Prediction Markets and the Price of Foresight

Prediction markets interest me because they sit at the edge between knowledge and price. They take something vague, disputed, and social, and force it into a number. That move has always been intellectually attractive. It promises a cleaner interface between belief and reality.

What changed recently is scale. Since 2024, prediction markets have moved out of their old niche and into a much more visible position between finance, media, gambling, and online politics. The catalyst was the US election, but the shift was broader than that. Better infrastructure, stablecoins, mobile onramps, and larger audiences gave the whole sector a different weight.

Two platforms defined this phase. Kalshi is the regulated domestic version, operating inside federal commodities law and now connected to Robinhood's twenty-seven million accounts. Polymarket is the crypto-native version: offshore, faster, more frictionless, built on Polygon and settling in USDC. One looks more like a financial exchange. The other looks more like an internet-native information market.

Both grew fast. Volume across the sector expanded from roughly sixteen billion dollars in 2024 to an estimated sixty billion in 2025. Monthly active users on the major platforms went from around 1.7 million in early 2025 to over eight million by early 2026. At that point the question is no longer whether prediction markets are interesting. The question is what exactly they are becoming, and what they reveal when they become large enough to matter.

Why the idea has real force

The underlying intuition goes back to Friedrich A. Hayek's essay "The Use of Knowledge in Society": in complex systems, the relevant knowledge is dispersed.¹ No committee holds it all. No expert panel sees the whole field. Different people notice different things, at different times, from different positions.

That is what makes aggregation hard. Committees drift toward consensus. Pundits are rewarded for confidence. Polls capture declared preferences, not necessarily belief. People adjust for social desirability. They also adjust for loyalty, fear, status, tribe, convenience, and many other pressures that have little to do with truth.

A prediction market approaches the problem differently. If you think something will happen, you can put capital behind it. If you are wrong, you pay for it. That changes the informational atmosphere. It becomes more expensive to bluff. At its best, the price reflects the judgment of people who have enough conviction to risk being wrong.

That is why internal forecasting markets have sometimes worked surprisingly well. Hewlett-Packard found that small internal markets could outperform official sales forecasts.² Not because the market was magical, but because it could aggregate knowledge already present inside the firm that hierarchy was not processing cleanly.

That is the serious case for prediction markets. Under the right conditions, they can reward accuracy more effectively than prestige systems, bureaucracies, or public commentary.

Why this wave happened now

The earlier era hit the same ceiling again and again. Payment friction. Regulatory hostility. Thin liquidity. Prediction markets kept sounding smarter than they felt in practice.

What changed between roughly 2023 and 2026 was infrastructure. Crypto rails reduced friction. Stablecoins created a settlement layer that did not require a bank account or regulated broker. Onramps moved directly into the product, so a mainstream user could enter a market with a debit card and almost no conceptual overhead. Robinhood's Kalshi integration did the same thing from the regulated side.

The legal environment shifted too. Kalshi fought the CFTC over whether political event contracts counted as illegal election gambling or federally regulated financial derivatives. The courts sided with Kalshi. That was the moment election markets became legally durable in the United States under commodities law.³

The 2024 US election then supplied the cultural trigger. Polymarket moved faster than the polls, a French trader made headlines with an enormous Trump position, and odds began circulating as news. At that point the sector stopped looking like a niche hobby for forecasters and started looking like part of the media-financial surface of politics.

The more interesting part is what came next. The new infrastructure did not only unlock more volume. It changed the character of the volume. Polymarket introduced short-duration crypto contracts and generated hundreds of millions in weekly turnover from them. So the story is not simply that prediction markets returned. It is that they returned on top of a much thicker speculative substrate.

Price is not truth

A prediction market price is not belief in purified form. It is a price. It comes out of a trading structure, with specific liquidity conditions, capital distributions, incentives, and rules.⁴

That sounds obvious, but it is easy to forget because the output looks so clean. A contract trading at sixty cents seems to say: the market assigns a sixty percent probability. But under the surface, a thin book, a few well-capitalized traders, or a badly designed market can produce something that looks more objective than it is.

Charles Manski has argued for years against casually equating price with probability. Under favorable conditions, deep liquidity, diverse participation, no dominant actors, market prices can approximate collective probability judgments. Outside those conditions, they become harder to interpret. The number can still be informative. It is just a stranger object than enthusiasts usually admit.

Once belief becomes tradable, it becomes a game

Once beliefs become tradable, the situation changes. People are no longer only expressing views. They are positioning against other people.

Then timing matters. Bluffing matters. Signaling matters. Sometimes the point is not to express what you believe. Sometimes the point is to shape what others believe, or what they think everyone else believes.

That does not make prediction markets useless. It makes them adversarial. That is a more accurate description.

The Romney Whale episode still captures this cleanly. In 2012, a single trader placed a third of all bets on Mitt Romney on Intrade and pushed the market in ways that looked irrational relative to broader evidence.⁵ The trade did not change reality in the end, and the trader reportedly lost money. But before that happened, it helped shape narrative, donor confidence, and momentum. Short-term manipulation can fail and still matter.

The new wave has the same dynamic at larger scale. On Polymarket, a tiny number of accounts appear to drive most meaningful price discovery. Research suggests that roughly a tenth of a percent of accounts capture two-thirds of the profits. One French trader reportedly moved election odds globally with positions exceeding thirty million dollars. That is not some abstract crowd speaking. It is a capitalized actor using the market as an instrument.

The gambling substrate

Here is what the growth story does not say cleanly: a large part of the expansion is not epistemic. It is gambling.

On Kalshi, sports contracts consistently account for roughly eighty percent of weekly volume. When Robinhood integrated the platform, Super Bowl trading alone exceeded one billion dollars. By early 2026, sports outcomes represented eighty-seven percent of Kalshi's total volume.

On Polymarket, the picture is more mixed but not cleaner. Roughly forty percent of weekly volume is sports-related. Another large share is fast cryptocurrency markets, short-duration contracts on Bitcoin price moves that have nothing to do with forecasting and everything to do with speculation.

The theoretical promise that prediction markets would aggregate dispersed knowledge across an infinite long tail of events has largely not materialized. In practice, what scaled first was a very effective sports betting and crypto speculation platform wearing the clothes of epistemic infrastructure.

This matters for a structural reason as much as a moral one. The platforms need volume to sustain market makers and maintain liquidity. The volume that is actually available, reliably and at scale, comes from sports and crypto. That is where the capital flows. That is what the platforms optimize for. The long-tail forecasting fantasy is not driving the growth. It is the cover story.

There is also the question of who is being served. Traditional sportsbooks are heavily regulated: minimum age twenty-one, self-exclusion registries, responsible gaming prompts, addiction treatment funding via gaming taxes. Prediction markets operate outside most of that framework. Platforms like Polymarket allow registration at eighteen. They do not integrate with state self-exclusion systems. Recovering gambling addicts who have legally banned themselves from DraftKings can find these platforms on their phone without friction.

This is not a moral panic. It is a design choice. The same interface that disciplines belief also monetizes compulsion. Both things are true at the same time.

When prediction becomes power

The deepest issue appears once a prediction market becomes visible enough to matter.

A widely watched price stops merely measuring expectations. It starts influencing them. Journalists cite it. Donors react to it. Traders use it as a signal. People begin coordinating around the number instead of merely observing it.

Major news organizations, including CNN, CNBC, and the Associated Press, integrated prediction market data into coverage during the 2024 election cycle. A ticker showing odds shifting was not treated as a speculative wager. It was treated as a report on reality. The market stopped mirroring the future and started participating in the production of it.

That creates a feedback loop. A price moves. Media reports the move as momentum. That reporting changes behavior. Donors give, volunteers show up, undecided voters adjust. The changed behavior feeds back into the market. What looked like a prediction becomes a cause.

Politicians know this. A candidate whose odds are collapsing may lose energy even if the fundamentals have not changed. A project that a market prices as likely to fail may trigger internal panic that makes the failure more likely. The market can create self-fulfilling dynamics, but it can also create self-negating ones. Sometimes a warning changes behavior early enough to prevent the predicted outcome.

Either way, the line between observation and intervention breaks down. The number stops being a mirror and becomes an actor.

This is where reflexivity enters. Once a forecast starts steering behavior, the meaning of accuracy itself gets more complicated. The market may be right in one sense and causally active in another.

A laboratory for market knowledge and distortion

Prediction markets make something visible in unusually concentrated form. They show what markets do to knowledge.

You can see information asymmetry more clearly there than in many ordinary markets. You can see how capital amplifies voice, how a few traders can shape what later gets interpreted as the crowd's judgment, and how a public number starts impersonating objectivity.

Market microstructure research adds another layer. Analysis of Polymarket activity suggests that reported volume contains significant noise, partly from accounting quirks and partly from wash trading practices that would look unacceptable in traditional equity markets. So even the apparently clean surface can contain a great deal of hidden structure.

The observer effect becomes sharper too. Once AI systems ingest market data and redistribute it through dashboards, feeds, and trading systems, a manipulated move can travel quickly and trigger reactions far outside the platform itself. Then the issue is no longer only epistemic. It becomes infrastructural.

That is why prediction markets are interesting beyond forecasting. They compress dispersed knowledge, but they also compress away context. They reward signal, but they can also reward narrative manufacturing. They clarify some things by flattening others. The current ecosystem makes both sides easier to see.

Why the market form should not be overrated

Austrian economics gives a useful lens here, but it cuts both ways.

Hayek's insight about dispersed knowledge is exactly why prediction markets seem so attractive. If relevant knowledge is distributed, then a mechanism that rewards people for acting on their private information is appealing. Prices as signals and markets as discovery processes: that is a serious argument in their favor.

But the same tradition also warns against treating markets as if they resolve uncertainty cleanly. Frank Knight distinguished between risk, situations where probabilities can be known, and radical uncertainty, situations where the future remains unknown in a deeper sense and cannot be modeled in advance.⁶ Prediction markets translate open-ended human situations into sharp-looking percentages. That can be useful. It can also create false confidence. A number can make an event feel more knowable than it really is.

There is another problem too. Not all relevant knowledge enters markets equally well. Some knowledge is tacit. Some is moral. Some is contextual in ways that resist compression into a bid. Some belongs to people without capital.

This creates a tension at the center of the whole picture. Prediction markets are often praised as spontaneous order, but in practice they are highly engineered systems. Their epistemic quality depends on contract design, liquidity structure, resolution rules, and platform governance. That does not make them worthless. It does mean they are not simply pure market truth bubbling up from below.

If you take market epistemology seriously, you should become less naive about markets, not more.

There are also other tools. Platforms like Metaculus and Good Judgment run structured forecasting tournaments where participants provide probabilistic estimates together with reasons. On some kinds of questions, especially long-horizon and structurally complex ones, trained superforecasters seem to outperform both individual experts and financial prediction markets. They preserve argument rather than only position. They do not have whales. They do not have wash trading.⁷

None of this makes prediction markets worthless. It means they should be kept in proportion. They are one way of organizing collective foresight, not the final form of it. In a world increasingly shaped by coordination systems, machine assistance, and institutional design, the deeper question may not be whether we can price expectations more efficiently. It may be what kind of knowledge institutions we want to build around uncertainty.

That question matters beyond forecasting. A society can accumulate knowledge through markets, but also through archives, deliberation, science, reputation systems, expert networks, and institutions that preserve reasoning over time. A price is excellent at compression. It is much worse at memory. It cannot explain itself. It cannot preserve disagreement in a rich form. It cannot tell you which part of a signal came from insight, which part from fear, and which part from capital. If we care about collective intelligence rather than spectacle, those limits should push us to think more seriously about what comes after the market form, or beside it.

The limit of legibility

Prediction markets are powerful because they make uncertainty legible.

Their danger is that legibility starts to masquerade as truth.

The recent wave showed both sides clearly. Prediction markets can reveal real information. They can aggregate dispersed knowledge, punish cheap talk, and respond faster than polls or panels. That part is real.

But the same wave also showed how quickly legibility can slide into monetization, gamification, and narrative control. When market prices circulate as shorthand for reality, when whales move global probabilities with eight-figure positions, and when sports contracts dominate a platform that was supposed to help us understand the world, the epistemic promise starts to look less pure.

A price on a screen feels clean. Sometimes it really is more informative than an argument. But it is never the future speaking directly. It is information filtered through incentives, capital, visibility, and strategy.

That is why I would neither romanticize prediction markets nor dismiss them. They are useful precisely because they are not oracles. They are interfaces. They reveal something real about expectation under conditions of money and risk. They also reveal the limits of price as a vessel for knowledge.

And that opens the more interesting question. If price is such a narrow vessel, what other ones should we build? Systems that preserve reasons, track uncertainty without flattening it too early, accumulate judgment over time, and let collective intelligence become something richer than tradable sentiment.

Footnotes