Paper: arXiv 2505.18895

Authors: Fei Huang, Silvana M. Pesenti

Abstract

This paper introduces marginal fairness, a new individual fairness notion for equitable decision-making in the presence of protected attributes such as gender, race, and religion. This criterion ensures that decisions based on generalized distortion risk measures are insensitive to distributional perturbations in protected attributes, regardless of whether these attributes are continuous, discrete, categorical, univariate, or multivariate. To operationalize this notion and reflect real-world regulatory environments (such as the EU gender-neutral pricing regulation), we model business decision-making in highly regulated industries (such as insurance and finance) as a two-step process: (i) a predictive modeling stage, in which a prediction function for the target variable (e.g., insurance losses) is estimated based on both protected and non-protected covariates; and (ii) a decision-making stage, in which a generalized distortion risk measure is applied to the target variable, conditional only on non-protected covariates, to determine the decision. In this second step, we modify the risk measure such that the decision becomes insensitive to the protected attribute, thus enforcing fairness to ensure equitable outcomes under risk-sensitive, regulatory constraints. Furthermore, by utilizing the concept of cascade sensitivity, we extend the marginal fairness framework to capture how dependencies between covariates propagate the influence of protected attributes through the modeling pipeline. A numerical study and an empirical implementation using an auto insurance dataset demonstrate how the framework can be applied in practice.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces a novel concept of marginal fairness for risk measures, extending machine learning fairness to technical pricing in insurance. It rigorously formulates the mathematical framework using generalized distortion risk measures and cascade sensitivity while validating the theoretical framework with empirical auto insurance data.

Research Flowchart

  flowchart TD
    A[Research Goal: Achieve fair decision-making under risk measures insensitive to protected attributes] --> B[Input Data: Auto Insurance Dataset & Covariates<br><i>Protected vs Non-Protected</i>]
    
    B --> C[Stage 1: Predictive Modeling<br>Estimate target variable e.g., insurance losses]
    
    C --> D[Stage 2: Fair Decision-Making & Risk Adjustment<br>Apply Distortion Risk Measures & Cascade Sensitivity]
    
    D --> E[Numerical Study & Empirical Implementation]
    
    E --> F[Outcomes: Operational Marginal Fairness<br>Compliant with regulations e.g., EU gender-neutral pricing]

Cited References (50)

  1. Generalized linear models for insurance rating (2016)
  2. Fairness seen as global sensitivity analysis: A general approach for explainable fair classification (2022)
  3. Learning Fair Decisions with Factor Models: Applications to Annuity Pricing (2025) — link
  4. Fairness risk measures (2019)
  5. The dual theory of choice under risk (1987)
  6. Pricing insurance risk: Theory and practice (2022)
  7. Fairness of ratemaking for catastrophe insurance: Lessons from machine learning (2024)
  8. Equality of opportunity in supervised learning (2016)
and 42 more
  1. Casualty Actuarial and Statistical (C) Task Force Price Optimization White Paper (2015)
  2. General insurance pricing practices market study Feedback to CP20/19 and final rules (2021)
  3. Price Discrimination with Fairness Constraints (2022)
  4. Fairness Regulation of Prices in Competitive Markets (2024)
  5. Implementing anti-discrimination policies in statistical profiling models (2011)
  6. To split or not to split: Capital allocation with convex risk measures (2009)
  7. Differential Sensitivity in Discontinuous Models (2025)
  8. Counterfactual Fairness (2017)
  9. Training Individually Fair ML Models with Sensitive Subspace Robustness (2023)
  10. Sensitivity-Based Measures of Discrimination in Insurance Pricing (2024)
  11. On statistical criteria of algorithmic fairness (2021)
  12. Inherent trade-offs in the fair determination of risk scores (2016) — link
  13. Insurance, biases, discrimination and fairness (2024)
  14. R Package ‘CASdatasets’ (2020)
  15. A fair pricing model via adversarial learning (2022) — link
  16. Fairness and Machine Learning (2019)
  17. Cascade sensitivity measures (2021)
  18. Welfare implications of fair and accountable insurance pricing (2025)
  19. Probabilistic sensitivity measures as information value (2021)
  20. Global sensitivity analysis via optimal transport (2024)
  21. A common rationale for global sensitivity measures and their estimation (2016)
  22. Algorithmic Fairness: Contemporary Ideas in the Insurance Context (2018)
  23. Fairness through awareness (2012)
  24. On the apparent conflict between individual and group fairness (2020)
  25. Big data’s disparate impact (2016)
  26. Fairness in criminal justice risk assessments: The state of the art (2018)
  27. On a representation of random variables (1977)
  28. Mathematical risk analysis (2013)
  29. Antidiscrimination Insurance Pricing: Regulations, Fairness Criteria, and Models (2024)
  30. The Discriminating (Pricing) Actuary (2023)
  31. A Discrimination-Free Premium under a Causal Framework (2024)
  32. A Fair price to pay: exploiting causal graphs for fairness in insurance (2024)
  33. Discrimination-free insurance pricing (2022)
  34. Distortion riskmetrics on general spaces (2020)
  35. Premium calculation by transforming the layer premium density (1996)
  36. Remarks on quantiles and distortion risk measures (2012)
  37. Sensitivity analysis using risk measures (2016)
  38. Risk budgeting allocation for dynamic risk measures — → 2305.11319
  39. On law invariant coherent risk measures (2001)
  40. Characterization, robustness, and aggregation of signed Choquet integrals (2020)
  41. To split or not to split: Capital allocation with convex risk measures (2009)
  42. Sensitivity measures based on scoring functions (2023)

1 of 50 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.