Due to the continuous rise in the amount of data over the recent years, insurance companies regularly seek for an improvement in their statistical analysis of the insurance claims data. Our main goal in this thesis is to investigate D-vine quantile regression, introduced by Kraus and Czado (2017), as a modelling approach for motor insurance severity rate. For that purpose, we present three additional regression methods; lognormal and gamma regression which are standard approaches in modelling positive, right-skewed data, and linear quantile regression which can be easily compared to D-vine regression since both regression methods predict conditional quantiles.
After laying the necessary fundamentals and the framework of the four regression methods, we perform an extensive exploratory data analysis for lognormal and gamma regression on two real-life motor insurance claims data sets. Then, we proceed with model fitting using the different regression methods. Finally, we evaluate and compare the resulting models based on several performance measures, some of which are the log likelihood, the training and test error and the interval score.
«
Due to the continuous rise in the amount of data over the recent years, insurance companies regularly seek for an improvement in their statistical analysis of the insurance claims data. Our main goal in this thesis is to investigate D-vine quantile regression, introduced by Kraus and Czado (2017), as a modelling approach for motor insurance severity rate. For that purpose, we present three additional regression methods; lognormal and gamma regression which are standard approaches in modelling p...
»