NOTE ON MISMODELLING OF POLICYHOLDER’S AGE IN CLAIM FREQUENCY MODEL: A MATTER OF GENDER IN VEHICLE INSURANCE

Abstract

Using motor hull insurance data from a Czech insurer, this paper examines how misspecification of policyholder age can lead to misleading conclusions about gender differences in claim frequency in vehicle insurance. The study is based on individual-level data with unit policy duration and emphasizes the correct modeling of the functional form of age, demonstrating that misspecification as well as categorization can produce biased inferences and that results depend strongly on the chosen categorization. The findings show that both linear specifications and categorization increase the likelihood of Type I error in detecting the interaction between gender and age. By applying fractional polynomials, the results partially support the judgment of the European Court of Justice to ban the use of gender as a rating factor, particularly for young policyholders. The study concludes that, in the absence of other relevant data, gender and its interaction with age should be included in claim frequency models, although such models cannot be directly used for premium setting.

Data:
Received: 9 Apr 2020
Revised: 23 May 2020
Accepted: 6 Jun 2020
Published: 20 Jun 2020