Ensemble Mortality Improvement Neural Network Model

Friday, 9 February 2024

10.30 am - 12 noon

Venue: NTU Gaia, Lecture Theatre 5, at Level 2 (ABS-02-LT5)

Chairperson: Asst Prof Wenjun Zhu

Abstract

In the context of a globally aging population and the ongoing increase in average life expectancy, longevity risk management has emerged as a key challenge for countries worldwide, making accurate prediction of long-term mortality rates crucial. Consequently, this study three sub-projects aimed at exploring the application of deep learning for providing a partial solution to deal with longevity risk management for pension funds and insurance companies. The paper focuses on utilizing neural networks to establish an interpretable model for mortality improvement. Extending the framework of Hsiao et al. (2023), this study constructs corresponding mortality neural network models for traditional mortality rate models. The advantage of these models lies in their ability to directly provide all the time, period and cohort effect parameters, eliminating the need for additional time series models to predict future long-term mortality rates. Furthermore, this study also employs ensemble modeling techniques to combine these mortality neural network models, forming an ensemble neural network model for mortality improvement, with the aim of providing more accurate long-term mortality rate predictions.

About the Speaker

Chou-Wen Wang is a professor in National Sun Yat-Sen University, Taiwan. The original area of research is on derivative pricing, derivative pricing, quantitative finance, credit risk and interest rate modeling, stochastic mortality models and mortality-linked derivatives, high dimensional asset models, portfolio performance measures, and dimension reduction in portfolio selection. I have been quite productive over the years, with many of the scholarly papers appeared in top finance and actuarial journals, including Journal of Risk and Insurance, Insurance: Mathematics and Economics, Journal of Banking and Finance, ASTIN Bulletin, Journal of Derivatives, Journal of Futures Markets and Quantitative Finance.