Should I Use That Rating Factor? A Philosophical Approach to an Old Problem
Should I Use That Rating Factor? A Philosophical Approach to an Old Problem
Chris Dolman, Seth Lazar, Tiberio Caetano, and Dimitri Semenovich
All-Actuaries Virtual Summit (04 August 2020)
A common question asked of insurance professionals is whether or not a rating factor ought to be permitted for price setting. To answer this question one often falls back on a set of heuristics, for example, is the factor commonly used in the market, does it predict well, would people generally expect it to be used, or is it visible to customers. However, the underlying principles behind such intuitions are not necessarily appreciated or understood.
In this paper, we draw on contemporary ideas in procedural and distributive justice to decompose the classical notion of risk into risk types that in turn allow development of semiformal criteria by which rating factors may be evaluated. We hope that this new set of conceptual tools will allow practitioners to reason with greater clarity about ethical questions surrounding pricing systems.This paper is a collaboration between HMI, IAG and Gradient, and reflects our broader concern that new methods that use machine learning to influence risk predictions to determine insurance premiums won't be able to distinguish between risks the costs of which people should bear themselves, and those that should be redistributed across the broader population, and might also involve using data points that it is intrinsically wrong to use for this purpose. It was peer-reviewed for the All-Actuaries Virtual Summit, and presented to a live zoom audience of 200 on August 4th 2020.
Read the paper here.