Antidiscrimination Insurance Pricing: Regulations, Fairness Criteria, and Models
Open accessThe DOI matches its Crossref record; 87 of the 87 references it cites were found in a registry or an index; 9 of the 56 sources its claims cite were checked for support.
Abstract
On the issue of insurance discrimination, a grey area in regulation has resulted from the growing use of big data analytics by insurance companies: direct discrimination is prohibited, but indirect discrimination using proxies or more complex and opaque algorithms is not clearly specified or assessed.This phenomenon has recently attracted the attention of insurance regulators all over the world.Meanwhile, various fairness criteria have been proposed and flourished in the machine learning literature with the rapid growth of artificial intelligence (AI) in the past decade and have mostly focused on classification decisions.In this article, we introduce some fairness criteria that are potentially applicable to insurance pricing as a regression problem to the actuarial field, match them with different levels of potential and existing antidiscrimination regulations, and implement them into a series of existing and newly proposed antidiscrimination insurance pricing models, using both generalized linear models (GLMs) and Extreme Gradient Boosting (XGBoost).Our empirical analysis compares the outcome of different models via the fairness-accuracy trade-off and shows their impact on adverse selection and solidarity.
Overview
CoChat Summary
TL;DR
The paper links insurance antidiscrimination regulations to machine-learning fairness criteria and implements them in pricing models using GLMs and XGBoost. In a French motor insurance dataset, the full model had the best accuracy but worst group fairness, XGBoost generally outperformed GLM in accuracy, and MU/MDP/MCDP/MC satisfied the four-fifths rule across scenarios.
Key findings
- M0 had the best prediction accuracy and worst group fairness in all scenarios, while XGBoost had better prediction accuracy than GLM. MU, MDP, MCDP, and MC all satisfied the four-fifths rule in all scenarios considered.
- Best fairness-accuracy trade-offs varied by scenario: XGBoost MCDP in Scenarios 1 and 3; XGBoost MU and MC in Scenario 2; XGBoost MU in Scenario 4.
- Table 3 reports mean predicted pure premiums by gender: GLM M0 male 130.47 vs female 95.66; MU male 114.03 vs female 124.05; MDP male 117.38 vs female 118.23; MCDP male 115.26 vs female 121.90; MC male 113.95 vs female 124.18, with similar XGBoost patterns.
- Fairness was achieved mostly via subsidy from young females to young males when using actual claims cost as benchmark; insurers implementing a fair model generally lost relatively high-risk customers and gained low-risk customers, with limited adverse selection effects.
Methods
- The paper reviews and compares insurance antidiscrimination laws in the United States, European Union, and Australia, matches regulations to individual and group fairness criteria, and proposes or implements models M0, MU, MDP, MCDP, and MC.
- Empirical analysis uses the French private motor insurance dataset from the R package CASdatasets (pg15training), containing 100,000 third-party liability policies observed from 2009 to 2010; after removing 21 duplicate records, there are 50,000 policies each year. The study focuses on third-party material claims, uses gender as the protected variable, and uses Age, Bonus, Group1, Density, and Value as nonprotected variables, with an Insurance Score created from Type, Category, Occupation, Group2, and Age.
- Models are estimated with frequency-severity GLMs (Poisson for frequency, gamma for severity) and XGBoost (Poisson deviance for frequency, gamma deviance for severity), across four scenarios; evaluation uses RMSE, normalized Gini index, disparate impact ratio, and double lift charts.
Limitations
- Stated: the paper considers only cost modeling (technical pricing), not the entire insurance pricing process; it focuses primarily on general insurance; and practical difficulty remains in collecting protected policyholder information.
- Stated: MDP using the disparate impact remover is not feasible when the nonprotected variable is categorical; orthogonal predictors only remove linear correlation; MDP loses information and fails to remove interaction effects when considering multiple protected attributes; M0, MDP, and MCDP require protected variables in both training and prediction, while MC requires them only in training.
- Stated and evident: the empirical analysis uses one French motor dataset and an artificially created gender proxy, with gender as the only protected variable; the paper also notes observational fairness criteria have inherent limitations such as indistinguishability, and results may not generalize across lines of business or jurisdictions.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
7 Valid · 17 Not valid · 35 Not checked · 99 Own results and interpretations
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“However, there is little research on insurance applications, particularly on insurance pricing as a regression problem; see Lindholm et al. (2022a), Araiza Iturria, Hardy, and Marriott (2022), and Grari et al. (2022) as examples of recent research in this area.”
What the sources say
- Not matchedcites Lindholm et al. (2022a)
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not checked — no full text of the sourcecites Araiza Iturria, Hardy, and Marriott (2022) source
- Supportedcites Grari et al. (2022) source
The citation presents Grari et al. as an example of recent research on insurance pricing, rather than as a source for a count of studies. The Abstract (p. 1) calls fairness in insurance pricing a relatively new topic, and §1.2 Motivation (p. 2) describes the paper’s model for predicting pure premiums in private car insurance. (LLM, conf=0.90)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
Every source the paper cites for this claim was checked and supports it.
In the paper
“Interestingly, if the protected attribute X P as a subset of X is also the parent (or direct cause) of random variables X j in X in a causal model and strong level 3 assumption in Kusner et al. (2017) is met, MDP further satisfies the counterfactual fairness criterion in Section 3.”
What the sources say
- Supportedcites Kusner et al. 2017 source
Section 4.2 (p. 7) describes Level 3 as a fully deterministic causal model with latent variables, matching the assumption attributed to Kusner et al. Section 3 (p. 3) defines the counterfactual fairness criterion. The statement that MDP satisfies that criterion is the citing authors’ own conditional conclusion, not a result attributed to this source. (LLM, conf=0.88)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on 2 assumptions not shown here.
- Supported
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As a better alternative to MU that also focuses on individual fairness, MC performs similar to MU when there is no strong gender proxy in the training data.”
- Also stated in: § 5.3. Model Comparison 5.3.1. Model Comparison with the Effects of Gender Proxy · plain
“MC performs similar to MU when there is no strong gender proxy in the training data.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 5.3. Model Comparison 5.3.1. Model Comparison with the Effects of Gender Proxy · plain
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“XGBoost MU and MC (MDP comes the third) achieve the best trade-off in Scenario 2.”
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“their model performance is different only when there is at least one moderate gender proxy in the training data”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
50 more claims
Create a free account to see all 158 claims and to verify the 1 reference still open.
- Probably not a substantial claim. 74
- It may state two separate claims as one. 64
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 5.3.2. Scenario Analysis: Group Fairness and Prediction Accuracy
“a smaller value of the normalized Gini index represents a more accurate model.”
§ 5.3.2. Scenario Analysis: Group Fairness and Prediction Accuracy
“a larger normalized Gini index indicates better model predictions.”
Citations
Cites
XGBoost: A Scalable Tree Boosting System
Chen, T., & Guestrin, C. (2016) · Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2939672.2939785 · Cited by 54,569 · cited 5×
“XGBoost was proposed by Chen and Guestrin (2016) as a novel gradient tree boosting method and has rapidly gained popularity because of its high efficiency in computational speed and predictive performance in applications in many fields.”
BackgroundMethodInfluentialAdvances in Neural Information Processing Systems 29
Teymur, O., Zygalakis, K., & Calderhead, B. (2016) · Cited by 13,887
Justice as Fairness: A Restatement
Rawls, J. (2001) · doi:10.2307/j.ctv31xf5v0 · Cited by 5,761
Rawls, J. (1958) · The Philosophical Review · doi:10.2307/2182612 · Cited by 5,565
Dwork, C., Hardt, M., Pitassi, T., Reingold, O., & Zemel, R. (2012) · Proceedings of the 3rd Innovations in Theoretical Computer Science Conference · doi:10.1145/2090236.2090255 · Cited by 3,501 · cited 6×
“In particular, we demonstrate the appealing potential of antidiscrimination pricing models for rate-making compared to common industry practice (fairness through unawareness; Dwork et al. 2012).”
BackgroundInfluentialBarocas, S., & Selbst, A. D. (2014) · doi:10.2139/ssrn.2477899 · Cited by 3,048 · cited 1×
“Barocas and Selbst (2016) analyzed the instances of discriminatory data mining under Title VII jurisprudence for employment discrimination taking into account both disparate treatment and disparate impact theories of liability and provided a bridge between computer science literature and existing…”
BackgroundEquality of Opportunity in Supervised Learning
Hardt, M., Price, E., & Srebro, N. (2016) · arXiv · doi:10.48550/arxiv.1610.02413 · Cited by 2,536 · cited 2×
“A major improvement in demographic parity is equalized odds proposed by Hardt, Price, and Srebro (2016; or separation criterion in Barocas, Hardt, and Narayanan 2019) requiring the predictor ^ Y and the protected attribute X P to be statistically independent conditional on Y .”
BackgroundFair prediction with disparate impact: a study of bias in recidivism prediction instruments
Chouldechova, A. (2017) · Big Data · doi:10.1089/big.2016.0047 · Cited by 2,159
xgboost: Extreme Gradient Boosting
Chen, T., He, T., Benesty, M., Khotilovich, V., Tang, Y., Cho, H., Chen, K., Mitchell, R., Cano, I., Zhou, T., et al. (2014) · CRAN: Contributed Packages · doi:10.32614/cran.package.xgboost · Cited by 2,025
Certifying and Removing Disparate Impact
Feldman, M., Friedler, S. A., Moeller, J., Scheidegger, C., & Venkatasubramanian, S. (2015) · Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2783258.2783311 · Cited by 1,827
Research works printed but not linked: 15
- Alexander v. sandoval (No. 99-1908; Vol. 532 p. 275). (2001). (Vol. 532). Supreme Court. https://www.supremecourt.gov/opinions/boundvolumes/532bv.pdf
- AMERICAN INS. V. US DEPT. OF HOUSING AND URBAN (Civil Case No. 13-00966 (RJL); Vol. 74 p. 30). (2014). (Vol. 74). Dist. Court Dist. of Columbia. https://ecf.dcd.uscourts.gov/cgi-bin/show_public_doc?2013cv0966-47
- Avraham, R., K. D. Logue, and D. Schwarcz. 2014b. Understanding insurance antidiscrimination laws. Southern California Law Review 87 (2):195–274.
- Barocas, S., M. Hardt, and A. Narayanan. 2019. Fairness and machine learning. fairmlbook.org.
- Dolman, C., and D. Semenovich. 2019. Algorithmic fairness: Some practical considerations for actuaries. Paper presented at Actuaries Summit 2019, Sydney, New South Wales, Australia, June 3–4.
- Goldburd, M., A. Khare, D. Tevet, and D. Guller. 2016. Generalized linear models for insurance rating. Arlington, VA: Casualty Actuarial Society, CAS Monographs Series, 5.
- Griggs v. Duke Power Co. No. 124; Vol. 401 p. 424 (1971). Supreme Court. https://tile.loc.gov/storage-services/service/ll/usrep/usrep401/usrep401424/usrep401424.pdf.
- Ojo v. Farmers group inc. (No. 10-0245; Vol. 356 p. 421). (2011). (Vol. 356). Tex: Supreme Court. https://www.txcourts.gov/media/819896/OpinionsFY2011.pdf
- Pearl, J. 2000. Models, reasoning and inference. Cambridge, UK: Cambridge University Press.
- Ricci v. DeStefano (No. 07-1428; Vol. 557 p. 557). (2009). (Vol. 557). Supreme Court. https://www.supremecourt.gov/opinions/boundvolumes/557bv.pdf
- Smith v. City of Jackson. No. 03-1160; Vol. 544 p. 228 (2005). Supreme Court. https://www.supremecourt.gov/opinions/boundvolumes/544bv.pdf.
- Texas Department of Housing and Community Affairs v. Inclusive Communities Project Inc. No. 13-1371; Vol. 576 p. 519 (2015). Supreme Court. https://www.supremecourt.gov/opinions/boundvolumes/576BV.pdf.
- Wards cove packing co. V. atonio (No. 87-1387; Vol. 490 p. 642). (1989). (Vol. 490). Supreme Court. https://tile.loc.gov/storage-services/service/ll/usrep/usrep490/usrep490642/usrep490642.pdf
- Willis C. J. R. J. Andreano and L. Sommerfield. 2021. President Biden issues executive order directing HUD to review fair housing act disparate impact rule. https://www.consumerfinancemonitor.com/2021/02/03/president-biden-issues-executiveorder-directing-hud-to-review-fair-housing-act-disparate-impact-rule/.
- Xenidis, R., and L. Senden. 2019. EU non-discrimination law in the era of artificial intelligence: Mapping the challenges of algorithmic discrimination. In General principles of EU law and the EU digital order, eds. Ulf Bernitz, Xavier Groussot, Jaan Paju, Sybe Alexander de Vries, 151–82. Alphen aan den Rijn, NL: Kluwer Law International.
Cited by
Kuppan, K., Acharya, D. B., & Divya, B. (2024) · IEEE Access · doi:10.1109/access.2024.3509918 · Cited by 62
A multi-task network approach for calculating discrimination-free insurance prices
Lindholm, M., Richman, R., Tsanakas, A., & Wüthrich, M. V. (2023) · European Actuarial Journal · doi:10.1007/s13385-023-00367-z · Cited by 19
“, Frees–Huang [6] and Xin–Huang [16], or indirect discrimination, see Lindolm et al.”
BackgroundFrom bias to black boxes: understanding and managing the risks of AI – an actuarial perspective
du Preez, V., Bennet, S., Byrne, M., Couloumy, A., Das, A., Dessain, J., Galbraith, R., King, P., Mutanga, V., Schiller, F., et al. (2024) · British Actuarial Journal · doi:10.1017/s1357321724000060 · Cited by 15
A fair price to pay: Exploiting causal graphs for fairness in insurance
Côté, O., Côté, M., & Charpentier, A. (2025) · Journal of Risk & Insurance · doi:10.1111/jori.12503 · Cited by 24
From point to probabilistic gradient boosting for claim frequency and severity prediction
Chevalier, D., & Côté, M. (2025) · European Actuarial Journal · doi:10.1007/s13385-025-00428-5 · Cited by 18
Metadata
- Authors
- Xin, X., & Huang, F.
- Affiliations
- UNSW Sydney (AU)
- Corresponding author
- Huang, Fei
- ORCID
- Huang, Fei
- Venue
- North American Actuarial Journal
- Publisher
- Informa UK Limited
- Published
- 2023-06-08
- Volume
- 28
- Issue
- 2
- Pages
- 285-319
- ISSN
- 1092-0277 (ISSN-L)
- 2325-0453
- OpenAlex
- W4379928685
- Semantic Scholar
- ffeb90bdaf8a78250cd8b1288e0e83f362954213
- Cited by
- 43
- References
- 70
- Citations by year
- 2026: 15 · 2025: 9 · 2024: 11 · 2023: 5
- FWCI
- 9.99
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- North American Actuarial Journal (publishedVersion, cc-by)
- SSRN Electronic Journal (acceptedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Insurance and Financial Risk Management
- Social Sciences › Law › Discrimination and Equality Law
- Keywords
- insurance pricing, anti-discrimination, indirect discrimination, big data analytics, algorithmic fairness, fairness criteria, insurance regulation, fairness-accuracy trade-off, adverse selection, solidarity, generalized linear models, XGBoost, actuarial science
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Antidiscrimination Insurance Pricing: Regulations, Fairness Criteria, and Models. https://app.cochat.ai/papers/10.1080/10920277.2023.2190528
- The paper
- Xin, X., & Huang, F. (2024). Antidiscrimination Insurance Pricing: Regulations, Fairness Criteria, and Models. North American Actuarial Journal. https://doi.org/10.1080/10920277.2023.2190528
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.