Associate Professor Jarad Niemi joined the COVID-19 Forecast Hub in early April. The Hub aims to collect COVID-19 forecasts in a central location with a standardized format. The Hub serves three primary purposes: 1) an ability to compare model forecasts, 2) a gatekeeper for the Centers for Disease Control and Prevention (CDC) COVID-19 forecasts, and 3) an ensemble forecast that combines all individual model forecasts into a single aggregate forecast.
The visualization of all forecasts are available on the Hub dashboard. The visualization provides forecasts for all 50 states, the District of Columbia, and a combined national forecast. The primary focus of these forecasts are COVID-19 deaths on a weekly time scale. The Hub provides the ability to “turn back the clock” and see how models performed historically compared to data from the Johns Hopkins University Center for Systems Science and Engineering.
Once per week, the Hub sends model forecasts to the CDC for inclusion on the CDC COVID-19 Forecasting site. The CDC provides a national forecast on its website as well as state-level forecasts, including US territories, as a downloadable pdf.
On both the Hub dashboard and the CDC forecasting site, an ensemble forecast is provided. This ensemble forecast combines the probabilistic forecasts for each individual model into a single probabilistic forecast. Construction of this ensemble is an interesting statistical question as we have limited history to evaluate models.
All models have underlying assumptions and by constructing a Hub we can compare how these assumptions affect forecasts.