Using Mixture Distributions in Collaborative Forecasting of COVID-19

Mon, 01 Jan 0001 00:00:00 +0000

The COVID-19 Forecast Hub was established in 2020 as a collaborative effort to forecast the coronavirus pandemic. Dozens of teams have been involved, each submitting weekly probabilistic forecasts of the spread of the virus. From the weekly forecasts, a single ensemble forecast is constructed and shared with the CDC for use in public reports. In order to assess the various forecasts against one another and to construct from them an ensemble forecast, each submitted forecast must share a common format. Each forecast is submitted as a set of 23 quantiles and their corresponding values.

Other collaborative forecast projects use different formats for representation such as binned probabilities used in the CDC’s annual flu forecasting competition. Other formats include parametric distributions and sample distributions. Introduction

We propose that collaborative forecast projects utilize a discrete mixture distribution format as a preferable alternative to the four mentioned above.

Download poster

Presentations | Spencer Wadsworth

Using Mixture Distributions in Collaborative Forecasting of COVID-19