Discrimination-free Insurance Pricing with Privatized Sensitive Attributes

2026-06-16 · Source: stat.ML updates on arXiv.org · Field: Finance & Economics — Insurance & Risk Management, Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Expert, extended

Summary

This paper introduces a multi-party training framework for discrimination-free insurance pricing, addressing regulatory demands for fairness and privacy. The proposed method allows insurers to create fair models using only privatized sensitive attributes, such as gender or race, without direct access to the true data. It leverages a Trusted Third Party (TTP) to combine transformed non-sensitive attributes from the insurer with noisy sensitive attributes to estimate a discrimination-free premium. The framework provides statistical guarantees for scenarios where the privatization noise rate is both known and unknown. Experiments on a synthetic dataset of 5000 observations and real-world US Health Insurance (1338 observations) and Auto Insurance (8150 observations) datasets validate that higher noise levels increase error, while data transformations and larger sample sizes improve model performance and robustness.

Key takeaway

For AI Scientists and Legal Professionals developing insurance pricing models, you should adopt multi-party training frameworks to comply with evolving privacy regulations like the EU Gender Directive and Colorado SB 21-169. Prioritize methods that utilize privatized sensitive attributes via a Trusted Third Party, as this ensures discrimination-free premiums without direct access to sensitive data. Be particularly cautious about underestimating noise rates in privatized data, as this significantly impacts model convergence and accuracy.

Key insights

A multi-party framework enables discrimination-free insurance pricing using only privatized sensitive attributes, ensuring regulatory compliance and statistical guarantees.

Principles

• Actuarial fairness and algorithmic fairness may not coexist.
• Underestimation of privacy noise rate is more detrimental than overestimation.
• Data transformations can enhance model robustness against noise.

Method

Insurer transforms non-sensitive data and sends it to a TTP. The TTP combines this with privatized sensitive attributes to train group-specific models, minimizing a population-equivalent risk, then computes and returns the discrimination-free premium.

In practice

• Employ a Trusted Third Party for sensitive attribute handling.
• Apply data transformations to non-sensitive features for robustness.
• Carefully estimate noise rates, avoiding underestimation.

Topics

• Discrimination-Free Pricing
• Algorithmic Fairness
• Insurance Actuarial Science
• Local Differential Privacy
• Multi-Party Computation
• Sensitive Data Privatization
• Regulatory Compliance

Best for: AI Scientist, Research Scientist, Legal Professional

Related on AIssential

• How probability models protect privacy — Differential privacy uses calibrated randomness to protect individual data while enabling accurate aggregate analysis.
• The Trump Administration’s AI Policy Framework Has an Ideology. It Just Won't Admit It. — The Trump Administration's AI policy framework prioritizes unchecked innovation over critical safeguards, driven by an unstated ideological worldview.
• Yes to California's Bill to Ban Surveillance Pricing — Surveillance pricing, customizing product costs based on personal data, harms privacy, equity, and transparency, necessitating legislative bans.
• Fairness Beyond Demographics: Optimizing Performance Across Appearance-Based Hidden Cohorts in Medical Imaging — Optimizing fairness across appearance-based hidden cohorts improves medical image analysis models without relying on demographic labels.
• Benchmarking Empirical Privacy Protection for Adaptations of Large Language Models — Distribution shifts critically impact practical privacy in differentially private LLM adaptations, often undermining theoretical guarantees.
• Differential Parity: Relative Fairness Between Two Sets of Decisions — Differential parity assesses relative fairness between decision sets by measuring independence of differences from sensitive attributes.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.