There are a number of ways of assigning just a portion of an observed effect to a model, rather than the full effect. These techniques have the potential to improve model accuracy in a variety of context, most notably when there are either too many parameters or insufficient data to estimate parameters well. Among the most popular approaches are credibility models, penalised regression and boosted models. This talk will provide an introduction to all three, and point out some of the striking similarities and differences between them. There will also be some discussion on applications, including examples where Taylor Fry has used these approaches to deliver significant model improvements.

Moderator/Panelist:
Peter J. Mulquiney, Actuary, Taylor Fry Consulting Actuaries