Implementing claims experience in risk premium calculations

Specialeforsvar ved Astrid Mernøe Skriver

Titel: Implementing claims experience in risk premium calculations

Abstract: The aim of this thesis is to find a risk premium based on the individual claim experience. This was done by using credibility theory based models. To deal with the application of credibility models data from a Danish insurance company will be used. The data contains information about a commercial motor insurance in the years 2008 to 2015, which has a lot of heterogeneity. Four different credibility models will be introduced: Two ordinary Bühlmann- Straub models with different risk specification, a hierarchical credibility model and a model combining generalized linear models and the Bühlmann-Straub model. The Bühlmann-Straub model where individual policies are modelled will be considered the base model, which the three other models are extensions or developments of. The models will be fitted on the whole data and the subset where the last observation is dropped. The comparison of the model will be based on their ability to predict the last observation. It is showed that in all models there were an effect of using credibility models, but the effect was difficult to visualise and interpret. Even though there was an effect from the credibility models, none of the models were able to contain the heterogeneity of the data set. Thus all prediction values had large errors and should not be implemented directly before further adjustments. Of the four models the combining model had the best predictive abilities and the most possibility for further development

Vejledere:  Thomas Mikosch / Michael Heldrup, Codan
Censor:      Mette Havning