Insurance premiums could be set at the same price for everyone, regardless of risk. That works when insurance has universal participation without influence over the benefit. U.S. Social Security Disability Insurance and medical expense insurance approximate this structure.

Individual Life insurance and Individual Disability insurance, among others, do not fit. Participation is voluntary and the buyer selects the amount of benefit. Since risk varies, these products use underwriting and actuarial pricing to match the risk to the price. That means insurers discriminate within parameters defined in statute, regulation and professional standards of practice. In fact, these ground rules obligate insurers to discriminate and set rates appropriate for the risk.

The basic justification for actuarially fair discrimination is that sound actuarial principles and reasonably anticipated experience justify differences in premium. Actuarially unfair discrimination treats the same risks differently and is illegal.

Regulations prohibit some forms of discrimination even when they satisfy the actuarially fair standard. Insurance can’t discriminate against a protected class (e.g., race, religion and national origin). Yet, among protected classes, some discrimination is actuarially fair, notably by age and gender. Examples include:

• Older insureds pay more for life insurance than younger insured.
• Males typically have higher mortality premiums than females.
• Healthy insureds pay less than unhealthy.
• Smokers pay more than nonsmokers.

Such risk-based pricing is crucial to the private voluntary insurance market. That said, unisex insurance pricing is mandatory within the European Union since the end of 2012. Given the universal need for insurance protection across genders, this did not result in significant changes to consumer behaviors and thus anti‑selection.

These permitted forms of discrimination represent disparate treatment. Insurers consciously and deliberately use these parameters to set the price. Insurers refrain from using or even knowing the parameters of protected classes where regulation prohibits disparate treatment.

The increased use of data science to create risk models with more expansive input parameters has generated concern over invisible actuarially unfair discrimination against protected classes. Of special concern is evidence that has been called “external consumer data information sources” (ECDIS). Insurers have adopted risk models from several vendors built in part or in whole from ECDIS. These complex models lack transparency as to the underlying mechanism that differentiates risk.

If the only accessible measure of these models is the decision output, how is assessment of actuarially unfair discrimination possible? One solution is to compare the distribution of risk scores or underwriting actions among members and nonmembers of a protected class. A material difference in the outcome is called disparate impact, in which some of the evidence in the model acts as a surrogate for direct ascertainment of membership in a protected class. The model might thus produce proxy discrimination against the protected class.

While any protected class could be a focus of disparate impact or proxy discrimination, we will use race as a template. Political attention is primarily focused on race. Race may be easier to assess than other classes, such as religion.

This raises numerous concerns and questions for insurers, for example:

We’ll have another blog soon on this topic that will address aspects of the evolving regulatory environment.