Estimating non-life insurance reserves

Specialeforsvar: Marnus Parum

Titel: Estimating non-life insurance reserves
by combining machine learning and hierarchical modelling

Abstract: In claims reserving detailed information collected over the lifetime of a claim is often disregarded when aggregating and structuring claims data in run-off triangles for loss reserving. This consists of covariates describing the policyholder and the policy itself. In this thesis, it is proposed to combine granular methods, in which this information is not neglected, but instead deemed valuable in order to estimate claim reserves. A machine learning method, able to estimate hazard functions and utilizing the subsequent hazard rates to compute development factors to predict the incurred but not reported claims, is combined with a
hierarchical framework estimating the reserve of reported claims.
Combining these frameworks allows for estimation of the total reserve consisting of both reported claims and claims not yet reported. The thesis explores the depths of machine learning techniques such as Feedforward Neural Networks and eXtreme Gradient Boosting to ultimately estimate the hazard function. We furthermore delve into the hierarchical model structure
for claims reserving using layered Generalized linear models to estimate the total reserve by simulating future payments.

Vejleder: Munir Hiabu
Censor:    Martin Møller Svensson,