New Report on Car Insurance Redlining

Empirical studies have shown that minorities pay more for goods and services, and that they pay more to finance their purchases of those goods and services -- for instance, through subprime home and auto loans. Machine Bias, a new study from ProPublica and Consumer Reports, adds car insurance premiums to the list of what minorities can expect to pay more for. The study uses zip codes to analyze auto insurance premiums and payouts in four states, California, Illinois, Texas, and Missouri. It finds that major insurers charge up to 30% more in minority neighborhoods as compared to white neighborhoods with the same risk profile. The results mean that where someone lives matters even more, and could have devastating consequences on upward mobility. When faced with budget-busting car insurance bills, do people give up the cars they need to get to work? Or do they go out without necessities, such as food and medicine, so they can pay their car insurance premiums?

Comments

[Tom @]

The correct question is if insurers are making more money in these zip codes than in other zips. Since we are in a multi year price war in private passenger automobile insurance, it is highly unlikely that black zip codes are more profitable (or lose less) than white zip codes.

State Farm just posted a $7 billion underwriting loss on this line of business for 2016. Costs are up and there is a market share war and insurers have been slow to raise rates, in spite of a sharp increase in loss costs.

Red Lining has traditionally meant that financial services were not available in a geographical area. No one is arguing that.

If blacks are being overcharged, then they should be more profitable. Provide evidence of that.

Or else this is just another example of an outcome that some people decided to label unfair. Not a process that involves racial bias.

[FJP]

Tom, if the rates are higher in minority neighborhoods after you control for claims payout experience, which I think is what the study says, doesn't that answer your question? Recognizing that loss payments are only one line on the expense side of the income statement (albeit a big one), is there any reason to believe that other S, G & A type routine operating expenses are consistently higher per dollar of premium in minority areas?

[Tom @]

You are right. They use industry loss cost data by zip. They don't have more granular loss data by insurer, nor do they have associated premiums. I will look at it in more detail. My initial comment was a little harsh and also off the mark. Sorry.

[ltk]

ProPublica used rates for a specific demographic, but average losses over everyone in the zip. This may cause a false effect to appear. (Like, let's say that boys in Zip A are more reckless than boys in Zip B. Zip A is mostly women with a few boys, and Zip B is mostly boys with a few women. It can simultaneously be true that (a) the fair rate for boys is higher in Zip A, and (b) Zip A shows lower claims on average, assuming the women incur much lower claims than the boys.)

The actuaries all pointed that out. ProPublica's rebuttal (linked below) seems to claim that they didn't do that, but their methodology says that "In order to control for factors outside of geography, we limited our analysis to a single profile – a 30-year-old female safe driver who is ...". They say in their rebuttal that they repeated the process for other single profiles, but that's not the point. The correct comparison is a weighted average over all the profiles by the demographics of the zip.

It wouldn't shock me if insurance rates were racially biased; but with this level of statistical ignorance or obfuscation, we'll never know. ProPublica also botched their article on scores used to predict the risk that defendants jump bail, and tried to recover by treating as a new discovery that a score can be (a) fair in that the score doesn't explicitly consider race, or (b) fair in that when the score is applied, defendants of different races jump bail at the same rate, but in general not both. Meanwhile, the fact that those scores openly, explicitly, and deliberately make it less likely that anyone with relatives in prison--of any race, though this is obviously the kind of question that creates the disparate racial impact too--makes bail is basically ignored, or at least treated as less outrageous than their half-baked statistical conclusions.

https://www.propublica.org/article/the-car-insurance-industry-attacks-our-story-our-response

[Tom @]

I made a couple of detailed comments on their web site.

I will only add that Private Passenger Auto is the most price competitive insurance product, if not consumer financial product. If a group is systematically overcharged, they would be more profitable and it would be surprising if this were discovered, with at least most of the excess profit arbitraged away through the market.

It also might be true that minorities in bad neighborhoods may need to shop more aggressively to get price competitive rates. It wouldn't surprise me if there were more price variance in quoted rates in some areas than others. I consider this analogous to situations where eggs or milk are more expensive in bad neighborhoods -- unfortunate, but not bias.

As far as the general thrust of their series -- the pricing algorithms are not new, they are filed with state insurance departments, and they have been regulated for decades -- frequently by liberal or democratic state insurance commissioners. In comments to a reply by an insurance industry group, a former Minnesota regulator stated that he enforced a reduction between the cost based rate differential between the minority urban neighborhood and the lower cost territories.