|
|
 |
| |
|
|
|
Modeling Extreme Risks in Non-Life Insurance, from Threshold Estimation to AI Premiums |
|
|
|
PP: 1029-1048 |
|
|
doi:10.18576/amis/200413
|
|
|
|
Author(s) |
|
|
|
Yassine Kouach,
Abderrahim EL Attar,
|
|
|
|
Abstract |
|
|
| Pricing extreme risks is difficult given the rarity of these events, the small volume of severe observations, and the peculiarity of their statistical distribution. Thus, traditional methods such as generalized linear models are unable to model the frequency and severity of extreme claims more accurately. On the other hand, most of the work put forward in the literature has required the formulation of assumptions on the distribution of extreme data, which depend on the threshold of severe claims used. In this regard, this study proposes a new approach to pricing extreme disasters based on artificial intelligence instead of the models traditionally used in this type of modeling. The proposed approach makes it possible to exploit artificial intelligence throughout the pricing process for extreme risks, from identifying the threshold to determining the pure premium. To achieve this, a k-means algorithm is used to determine the threshold for extreme losses. The occurrence of these losses is predicted by using machine learning methods adapted to classification problems. An actuarial model of pure premium insurance is developed to accommodate extreme losses, while the frequency and average cost of major losses are estimated using machine learning algorithms. The results of this study demonstrate the effectiveness of the proposed approach, based on artificial intelligence algorithms, in offering fair prices tailored to the high level of risk involved. This approach allows insurance companies to optimize the pricing of high-risk insurance contracts. |
|
|
|
|
|
|
|
