Given how easy it would be to data mine this out of the claim history I find this explanation hard to buy. There must be a whole set of specific claim types that show no correlation with future claims. If you can’t find these in your data you are not looking very hard.

Of course if no one else is looking why give up the easy money - after all it is not like the customer can buy a new policy without the massive markup. Collusion does not need to be explicit, it can be accomplished just through everyone using the same systems and rules.