The Convergence and Robustness of Cohort Extensions of Mortality Models

Abstract

The prediction of future of future lifespans of society is known to be very difficult to predict. The most efficient and effective method at the moment of writing is known as the Lee Carter model. However, it has been debated that the intrinsic simplicity of the Lee Carter model and its non-usage of societal trends is not too simple for a complex process such as the average lifespan of currently living people. A strong argument and example can be found in (un)healthy habits of society like smoking and exercise. The simplicity of the Lee Carter model extended with a more abstract form of this structural information gives rise to the cohort extended mortality models. And, although the explanatory power and the underlying rationality might theoretically be valid, in practice it appears that this extension of the model makes the fitting procedure more complex and thus affects the robustness of the model negatively. A possible solution was proposed by Hunt and Villegas (2016) and applied in this paper. Although the theoretical aspect of their solution seems valid, the empirical results show that it is a step in the right direction, but not yet a perfect solution.

Mortality forecasting in Colombia from abridged life tables by sex

“…In this study, the models presented problems of convergence with the cohort effect with both R-packages for men, except the APC model. The convergence problem for mortality models with cohort effect has been pointed out by other authors such as Debón et al (2010 ), Hunt and Villegas (2015) , and Kennes (2017) . In this study, we would like to remark that the cohort effect presents problems of estimation of the parameters on abridged life tables as cohorts represent subsets of five cohorts with different numbers of observations.…”

“…In our study, convergence problems were presented for the LCC model using the gnm R-package and for the RH and M8 models with the StMoMo R-package using the data for men. The convergence problem for mortality models with cohort effect has been pointed out by other authors such as Debón et al (2010 ), Hunt and Villegas (2015) , and Kennes (2017) . On the other hand, the CBD assumes that mortality is linear on the logit scale, so it only works well for advanced ages, causing very high residuals at early ages and poor behavior in general for residuals (see Fig.…”

Ajuste y predicción de la mortalidad. Aplicación a Colombia

“…Debón, Martínez-Ruiz y Montes (2010),Hunt y Villegas (2015) yKennes (2017). Es importante señalar que el efecto de cohorte presenta problemas de estimación de los parámetros en las tablas de vida abreviadas, ya que las cohortes representan subconjuntos de cinco cohortes con diferentes números de observaciones.…”