Context and aims
The viability of an insurance portfolio depends on the number of future claims. The premiums that are collected must be sufficient to absorb the cost of future events. By absorb, one means that the available funds must be sufficient even for pessimistic scenarios that occur with very low probability (0.5% under the European Solvency II regulation). In this context, the problem is not only to predict the most likely outcome but also to understand the tail of the distribution of the loss.
The aim of this project is to design Generative AI methods to generate scenarios of catastrophic events, taking into account the spatial dependence of the losses and the fact that the distributions are heavy tail.
Project domain: actuarial science, machine learning
Sponsor: EQUIPEX PLADIFES (Institut Louis Bachelier)