Florida Insurance Regulators Claim Use of Occupation and Education in Rating is Almost Unfairly Discriminatory

You’d think that the regulators had their hands full dealing with the filings arising from property insurance reform. However, they must be keeping up well if the OIR is able to challenge modern auto insurance rating algorithms.

The Florida Office of Insurance Regulation (Office) today issued a report analyzing the insurance industry’s use of occupation and education for the underwriting and rating of auto insurance policies. The report finds the use of these practices unintentionally harms minorities and low-income individuals in determining auto insurance premiums and insurance eligibility.

“Let me be perfectly clear,” states Commissioner Kevin McCarty, “this practice is legal under current Florida law. However, similar to insurance companies’ past use of credit scoring, this practice creates unintended effects that policymakers may find unacceptable.” In 2003, the Florida Legislature passed Section 626.9741 severely limiting the use of credit scoring in insurance underwriting after this practice was also shown to disproportionately impact minorities and low-income individuals.

The cited report is available on the OIR website. Some of the criticism raised is, I think valid. (Although, I should remind readers of my standard disclaimer that I’m expressing my own opinions here, not those of past or present employers.) A couple of good points, in particular:

• Insurance industry representatives all claim ignorance of the relationship between occupation, education and income-level and race despite the existence of publicly available U.S. Census Bureau Data.
   
• Insurers [...] refuse to study the impact of underwriting practices on minority and low-income consumers.

It is an unfortunate side-effect of the litigiousness in today’s society that insurers are driven to adopt a see-no-evil, hear-no-evil stance on certain subjects for fear of how a failure to do so could inflame some silly, yet expensive (both financially and in terms of bad PR) lawsuits.

However, I fear that Florida regulators may be going above-and-beyond here in damning rating variables that apparently have significant predictive power for simple correlation with income and skin color.

Insurers are prohibited from making rates or underwriting on certain inappropriate variables, including ethnicity. There are certain societal reasons why such a prohibition is appropriate. Since this prohibition isn’t believed to open the door to abuse of the insurance system through adverse selection and other gaming, I have no problems with such a restriction.

However, under current laws, insurers are not denied access to variables that are simply correlated with prohibited attributes. This is a good thing, since every rating/underwriting variable I’m aware of in personal lines insurance is likely to have some correlation.

That some level of correlation is acceptable is a fact that has been tested in the courts. If memory serves, as long as there is a cost-based rationale for the use of a particular variable, and the insurer is adopting the distinction in a manner that is least likely to be borne by particular protected groups, it’s generally OK in the eyes of antidiscrimination law. (Note, I’m not a lawyer, so I could very easily be fumbling the specifics.)

Consider, for example, health or life insurance, where it is common practice for the underwriter to charge more for when the insured smokes. However, it’s also commonly known that there is a correlation between the likelihood of being a smoker and of belonging to a particular ethnic group or of having a lower income.

Yet I don’t hear too many insurance commissioners complaining that the use of smoking as a rating/underwriting variable is tantamount to illegal discrimination against minorities or poor people.

I am not immediately bothered by auto insurers looking at profession or education, even though those variables are correlated to income and ethnicity. Although I didn’t get into those variables back when I was more actively working with personal auto insurance, I suspect that there are loss-cost based reasons why insurers find those variables attractive.

I think that consumer advocates seeking to protect historically disadvantaged groups would do a far greater service by questioning whether those groups are paying in general more (or less) for their insurance in rating/underwriting systems using education and/or profession than they would in analogous systems that don’t rely on those elements.

Consider, for the sake of argument, that we live in a world where everyone is an identical auto insurance risk — purchases identical limits, drives identical cars, have identical driving experience — and that the only way we can differentiate among those risks is by where they live (”country club” versus “ghetto”) or by their occupation (”doctor” versus “janitor”). We also know that people are different by income level (”rich” versus “poor”), that rich people are more likely to live in the country club and be doctors, and that poor people are more likely to be janitors living in the ghetto.

Assume, for the sake of argument, that our little world, made up of 1000 auto insurance risks, looks something like this:

First, let’s assume that I can only do ratemaking by looking at territory, a rating variable that we know is legal, if somewhat unpopular. I might see the following data:

Therefore, I’d want to charge my country club risks $1,258 plus expenses, while I’d want to charge my ghetto risks $2,817, because that is the most accurate reflection of exposure to risk I can make based on just looking at territory.

On average, a poor person will pay $2,747 plus expenses — 15 are paying $1,258, while 318 are paying $2,817.

Now, let’s say just for the sake of argument, I introduced occupation into the mix. I might see the following data:

…and accordingly I’d seek to charge country club doctors $1,000, country club janitors $3,000, ghetto doctors $3,300 and ghetto janitors $2,200…plus expenses.

Enter the consumer advocates. They note that the average loss cost for a janitor is $2,363, while the average loss cost for a doctor is $1,978…and since occupation is correlated to income, we must have unfair discrimination at work

However, crunching the numbers, I note that on average, a poor person is being charged for an average loss cost of $2,536 (5 @ $1,000; 10 @ $3,000; 100 @ $3,300; and 218 @ $2,200), which is less than the $2,747 average loss cost that would be charged to poor people if I didn’t consider occupation.

Even though occupation is correlated to income, and even though the “poorer” occupation would be charged on average more than the “richer” occupation, I submit that, at least within this little example of ours, it would be unfairly discriminatory to not factor in occupation.

The relationship among demographic and rating variables are rarely clean, and oddities will arise in analysis. This means that you can’t simply bless or condemn a particular rating variable based on simple correlation.