Discussion about this post

User's avatar
Ben Wolfson's avatar

Option three at least seems to be the most honest: the point is (isn't it?) that you're pretty sure that the actual expected loss figure is too low, so it should be higher; you don't have a good reason to fudge the inputs in any specific way, so any fudging of the inputs you do is in service of making the output higher. So why not just make the output higher directly?

Expand full comment
Doug's avatar

Isn't there a sensitivity element to this that suggests 1 and 2 are the right place? If there's a parameter with a lot of uncertainty (of whatever sort), and that has a nonlinear effect on the magnitude of the loss, you're going to want to dick about with that to get a sense of how big a mistake it could be.

Expand full comment
11 more comments...

No posts