Prediction markets for climate scenario analysis

Prediction markets are a potentially useful tool for climate scenario planning.

Scenario analysis

Scenario analysis is a strategic planning tool that involves identifying potential future scenarios and studying what implications they may have for an organisation. Imagining possible futures and preparing for them has been practiced by leaders throughout history but formal scenario analysis is most closely associated with the Cold War work of Herman Kahn as well as Royal Dutch Shell, who have been using it for 50 years.

Shell first incorporated climate change and sustainability into their scenario approach in the 1990s. Scenario analysis is now advocated as a key tool for companies to use when assessing their climate related risks and opportunities by the Task Force on Climate Related Financial Disclosures (TCFD) as well as by the Bank of England’s Prudential Regulation Authority.

“Since plausibility is a great virtue in a scenario, one should, subject to other considerations, try to achieve it.” — Herman Kahn and Anthony J. Wiener (1967)

Scenarios or forecasts?

A recurring claim in the field of scenario analysis is that scenarios are not predictions. However, a vital characteristic of scenarios is that they are plausible. Therefore, to include a scenario in a planning exercise is to make a judgment about its plausibility. To say that a scenario is plausible is not the same as saying it is probable but saying scenario A is more plausible than scenario B is an assertion that A is more probable than B. Inherent in such a statement is a prediction. The prediction is a probabilistic one but all predictions are probabilistic, whether this is acknowledged or not.

It is possibly because many people fail to understand the probabilistic nature of prediction that advocates of scenario analysis want to distance it from forecasting but anyone who is familiar with probability will recognise an element of prediction is implicit in scenario analysis.

Prediction markets

Prediction markets are markets specifically designed to obtain information, rather than to allocate risk or capital. They share many features with betting markets but the bettors should be viewed as a source of information rather than revenue and so the market maker can expect to lose money in return for obtaining this information. This is not how traditional bookmakers operate.

In 2018 Winton Group sponsored a prediction market in which academics in subjects such as meteorology, economics, statistics and engineering predicted monthly average temperatures and rainfall for the U.K. This year Hivemind, which spun out of Winton last June, is sponsoring a prediction market in which academics are forecasting the state of El Niño-Southern Oscillation.

These experiments demonstrate that experts in climate science and related disciplines can engage with prediction markets and inject their expert knowledge and judgment into the probabilistic predictions that emerge from them. Indeed, it is impressive how sophisticated the trading strategies of the participants become over the duration of a market.

Prediction markets could have an important role to play in the application of scenario analysis to climate-related financial risks. The TCFD divides these risks into two categories: transition and physical. Transition risks arise due to a shift away from carbon emitting technologies to mitigate climate change. This is a risk for firms that own fossil fuel assets, for example. Physical risks come from climate change itself, such as a reduction in the value of coastal property due to more frequent flooding associated with rising sea-levels.

The TCFD recommends that a starting point for climate-related scenario analysis is to use scenarios provided by organisations such as the International Energy Agency (IEA) and the Intergovernmental Panel on Climate Change (IPCC). These scenarios include “transition scenarios” and “physical climate scenarios”. While these scenarios are widely used and thus allow comparisons between organisations they are global scenarios and may not be sufficiently granular for a specific organisation. The TCFD suggests, at the minimum, including a “2°C scenario” in any analysis but determining the impact of a 2°C rise in global average temperatures on a particular company is an incredibly complex undertaking. A more tractable approach is to identify the climate variables and locations to which the firm is most sensitive and design scenarios which focus on these. A prediction market targeting the required variables can be used as a tool to elicit and aggregate expert judgments to generate scenarios that are both plausible and relevant.

Third-party scenarios which market participants deemed credible would be incorporated into the probabilistic predictions generated by a prediction market in the same way that third-party forecasts of U.K. weather and ENSO were incorporated into the Winton climate and Hivemind ENSO markets respectively. Using a prediction market has advantages over using third-party scenarios directly: It reduces the risk of placing too much emphasis on a particular scenario simply because of its availability and prediction markets can respond to new information more quickly than organisations can update their scenarios.

Participation and incentivisation

The Winton climate market and the Hivemind ENSO market both involved a limited number of experts who did not have to pay to take part. Instead they were provided with initial endowments of credits with which to bet. At the end of the market the final credit balances were used to award cash prizes: The Winton market awarded cash prizes to the participants with the most credits while the ENSO market converts credit balances to tickets in a lottery for cash prizes. This free participation approach would make little sense to a conventional bookmaker. It is viable in a prediction market because the participants are not an intended source of revenue but of information. An advantage of these types of arrangement is that they are not classed as gambling. The disadvantages are that they do not scale well and that the market organiser must decide who counts as an “expert” qualified to participate.

The most inclusive prediction market would allow anyone to participate. For such a market to be scalable participants would have to bet their own money, even if the market was subsidized. This would be classed as gambling which, under U.K. law, requires a license and regulatory compliance introducing additional costs.

One issue with using prediction markets for long-term questions is that the incentive structure needs to account for the time value of money. If the market is invitation only, with participants being provided with their initial stake, this is not a problem. In a market in which people bet their own money, however, discounting will erode the net present value of payouts making nominally underpriced contracts unattractive. This will prevent the market from properly capturing the collective judgment of participants. This problem can be mitigated by denominating the market in interest-bearing assets such as units of a fund invested in low-risk interest-bearing securities.

What type of participants would want to place bets on events years, even decades in the future? Individuals invest in pensions on these time horizons so those who believe they have informed views about climate change might be prepared to stake a portion of their wealth on these views. Institutions, such as universities, could also participate. Using government research grants to participate in a climate market would be problematic but there is nothing to stop philanthropists earmarking donations for the market. The idea that universities with climate research programmes could have part of their endowments riding on their predictions is an attractive one.

A platform for climate prediction markets

Prediction markets targeting variables of interest to climate scenario planners would be a mechanism to elicit the knowledge and judgment of relevant experts and to aggregate it into probabilistic forecasts from which plausible yet distinctive scenarios could be constructed. Even if different organisations are interested in different variables, based on their own sensitivity, there are strong arguments for hosting the markets on a common platform. A single platform would spread fixed costs, including any costs of complying with gambling regulations. A single platform is also more likely to attract a critical mass of experts, many of whom would have expertise relevant to several markets.

A centralized platform for climate prediction markets would be a useful resource for all companies engaged in scenario analysis as part of their climate-related financial risk disclosure process. There are challenges about the best way to structure such an institution. The information produced by a public market would become a public good so there is a free-rider problem. Also, given the long-range nature of the predictions, there would have to be confidence in the longevity of the platform. One possibility would be joint ownership by the companies using it but, whatever the structure, it would have to adhere to high levels of transparency and governance.