Forecasting cause-of-death mortality with single- and multi-population models in Hungary

The aim of this study is to analyze and forecast cause-specific mortality in Hungary using different models from the Lee–Carter model family. The stochastic mortality models were fitted to time series from 1970 to 2021 by sex, main cause of death and age 60 and over. We fitted single- and multi-population models that differed in their assumptions about the distribution of deaths, the number of mortality indices describing the trend in mortality change, and the age-varying coefficient(s). The use of the dynamic time warping algorithm to cluster the period effects of the single-population models and to build multi-population models on this basis is a new approach. The clustering of mortality indices by cause of death appeared to be different for men and women, reflecting the different lifestyles of the sexes. For each model and each main cause of death, we made a projection by sex up to 2050. The reduced Plat model with three mortality indices was found to be the best predictive model, and its multi-population version also performed well for some causes of death.