Insurance: an R-Program to Model Insurance Data

Abstract

Data sets from car insurance companies often have a high-dimensional complex dependency structure. The use of classical statistical methods such as generalized linear models or Tweedie’s compound Poisson model can yield problems in this case. Christmann (2004) proposed a general approach to model the pure premium by exploiting characteristic features of such data sets. In this paper we describe a program to use this approach based on a combination of multinomial logistic regression and epsilon-support vector regression from modern statistical machine learning.