2775. Premium calculation using Parametric Quantile Regression for insurance count data
Invited abstract in session MB-9: Data Science in Insurance and Finance: New perspectives and Applications, stream OR in Finance and Insurance .
Monday, 10:30-12:00Room: Clarendon SR 2.01
Authors (first author is the speaker)
| 1. | Fabio Baione
|
| Università La Sapienza | |
| 2. | Davide Biancalana
|
| Sapienza University of Rome | |
| 3. | Aurora Ferri
|
Abstract
The measurement of the risk premium in insurance is a fundamental problem that is typically faced by evaluating separate models for frequency and severity and applying a safety loading to compensate for unexpected losses. In actuarial literature, recent studies have explored premium calculation using quantile-based principles, where quantile regression is employed to compute conditional quantiles, typically using a continuous response variable to model severity.
Our objective is to measure the safety loading using a premium principle based on the quantiles of the number of claims (frequency). However, applying quantile regression to count data presents both theoretical and practical challenges, which are usually addressed by artificially smoothing the discrete response variable through jittering. For these reasons, quantile regression has been rarely applied to count data in the actuarial field.
In this paper, we employ parametric quantile regression for count data, as introduced by Frumento and Bottai, to derive a quantile-based premium. Estimation is performed by minimizing a "simultaneous" version of the loss function used in standard quantile regression.
We provide an application to “daily disability benefit insurance.” This type of insurance contract falls under non-life insurance policies, where the source of uncertainty lies exclusively in claim frequency, as the benefit amount per day (severity) is contractually fixed at the inception date.
Keywords
- Risk Analysis and Management
- Machine Learning
Status: accepted
Back to the list of papers