Forecasting Covid-19 in the US

This repository is set up to contribute to the Covid-19 Forecast Hub, which is hosted by Reich Lab. See the docs/SMG.md for technical submission details, or see our fork of the project.

Models

We use an ensemble of models to forecast deaths from Covid-19 in US states. These include:

• Rt estimation (EpiNow2): models/rt
• Deaths as a convolution of cases (EpiNow2): models/deaths-conv-cases
• A mean ensemble of time series approaches: models/timeseries

These models are then ensembled using quantile regression averaging over a range of horizons and training windows. The final submission is then selected using proper scoring rules from scoringutils.

We output 1 to 4 week ahead forecasts of weekly cumulative and incident deaths, for the US and selected US states. See the project board for ongoing work.

For national and sub-national Rt estimates and forecasts in other countries, see our website.

Contributors

LSHTM, Centre for Mathematical Modelling of Infectious Disease:

• Sam Abbott (@seabbs)
• Kath Sherratt (@kathsherratt)
• Nikos Bosse (@nikosbosse)
• Sophie Meakin (@sophiemeakin)
• Sebastian Funk (@sbfnk) and Epiforecasts team