Application of spatial regression models in general insurance

Desjardins General Insurance proposes different types of P&C (property and casualty) coverage to its clients. Properly rating an insurance premium to cover for P&C losses requires complex modelling that uses variables that are both at the individual and at the spatial level. In general, including a spatial component in the modelling is challenging. At Desjardins, the current approach generally produces good results, but they have shown inconsistencies over time. With a goal of constant improvement, we aim to question and revisit this approach, like it is regularly done with the various approaches in use at Desjardins. In particular, we aim to effectively include a spatial component in a way that ensures a smoothly varying rating over the territory (for similar individuals), the latter being a quality that the current approach lacks. While achieving this, we will make sure that the resulting approach is easy to implement on Desjardins’ platform. Our strategy to achieve those goals is to revisit the current model and explore different solutions, from robust estimation to highly flexible spatial models. Properly rating insurance premiums is beneficial for Desjardins as it positively affects its financial results.

Faculty Supervisor:

Philippe Gagnon

Student:

Partner:

Desjardins Assurances Générales

Discipline:

Mathematics

Sector:

Finance and Insurance

University:

Université de Montréal

Program: