Can Combining Forecasting Models Better Measure Excess Mortality During a Pandemic?

What is it about?

Dynamic model selection and combination can improve excess-mortality forecasts when countries and mortality patterns differ substantially. Estimating excess mortality during a major health crisis is difficult because countries differ in population size, long-term mortality trends, seasonal patterns, and data quality. This study develops a dynamic ensemble-learning strategy for forecasting monthly respiratory-disease deaths across 61 countries. Instead of relying on one forecasting model, the method combines several time-series models, selects the best-performing forecasters, chooses an appropriate holdout period for each model, and assigns weights according to out-of-sample predictive accuracy. The results show that this flexible ensemble strategy improves forecasting accuracy. The resulting respiratory-disease death forecasts were also highly correlated with reported COVID-19 deaths in 2020, providing a useful way to assess excess mortality during the pandemic.

Featured Image

Photo by Adam Blacke on Unsplash

Photo by Adam Blacke on Unsplash

Why is it important?

Reliable estimates of excess mortality are essential for understanding the true impact of pandemics and other large-scale crises. Simple comparisons with historical averages may be misleading because mortality differs substantially across countries and follows long-term and seasonal patterns. This research shows how dynamic ensemble learning can produce more accurate mortality forecasts by combining the strengths of different time-series models rather than relying on a single forecasting technique. More accurate expected-mortality estimates can help policymakers, public-health authorities, and researchers assess the severity of crises, compare impacts across countries, and support better evidence-based decisions.

Perspectives