Hacker News Reader

Comment by MivLives | Hacker News Reader

I'd love to see the weird trends they find on some of the bigger more complex algorithms. Like people who drive red cars born July 2000 with a pet bird are 8 times more likely to get in an accident if they primarily drive near Chicago or Miami. I'm sure it's not that granular but I can dream.

View on HN

People who drive red cars born July 2000 with a pet bird are 8 times more likely to get in an accident if they primarily drive near Chicago or Miami

Hopefully they know their stats modelling well enough to know that's not an actual signal.

Heh. “I think you're overfitting here, no?”

“That may be, but as your insurance agent, I recommend getting rid of the bird to save money on your premiums.”

One of the things that swings the price a fair bit in the UK is your job title which has always intrigued me as a signal. I could understand if it's someone using their car for deliveries or as a race car, but I'm unsure why a "counsellor" could be paying double the premiums of a "Web developer" on standard insurance.

I've been told by an agent before that a "web developer" at a contractor/wordpress body shop-type will be less likely to take risks, while a "web developer" at FB or G would be more likely to do things like rock climb, drive into the middle of nowhere, drive faster.

Ach, there may well there be signal in there (and the insurance companies do enough stats modelling to tell if there is) - I mean, isn't the car color a well known example where the people's tendency to drive recklessly is correlated with their choice of buying a boring color or a flashy one?

Most state insurance commissions have strict rules that limit the factors which insurers are allowed to use for setting auto insurance premiums.

Reply on news.ycombinator.com