If insurance companies are content to force good drivers with lower socioeconomic profiles to pay more for insurance, as a report published this week reveals, then I believe that big data will punish them.
New research out this week shows that conventional auto insurance underwriting has seized upon what some would
consider to be a socially-suboptimal model. Insurers are letting analytics make decisions on pricing. Their machines are telling them that a driving record is not very important in predicting the likelihood of a future claim - or at least it is not enough to underwrite based on that information all by itself. The Consumer Federation of America is pushing back against those practices.
Bob Hunter, CFA's Director of Insurance and the former Texas Insurance Commissioner, says that "American public knows that it is unfair for auto insurers to use factors like education and occupation in setting rates. In effect, auto insurers are discriminating on the basis of income and race."
Such is the current state of affairs in insurance pricing. While it is new, it is not the most modern approach out there. Usage-based pricing, which I will go into in more detail later on in this entry, is the state of the art. But the sticking points on the whole issue seem to depend on these two questions:
a) If insurers can find analytics models that improve on predicting risk but that have the unfortunate result of imposing higher pricing on the poor, is it OK? Should it be acceptable for an insurance company to decide that it is going to put less weight on driving records and more on educational attainment?
b) Is it still acceptable if there is another way out there to do it better?
Insurers tell us that good drivers pay less. Allstate advertises that it offers a "safe-driving bonus check." State Farm and GEICO are among many insurers that offer discounts for good driving records. The State of California's Proposition 103 put that principle in to its law books when its voters approved a referendum requiring that all insurers give a twenty percent discount to drivers who have not had a moving violation in the previous three years.
But perhaps those discounts are only a modest complement to a rate card that is stacked against the poor.
Current risk-based auto insurance pricing often hinges on factors that have nothing to do with driving records. Here are a few data points that CFA says are used to set premiums:
- Marital status
- Educational attainment
- credit score
- tenancy (rent versus own)
In breaking out pricing, CFA estimated that a Baltimore factory worker would pay $450 less for a yearly premium at Progressive if he or she could earn a college degree.
One result is that a lot of people are uninsured. Across the country, more than ten percent of people lack for liability coverage. The numbers are especially bad where poverty abounds. Poor states have more uninsured drivers: the correlation between the percentage of people in poverty in a state and its share of uninsured drivers was high (+0.57). In lower-income zip codes, more than thirty percent of registered vehicles were not covered. In Mississippi, more than one of every four drivers has no coverage.
The legal response is to outlaw underwriting based on certain kinds of inputs. But that is a game of whack-a-mole. Perhaps the better solution is to introduce even more data into the equation.
I am intrigued by Progressive's new SnapShot policy that prices risk based upon a person's driving habits: don't drive in the middle of the night, drive less frequently, and don't slam on the brakes and they discount the premium. Usage-based insurance does a lot to remedy under-writing inequity. A small insurer called MetroMile is now charging on a per-mile basis.
Even traditional underwriting is a lot different than it used to be. Nowadays, many companies pull from a wide swath of information sources. I think this is particularly the case in the kinds of non-bank financial products that are extended to under-banked consumers. The math geeks who package asset-backed securities for subprime car loans insist that credit scores have very little to do with loan payment performance. Frequently, they build models based on all kinds of unrelated information - your cell phone account history, the brand and age of your vehicle, or even from information about the broader economic picture of your neighborhood. The insurers have to make the same guesses. Indeed, given that they are asked to insure individuals from across the entire socioeconomic spectrum, they often have to find creative sources for data.
But telematics will prove to be much better. Once UBI gets rolling, its users will have better analytics than does any company that is still trying to predict risk based upon the duration of your cell phone contract. Progressive says that it now has 9 billion miles of data from SnapShot. With each additional mile, they are probably finding new independent variables. Ultimately, better data should push their loss ratios down.
Given that miles driven is inversely correlated to population density, many of the constituencies left in the cold by the techniques that CFA is deriding are going to fare better. To the extent that inner cities house more people of lower socioeconomic status, the new data is going to reward them because they are likely to be driving fewer miles. It is going to punish the suburbanite that commutes forty miles each way to work.
Part of the push-back on UBI is its impact upon our privacy. But let's be honest: that train has left the station. Individuals that value privacy should avoid using smart phones and browsers. If they must have their devices, then they should accept that part of the price for their significant functionality is personal privacy.
The thrust of CFA's report should frustrate a lot of people. It reveals that insurance companies are not being truthful with how much they value safe driving habits. More than two of three respondents in a 2012 survey said that it was unfair for auto insurers to base premium prices upon educational attainment. Almost as many said the same thing about using occupation as a factor.
Given its impact upon consumer privacy, many people will dislike usage-based insurance. But the privacy-benefit trade-off is the way of the future. Indeed, in any industry where companies use big data to change pricing, consumers usually have a chance to win from the arrangement. In fact, I would argue that this is a fairly good frame for gauging the impact of big data. If all of the savings go to the supplier (think online payday lending), then a shift to big data is a bad deal. The shift to mobile payments will persuade people to sell their privacy in exchange for offers. Most will accept the deal, just as many have already accepted the idea of swiping their customer card at the grocery store.