note: cross-posted from LessWrong
Next Monday (May 24th 2021) is the official begin of the UK Covid-19 Crowd Forecasting Challenge. Over the course of 12 weeks, human forecasters will be asked for their prediction of Covid-19 case and death numbers. The (aggregated) forecasts will then be submitted to the European Forecast Hub and compared to models submitted by other research institutions.
How to sign up
In order to sign up, checkout this post that explains everything in detail. You can also go directly to app.crowdforecastr.org or rt-app.crowdforecastr.org where you can directly make a forecast. Forecasts can be made every week from Sunday 10am to Monday 10pm UK time.
Why is this interesting
This challenge is part of an ongoing research project by the Centre for Mathematical Modelling of Infectious Diseases at the London School of Hygiene and Tropical Medicine in collaboration with Imperial College London and Public Health England.
At the moment, the UK is at a critical juncture: on the one hand we see a very successful vaccination campaign. On the other hand, new variants are emerging and spreading and easing restrictions is not without risk. Many of these things are hard to model and hard to predict. The next weeks may therefore hold important lessons for other countries going out of lockdowns.
In the past we submitted forecasts to the European Forecast Hub as well as the German and Polish Forecast Hub - with very promising results. The crowd forecasts usually were among the top performers, beating most other computer-based prediction models. Another thing that we found is that humans had a comparative advantage when it came to forecasting cases. Case numbers are extremely hard to model and depend on chance as well as shifting policies. Deaths, on the other hand, are a lagged quantity that depend on previous cases through some complicated mechanism. Models often were slightly better at capturing this.
The main goal of the Crowd Forecast Challenge is to
- learn more about how well humans (as opposed to computer models) are able to foresee these future developments
- provide valuable insight to policy makers through the European Forecast Hub
- Investigate an exciting new way of making forecasts. In addition to making a direct forecasts, participants can forecast the trajectory of Rt (i.e. the average number of people each infected person will infect in turn). The trajectory of Rt will then be used to infer future case and death numbers (This is done using the so-called 'renewal equation', where future cases are modeled as a weighted sum of past and present cases times Rt, as implemented in the R package EpiNow2)
Why this could be interesting for you
Of course there is a prize, but the interesting thing really is the exercise in forecasting. We strive to give forecasters detailed feedback about their performance and calibration (you can see some current evaluations here). Being able to make well calibrated predictions is an important skill that is tremendously helpful in a lot of situations. Quantifying one's own uncertainty is also surprisingly hard as anyone can testify who has tried some serious forecasting.
This is also an opportunity for your predictions to make a meaningful contribution - both in terms of the research we do and in terms of the potential impact it can have on policy makers.
We'd be very happy about your participation in the UK Covid-19 Crowd Forecast Challenge. Also please tell your friends and colleagues who might be interested!