Numerical rating system for motor vehicle insurance rating.

Abstract

Academicians and insurance industry practitioners alike have always tried to come up with a premium rating structure that charges each buyer of insurance products according to the risk he exhibits. Private motor insurance, in particular, has been the focal point of such attention. This is because it is a large class (often the largest for any general insurer) which is often inadequately priced. With investment returns spiralling downward in most Asian countries due to the recent financial turmoil, insurers can no longer rely on high investment income to subsidise underwriting profit, and thus need to ensure that their premium rates are adequate to maintain profitability. In this study, an empirical analysis of motor vehicle insurance data is performed to predict pure risk premiums to be charged. The rating factors being studied comprise of those commonly used in the insurance industry. In arriving at the risk premium rates, three models have been fitted. A log-linear model based on the multinomial distribution is used for modelling claim frequency while a general linear model is fitted to claim severity. Predicted claim frequency probabilities are then combined with predicted claim severity amounts to arrive at a pure risk premium. A points rating model is then introduced to enable calculation of risk premiums from a set of rating points.