Chapters from modeling mortality in Hungary [védés előtt]

Abstract

This thesis attempts to stochastically model and forecast mortality in Hungary, and aims to support the development of the population projection model of the Hungarian Central Statistical Office. From the Lee–Carter model family, we have selected both those used to forecast mortality in a single population, and those used to study the mortality of different populations together. We focused on four topics, each of which can improve the modeling of mortality in Hungary. These four topics are: 1) regional mortality projection, 2) prediction of mortality by cause of death, 3) investigation of the time-dependence of age-varying parameters, i.e., rotation and their incorporation into the prediction of some mortality models, and 4) application of tree-based machine learning methods to mortality projections. For these research topics, we fitted different but related mortality models, always comparing their results and predictive accuracy, and selecting the best performing model. We used the Lee–Carter model as the main benchmark for all our research topics. The topics are also linked in that we have always fitted multi-population models. We used long-term time series to forecast Hungarian mortality rates by region, main cause of death and sex up to the year 2050.