A prediction market for hurricane activity: call for participants

Mark Roulston, Senior Data Scientist •

Blog 7


Hivemind has been invited to be part of the fifth cohort of Lloyd’s Lab (the home of InsurTech at the world’s most famous insurance and reinsurance market). Throughout this 10-week programme, we will be investigating the use of prediction markets—using AGORA, our prediction market tool—as a method for aggregating expert judgments into consensus forecasts relevant to the insurance industry.

One of the experimental prediction markets we are running will be to predict the number of hurricanes occurring during the remainder of the 2020 Atlantic hurricane season, and also the number of these hurricanes that will make landfall in the United States. The Atlantic hurricane season is traditionally defined as 1st June to 30th November. Historically, the number of hurricanes occurring during a season has ranged from zero in 1907 and in 1914, to a record 15 during the 2005 season (although the last of these actually only attained hurricane status “out-of-season” on 2nd December). How active a hurricane season is likely to be is influenced by factors such as sea surface temperatures and vertical wind shear. Several models exist for forecasting the seasonal level of activity.

Our prediction market will be denominated in virtual on-platform credits. Participants will be given initial endowments of credits with which they will be able to trade contracts for specific numbers of hurricanes. After the final numbers are known, at the end of the season (30th November), the market will be settled. Participants who accumulate the most on-platform credits will be eligible for performance rewards totalling £5,000, courtesy of Lloyd’s Lab.

The hurricane prediction market is invitation-only and we are now seeking 20 UK residents with a diverse range of expertise in either meteorology, climatology, general data science, statistics or machine learning to participate.