They use a heck of a lot more than that. They use your name and DOB to run a backdoor credit pull and it turns out that is a pretty reliable indicator of what risk pool you belong in. There is also a "credit check" for insurance (You can get your report from LexisNexis) and they can see how many times you applied for other insurance products and how long you had them, which is another indicator.

Also your address, which is surprisingly useful in predicting likely outcomes (even if the range of outcomes in a neighborhood are wide at the extremes).

It really is depressing just how much information, from life expectancy to health outcomes, can be gathered from a simple five digit zip code. Very accurate too.

They're not "supposed" to do that (at least in the US) because the practice of Redlining was made illegal by the Fair Housing Act of 1968.

https://en.wikipedia.org/wiki/Redlining

https://web.archive.org/web/20170317143050/https://fairhousi...

Insurance companies apparently seem to think it’s useful, but I find it extremely hard to believe it would be more useful than, say, even the most basic information about one’s driving behavior, like whether one drives 1,000 miles a year or 50,000.

They mostly deal with numbers on the macro scale. If 10,000 miles per year is a good representation of how much the general populous drives, it’s good enough for them. This is why we’re seeing more niche insurance companies pop up (MetroMile, et al) who can offer “smarter” rates based on a more granular assessment of risk.

All else equal a driver who does 5000 miles a year will have half the adverse events of Someone driving 10,000 miles/year. That’s a huge difference that makes no sense of lump together. And yet only a few (non-commercial) insurers, like metromile, account for it. It makes no sense.

I don't think that'd be less useful, but they can't trust what you say: "sure, I only drive very little, very carefully, and very slowly" is not something they can check.

If they can, they'll use it: that is why black box car insurance exists.

If you’re just going by mileage count, you can look at the odometer.

My car insurance definitely asks for that sort of information, so I’m not sure why yours doesn’t.

They do ask it, but it doesn’t affect the rates and they don’t have any way of verifying how much I actually drive.

I also think the simpler the algorithm, the harder it is to game.

“Oh wait, I almost never drive, but when I do, I’m either drunk or speeding. Well, I’ll sign up with this insurance company which offers me a discount because I drive so little!”

To use more information, you make more granular risk pools. The pools need to ne large enough to actually pull risk.

If you try to ise 10 pieces of information with 10 possible values each, that could give you 10 billion risk pools, i.e. individual person. Alternatively 33 Yes/No questions alsp let you identify an insividual person on planet earth.

Having "individual" insurance has no real meaning, because at that point its not spreading risk.

Two things to this:

- That bit of info you give can be used to infer other info. Also it's possible those first three things explain most of the variation. We've all heard that young men in their 20s crash their cars a lot, I get the feeling it's true though I've never checked.

- You only need the risk groups to be granular enough that it isn't totally obvious that people are different risks (footballers and grannies).