Fitting and forecasting multi-population mortality models based on Hungarian regional data

Abstract

The stochastic mortality model of Lee–Carter (1992) has become extremely popular over the past decades. As a result, many extended versions, for example, multi-population models, exist nowadays. In the original model, the logarithm of the mortality rates is a linear function of a time-dependent factor called the mortality index, which can be considered a long-term trend describing the improvement in mortality. The prediction of mortality rates is based on forecasting thislong-term trend. Multi-population mortality models were developed to jointly model related populations. This research fit five multi-population mortality models to Hungarian regional data by sex, and compared the models’ results to each other and to the results of the original Lee–Carter model. This paper is the first to examine the possibility of applying multipopulation mortality models to Hungarian regional data. The selected mortality models are discussed in a standardized methodological framework. The author use maximum likelihood estimation to observe the models on the base period of 1970–2021. Regarding the number of deaths, a Poisson distribution is assumed. As a result, the age-specific mortality rates are forecasted until 2050, and the future life expectancies of the regions of Hungary are estimated by sex. The author consider the presence of the COVID-19 pandemic during fitting and predicting the models similarly to Lee and Carter, who treated the effect of the Spanish flu in their study.