We're hiring!

AGORA use case: Predicting COVID-19

Mark Roulston, Senior Data Scientist •

IMG Coronavirus

Problem

The COVID-19 pandemic began early in 2020 as the SARS-Cov-2 virus, that causes the disease, spread around the world. The United States was particularly hard hit and by July was reporting over 70,000 new cases every day. Case counts began to fall at the end of the summer but it was uncertain whether they would keep falling or whether there would be a resurgence. The evolution of the pandemic had major implications for US public health as well as the economy.

Approach

As part of its participation in the Lloyd’s Lab InsurTech accelerator Hivemind used AGORA to produce expert consensus forecasts of how many new cases of COVID-19 would be reported in the US. Experts with relevant knowledge were invited to participate in a prediction market on AGORA to predict the total number of new monthly cases in the US during each of the last four months of 2020. With on-platform credits they were able to buy and sell contracts corresponding to different numbers of cases.

Outcome

Just over 20 experts with backgrounds in statistics, data science, machine learning and epidemiology participated in the market. During the course of the market they made over 1,000 trades. From their trading activity, AGORA generated implied probability distributions for the numbers of new monthly US COVID-19 cases that evolved as participants updated their positions in response to new information. The AGORA predictions produced better estimates of tail risk than the consensus forecast produced by the COVID-19 ForecastHub: AGORA predicted a 5.8% probability that October cases would be as high as their actual value of 1.89 million, while the ForecastHub ensemble indicated almost no chance of this happening.