Unfair Discrimination vs. Actuarial Fairness in Pricing

Actuarial Fairness Defined

The actuarial principle of fairness holds that each policyholder's premium should reflect their expected cost to the insurer. Under this principle, rating variables that are statistically related to expected losses are actuarially justified, and charging different premiums to groups with different expected costs is not only fair but necessary to prevent cross-subsidization. Without risk classification, low-risk individuals effectively subsidize high-risk individuals, potentially leading to adverse selection as low-risk individuals leave the pool. Actuarial Standards of Practice (particularly ASOP No. 12, Risk Classification) provide guidance on the selection and use of rating variables.

ASOP No. 12 establishes that a risk classification system should reflect expected cost differences, be practical to administer, be responsive to emerging cost differences, and be consistent with applicable law. The standard recognizes that actuarial considerations alone do not determine the acceptability of a rating variable; legal, regulatory, and social factors also play a role. This creates a framework in which the actuary provides the technical analysis while regulators and society determine the boundaries of acceptable practice.

The Legal Standard of Unfair Discrimination

Insurance regulation prohibits "unfair discrimination," which means charging different premiums to similarly situated risks or using rating criteria that violate public policy. The distinction between permissible risk classification and unfair discrimination is not always clear. The legal standard typically requires that rate differentials be based on "actuarial justification" or "sound actuarial principles," but these terms are subject to interpretation and can mean different things in different regulatory contexts.

The key distinction is between "fair discrimination" (charging different rates to risks with different expected costs, which is permitted and indeed required for sound insurance practice) and "unfair discrimination" (charging different rates based on criteria that are not related to expected costs or that violate public policy, which is prohibited). The challenge arises when a rating variable is statistically predictive of expected costs but is also correlated with a protected characteristic, creating tension between actuarial accuracy and social equity.

Controversial Rating Variables

Several rating variables have generated significant debate about the boundary between actuarial fairness and unfair discrimination. Gender has been used as a rating variable in auto insurance (young male drivers have higher expected costs than young female drivers) and life insurance (females have lower mortality rates), but the European Union banned gender-based insurance pricing in 2012, and several U.S. states restrict its use in various lines. Age is widely accepted as a rating variable in life and health insurance but is restricted in auto insurance in some jurisdictions.

Credit-based insurance scores are among the most contentious rating variables. Multiple actuarial studies have demonstrated a strong statistical relationship between credit scores and insurance loss experience across multiple lines of business. However, critics argue that credit scores are correlated with race and income, potentially serving as proxies for protected characteristics. Several states have restricted or prohibited the use of credit scores in insurance pricing, while others have allowed their use subject to specific regulatory requirements.

Geographic rating (ZIP code or territory) is another area of debate. Location is one of the strongest predictors of insurance losses, reflecting local hazards, crime rates, litigation environments, and cost of living. However, geographic rating can result in significantly higher premiums for residents of urban, low-income, or predominantly minority communities, raising concerns about disparate impact. Some states have required insurers to demonstrate that geographic rating does not produce unfairly discriminatory outcomes, adding a layer of fairness analysis to the traditional actuarial justification.

Algorithmic Fairness and Machine Learning

The increasing use of machine learning and predictive analytics in insurance pricing has intensified the debate about fairness. Machine learning models can identify complex, non-linear relationships in large datasets that traditional rating methods might miss. However, these models can also learn and amplify patterns in historical data that reflect past discriminatory practices or societal inequalities. A model trained on historical data may produce predictions that systematically disadvantage certain demographic groups, even when protected characteristics are not explicitly included as input variables.

The concept of "proxy discrimination" is central to this debate. Even if a model does not use race, gender, or other protected characteristics as inputs, it may use variables that are strongly correlated with these characteristics (such as ZIP code, occupation, or consumer behavior patterns), effectively achieving a similar result indirectly. Regulators and researchers have proposed various fairness metrics and testing procedures to detect and mitigate proxy discrimination, but there is no consensus on which metrics best capture the concept of fairness in insurance pricing.

Regulatory Approaches to Fairness

Regulators are developing new frameworks to address fairness concerns in insurance pricing, particularly as algorithmic and data-driven pricing becomes more prevalent. Some states have enacted or proposed legislation requiring insurers to test their pricing models for disparate impact on protected groups. Colorado's SB 21-169, for example, requires insurers to demonstrate that their use of external consumer data and information sources does not unfairly discriminate against consumers based on protected characteristics.

The NAIC has established working groups to study the use of artificial intelligence and machine learning in insurance, including their implications for fairness and discrimination. These efforts are producing guidance on how insurers should develop, test, and monitor algorithmic pricing models to ensure compliance with unfair discrimination standards. The regulatory landscape is evolving rapidly, and actuaries must stay current with these developments to ensure that their pricing models meet both technical actuarial standards and emerging fairness requirements.

The Actuary's Role

Actuaries are uniquely positioned to contribute to the fairness debate because they understand both the statistical methods underlying insurance pricing and the regulatory framework governing their use. The actuarial profession's emphasis on ethical practice, embodied in the Code of Professional Conduct and the Actuarial Standards of Practice, provides a foundation for engaging constructively with fairness concerns. Actuaries can contribute by developing transparent models, testing for disparate impact, communicating the trade-offs between statistical accuracy and social equity, and advising regulators on the technical feasibility and implications of proposed fairness standards.

Study Relevance

Fairness and discrimination topics appear on CAS exams covering ratemaking and regulation, and on SOA exams covering ethics and professionalism. Key concepts include the distinction between fair and unfair discrimination in insurance pricing, the requirements of ASOP No. 12, the debate over controversial rating variables (credit scores, geography, gender), the emerging field of algorithmic fairness, and the regulatory frameworks being developed to address proxy discrimination in data-driven pricing models. These topics are increasingly important for actuarial practice as the profession grapples with the societal implications of its pricing decisions.