- View all blogs
- Prediction markets are not a free lunch
Prediction markets are not a free lunch
Mark Roulston, Senior Data Scientist •
Participants in prediction markets should be a source of information, not money
Prediction markets are markets whose primary purpose is the discovery and aggregation of information. Participants can buy and sell contracts that pay out a pre-determined amount if a specified outcome occurs. The purpose of other types of financial markets is the transfer of ownership or risk and information discovery is just a useful side-effect, but for prediction markets it is the reason for their existence.
Markets can be categorized as positive-sum, negative-sum or zero-sum depending on whether, in aggregate, participants make money, lose money or money is conserved. Although many existing prediction markets are zero or negative-sum these arrangements are unlikely to be optimal for information discovery.
In a zero-sum market losses balance gains
Conventional futures markets are an example of a zero-sum market. If a participant sells a futures contract to deliver oil at $50 a barrel next month and the spot price next month turns out to be $60 then they lose $10, but the person on the other side of the trade gains that $10. Some traders in futures markets enter into trades because they think they know more than others but many enter into trades to hedge risk. For example, farmers are happy to lock in a price for their crops that is below the expected price because doing so removes exposure to potentially catastrophic price drops. This is analogous to buying home insurance: I know my premium is more than the expected payout of the policy (the probability of loss multiplied by the loss) but I am happy to pay because, if the loss did happen, it would be crippling. In both cases transactions occur because of the differing risk preferences of the parties involved.
Gambling is a negative-sum activity: the house always wins
Traditional gambling is a negative-sum activity, colloquially expressed as, “The house always wins”. Bookmakers strive to make odds in their favour so that, on average, punters lose money. Betting exchanges, in which participants can both place and lay bets, are also negative-sum thanks to the commissions the exchanges charge. The negative-sum nature of gambling raises the question: why do people gamble? The answer, at least in part, is that the gaming industry encourages customers’ irrationality and benefits from them overestimating the probability of outcomes occurring.
Wealth creation means the stock market is positive-sum
The stock market is positive-sum. Every investor can’t “beat the market” — because this is a relative benchmark — but every investor can, in principle, make a positive absolute return thanks to the wealth created by listed companies. Investment brokerages encourage clients to participate in the stock market in the belief that the clients can make money. Some clients may lose money but this is not a requirement of the business model of reputable investment firms. In contrast, “boiler room” firms persuade people to buy shares knowing that they will lose money because the firm earns its profits from the losses of these victims.
Prediction markets can be negative, zero or positive-sum
In a zero-sum (or negative-sum) prediction market the gains of traders who provide good information must come from the losses of other traders. The viability of such a market depends upon the recruitment of participants who will lose money. It might not be apparent who the losers will be in advance but, since collectively participants can’t make a positive return, the existence of losers is integral to the design of the market.
In the same way that a boiler room must attract marks to provide its profits, zero-sum prediction markets must attract people to subsidize information discovery; this is problematic for practical and ethical reasons. Attracting less informed participants to subsidize a zero-sum market will mean, implicitly or explicitly, adopting strategies used by the gaming industry: nudging people towards irrational behaviour and misestimation of probabilities. But these strategies are not desirable in a prediction market with the aim of eliciting and aggregating accurate information.
Subsidizing a prediction market is more likely to succeed than hoping information will be paid for by the uninformed and unlucky
In a positive-sum prediction market a sponsor provides a subsidy, removing the need to recruit uninformed participants. The subsidy means that it is possible for all those who take part to earn money by improving the collective forecast of the market. The sponsor is paying for information and the market can be designed to align this payment with the quality of the information. This arrangement is more likely to succeed than hoping information will be paid for by the uninformed and unlucky. For specialist topics, it might be difficult to attract uninformed participants — some domain knowledge might be required to merely understand the question being asked — but the resulting forecasts can be of much higher value than the questions addressed by prediction markets with wide appeal.
The mechanism for delivering the subsidy is important: A badly designed scheme may reward people who have made no contribution to the forecast. A subsidized market maker based on a proper scoring rule, such as the logarithmic market scoring rule proposed by Hanson, distributes the subsidy in proportion to the improvement made to the forecast by each participant.
Prediction markets have not become as widespread as their advocates hoped. Some prediction markets have established themselves as predictors of political elections but attempts to extend the concept to more general questions have been less successful. Outside of areas like sports and politics, prediction markets often struggle to attract meaningful levels of activity. To change this we should remember the adage, “There’s no such thing as a free lunch”, and not try to obtain information without providing compensation in return.