Work

Accounting for temporal and spatial dependencies in multi-population forecasts: the Transformers approach

Year: 2025

Type: preprint

Abstract: Since the advent of machine learning, there have been significant advances in mortality modelling. The introduction of the Transformer model Wang et al. (2024), which is highly effective in processing long sequences, has notably enhanced the precision of mortality forecasts. Our work aims to extend the Transformer architecture by incorporating structural dependencies between countries. This involves constructing a similarity matrix that combines death data and exogenous variables impacting mortality, using the method proposed in Gouthon and Milhaud (2025). This enables the proposed model to simultaneously capture the long-term temporal dependencies specific to each country and the structural correlations between them. This new approach is highlighted as performing well in a comprehensive comparative study.

Source: HAL

Authors: Gouthon, Jean-Luc, Milhaud, Xavier, Garrido, José

Institutions: Aix-Marseille Université, Institut de Mathématiques de Marseille, Concordia University

Language: English

Topic: Insurance, Mortality, Demography, Risk Management

Subfield: Demography

Field: Social Sciences

Domain: Social Sciences

SDG: Good health and well-being

Open Access status: green

OpenAlex ID: W7117196366

Cites: 23

Cited by:

Related to: 10