A Mortality Model for Pandemics and Other Contagion Events

Abstract

AbstractThe crisis caused by COVID-19 has had various impacts on the mortality of different sexes, age groups, ethnic and socio-economic backgrounds and requires improved mortality models. Here a very simple model extension is proposed: add a proportional jump to mortality rates that is a constant percent increase across the ages and cohorts but which varies by year. Thus all groups are affected, but the higher-mortality groups get the biggest increases in number dying. Every year gets a jump factor, but these can be vanishingly small for the normal years. Statistical analysis reveals that even before considering pandemic effects, mortality models are often missing systemic risk elements which could capture unusual or even extreme population events. Adding a provision for annual jumps, stochastically dispersed enough to include both tiny and pandemic risks, improves the results and incorporates the systemic risk in projection distributions. Here the mortality curves across the age, cohort, and time parameters are fitted using regularised smoothing splines, and cross-validation criteria are used for fit quality. In this way, we get more parsimonious models with better predictive properties. Performance of the proposed model is compared to standard mortality models existing in the literature.