Arbeitsbereich

Bevölkerungsdynamik und Nachhaltiges Wohlbefinden

Auf einen Blick Projekte Publikationen Team

Projekt

Novel Approaches to Forecasting Population-Level Mortality

Ugofilippo Basellini; in Zusammenarbeit mit Carlo Giovanni Camarda (French National Institute for Demographic Studies, Paris, Frankreich)

Ausführliche Beschreibung

Mortality modeling and forecasting play central roles in demographic and actuarial analyses. This area of research has recently attracted renewed and increased attention in response to two pressing challenges faced by modern societies: population aging and longevity risk. Virtually every country is experiencing increases in population aging as a result of continuous declines in mortality and fertility. Furthermore, unanticipated improvements in longevity have generated an enormous global longevity risk market.

Over the last few decades, there have been significant advances in mortality forecasting, including a shift from deterministic to stochastic approaches. However, more than 30 years after its introduction, the Lee-Carter method – in its original or modified version – is still the benchmark approach for forecasting mortality (Basellini et al. 2023). Lee-Carter types of mortality forecasts have repeatedly failed to anticipate the sustained rate of mortality improvements observed in many low-mortality countries before the COVID-19 pandemic. At the same time, these models appear to be too rigid to capture the post-COVID-19 life expectancy stagnation observed in a few high-income countries. Hence, there is a pressing need for innovative models that can predict the future course of mortality more accurately than previous approaches could.

This project aims to introduce innovative statistical methods that can improve the interpretability, robustness, and accuracy of population-level mortality forecasts. On the one hand, a key measure used in this project is the age-at-death distribution, which is a very useful measure of age-specific mortality that has so far received limited attention in the context of mortality forecasting. As well as offering a promising alternative approach for forecasting mortality, modeling age-at-death distributions enhances our understanding of mortality changes by providing an alternative, yet still informative, perspective on mortality changes. On the other hand, the project builds on the P-splines framework pioneered by Eilers and Marx (1996) as a powerful data-driven tool for modeling mortality within a nonparametric smoothing framework using a penalized generalized linear model approach in which forecasting is achieved as a natural consequence of the smoothing process.

Three additional focus areas characterize this project. First, while the majority of mortality forecasting approaches focus on the period perspective, this project also develops methodologies for cohort mortality forecasts, i.e., for completing the mortality experience of real birth cohorts. From a demographic perspective, this is relevant because cohort measures of mortality are not affected by “tempo effects,” which are distortions of period measures caused by not considering the mortality history of the population. Second, we propose and compare three approaches for the modeling and short-term forecasting of seasonal mortality based on the framework of the Serfling-Poisson model. Based on multiple out-of-sample validation exercises, the Serfling-Poisson model that includes a smooth trend and parametric seasonal effects was found to have higher forecast accuracy than other Serfling-Poisson model variants. Finally, the project also proposes an approach for forecasting mortality during mortality crises that is based on a combination of multiple fitting periods and scenarios for post-crisis mortality levels, exemplified with an application to São Paulo during the COVID-19 pandemic.

Beobachtete und prognostizierte Sterberaten für Schweizer Männer

Actual, estimated and forecast death rates by Three-Component smooth Lee-Carter model over age (left) and time (right) on a log scale for Swiss males. Unlike in the original Lee-Carter method, mortality forecasts over ages are smooth, and rates of mortality improvement over time are not constant. © Camarda and Basellini (2021)

Publikationen

Basellini, U.; Camarda, C. G.:

Forecasting cohort mortality: Lee–Carter methods and CCP-splines

SocArXiv papers. unpublished. (2025)    

Leger, A.-E.; Rizzi, S.; Basellini, U.:

Modelling and short-term forecasting of seasonal mortality

arXiv e-prints 2502.10787. unpublished. (2025)    

Miranda, M. L.; Turra, C. M.; Basellini, U.:

Forecasting life expectancy in São Paulo City, Brazil, amidst the COVID-19 pandemic

Population Health Metrics 23:36, 1–11. (2025)    

Li, Z. R.; Wu, Z.; Chen, I.; Clark, S. J.:

Bayesian nested latent class models for cause-of-death assignment using verbal autopsies across multiple domains

Annals of Applied Statistics 18:2, 1137–1159. (2024)    

Basellini, U.; Camarda, C. G.; Booth, H.:

Thirty years on: a review of the Lee-Carter method for forecasting mortality

International Journal of Forecasting 39:3, 1033–1049. (2023)    

Basellini, U.; Camarda, C. G.:

A three-component approach to model and forecast age-at-death distributions

In: Developments in demographic forecasting, 105–129. Cham: Springer International Publishing. (2020)    

Bergeron Boucher, M.-P.; Kjærgaard, S.; Pascariu, M. D.; Aburto, J. M.; Alvarez Martinez, J. A.; Basellini, U.; Rizzi, S.; Vaupel, J. W.:

Alternative forecasts of Danish life expectancy

In: Developments in demographic forecasting, 131–151. Cham: Springer International Publishing. (2020)    

Pascariu, M. D.; Basellini, U.; Aburto, J. M.; Canudas-Romo, V.:

The linear link: deriving age-specific death rates from life expectancy

Risks 8:4, 109.1–109.18. (2020)