One-step mortality projection method – more than one application
Applying the method of Generalised Linear Mixed Models to the Lee-Carter or APC model smooths out the COVID-19 mortality spike, or volatility caused by a small sample size. Ad Res student Annika Ziegler presented her work at the 2022 HMD Symposium in Paris, showing smooth results even for a population as small as Iceland.
In a paper presented at the 2021 Longevity 16 Conference in Copenhagen, Prof. Rui Zhou and her co-author Prof. Johnny Li proposed a one-step mortality projection method, which combines fitting a Lee-Carter model with the simultaneous estimation of a random walk process for the resulting mortality index. This method makes use of the theory of Generalised Linear Mixed Models. In their work, Zhou and Li intended use the GLMM one-step method to remove the COVID-19 spike from the mortality data series in a controlled way, before adding back a parametric mortality spike based on expert opinion.
Annika took up the idea of the one-step mortality model fitting and applied it to the APC model as well as testing different time series processes. For large populations, there is little difference between the one-step method and the traditional two-step approach of first fitting the mortality model and then a time-series model, except of course of the largest spikes. Applying the one-step method to smaller and smaller populations sizes, Annika found that it may also lend itself to modelling the future mortality of small populations, such as regions within a country, or socio-economic classes.
