Detariffed premium estimation for automobile insurance: modelling frequency and severity using Poisson, Lognormal, Pareto and Generalized Linear Model

The transition from a tariff-based to a de-tariffed automobile insurance market in Malaysia has significantly reshaped premium estimation thus requiring for more accurate risk assessment among insurance companies. This project provides significant milestones by investigating the effects of de-tariffication by modelling claim frequency and severity using real-world insurance data. Advanced statistical models have been applied for robust premium pricing in which Poisson and Zero-Modified Poisson (ZMP) models were employed to model claim frequency. Moreover, Lognormal and Pareto distributions models were adopted to model the respective claim severity. The Lognormal distribution best captured small to moderate claims, whereas the Pareto model better represented extreme losses. Model evaluations favored the Lognormal distribution due to its superior goodness-of-fit metrics. We propose Generalized Linear Models (GLMs) to estimate pure premiums. The tariff model included factors like Cylinder Capacity and vehicle value, while the de-tariff model incorporated additional risk variables such as driver age, experience, vehicle age and power, second driver availability, area classification, and vehicle value. Findings confirm that risk-based pricing allows for fairer, more personalized premiums but also introduces new challenges like price competition and adverse selection. This research offers important insights for insurers, actuaries, and regulators striving for pricing sustainability.