This study introduces a novel Generalized Additive Mixed Model (GAMM) for mortality modelling, employing mortality covariates $k_t$ and $k_{ct}$ as proposed by Dastranj-Kolář (DK-LME). The GAMM effectively predicts age-specific death rates (ASDRs) in both single and multi-population contexts. Empirical evaluations using data from the Human Mortality Database (HMD) demonstrate the model’s exceptional performance in accurately capturing observed mortality rates. In the DK-LME model, the relationship between log ASDRs, and $k_t$ did not provide a perfect fit. Our study shows that the GAMM addresses this limitation. Additionally, as discussed in the DK-LME model, ASDRs represent longitudinal data. The GAMM offers a suitable alternative to the DK-LME model for modelling and forecasting mortality rates. We will compare the forecast accuracy of the GAMM with both the DK-LME and Li-Lee models in multi-population scenarios, as well as with the Lee-Carter models in single population scenarios. Comparative analyses highlight the GAMM’s superior sample fitting and out-of-sample forecasting performance, positioning it as a promising tool for mortality modelling and forecasting.
Keywords: Longitudinal data analysis, Nonparametric modelling, Smoothing functions, Restricted maximum likelihood, Random walks with drift.
For more details, refer to the related paper: Generalized Additive Forecasting Mortality:
https://doi.org/10.48550/arXiv.2311.18698
https://rezadastranj.github.io/LME-ASDRs
https://doi.org/10.48550/arXiv.2311.18668