The DOI matches its Crossref record; 80 of the 80 references it cites were found in a registry or an index; 7 of the 141 sources its claims cite were checked for support.
The Discriminating (Pricing) Actuary
Open accessAbstract
The insurance industry is built on risk classification, grouping insureds into homogeneous classes. Through actions such as underwriting, pricing, and so forth, it differentiates, or discriminates, among insureds. Actuaries have responsibility for pricing insurance risk transfers and are intimately involved in other aspects of company actions and so have a keen interest in whether or not discrimination is appropriate from both company and societal viewpoints. This article reviews social and economic principles that can be used to assess the appropriateness of insurance discrimination. Discrimination issues vary by the line of insurance business and by the country and legal jurisdiction. This article examines social and economic principles from the vantage of a specific line of business and jurisdiction; these vantage points provide insights into principles. To sharpen understanding of the social and economic principles, this article also describes discrimination considerations for prohibitions based on diagnosis of COVID-19, the pandemic that swept the globe in 2020.Insurance discrimination issues have been an important topic for the insurance industry for decades and are evolving in part due to insurers’ extensive use of Big Data; that is, the increasing capacity and computational abilities of computers, availability of new and innovative sources of data, and advanced algorithms that can detect patterns in insurance activities that were previously unknown. On the one hand, the fundamental issues of insurance discrimination have not changed with Big Data; one can think of credit-based insurance scoring and price optimization as simply forerunners of this movement. On the other hand, issues regarding privacy and use of algorithmic proxies take on increased importance as insurers’ extensive use of data and computational abilities evolve.
Overview
CoChat Summary
TL;DR
This paper reviews social and economic principles for assessing insurance discrimination, emphasizing that fairness judgments vary by jurisdiction and line of business and are complicated by Big Data, proxy discrimination, and COVID-19. It illustrates pricing/proxy issues with an Australian auto claims example showing gender is a significant predictor and that dropping or orthogonalizing it changes model fit and fitted values.
Key findings
- Insurance is inherently based on discrimination through risk classification; whether prohibitions are appropriate depends on viewing insurance as an economic commodity versus a social good, and prohibitions vary by jurisdiction and line of business.
- Big Data does not change the fundamental issues of insurance discrimination, but it increases the importance of privacy and algorithmic proxy discrimination; the paper notes that machine learning fairness has inherent trade-offs, citing Kleinberg et al. that three core fairness conditions cannot generally be satisfied simultaneously.
- In the empirical example using 4,624 Australian automobile claims, a gamma regression finds `Female` has a coefficient of -0.179 (t = -3.450, p = 0.001), indicating lower claim amounts for females; Model 1 has the best AIC at 79,365.63, versus 79,391.08 for Model 2 and 79,390.49 for Model 3, while adding a proxy improves Models 4 and 5 to AIC 79,382.40 and 79,381.38.
- For COVID-19, the paper concludes that societal concerns dominate and that a prohibition based on current or suspected diagnosis is warranted; it also notes that because COVID-19 affects African-Americans more than other ethnic groups and mortality is related to age, insurer discrimination based on COVID-19 could be an indirect path to race and age discrimination.
Methods
- Conceptual/review paper synthesizing social justice, economic, actuarial, and legal/regulatory perspectives across jurisdictions (including the U.S., EU, China, Japan, and Australia) and lines of business, with special discussions of Big Data, proxy discrimination, genetic testing, price optimization, credit-based insurance scores, and COVID-19.
- Empirical illustration analyzes 4,624 claims from Australian automobile insurance drawn from De Jong and Heller (2008), using the R package CASdatasets; claim severity (`ClaimAmount`) is modeled with a gamma GLM using a logarithmic link, with predictors including `VehValue`, `DrivAge`, and `Female`.
- Six models are compared: full model with gender, gender-excluded model, orthogonalized predictors uncorrelated with gender, proxy-added models, and Pope-Sydnor predictors; analysis compares coefficients, AIC, fitted means by gender, and fitted-value distributions.
Limitations
- Stated: the paper says strategies for mitigating proxy discrimination are “none of which are ideal”; discrimination-aware machine learning is “still in a nascent state, especially in the context of insurance practice”; and there is limited actuarial/insurance literature on measuring and managing discrimination with machine learning. It also states that no study has confirmed or disproved whether mutuals discriminate differently than stock insurers.
- Stated/evident: the empirical example is explicitly “for illustrative purposes” and uses one Australian auto claims dataset focused on claim severity, with gender as the protected variable; the models are predictive and do not establish causality.
- Evident: the paper is largely a conceptual and regulatory review rather than a systematic empirical test, and its regulatory descriptions are current as of the writing, so they may not capture later legal or market changes.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
6 Valid · 49 Not valid · 125 Not checked · 110 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“For other lines of business such as personal auto, it is possible to find variables that provide a suitable substitute for gender, c.f., Ayuso, Guillén, and Pérez-Marín (2016).”
What the sources say
- Supportedcites Ayuso, Guillén, and Pérez-Marín (2016) source
The Conclusions state that driving patterns and vehicle usage can substitute for gender as rating variables in pay-as-you-drive auto insurance. This supports the citing sentence’s claim that suitable substitute variables are possible in personal auto insurance; the paper’s evidence comes from a sample of drivers under 30. (LLM, conf=0.92)
hedgedstates an association- 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.
- Supported
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
“removing effects of gender from the other predictor variables does not do that much damage to the overall model fit.”
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.
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
“From Model 1 , females have significantly lower claims.”
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.
None of the sources cited for this claim have been checked yet.
In the paper
“Some standard matrix algebra shows that b ⋆ 2 = b 2 , that is, the regression coefficients from the transformed variables equals the regression coefficients in the full model.”
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.
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
“By construction, these predictors are uncorrelated with the sensitive, protected, variables.”
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.
62 more claims
Create a free account to see all 290 claims and to verify the 17 references still open.
- Probably not a substantial claim. 204
- It may state two separate claims as one. 70
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 major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ 6.3 Empirical Example
“we fit a model using the orthogonalized versions of the VehValue and DriveAge variables, making each variable uncorrelated with Female . This results in Model 3 .”
§ Gamma Regression Summary
“VehValueFemale <-RemFem (ClaimsData $ VehValue) DrivAge1Female <-RemFem (1 * (ClaimsData $ DrivAge == "old people"))”
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
§ 6.3 Empirical Example
“These fits are rescaled so that they have the same mean, thus promoting comparability. Figure 2 expands upon this by showing the corresponding distributions.”
§ Plot by Gender
“val= c (AmtGamma1 $ fitted.values, AmtGamma2 $ fitted.values, AmtGamma3 $ fitted.values, AmtGamma4 $ fitted.values, AmtGamma5 $ fitted.values, PSFitted)”
Citations
Cites
Filders, R. (1992) · International Journal of Forecasting · doi:10.1016/0169-2070(92)90130-2 · Cited by 16,165
(2012) · American Economic Review · doi:10.1257/aer.102.2.i · Cited by 13,709
A Survey on Bias and Fairness in Machine Learning
Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2019) · arXiv · doi:10.48550/arxiv.1908.09635 · Cited by 5,812 · cited 3×
“Following Mehrabi et al. (2019), we summarize three ways for discrimination prevention: preprocessing, in-processing, and postprocessing.”
Official Journal of the European Communities
(1984) · Analytical Proceedings · doi:10.1039/ap9842100196 · Cited by 4,027
Barocas, S., & Selbst, A. D. (2014) · doi:10.2139/ssrn.2477899 · Cited by 3,048
Data preprocessing techniques for classification without discrimination
Kamiran, F., & Calders, T. (2012) · Knowledge and Information Systems · doi:10.1007/s10115-011-0463-8 · Cited by 1,305
Inherent Trade-Offs in the Fair Determination of Risk Scores
Kleinberg, J., Mullainathan, S., & Raghavan, M. (2016) · arXiv · doi:10.48550/arxiv.1609.05807 · Cited by 941
Fairness-Aware Classifier with Prejudice Remover Regularizer
Kamishima, T., Akaho, S., Asoh, H., & Sakuma, J. (2012) · Lecture Notes in Computer Science · doi:10.1007/978-3-642-33486-3_3 · Cited by 694 · cited 3×
“For example, Loi and Christen (2019) provided an ethical analysis of private insurance discrimination and fairness in machine learning.”
Veale, M., & Binns, R. (2017) · Big Data & Society · doi:10.1177/2053951717743530 · Cited by 512 · cited 1×
“Veale and Binns (2017) introduced and discussed three potential approaches to deal with this problem, including (1) having trusted third parties to store data necessary for incorporating fairness constraints in modeling, (2) building collaborative online platforms to allow diverse organisations to…”
BackgroundVeale, M., & Binns, R. (2017) · doi:10.31235/osf.io/ustxg · Cited by 512
Research works printed but not linked: 12
- Baker T., 2002, In Embracing risk: The changing culture of insurance and responsibility, 33
- Barocas S., 2016, California Law Review, 104, 671
- Borselli A., 2011, Connecticut Insurance Law Journal, 18, 109
- European Union, 2004, Official Journal of the European Union
- European Union., 2012, Official Journal of the European Union
- Federal Trade Commission and others, 2007, A report to Congress by the Federal Trade Commission
- Friedland J., 2013, Fundamentals of general insurance actuarial analysis
- Klein R., 2017, Genetics and life insurance. A view into the microscope of regulation
- Maysami R. C., 1999, Journal of Insurance Regulation, 18
- Swedloff R., 2014, Connecticut Insurance Law Journal, 21
- Thouvenin F., 2019, Journal of Intellectual Property, Information Technology and E-Commerce Law, 10
- Werner G., 2016, Basic ratemaking
Cited by
Data Analytics for Non-Life Insurance PricingPreprint
Wüthrich, M. V., & Buser, C. (2017) · SSRN Electronic Journal · doi:10.2139/ssrn.2870308 · Cited by 134
Discrimination-Free Insurance Pricing
Lindholm, M., Richman, R., Tsanakas, A., & Wüthrich, M. V. (2022) · ASTIN Bulletin · doi:10.1017/asb.2021.23 · Cited by 60
“A recent wide-ranging discussion of several issues connected with the topic of discrimination in insurance is found in Frees and Huang (2021), who also address the issue of indirect discrimination.”
BackgroundEllili, N. O. D., Nobanee, H., Alsaiari, L., Shanti, H. Z., Hillebrand, B., Hassanain, N., & Elfout, L. (2023) · The Journal of Finance and Data Science · doi:10.1016/j.jfds.2023.100102 · Cited by 66
Antidiscrimination Insurance Pricing: Regulations, Fairness Criteria, and Models
Xin, X., & Huang, F. (2024) · North American Actuarial Journal · doi:10.1080/10920277.2023.2190528 · Cited by 43
Self-citationWhat is fair? Proxy discrimination vs. demographic disparities in insurance pricing
Lindholm, M., Richman, R., Tsanakas, A., & Wüthrich, M. V. (2024) · Scandinavian Actuarial Journal · doi:10.1080/03461238.2024.2364741 · Cited by 37
“…include: a counter factual approach drawing from causal inference, see see (Araiza Iturria et al., 2022; Charpentier, 2022; Kusner et al.,2017 ); the probabilistic approach of Lind-holm et al. (2022) focussing specifically on implicit inferences; and the projection method of Frees and Huang (2022).”
Metadata
- Authors
- Frees, E. W. (., & Huang, F.
- Affiliations
- Australian National University (AU)
- UNSW Sydney (AU)
- Corresponding author
- Frees, Edward W. (Jed)
- Venue
- North American Actuarial Journal
- Publisher
- Informa UK Limited
- Published
- 2021-08-06
- Volume
- 27
- Issue
- 1
- Pages
- 2-24
- ISSN
- 1092-0277 (ISSN-L)
- 2325-0453
- OpenAlex
- W3195250490
- Semantic Scholar
- 65d55cd4395a4f79924e8f947a56ba0ef6af5dd4
- Cited by
- 56
- References
- 66
- Citations by year
- 2026: 15 · 2025: 9 · 2024: 15 · 2023: 8 · 2022: 5 · 2021: 3
- FWCI
- 5.92
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- North American Actuarial Journal (publishedVersion, cc-by-nc-nd)
- SSRN Electronic Journal (acceptedVersion)
- ANU Open Research (Australian National University) (submittedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- Topics
- Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Insurance and Financial Risk Management
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Economics, Econometrics and Finance › Economics and Econometrics › Healthcare Policy and Management
- Keywords
- insurance discrimination, insurance pricing, risk classification, actuarial practice, privacy, price optimization, underwriting
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.
A language model drew the claims from the paper on 2026-11-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-11 (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: The Discriminating (Pricing) Actuary. https://app.cochat.ai/papers/10.1080/10920277.2021.1951296
- The paper
- Frees, E. W. (., & Huang, F. (2023). The Discriminating (Pricing) Actuary. North American Actuarial Journal. https://doi.org/10.1080/10920277.2021.1951296
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.