Prediction Markets Are Not Gambling: The Case for Forecasting as a Public Good

Key Takeaways

• Prediction markets and forecasting tournaments are distinct from gambling because they reward accuracy, calibration, and updating rather than chance.
• The strongest empirical evidence shows that forecasting skill is real and improvable: top performers continue to outperform over successive years, with no regression to the mean.
• The largest single driver of forecast accuracy is not individual talent but aggregation algorithms—converting individual probabilities into log-odds and extremizing them.
• Forecasting tournaments function as a form of adversarial collaboration, forcing competing camps to place reputational bets before seeing data.
• The evidence base has clear boundaries: most rigorous findings come from geopolitical questions resolvable within 12–24 months, not longer-horizon structural questions.

1. Introduction

When people hear "prediction market," they often imagine a betting parlor: odds, stakes, winners, losers, and the emotional whiplash that accompanies games of chance. That association is understandable, but it misses something important. Prediction markets and structured forecasting tournaments are not primarily about winning money or beating the house. They are about producing a scarce and valuable resource: calibrated, accountable knowledge about the future.

The distinction matters for everyone—not just researchers. Businesses allocate capital, governments plan for risks, and individuals make career and family decisions under uncertainty. The question is whether there is a systematic way to improve those decisions, or whether "expert judgment" is just another form of guesswork dressed in credentials.

This article addresses that question directly. It draws on evidence from one of the most rigorous programs ever conducted on human forecasting: the multi-year forecasting tournaments led by Philip Tetlock and colleagues, which ran under the Good Judgment Project and the Intelligence Advanced Research Projects Activity framework. The findings challenge both the gambling analogy and the comfortable assumption that expertise guarantees foresight. They also point toward a practical conclusion: forecasting, when structured properly, behaves less like a casino and more like a public utility—a tool for producing better collective judgment.

2. What Forecasting Tournaments Actually Test

The core misunderstanding behind the "gambling" label is the assumption that forecasting tournaments reward luck. Gambling games are designed to be random or near-random in expectation; the house edge ensures that skill plays a marginal role. Forecasting tournaments are different. They score participants on the accuracy of their probability estimates over hundreds of questions, using proper scoring rules such as Brier scores and calibration curves.

The evidence from Tetlock, Mellers, Rohrbaugh, and Chen makes this clear. In the multi-year tournament run through the Good Judgment Project, the top performers did not get lucky once. Each year, the strongest forecasters—labeled superforecasters—were selected to form elite teams. Skeptics expected regression to the mean: the idea that past success would wash out as randomness averaged away. That did not happen. Year after year, the superforecasters continued to outperform the threshold set for them, improving on calibration, discrimination, and area under the curve metrics. In fact, by the second year of the program, these teams had already surpassed the accuracy target that had been set for the fourth year.

This pattern is hard to reconcile with a gambling model. If forecasting were equivalent to betting on coin flips, the best performers would eventually fall back toward average skill. They did not. The persistence of performance suggests that forecasting skill is real, measurable, and—critically—trainable.

The practical takeaway for organizations is straightforward: structured forecasting can identify people who are genuinely better at assigning probabilities under uncertainty. Those people are not simply lucky; they are using cognitive strategies that can be studied, taught, and scaled.

3. The Role of Aggregation: Algorithms Over Individual Genius

One of the most counterintuitive findings from the forecasting tournament evidence concerns where accuracy actually comes from. The intuitive assumption is that forecasting quality is mainly a function of who is doing the forecasting—hire smarter people, get better predictions. The tournament data complicates that picture.

Within the Good Judgment Project, factorial experiments were used to decompose the drivers of forecast accuracy. Recruiting and retaining better forecasters contributed roughly 10% to accuracy gains. Cognitive debiasing training contributed another 10% or so. Collaborative teams and a prediction-market-like work environment contributed a similar amount. But the single largest lever was the aggregation algorithm: converting individual probabilities into log-odds and stretching them toward extremes. That algorithmic step contributed roughly 35% to accuracy improvement—more than any of the human factors.

This is a profound result. It means that the biggest gains in forecast quality do not come from finding oracles. They come from how you combine judgments. The specific technique—log-odds extremizing—works because group averages tend to be systematically underconfident. By converting estimates to log-odds and pushing them away from the middle of the probability scale, aggregators can correct for that bias and produce sharper, more informative forecasts.

For decision-makers, this reframes the value proposition of prediction platforms. The public good is not just in the raw opinions of forecasters; it is in the institutional machinery that turns messy, heterogeneous judgments into well-calibrated probabilities. That machinery—transparent scoring, algorithmic aggregation, and continuous feedback—is what separates forecasting from both unaided expert opinion and from gambling.

4. Adversarial Collaboration: Forcing Ideas to Take Risks

Another way forecasting tournaments differ from gambling is epistemological. Gambling is a closed game: probabilities are defined by the rules, and outcomes are independent of belief. Forecasting, by contrast, is open-ended and adversarial in a productive sense. It forces competing intellectual camps to commit to testable positions before the evidence resolves.

The tournament format, as the evidence documents, imposes reputational penalties on poor updating. Theorists can no longer retreat into post hoc explanations when things go wrong. They must state what they think will happen, with a probability attached, and then live with the score. This mechanism—what Tetlock and colleagues describe as adversarial collaboration—has already adjudicated several longstanding debates.

One debate concerned cognitive style. The tournament data supported the claim that actively open-minded thinking correlates significantly with forecast accuracy. Another debate concerned collaboration: interactive collaboration outperformed independent work. A third concerned the comparison between collaborative teams and prediction markets, where the two approached similar levels of improvement. A fourth confirmed that debiasing training yields substantial gains. And a fifth produced a striking finding: conservatism—persistent underreaction to new information—was more harmful to accuracy than overreaction.

These findings matter because they show that forecasting tournaments are not just competitions; they are epistemic tools. They resolve disputes that ordinary argument cannot. When a theorist claims to know something about the future, the response is no longer "prove it" in the abstract. It is "state the probability, and we will score you."

5. Key Comparison: Gambling vs. Forecasting as a Public Good

The table below summarizes the structural differences between gambling and forecasting tournaments as revealed by the evidence.

The comparison clarifies the stakes. Gambling is a consumption good; forecasting is a production good. The former redistributes money among participants. The latter produces something new: a calibrated probability estimate that did not exist before and can inform decisions.

6. Boundary Conditions: What the Evidence Does Not Show

Any honest case for forecasting as a public good must acknowledge what the evidence does not establish. The tournament data, while robust for its domain, comes with important caveats.

First, the lack of regression among superforecasters was not fully isolated in attribution. The authors acknowledge that the "super" label itself may have had a self-fulfilling effect, and that elite team collaboration may have conferred environmental advantages that individual skill alone cannot explain. The persistence of high performance could reflect a mix of ability and advantageous conditions, not pure individual talent.

Second, and more importantly, the evidence base has an extrapolation boundary. All of the tournament results come from geopolitical questions resolvable within 12 to 24 months. That is a meaningful and practical time horizon, but it is not universal. For longer horizons or structural questions—the kind that involve slow-moving transformations, complex feedback loops, or regime changes—these findings are extrapolations, not direct evidence.

This caution matters for anyone hoping to apply forecasting methods to problems like climate policy, technological disruption, or demographic change over decades. The methods are promising, and the underlying logic of scoring—state your probability, resolve, update, repeat—is sound. But the empirical warrant for claiming skill transfers to those domains is thinner. Decision-makers should treat long-horizon forecasts with corresponding uncertainty.

7. FAQ

Q1. If forecasting is not gambling, why do prediction markets look so similar to betting markets?

The surface similarity is real: both involve probabilities and potential payoffs. But the function is different. In gambling, the odds are set to ensure the house profits regardless of outcome, and the player's expected return is typically negative. In forecasting, the scoring mechanism is designed to measure accuracy, not to extract a margin. The payoff, when it exists, is subordinate to the informational output. A market that produces well-calibrated probabilities is producing a public good, even if its interface resembles a sportsbook.

Q2. Can ordinary people learn to forecast better, or is this only for elite specialists?

The tournament evidence strongly suggests that improvement is possible through training. Cognitive debiasing training contributed meaningfully to accuracy gains, and the identified drivers—such as actively open-minded thinking and willingness to update—are not fixed traits. They are practices. The challenge is not raw intelligence alone; it is the discipline of assigning probabilities, tracking them, and revising them when evidence arrives.

Q3. What is the single biggest factor in improving forecast accuracy?

According to the factorial experiments within the Good Judgment Project, the largest measurable driver was the aggregation algorithm—specifically, converting individual probabilities to log-odds and extremizing them. That step contributed roughly 35% of the accuracy improvement, more than recruiting, training, or collaboration individually. It is a reminder that institutional machinery can matter more than individual brilliance.

Q4. Does the evidence apply to long-term forecasts?

Caution is warranted. The most rigorous tournament findings are based on questions resolvable within 12 to 24 months. For longer horizons or structural questions, the evidence is an extrapolation rather than a direct demonstration. The methodological principles—transparency, scoring, adversarial commitment—are portable, but the specific accuracy results should not be assumed to transfer automatically.

8. Conclusion

The case for treating prediction markets as a public good rather than as gambling rests on a clear distinction. Gambling consumes risk for entertainment; forecasting harnesses risk to produce knowledge. The evidence from multi-year forecasting tournaments shows that forecasting skill is real, persistent, and improvable. It reveals that aggregation algorithms—not lone experts—are the largest lever for accuracy. And it demonstrates that forcing people to state probabilities before evidence resolves can settle debates that argument alone cannot.

But the case also has limits. The strongest evidence covers a specific window: geopolitical questions with near-term resolution. Extending those findings to longer horizons requires intellectual humility, not just methodological enthusiasm.

The practical implication is worth stating plainly. Organizations, governments, and communities that face recurring decisions under uncertainty should consider investing in structured forecasting infrastructure—not because it is glamorous or because it resembles betting, but because it is one of the few mechanisms we have for generating accountable, testable knowledge about the future. That, in the end, is what a public good looks like: something that produces value beyond any single participant, and that gets better the more people use it.