Can Prediction Markets Forecast the Future of Science?

Prediction prices as collective beliefs—not scientific probabilities
Useful supplements to models and experts, but unreliable substitutes
Non-monetary markets as laboratories for forecasting and human behaviour

Prediction markets have moved rapidly from specialised experiments in information aggregation to widely visible digital services. Platforms such as Kalshi and Polymarket now offer contracts linked to elections, interest rates, climate records, disease outbreaks, technological milestones, and other events whose outcomes remain uncertain. Instead of asking participants to state an opinion in a conventional survey, these services require them to buy and sell claims whose value depends on what eventually occurs. The resulting price is then presented as a continuously updated estimate of the likelihood of the event.

This mechanism appears to offer something that conventional scientific communication often lacks: a single numerical signal that changes whenever new information arrives. Scientific assessments are commonly expressed through reports, confidence intervals, competing models, and qualified expert statements. Prediction markets compress this complexity into an immediately legible number. When a contract trades at 0.60 and pays one unit if an event occurs, observers commonly interpret the price as indicating a 60 per cent probability. That apparent simplicity has encouraged media organisations, investors, and policymakers to treat market prices as real-time measurements of collective expectations.

The scientific question is not whether such markets can occasionally forecast correctly. They clearly can. The more important question is what a market price actually measures. It may reflect informed analysis, but it may also incorporate public anxiety, speculative demand, strategic trading, unequal access to information, and ambiguity in the wording of the contract. Prediction markets should therefore be assessed not as automatic mechanisms for discovering truth, but as institutions that transform heterogeneous beliefs and incentives into observable data.

How Prediction Markets Convert Beliefs into Prices

A basic binary prediction contract pays a fixed amount if a specified event occurs and nothing if it does not. Participants who believe the event is more likely than the current price suggests have an incentive to buy, while those who believe it is less likely have an incentive to sell. Through repeated transactions, dispersed information may become incorporated into the market price. The economic attraction of the mechanism is that participants are not merely asked what they believe; they are required to act consistently with that belief under an incentive structure. Prediction markets have consequently been described as tools for aggregating information that is distributed across individuals and may not otherwise be revealed.

The familiar interpretation of price as probability nevertheless requires caution. A market price is produced through transactions, not through direct measurement of an objective probability. It may depend on participants’ risk preferences, wealth, trading constraints, transaction costs, available liquidity, and beliefs about the behaviour of other traders. Charles Manski demonstrated that even under simplified assumptions, an equilibrium prediction-market price does not reveal the complete distribution of participants’ beliefs and only partially identifies their central tendency. Calling the price a “market probability” can therefore suggest a degree of statistical meaning that the observed price does not necessarily possess.

This distinction is especially important when the market concerns a complex scientific event. A price of 20 per cent does not establish that a properly specified scientific model would assign the event a probability of 20 per cent. It establishes that the market reached a particular exchange price under its prevailing rules, participants, information, and incentives. The price may still be highly informative, but its meaning is conditional on the process that generated it. The scientific task is therefore not simply to record the latest price, but to investigate whether that price is calibrated, stable, liquid, and responsive to relevant information.

When the Wisdom of Crowds Becomes Informative

The intellectual foundation of prediction markets is closely related to the wisdom-of-crowds hypothesis. A group can outperform many of its individual members when participants possess diverse information, make sufficiently independent judgments, and contribute to an effective aggregation mechanism. Prediction markets add an incentive structure to this process. A person with relevant knowledge can trade against an inaccurate consensus, potentially earning a reward while moving the price toward a more defensible estimate.

Research has found that prediction markets can perform well in several settings. Earlier studies reported that market-generated forecasts were often competitive with, and sometimes superior to, moderately sophisticated forecasting benchmarks. Markets have also been used inside organisations to forecast sales, project completion, and other operational outcomes. Their advantage is most plausible when relevant information is distributed among many participants and no single expert or model has access to the entire information set.

Yet the wisdom of crowds is not an automatic property of a large number of participants. The crowd must contain useful and sufficiently independent information. If participants copy one another, respond to the same misleading news, or trade primarily for entertainment, the market may aggregate correlated error rather than dispersed knowledge. A large trading volume can make a price look authoritative without making the underlying judgments diverse. The central scientific question is thus not how many users participate, but what information they possess, how independently they interpret it, and how strongly the market permits knowledgeable participants to correct collective error.

Why Scientific Questions Are Particularly Difficult

Scientific forecasting presents conditions that differ substantially from those found in elections or frequently repeated economic events. Many scientific outcomes are rare, technically specialised, or dependent on definitions that remain contested. Whether a particular year becomes the warmest on record can be resolved using established datasets and a clearly specified measurement procedure. Whether a quantum computer has “broken” an encryption system, whether an outbreak qualifies as a pandemic, or whether an experimental result has been successfully replicated may depend on technical standards that ordinary participants do not fully understand.

Recent prediction markets have offered contracts concerning disease emergencies, future climate records, artificial intelligence, and quantum-computing milestones. In some climate-related markets, prices have remained reasonably close to estimates produced by specialist models. In other areas, researchers have argued that prices appear overly responsive to dramatic headlines or technologically optimistic interpretations. These differences suggest that a market may perform well when participants can draw upon reliable public models and clearly defined data, while performing poorly when the relevant scientific uncertainty is highly specialised or the event definition is imprecise.

Scientific expertise cannot therefore be replaced merely by creating a tradable contract. Epidemiological forecasting relies on disease surveillance, transmission models, genomic data, clinical reporting, and assumptions about human behaviour. Climate projections rely on physical models, observational systems, and established statistical procedures. A prediction market may aggregate how participants interpret those sources, but it does not independently generate the underlying scientific knowledge. Its strongest role is likely to be supplementary: identifying changes in collective expectations, revealing disagreement, or providing an additional forecast against which expert models can be compared.

What Replication Markets Reveal about Scientific Judgment

One of the clearest scientific applications of prediction markets has involved forecasting whether published research findings will replicate. Researchers have invited scientists to trade contracts whose payoffs depend on the outcomes of subsequent replication studies. These markets convert informal doubts—often discussed privately among researchers—into explicit probabilistic assessments. Participants may consider the original effect size, statistical power, research design, field-specific practices, and credibility of the reported evidence.

Across several replication projects, prediction markets and structured forecasting surveys have contained meaningful information about which findings were likely to replicate. A pooled analysis of 103 findings reported that prediction markets classified replication outcomes correctly in approximately 73 per cent of cases. Markets have also been used experimentally to help select which studies should receive scarce replication resources. These results indicate that scientific communities possess distributed knowledge about research credibility that may not be visible in published articles alone.

The results are encouraging but do not show that markets can replace replication. A 2026 study examining forecasts for 28 classic findings concluded that prediction markets could anticipate replication outcomes to some extent, but that their forecasts remained far from perfect and that conducting the replication itself was considerably more informative. Markets may help prioritise attention, identify contested findings, or estimate community confidence. They cannot establish whether a claim is valid in the same way that new empirical evidence can.

Forecast, Sentiment Indicator, or Behavioural Measurement?

Prediction-market prices may remain scientifically useful even when they are not optimal forecasts. A sudden increase in the price of a pandemic contract following news of an isolated outbreak may reveal more about public anxiety than about epidemiological risk. From a narrow forecasting perspective, this response may appear to be an error. From the perspective of behavioural science, however, it constitutes valuable information about how people process salient threats, update beliefs after dramatic news, and overweight events that resemble recent collective trauma.

The complete trading history contains considerably more information than the final price. Researchers can observe how quickly participants react to a publication, government announcement, expert interview, or social-media discussion. They can examine whether an initial price movement persists or reverses, whether informed participants trade earlier than the general public, and whether disagreement increases when evidence becomes more ambiguous. Volume, spreads, order imbalance, price volatility, and the concentration of positions can reveal the structure of belief formation rather than merely its final numerical result.

This broader interpretation changes the purpose of the market. Instead of asking only whether the crowd correctly predicted an event, researchers can ask how the crowd reached its estimate. Did participants update gradually or abruptly? Did specialist information correct public sentiment? Did one influential trader move the price, or did many independent participants revise their beliefs simultaneously? A prediction market can thus operate as an observational instrument for studying attention, confidence, imitation, disagreement, and information transmission.

Manipulation, Insider Knowledge, and Market Design

The same mechanisms that make prediction markets informative also create ethical and methodological risks. A participant with privileged information may improve the apparent accuracy of the market while violating legal or institutional expectations. A well-funded trader may move a thin market without possessing superior knowledge. Participants may also attempt to influence the event itself, its reporting, or its settlement process. These concerns become particularly serious when contracts concern political decisions, armed conflict, public health, or outcomes controlled by identifiable institutions.

Experimental and theoretical studies offer mixed conclusions about manipulation. Some markets can resist attempted distortion because informed traders profit by trading against an artificial price movement. Under other conditions, however, a well-funded or persistent manipulator can impair information aggregation, especially when other participants exhibit herding behaviour or learn slowly. The possibility of manipulation therefore cannot be evaluated independently of liquidity, participant composition, position limits, transparency, and the specific mechanism used to form prices.

Contract wording is equally important. A market cannot provide a meaningful forecast when the event is vaguely defined or when the resolution source is vulnerable to dispute. Scientific questions often contain concepts whose definitions evolve as research advances. A carefully designed platform must specify the measurement source, deadline, threshold, treatment of revisions, and procedure for exceptional cases before trading begins. Otherwise, participants may be forecasting how the platform will interpret the contract rather than whether the underlying scientific event will occur.

The Case for Non-Monetary Research Markets

Real money is not the only mechanism capable of sustaining a prediction market. Research has documented useful forecasting performance in markets based on virtual currency, points, or reputation. Monetary stakes can increase attention and impose a cost on poorly supported confidence, but they also introduce gambling behaviour, unequal purchasing power, financial harm, and regulatory complications. Earlier comparisons of real-money and play-money markets produced mixed results, indicating that monetary incentives are not a prerequisite for information aggregation, although they may affect participation and accuracy under some conditions.

A non-monetary scientific market can instead reward calibration, consistency, research quality, and long-term forecasting performance. Participants might receive an identical virtual endowment, accumulate reputation through accurate forecasts, and be evaluated using proper scoring rules. The objective would not be to maximise trading excitement or transaction volume. It would be to elicit honest probabilistic judgments, preserve a traceable history of belief revisions, and compare the performance of individuals, expert groups, statistical models, and the aggregated market.

Such a system would also permit forms of experimentation that commercial platforms may have little incentive to conduct. Researchers could vary the information shown to participants, separate experts from non-specialists, test anonymous against attributed forecasts, or examine how market structure affects herding and overconfidence. Because no real money changes hands, the platform could be designed primarily around scientific validity and participant protection. The market would function less as a betting service than as a controlled infrastructure for collecting longitudinal data on human judgment.

Prediction Markets as Scientific Infrastructure

The strongest argument for prediction markets is not that they always outperform experts. It is that they create a measurable process through which uncertain beliefs are stated, revised, aggregated, and ultimately compared with observable outcomes. Traditional commentary leaves much of this process undocumented. Experts may express qualitative confidence, change their views without a permanent record, or disagree without specifying the magnitude of their disagreement. A forecasting market obliges participants to translate those judgments into quantities that can later be evaluated.

For scientific institutions, this opens several research directions. Market forecasts can be compared with formal statistical models, expert panels, surveys, and machine-generated predictions. Price movements can be linked to publications, news coverage, community discussion, and institutional announcements. Repeated participation can reveal whether individuals become better calibrated over time and whether particular forms of expertise generalise across domains. The objective is not merely to forecast a collection of isolated events, but to construct a dataset describing how scientific and social expectations evolve.

Prediction markets should therefore be treated neither as electronic oracles nor merely as gambling platforms. Their scientific value lies in the observable interaction among information, incentives, uncertainty, and human behaviour. A carefully governed, non-monetary forecasting environment could preserve the information-aggregation benefits of market design while avoiding many of the harms associated with financial wagering. Under such a model, the central product would not be the winning contract. It would be the accumulated evidence on how people interpret the future.

---

Selected References

Arrow, K. J., et al. “The Promise of Prediction Markets.” Science, 2008.

Manski, C. F. “Interpreting the Predictions of Prediction Markets.” Economics Letters, 2006.

Wolfers, J., and Zitzewitz, E. “Prediction Markets.” Journal of Economic Perspectives, 2004.

Camerer, C. F., et al. “Evaluating the Replicability of Social Science Experiments in Nature and Science between 2010 and 2015.” Nature Human Behaviour, 2018.

Chandrashekar, S. P., et al. “Using Prediction Markets and Forecasting Surveys to Predict 28 Replication Outcomes.” Royal Society Open Science, 2026.