Fairness-Aware Insurance Pricing: A Multi-Objective Optimization Approach
Tim J. Boonen, Xinyue Fan, Zixiao Quan
TL;DR
The paper tackles the challenge of balancing accuracy with multiple fairness notions in insurance pricing. It introduces a multi-objective optimization framework using NSGA-II to jointly optimize predictive accuracy and four fairness criteria (group, individual, counterfactual, plus overall performance), producing a Pareto front of solutions. A neural meta-learner and TOPSIS are employed to select a practical ensemble that leverages strengths of different models while respecting regulatory fairness constraints. Empirical results on private motor insurance data reveal inherent trade-offs and demonstrate that the proposed NSGA-II–driven ensemble achieves balanced performance across objectives, outperforming single-fairness baselines. The work provides actionable guidance for insurers and regulators on designing fair, accurate, and auditable pricing systems that navigate actuarial and ethical considerations.
Abstract
Machine learning improves predictive accuracy in insurance pricing but exacerbates trade-offs between competing fairness criteria across different discrimination measures, challenging regulators and insurers to reconcile profitability with equitable outcomes. While existing fairness-aware models offer partial solutions under GLM and XGBoost estimation methods, they remain constrained by single-objective optimization, failing to holistically navigate a conflicting landscape of accuracy, group fairness, individual fairness, and counterfactual fairness. To address this, we propose a novel multi-objective optimization framework that jointly optimizes all four criteria via the Non-dominated Sorting Genetic Algorithm II (NSGA-II), generating a diverse Pareto front of trade-off solutions. We use a specific selection mechanism to extract a premium on this front. Our results show that XGBoost outperforms GLM in accuracy but amplifies fairness disparities; the Orthogonal model excels in group fairness, while Synthetic Control leads in individual and counterfactual fairness. Our method consistently achieves a balanced compromise, outperforming single-model approaches.