A novel EVT-modified Lee-Carter model for mortality forecasting : An application to extreme mortality events. Issue 1 (2nd January 2022)

• Record Type: Journal Article
• Authors: Gungah, Gavakshi; Narsoo, Jason
• Is Part Of: Journal of statistics & management systems. Volume 25:Issue 1(2022)
• Page Start: 211
• Page End: 243
• Publication Date: 2022-01-02
• Subjects: 62P05; Mortality forecasting -- Age-Period-Cohort model -- Cairns-Blake-Dowd model -- Lee-Carter model -- Extreme mortality events -- Extreme value theory -- Hybrid model; Statistics -- Periodicals; Mathematical models -- Periodicals; Mathematical models; Statistics; Periodicals; 519.5
• DOI: 10.1080/09720510.2021.1878629
• Languages: English
• ISSNs: 0972-0510

Abstract

Abstract: This paper proposes a mortality model which aims to accurately forecast mortality during extreme mortality events such as war, earthquakes or pandemics. To the best of our knowledge, it is the first study to develop the EVT-modified Lee-Carter model which is an amalgamation of the classical Lee-Carter model and the Extreme Value Theory (EVT). The hybrid model aims to cater for mortality jumps which usually occur during extreme mortality events. The forecasting accuracy of the EVT-modified Lee-Carter model is investigated against that of the classical Lee-Carter model, the Age-Period-Cohort model and the Cairns-Blake-Dowd model. The mortality datasets from four European countries are considered for two modelling periods, (1) the unstable period (1900-1960) during which extreme mortality events such as World War I, World War II and the Spanish flu pandemic occurred and (2) the stable period (1961-2017) where no significant mortality jump is observed. The EVT-modified Lee-Carter model evidently outperforms the other benchmark models for the unstable period while for the stable period (1960-2017), the classical Lee-Carter model performs best. The findings highlight the adequacy of the novel EVT-modified Lee-Carter model to capture mortality jumps and leptokurtic features which may be caused by extreme mortality events. In the context of the recent Covid-19 pandemic, the novel model may be an interesting tool for life insurance companies in the valuation of insurance products, mortality derivatives and solvency requirements.