Communication Dans Un Congrès Année : 2025

Feature and quantile selection for the actuarial climate index: Everything, everywhere, all at once (1/2)

Résumé

The actuarial climate index (ACI) has traditionally been computed as the average of standardized anomalies across several climatic variables, a method that uniformly weights features and lacks a statistically robust mechanism for feature and quantile selection. In this presentation, we challenge these limitations by introducing an advanced statistical framework that leverages Time-Penalised Trees (TpT) to dynamically select and weight the features based on their predictive power for extreme events and climate-related mortality, while also rigorously assessing the importance of the climatic variables to refine the quantile selection process. By focusing on key risk areas and integrating a more adaptable, data-driven methodology, this approach enhances the interpretability and sensitivity of the index, thereby offering insurers and policymakers a more tailored tool for assessing and managing evolving climate risks. Preliminary results indicate significant improvements in the index’s ability to capture localized risk variations and provide a nuanced understanding of climate impacts on insurance claims, paving the way for further research and broader applications in actuarial science.

VALLA_GARRIDO_Slides_DSI_London.pdf (2.7 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05127832 , version 1 (24-06-2025)

Identifiants

  • HAL Id : hal-05127832 , version 1

Citer

Mathias Valla, Jose Garrido. Feature and quantile selection for the actuarial climate index: Everything, everywhere, all at once (1/2). Insurance Data Science Conference, Bayes Business School, Jun 2025, Londres, United Kingdom. ⟨hal-05127832⟩
72 Consultations
87 Téléchargements

Partager

  • More