London Market Monitor – 31 December 2021
Learn more on Solvency II discount rates, and the state of the local and global markets at month end.
The disproportionate adverse health and economic impact of COVID-19 on communities of color combined with the tragic murder of George Floyd and ensuing racial justice protests of 2020 have ignited widespread reexamination of current systems and how they may contribute to inequities. This reexamination has impacted the insurance industry as well, with many insurers making public statements about their commitments to diversity, equity, and inclusion and reaffirming their commitments to racial equity not only in their communities but in their business practices as well. Within the actuarial community, there is also renewed interest in reexamining what it means for rates to be “unfairly discriminatory” or result in a “disparate impact.”
Some regulators and state legislatures have also followed suit.
Most recently, the governor of Colorado signed Senate Bill 21-169, which beginning January 1, 2023, will require insurers to:
Many states already have statutes that prohibit insurers from using race, religion, or other variables in insurance rating. Unique to the Colorado bill, however, is the requirement for companies to demonstrate that procedures have been implemented to assess “unfair discrimination” and to share the results with the Colorado Division of Insurance. Historically, companies have largely demonstrated rates are not unfairly discriminatory by explaining they do not use prohibited variables or variables that serve as proxies for prohibited variables in their pricing or underwriting methodologies. In most cases, insurers do not have a need to collect protected class variables to underwrite and rate the policy, and therefore have not historically collected this data. The Colorado bill at face value suggests that more rigorous governance will be required for insurers writing business in Colorado.
An important feature of the Colorado bill is that, while it uses the term “unfair discrimination,” it defines the term to mean something that more closely resembles what traditionally has been disparate impact discrimination:
"Unfairly discriminate" and "unfair discrimination" include the use of one or more external consumer data and information sources, as well as algorithms or predictive models using external consumer data and information sources, that have a correlation to race, color, national or ethnic origin, religion, sex, sexual orientation, disability, gender identity, or gender expression, and that use results in a disproportionately negative outcome for such classification or classifications, which negative outcome exceeds the reasonable correlation to the underlying insurance practice, including losses and costs for underwriting.2
Historically, “unfair discrimination” has not always meant the same thing as “disparate impact.” In the context of “unfair discrimination,” the term “discrimination” traditionally refers to treating risks differently, usually in regard to pricing and underwriting, but not necessarily related to a federally protected class. The insurance industry’s ability to identify, differentiate, and quantify risk is hinged on the concept of fair discrimination, where risks are treated differently based on their expected insurance cost differences. For example, it is legal and fair to discriminate for auto insurance (charge different prices) based on the at-fault accident history of a driver, as drivers with prior at-fault accidents are more likely to cause accidents in the future, resulting in increased insurance costs. Unfair discrimination, therefore, refers to treating risks with similar expected costs differently. Disparate impact, on the other hand, typically refers to when a set of seemingly neutral practices disproportionately impact a protected class and are otherwise unjustified by legitimate rationale.3 The historical distinction between unfair discrimination and disparate impact is important because a practice that traditionally is not considered unfairly discriminatory could still result in a disparate impact.
Colorado’s expansion of the definition for unfair discrimination to include disparate impact means insurers that use external consumer data in Colorado will likely need to complete some form of a disparate impact analysis. The bill requires the commissioner to adopt rules for its implementation, so it is yet to be seen what kind of support will formally be required.
As consumers, regulators, and stakeholders demand more transparency and accountability with respect to how insurers’ business practices contribute to potential systemic societal inequities, insurers will need to adapt. One way insurers can do this is by conducting disparate impact analyses and establishing robust systems for monitoring and minimizing disparate impacts. There are several reasons why this is beneficial:
For insurers that are exploring disparate impact analyses for the first time, it is highly advantageous to engage outside counsel and an external actuarial consulting firm. Under this arrangement, the external actuarial consulting firm could be subcontracted by the insurer’s outside counsel. This model has several key benefits:
Of course, companies can conduct these analyses without the involvement of third-party experts. Companies that opt to conduct analyses internally should ensure safeguards are in place so that the data utilized cannot be accessed by unauthorized individuals or misused.
Prior to completing a disparate impact analysis, the necessary data must first be collected. The Colorado bill deals with effects on specified protected groups, but most property and casualty (P&C) insurers are not currently collecting information about those groups. How this information gap can be addressed is necessarily a part of any analysis. For example, if actual protected class data is unavailable, insurers can possibly estimate some of this data, but there are important questions about what sources or methods could be used to derive reliable data for some classes and characteristics.
After collecting the necessary data, there are several ways in which insurers can test whether disparate impacts exist. If an insurer is evaluating disparate impact in its rating plan, then at a high level the insurer should try to demonstrate the following:
The above testing procedures fit seamlessly into multivariate predictive modeling approaches, such as generalized linear models (GLMs), though the concepts could be applied using univariate methods or nonlinear methods. Similar concepts could be used to evaluate underwriting or other practices beyond rating.
It is yet to be seen how Colorado will implement the bill’s requirements and whether other states will follow in the footsteps of Colorado, though it is clear that there is renewed interest and focus on the topic of unfair discrimination and disparate impact due to recent events. Insurers that actively recognize this and plan accordingly will likely have an easier time adapting to regulatory changes, with opportunities to strengthen their reputations with consumers, regulators, and other stakeholders.
1The full text of Colorado Senate Bill 21-169 is available at https://s3.amazonaws.com/fn-document-service/file-by-sha384/0778a54759b619a1c1363b7b9215e3c6a2cc3c45073f5367e7182c0eaab126fe39317b6c90c6291ddb17dbd3e3cdc96e.
3Stead, S.T. Disparate Impact and Unfair Discrimination in Insurance Are Not the Same Thing. Federation of Regulatory Counsel. Retrieved September 26, 2021, from https://www.forc.org/Public/Journals/2020/Articles/Fall_2020/Vol31Ed3Article4.aspx.