Lesson 04·Foundations·8 min read

Why Prediction Markets Often Beat Polls

Prediction markets aggregate information faster than polls because traders are paid to be right and punished for being wrong.

It's a stylized but well-supported finding: prediction markets, in aggregate and over time, produce more accurate forecasts than expert polls. Wolfers and Zitzewitz showed this for elections; Berg, Forsythe, and Rietz showed it for political markets over decades; the Good Judgment Project extended similar findings to geopolitics. The insight isn't that markets are magic — it's that they use a specific mechanism (skin in the game) that polls structurally can't replicate.

Why markets aggregate information better

Polls ask people what they think. Markets ask people what they'd bet. The first answer is cheap; the second has skin in the game. Information that would never make it into a poll — a campaign internal document, a political donor's private read, a tracking model's latest run — finds its way into a market via someone willing to trade on it. Anyone with a real edge is financially motivated to act on it, and their trade shifts the price for everyone else to see.

Polls, by contrast, sample a fixed population using structured questions. Any information that lives outside that population (or outside the questions) is invisible. Markets have no such boundary — any participant with any information can contribute, weighted by how much they're willing to risk on it.

The academic record

The Iowa Electronic Markets, running since 1988, have outperformed contemporaneous polls in most presidential election cycles. Berg, Forsythe, and Rietz's decades-long comparison found IEM prices closer to actual outcomes than final Gallup poll averages. Wolfers and Zitzewitz's 2004 paper compiled a broad review showing prediction-market accuracy across multiple domains — sports, politics, box-office receipts.

The pattern isn't universal. Markets don't always win. But averaged across many events and many years, they beat both individual expert forecasters and aggregated polls.

When polls win

Polls outperform markets when the market is thin, manipulated, or biased by partisan retail flow. The 2024 US election briefly saw Polymarket pricing significantly higher for one candidate than poll aggregators predicted — a gap eventually closed by reality but at the time hotly debated. When markets are dominated by non-expert money, they can drift away from true probability.

Markets also fail at unknowable events. A poll of epidemiologists is more useful than a thin novelty market on a future pandemic, even though the poll is less rigorous in form. If nobody has real information, the market will reflect noise; if a specific expert community has real information, it's often faster to ask them directly.

Worked example: 2024 election night

Going into November 5, 2024, poll aggregators had the race within a percentage point. Polymarket had one candidate at roughly 60% Yes at close of trading. As results came in from key states, both signals updated — but Polymarket updated faster and more decisively, moving from 60% to 90%+ within a few hours as county-level results dropped. Poll aggregators can't update in real time; markets can and did.

In the days before the election, market and poll signals diverged. Some observers called it a warning that Polymarket had been manipulated; others called it real information the polls missed. The market was closer to the outcome. That's a single data point, but it fit a broader pattern.

How to use both

Use poll aggregators for ground-truth sampling of voter intent. Use markets to see how aggressively traders are willing to put money behind that intent. The gap between the two is itself information — it tells you where the smart money disagrees with the mainstream. If markets and polls agree, you have a confident forecast. If they disagree by a lot, dig deeper before acting on either.

Common pitfalls

Don't treat a single market print as gospel. Look at time-weighted average price, spread, and depth. A single trade at $0.65 doesn't mean the market believes 65% — check whether that price is supported by real depth.

Don't dismiss polls just because markets sometimes beat them. Polls are the source of a lot of the information markets aggregate — if pollsters stop polling, markets get less accurate, not more.

When to trust markets over polls

Trust markets more when the market is deep and actively traded, the event is well-defined, and the information relevant to the outcome is broadly distributed rather than expert-locked. Trust polls more when the market is thin, the event is technical or specialized, or the market shows signs of concentrated capital or manipulation.

Frequently asked

Have prediction markets ever been catastrophically wrong?
Yes. Brexit and Trump 2016 are the canonical examples. Markets priced both as unlikely; both happened. Markets fail when their participants share the same blind spots as pollsters — usually because both are drawing from the same informational pool.
Are prediction markets manipulable?
On thin markets, yes — but manipulation is expensive and usually leaves an obvious arbitrage opportunity for other traders. Deep markets like Polymarket's flagship contracts require millions of dollars to move by even one cent, which is a high bar for manipulation to be profitable.
How do markets aggregate information exactly?
By reflecting the weighted-average view of everyone willing to trade. A well-informed trader with $10k moves the price more than an uninformed one with $100. Over time, informed traders profit and gain capital while uninformed ones lose it, shifting the average toward informed views.
Which academic sources confirm markets beat polls?
Wolfers and Zitzewitz (2004, Journal of Economic Perspectives) is the standard review. Berg, Forsythe, and Rietz's Iowa Electronic Markets analyses across multiple election cycles are the most detailed political-market evidence. Hanson's work on prediction markets and information aggregation is the theoretical grounding.
Do markets predict everything better?
No. Markets need deep participation and unambiguous outcomes to work well. On technical, specialized, or subjective events, expert polls or Delphi-style aggregation often beat thin markets. Use each where it's strongest.
Why did Polymarket disagree with polls in 2024?
Because Polymarket's participants collectively believed information the poll models weren't capturing — partly early-vote analytics, partly ground-game reads, partly bettor conviction. Whether that was true insight or lucky retail flow is still debated. The market was closer to the outcome.
Can polls learn from markets?
Yes — some pollsters now weight their forecasts partly by market-implied probabilities as a real-time correction. Combining both signals tends to beat either alone. Nate Silver's models have used this hybrid approach in recent cycles.
What's the best way to use both in personal forecasting?
Look at both, note the gap, and let the gap guide your research. If markets and polls agree, take the number and move on. If they disagree, spend time understanding why — that's usually where the interesting forecasting question actually lives.

The .health network

Independent status monitors for the tools and markets we track.