B1515
Title: The accuracy of claim severity prediction using longitudinal data
Authors: Alicja Wolny-Dominiak - University of Economics in Katowice (Poland) [presenting]
Tomasz Zadlo - University of Economics in Katowice (Poland)
Abstract: In the field of casualty/property insurance, an important issue is the prediction of the pure premium for the policy for the next insurance period. Given the available longitudinal claim data, the classical Buhlmann-Straub credibility model represents the next period's claim as a weighted average of historical claims arising from the experience of each risk group and the experience of the entire portfolio of policies. In this model, the fundamental assumption is claim independence. The claim dependence occurring in time is taken into account by the use of the copula. The credibility predictor is then the best predictor (conditional expectation) in the sense of the mean squared prediction. A bootstrap estimator of the predictor measure of accuracy is proposed which is based on the quantile of absolute prediction error. To obtain the value of the predictor, a Monte Carlo simulation is applied. To illustrate the bootstrap procedure in practice the portfolio of MOD risks from an insurance company operating on the Polish market is analyzed.
