Multi-population mortality model in a state-space framework

Abstract

Recent developments in several countries have brought forth two critical issues which are unexpected increases in life expectancy and gap in life expectancy between populations. These two issues have led to the study of mortality rates modelling. It began with the single-population regression model which comprises of the single age and time effect parameters. The single factor is limited by the fact that, firstly, the parameter is unable to capture the variances across ages and therefore tend to produce inaccurate forecasts. Secondly, the indicated model does not incorporate the inter- relationship between populations, therefore fails to produce coherent forecast in the long run. Finally, there are two-steps of separate estimations involved in the modelling procedure which are estimation procedure for Poisson regression approach and estimation procedure for forecast model. To overcome these limitations, this study proposed an extended version of multi-population mortality model. This extended model takes into account the quadratic age-effect parameter which captures the relevant information across age-groups and will help to ensure more coherent way of forecast in the future. In addition, the proposed multi-population mortality model is reformulated into the state-space framework by combining the Poisson regression and forecasting procedures into a single estimate. The mortality model in a unified estimation is proposed to avoid unreliable forecasts. This study produces a cohort mortality life table based on the assumptions of future mortality improvements. The proposed methods were applied to the Malaysian mortality data for male and female. The proposed methods are then compared with the other existing methods. The results have shown that it is vital to incorporate accurate information across ages in order to obtain accurate mortality rates forecast. Other than that, this thesis has shown that the mortality model in a state-space framework could significantly improve the independent estimation method.