Actuarial Seminar talk: Prof. Filip Lindskog (University of Stockhom)

Abstract: In non-life insurance pricing, finding a suitable premium function from data is seen as a regression problem. Traditional inference-based GLM analysis has been the go-to method for decades, but it is gradually being replaced by machine learning methods focusing on prediction rather than inference. We consider the realistic setting where a customer is allowed to terminate a one-year contract before the end of the one-year coverage period. We discuss what premium function standard regression methods are estimating and whether that function coincides with what they should be estimating. We also take a critical look at whether the commonly encountered assumptions involving exponential dispersion models make sense, where these assumptions come from, and to what extent they are needed.

Seminar organized by Prof. HJ. Albrecher