Analysis of Finnish and Swedish mortality data with stochastic mortality models

Abstract

I have compared quantitatively nine stochastic models explaining the improvements in male mortality rates over ages 20–99 in Finland and Sweden. The study was divided in two main parts. In the first part I analysed the fit to historical data starting with the Bayes information criterion and the comparison of nested models. Subsequently, residual series that are predicted to be iid standard normal were assessed by different tests. In the second part I looked at the forecast of the nine stochastic mortality models and the plausibility of the forecasts. An ex-post evaluation of the forecast completed the study. For this I examined the consistency of the forecasts for a fixed future year and the accuracy of 20-years-forecasts against the realised mortality rates. For the considered dataset, there were some notable differences amongst the models, but none of the models performed well in all tests and no model clearly dominated the others. For insurance-related applications the model of Plat (Insur Math Econ 45(3):393–104, 2009) would be an appropriate choice.