scoring workloads
• SR 11-7 model governance: validating, documenting, and monitoring quantum credit models in a regulated environment