Evaluations of Covid-19 Forecasts

I have discontinued active forecasts related to Covid-19 as of July 2022, but I continue to try to learn lessons from evaluating the forecasts that were submitted.

An April 2022 paper in the Proceedings of the National Academy of Sciences found CovidComplete to be the fourth most accurate individual forecast model.

As time permits, I want to continue publishing longer-term evaluations of the forecasts, focusing on the 2021 calendar year, that provide insights into forecasting in general.

For now, you can find both summary coronavirus forecast evaluations and detailed evaluations of forecast accuracy, including assessments of national and state forecasts, as well as weekly updates on the accuracy of the pandemic forecasts submitted to the CDC. For a different high level summary, see the forecast scorecards page.

To see all the accuracy evaluations grouped by chart type, see the deep dive evaluation page. For the most detailed coronavirus accuracy information, check out the raw forecast evaluation data on my github site.

Long-term national pandemic forecast accuracy

These multi-period evaluations show which forecast models have been consistently reliable. They provide guidance about which models are most likely to be accurate for future forecasts. No model has consistently been the best, however, over the long term, any of the models that have averaged 10% error or less have been both accurate enough and consistent enough to provide reliable guidance for national level forecasts. Only models currently creating forecasts are included. For week-by-week forecast performance of individual models, see my Covid-19 national forecast model detail page.

Long term

Most recent 12 weeks

Most recent 4 weeks