City Research Online

Abstract

This dissertation addresses the challenge of jointly forecasting mortality rates across diverse populations and subgroups using a range of approaches, from traditional stochastic models to artificial neural networks. The first chapter introduces several model-averaging approaches to jointly forecast mortality rates for male and female populations within a single country, demonstrating the benefits of combining different models to improve forecasting accuracy and interval estimation. In the second chapter, the focus shifts to forecasting mortality rates for multiple populations, including both gender and country as input variables, through the application of multi-task neural networks. This approach allows for shared learning across different population groups while accounting for country specific dynamics. The third chapter explores mortality rate predictions by cause of death and socio-economic class and their interaction using feedforward neural networks of both single-task and multi-task types. The performance of these neural networks is compared to that of existing stochastic models and machine learning approaches. Together, the chapters highlight the potential of joint modelling of mortality rates to achieve higher forecasting accuracy while pointing out their limitations and the necessity, especially for
artificial neural networks, for further research and improvement.

Publication Type:	Thesis (Doctoral)
Subjects:	H Social Sciences > HA Statistics H Social Sciences > HD Industries. Land use. Labor > HD61 Risk Management Q Science > QA Mathematics
Departments:	Bayes Business School > Bayes Business School Doctoral Theses Bayes Business School > Faculty of Actuarial Science & Insurance Doctoral Theses

Export

Downloads