Filip Lindskog: Estimation of conditional mean squared error of prediction

Time: Wed 2019-10-16 15.15

Lecturer: Filip Lindskog, SU

Location: Cramér room, room 306, house 6 at Kräftriket

Abstract

I will talk about challenges one faces when wanting to estimate the conditional mean squared error of prediction (MSEP), conditional on what is known at the time of prediction, when out-of-sample prediction performance is not an option because of lack of data. The prediction assessment principle advocated here can be seen as a generalisation of Akaike’s final prediction error (FPE). When applying this principle, estimators of MSEP that have been proposed by other approaches emerge.

The motivating example is an insurer having to predict what remains to be paid to policyholders for accidents that have occurred but are only partially or not at all known to the insurer. By accurately assessing the prediction error the insurer can put an appropriate amount of money aside that is likely to cover the yet unknown future payments.

The talk is based on joint work with M. Lindholm and F. Wahl.

Page responsible:webmaster@math.kth.se
Belongs to: Department of Mathematics
Last changed: Oct 11, 2019