Centro de Documentacion de Fundación MAPFRE - Mortality forecasting via multi-task neural networks

Notes: Sumario: Understanding and projecting mortality dynamics has become increasingly critical for pension systems and life insurers. While traditional stochastic and deterministic models have long been used to extrapolate mortality patterns beyond observed data, recent advances in machine learning offer new modelling opportunities. In particular, neural networks (NNs) provide strong computational capabilities and the flexibility to capture complex relationships without relying on explicit probabilistic assumptions. In this study, they develop a multi-task neural network framework that jointly learns from multiple related populations, enabling shared information to enhance forecasting accuracy across countries. Using mortality data from 17 nations, they evaluate the performance of these multi-task NNs against both single-task NN architectures and established stochastic mortality models. Their findings highlight the potential of multi-task learning to improve long-term mortality projections and support actuarial applications.Related records: En: Astin bulletin. - Belgium : ASTIN and AFIR Sections of the International Actuarial Association = ISSN 0515-0361. - 12/05/2025 Volume 55 Issue 2 - may 2025 , p. 313 - 331Materia / lugar / evento: Mortalidad Tablas de mortalidad Redes neuronales artificiales Matemática del seguro Cálculo actuarial Estadística matemática Proyecciones Procesos estocásticos Otros autores: Mori, Luca De
International Actuarial Association
Other categories: 6

Rights: In Copyright (InC)

‌ See issue detail