Climate prediction markets and manipulation

Prediction markets for climate-related risks could be a prime target for manipulators wishing to create the impression that climate change is either less or more of a problem than the scientific consensus. How susceptible could a climate prediction market be to this type of activity?

It is possible that interested parties would attempt to distort prediction market forecasts.

Long-range prediction markets for climate change could be a useful tool for policy makers by aggregating expertise and modelling from different disciplines to produce probability forecasts of climate-related risks. Policy makers could use these forecasts to guide their decisions aimed at reducing emissions of greenhouse gases and adaptation. However, given the political and economic ramifications of such decisions it is possible that interested parties would attempt to distort prediction market forecasts. For example, fossil-fuel companies might want to underplay the risk of climate change to slow down the decarbonization of the economy. They might try to manipulate market prices by buying outcomes they want to appear more probable. They might lose money in the prediction market by doing this but consider it a price worth paying to steer policy makers towards policies that are good for their profits.

If prediction markets are to be used to inform major questions, such as climate change, it is vital that we understand how easy they are to manipulate and what the consequences of manipulation might be for decision making based on the market-generated forecasts. A study by Lawrence Choo , Todd Kaplan and Ro'i Zultan looks at this question in a laboratory setting. In their experiments, four participants took the role of “managers” — or policy makers — who had to vote on policy. Meanwhile, eight other participants traded in a prediction market to predict whether the world was in one of three states: X, Y, and Z. None of these traders knew the correct state but each was told one of the incorrect states so that collectively they had enough information to identify the correct state. The managers could observe the prices of three contracts that would pay out to traders if X, Y, or Z was the correct state respectively. Based on these observations the managers voted on a policy: X, Y, Z, or a "status quo" option. In some of the experiments two randomly chosen traders were given the role of “manipulators” who would be rewarded if the managers voted for a specified incorrect “fake” state, but lose rewards otherwise. ?The other traders, and the managers themselves, obtained rewards when the managers voted for the policy corresponding to the correct state but would lose rewards if either of the other states was the winner. If they voted for the status quo option they received a modest but certain reward, instead of risking the loss associated with getting it wrong. Some experiments were run with manipulators and some without and, furthermore, in some cases everyone (traders and managers) knew whether manipulators were present whereas in others they were only told there might be manipulators. In the latter case only the manipulators, if present, knew for sure.

Even a suspicion of manipulation increased the volatility of prices.

The experiments found that when there are no manipulators, and everyone involved knows this, the prediction market correctly identified the true state of the world. Even a suspicion of manipulation, however, increased the volatility of prices although, if the suspicion was unfounded, the market could still identify the true state. Unfortunately, if manipulators were present, it was harder for the market to get the right answer, particularly if traders didn’t know for sure about the manipulators.

But what effect did the possibility of manipulation have on the votes of the managers? In the absence of manipulation, managers overwhelmingly voted for the correct state, although the number opting for the status quo increased when they were told that manipulation was possible. When manipulators took part in trading only around half of managers picked the correct state. Although the correct option was selected more often than any other option, the fraction of managers who picked the state implied by market prices fell from 90 per cent to just under 60 per cent when markets were manipulated. These experiments demonstrate how even a suspicion of market manipulation can erode the usefulness of a prediction market for decision making by undermining trust, beyond the distorting effects manipulation may have on prices.

Regulating participation in a market might increase its usefulness.

The Choo et al. study has important implications for a climate prediction market, which could be a target of manipulators seeking to influence policy. The authors suggest that regulating participation in a market might increase its usefulness, not just by excluding potential manipulators, but also by reassuring policy makers that they had been excluded.

The prediction markets for climate-related risks run by CRUCIAL have restricted participation. The main reason for this is because they are subsidised and expert participants do not have to pay to take part. However, the Choo et al. study suggests that this restriction may increase the trust that decision makers have in the forecasts produced by the markets. Choo et al. also point out that, while there is research comparing prediction markets with other approaches to information aggregation, this tends to focus on the accuracy of different methods; there is little research on the perceived trustworthiness of different methods. When prediction markets are suggested as a way to forecast climate change people often raise the issue of manipulation, yet the vulnerability to manipulation of other methods of forming a scientific consensus is often overlooked.

Header image scales by FOX52. CC BY-SA 4.0