This paper presents a concrete solution to optimize input creation process for risk modeling while improving quality and thoroughness of model results by dealing with clustering on dataset liability. It suggests in addition an analysis of aggregation criteria influence on these dataset liabilities.
The study framework will be the Solvency 2 European prudential regulation applied to the case of a unit-linked life insurance contract integrated in a "Standard Formula" model.
In the first part, we will present a short reminder about classification and we will make some comparison of different clustering families in the case of an insurance application. Then, a presentation of the study setup will be performed. This presentation will be accompanied by a process proposal which objective will be to define the main steps of an insurance liability clustering. Finally we will observe aggregation criteria effects on capital requirements and model calculation time, and we will compare results between the different choices of clustering strategies and the per head case.