Ex-Post Evaluation of COVID-19 Mortality Forecast Models

Ricarda Duerst , Max Planck Institute for Demographic Research
Christina Bohk-Ewald, Max Planck Institute for Demographic Research

The mortality burden of COVID-19 has often been forecasted since the beginning of the coronavirus pandemic in early 2020. Such mortality forecasts are in high demand, as they are crucial to inform public health policies and to decide on effective pandemic regulations. However, it is unclear how suitable these models are to accurately forecast COVID-19-related mortality at different stages of the pandemic: e.g., at the beginning, the peak, and the end of subsequent waves of infections. That is why we systematically validate four popular models that are widely adopted to forecast daily deaths attributable to COVID-19 in countries worldwide. Overall, we find large differences in the forecast performance with respect to the pandemic phase, the forecast length, and the world region, within and between models. It is particularly challenging to forecast mortality at turning points such as the peak of an infection wave.

See extended abstract

 Presented in Session 78. Improvements in forecasting methods