| Abstrakt: |
The central function of an insurance company in the context of automobile insurance is to establish suitable insurance prices, often referred to as insurance pricing. Recently, multivariate predictive modeling techniques, such as classifications and regression, have been employed for this process. The pricing process involves determining appropriate insurance premiums for individual policies that accurately reflect the risks associated with a particular vehicle and driver. Currently, generalized linear models (GLMs) are the standard. However, GLMs only provide estimates of the expected values of the response variables, disregarding extreme claims or tail events. To address this issue, a quantile premium may be a viable option as it delivers information about the entire distribution of a given phenomenon. This paper proposes a method for choosing between two models for estimating the quantile premium. Both models utilize the two-stage modeling technique to estimate the quantile premium in the special case where the independent variables are multicategorical factors. The method employs a supervised learning procedure, and the leave-one-out cross-validation (LOOCV) algorithm is used to validate models and select the appropriate one. [ABSTRACT FROM AUTHOR] |
