What Are Prediction Markets, and Why Might They Be Useful?
The Best Argument Against Prediction Markets, §1: Introduction
This is the first section of a larger post, that will eventually present a critique of prediction markets. Here, I lay the groundwork by setting out how they work and explaining the case for their usefulness.
A prediction market is, at its core, a mechanism for turning private beliefs about uncertain events into publicly visible prices. In the simplest design, we have a security that will pay £1 to its owner if some event occurs, and £0 otherwise — for instance, “a Democrat will win the next U.S. presidential election”. If this security currently trades for £0.54, then the implied probability of the event is 54%1. The event resolves when an adjudicator or dispute mechanism confirms whether the specified criteria were met; at that point, all outstanding contracts pay out, and the market closes.
As in financial markets, traders who believe the value of the security is higher than the current market price (or equivalently, that the probability is higher than 54%) will buy, since the contract offers positive expected value for them. For instance, if one thinks the “true probability” is 70%, the expected value of the security is £0.70, and so buying at £0.54 yields an expected profit of £0.16 per contract2. Conversely, traders who think the probability is lower than 54% will sell short. In effect, traders who are more right than the price make money; traders who are less right than the price lose money.
These profit-seeking trades push the price towards a consensus forecast. Buyers increase demand and push the price (and implied probability) up; sellers increase supply and push it down. The process continues until the market reaches a point where, at the margin, buyers and sellers are indifferent: the price reflects the crowd’s risk-neutral expected value of the payoff. Under familiar conditions — adequate liquidity, low transaction cost, and a sufficiently large pool of informed traders — that marginal price is a reliable forecast.
Of course, the kind of “binary option” prediction market I’ve described isn’t the only possible contract type. Markets can be multinomial, with mutually exclusive outcomes whose prices sum to one (“which candidate will win the Republican nomination?”); continuous, that tracks an ongoing variable over time instead of resolving at a single future moment (“what will quarterly inflation be?”); range or time-to-event contracts, which slice a continuum into bands or dates; and, most usefully for decisions, conditional and combinatorial contracts (“if policy P is implemented, will Y occur?”).
The distinctive feature of prediction markets is simply that the underlying asset is not a stock or bond but a claim on a particular outcome. Just as the price of Apple shares reflects the market’s collective judgement about the company’s future profits, the price of an election security reflects the market’s collective judgement about the chances of a given result. The key insight is that traders have incentives to act on their private information — whether that information is explicit (a new poll, an inside view of how a policy will play out) or tacit (a hunch grounded in domain expertise). Prices aggregate all of this information into a single, continuously updated number.
The cleanest intuition pump3 for why aggregating many small, noisy judgements can yield reliable forecasts — the principle underlying prediction markets — is the example of Francis Galton4 observing a contest in which villagers were invited to guess the weight of an ox. The median estimate turned out to be extremely close to the true weight, and it is the opening example of James Surowiecki’s The Wisdom of Crowds. As Philip Tetlock recounts and explains in Superforecasting:
When Galton watched people guessing the weight of the doomed ox, he was watching them translate whatever information they had into a number. When a butcher looked at the ox, he contributed the information he possessed thanks to years of training and experience. When a man who regularly bought meat at the butcher’s store made his guess, he added a little more. A third person, who remembered how much the ox weighed at last year’s fair, did the same. And so it went. Hundreds of people added valid information, creating a collective pool far greater than any one of them possessed. Of course they also contributed myths and mistakes, creating a pool of misleading clues as big as the pool of useful clues. But there was an important difference between the two pools. All the valid information pointed in one direction — toward 1,198 pounds — but the errors had different sources and pointed in different directions. Some suggested the correct answer was higher, some lower. So they canceled each other out. With valid information piling up and errors nullifying themselves, the net result was an astonishingly accurate estimate.
Prediction markets are designed to harness this dynamic, and improve on it in several respects. They align incentives, so that unlike surveys or pundit panels, traders are rewarded for being right and moving the price towards reality, and penalised for adding noise. They self-correct at the margin, because any mispricing attracts those who can profit by correcting it: when the price drifts from reality, the next trader to act is disproportionately likely to be someone who recognises the error, and their trade pushes the market back towards truth. And cross-market arbitrage squeezes out inconsistency, ensuring for instance that the prices of mutually exclusive events sum to one and that conditional probabilities respect Bayes’ rule, indirectly improving each component forecast. The result is a system that, like Galton’s ox-weighing crowd, aggregates dispersed fragments of knowledge while letting errors cancel out, but with built-in mechanisms that strengthen the signal and discipline the noise.
A concrete example makes this vivid. Consider a binary market: “Will Policy X pass by 31 December?” It opens at 50%. A civil-service insider, seeing a whip count, believes 75% and buys shares, pushing the price to 68%. Journalists report new defections; some traders update down to 60% and sell. A legal analyst notes an obscure procedural hurdle, sells, and price moves to 56%. Over time, as evidence accrues through news, leaks, and analysis, the price traces a path that — ideally — looks like a sequence of Bayesian updates on the true state of the world. At resolution, holders are paid £1 or £0; ex-post, the market’s entire time-series is a transparent, monetised, continuously-updated forecast.
So, that’s the idea. Prediction markets give us a way to harness the wisdom of crowds, and turn it into useful forecasts for decision-making. But I know from past conversations that you’re probably not convinced by this; indeed, you probably shouldn’t be, just yet. You’ve likely identified at least some flaws in this story.
Good: let’s turn now to rebutting some common objections.5
At time of writing, this is in fact the live forecast on Polymarket, one of the largest existing prediction market platforms.
Of course, the realised payoff will be either +£0.46 or -£0.54. The £0.16 is the ex ante expectation the trader forms, given her subjective probability of 70%.
If you’re interested in a more rigorous justification of these ideas rather than just the intuition, I can strongly recommend Hayek’s 1945 article The Use of Knowledge in Society.
Unfortunately, per Wikipedia, “In recent years, he has received significant criticism for being a proponent of social Darwinism, eugenics, and biological racism; indeed he was a pioneer of eugenics, coining the term itself in 1883.“ Fortunately you’re all high decouplers!
Appreciate the Wikipedia links 🙏💯