The Canadian Institute of Actuaries (CIA) has updated its 2017 educational note, Use of Models, to address the rising use of machine learning (ML) models in property and casualty (P&C) pricing.
“This educational note supplement is a review of methods and guidance for the use of ML models in P&C pricing,” they write. The previous note focused on traditional statistical models. The new update fills in gaps where newer ML techniques have emerged, they add.
“Many of the topics covered then do not need to be expanded on in the context of ML models. For example, the discussions on model risk, choice of model, model limitations, documentation and use of models built by others are still relevant when the model used is an ML model,” they state in the note, Use of Machine Learning Models in P&C Pricing. “However the sections on sensitivity testing, model implementation, data validation and model validation necessitate some additional considerations that aren’t covered in the original educational note. This supplement is intended to provide some insights into those additional considerations.”
The supplement includes definitions, use cases and a discussion about the general workings of some common ML models.
“The actuarial profession has long used models for pricing and decision-making in P&C insurance,” the paper states. “The growing use of ML models in P&C pricing necessitates updated guidance.”
