eli@gs160.sp.cs.cmu.edu (Eli Brandt) writes:
>Since insurance has come up, there's something I've always wondered
>about -- maybe this list has somebody who knows. Why are insurance
>companies' models so crude? They lump risks into classes that I find
>astonishingly broad, throwing away a lot of data right up front.
>Compare with, say, credit-card issuers, who acquire all sorts of data
>by all sorts of means and shovel it into fancy machine-learning
>algorithms that try to form predictive models of their customers.
Credit card companies have a source of large amounts of unbiased data
that comes as a natural byproduct of their operations.
Insurors trying to get data directly from their customers are limited
by customers desire to mislead insurors about their risks, and are often
hampered by privacy concerns and lack of automation when trying to get
data from sources such as doctors.
Also, credit card companies tend to use the data for things such as
fighting theft, where success causes few complaints. Insurors use
data mainly to separate the risk aversion functions of insurance from
the wealth-redistribution effects, which many customers consider wrong.
This has probably created some legal restrictions on discrimination
against riskier customers.
-- ------------------------------------------------------------------------ Peter McCluskey | Critmail (http://crit.org/critmail.html): http://www.rahul.net/pcm | Accept nothing less to archive your mailing list [To drop AltInst, tell: majordomo@cco.caltech.edu to: unsubscribe altinst]Received on Thu May 7 12:51:55 1998
This archive was generated by hypermail 2.1.8 : Tue Mar 07 2006 - 14:49:12 PST