Non-Life Insurance Pricing Using the Generalized Additive Model, Smoothing Splines and L-Curves

Sammanfattning: In non-life insurance, almost every tariff analysis involves continuous rating variables, such as the age of the policyholder or the weight of the insured vehicle. In the generalized linear model, continuous rating variables are categorized into intervals and all values within an interval are treated as identical. By using the generalized additive model, the categorization part of the generalized linear model can be avoided. This thesis will treat different methods for finding the optimal smoothing parameter within the generalized additive model. While the method of cross validation is commonly used for this purpose, a more uncommon method, the L-curve method, is investigated for its performance in comparison to the method of cross validation. Numerical computations on test data show that the L-curve method is significantly faster than the method of cross validation, but suffers from heavy under-smoothing and is thus not a suitable method for estimating the optimal smoothing parameter.