Joint-outcome prediction markets for climate risks

predicting. This makes sense because seasonal forecasting for mid-latitudes is particularly chal- lenging, compared with the tropics where the presence of El Niño-Southern Oscillation can allow skillful forecasts to be made months in advance. Once the target month has started, how- ever, the flow of information to participants, in the form of medium-range weather forecasts and observations, increased and they injected this information into the market. September stands out due to the high number of trades—sometimes more than 1,000 in a day—although the overall volume of trades (in terms of credits) was not as exceptional. The average trade size fell to 6.7 credits in September, compared with 50.7 credits in August. Most of these September trades were made programmatically through the platform’s API. Fig 3 shows how the marginal distributions of prices for temperature and rainfall evolved throughout the duration of the markets. The April and May markets were held open until the 15th of the following month, shortly after the official numbers from the Met Office were avail- able. This was done to explore the behaviour of the markets during the period after the temper- ature and rainfall values were, in principle, knowable, but before publication of the official values for settlement. Some trading did occur in this period, but it did not affect the expected values implied by the market. Subsequent markets were closed on the last day of the month they referred to, before publication of the official numbers. The decision to close later markets

on the last day of the month was made to prevent participants being rewarded for trading on the publication of the official numbers. If the participants were rewarded proportionally to their credit accumulation such last minute trading would only be at the expense of the market maker but, because of the tournament-like structure of the rewards, participants who had made contributions to forecast accuracy earlier in the month could conceivably lose out to par- ticipants who only traded after the end of the month. Temperature markets for June and July both started trending upwards during May, antici- pating a continuation of that month’s warm weather. The August and September markets showed a less pronounced upward trend during May, presumably because of the lower corre- lation between monthly temperatures at intervals beyond a couple of months. All the markets showed convergence to the actual values of temperature and rainfall as they approached the end of the relevant month. Discussion The pilot markets clearly demonstrate the viability of large, two-dimensional outcome spaces when combined with an LMSR automated market maker. They also demonstrate that meteo- rologists and climate scientists—the types of experts who should participate in long-range cli- mate prediction markets—can quickly adapt to prediction markets as a mechanism through which to contribute their expertise. Neither the prediction market institution itself, nor the complexities associated with the joint-outcome market format, posed an impediment to par-

ticipation. The complexity did highlight the advantage of providing participants with access to an API. This allowed them to integrate their own quantitative models directly into trading, facilitating the synthesis of these models, and other tacit knowledge possessed by participants, into the collective forecasts of probabilities. The results from the trial markets provide a proof-of-concept for using prediction markets to address the problems of aggregation, circularity, and accountability, in longer-range climate prediction. Joint outcome prediction markets for GHG concentrations and global tempera- tures could provide evolving estimates of both quantities as well as transient climate sensitivity. A probabilistic prediction of GHG concentrations over the next two or three decades would allow relative likelihoods to be assigned to emissions scenarios produced by organizations such as the IPCC, International Energy Agency (IEA) and the Network for Greening the PLOS ONE Joint-outcome prediction markets for climate risks PLOS ONE | https://doi.org/10.1371/journal.pone.0309164 August 30, 2024 10 / 14