Who Really Makes Prediction Markets Accurate? The Skilled 3%

Prediction markets are often described as the ultimate expression of the “wisdom of crowds.” New research suggests something more interesting is happening: most of the crowd contributes volume, while a tiny minority contributes the information.

Prediction markets have a compelling premise.

Thousands of people bring different information, opinions and incentives to a market. Their trades combine those views into a single price. A contract trading at 70 cents effectively says the market thinks an event has roughly a 70% probability of happening.

The standard explanation for why this works is the wisdom of crowds.

A new academic paper based on almost the entire trading history of Polymarket challenges that explanation.

Researchers Roberto Gómez-Cram, Yunhan Guo, Theis Ingerslev Jensen and Howard Kung analyzed $13.76 billion of trading volume across 210,322 markets and 1.72 million accounts between 2023 and the end of 2025.

Their conclusion is striking:

Prediction markets are not primarily made accurate by the crowd. They are made accurate by a small minority of traders who consistently outperform it.

About 3% of Polymarket accounts were classified as persistently skilled. Their trades predicted future price movements and eventual market outcomes. More importantly, when these traders entered a market, prices tended to move closer to the correct value.

The remaining 97% played a different role.

They provided the liquidity, volume and mistakes that the skilled minority traded against.

 

Separating skill from luck

The difficult part of studying prediction-market traders is that profitability alone tells you surprisingly little.

Someone can make $500,000 because they are exceptionally good at forecasting.

They can also make $500,000 because they made three large bets and happened to win all three.

With more than a million accounts, some spectacular winners will inevitably emerge through chance alone.

The researchers therefore built a statistical test designed to separate skill from luck.

For every trader, they kept the trader’s actual trades but repeatedly randomized which side of each event they had taken. They then compared the trader’s real P&L with hypothetical versions of that trading history.

If the actual results were too good to plausibly come from random direction-taking, the account was classified as skilled.

The results divided Polymarket traders into a few distinct groups:

• 3.2% were skilled
• 28.9% were profitable, but their returns could not be statistically distinguished from luck
• 61.7% lost money without showing statistically significant “anti-skill”
• 6.2% were anti-skilled, meaning they lost so consistently that random trading would have performed better
• roughly 0.1% were classified separately as market makers

That distinction matters.

A trader who has made a lot of money is not necessarily a good trader.

 

Real skill persisted

The strongest evidence came when the researchers divided each trader’s history into separate training and test samples.

If the results were simply luck, the winners should disappear when tested on different markets.

They did not.

Around 46% of traders identified as skilled in the first sample were again classified as skilled in the second. Another 22% remained profitable but did not clear the statistical threshold for skill.

By comparison, only about 10% of mutual funds identified as skilled using gross returns remained classified as skilled when the researchers ran a similar exercise on fund managers.

Prediction markets therefore appear to contain an unusually persistent group of outperformers.

The same phenomenon exists at the other end.

Almost half of traders classified as anti-skilled remained anti-skilled out of sample.

Some people were not merely unlucky.

They were repeatedly making the wrong trades.

 

The crowd provides liquidity. The minority provides information.

This is where the paper becomes particularly relevant to how we think about prediction markets.

If prediction-market accuracy really came from the wisdom of crowds, information should be broadly distributed across traders.

Instead, the researchers found that information was heavily concentrated.

Skilled traders represented just over 3% of accounts and captured roughly 27% of aggregate dollar profits.

Their order flow also contained information that market prices had not yet fully incorporated.

When skilled traders were buying, subsequent prices tended to rise and the event was more likely to eventually resolve YES.

When they were selling, the opposite was true.

The researchers then measured something even more important: whether each group’s trading actually made prices more accurate.

Only the skilled group consistently did.

As markets approached resolution, the effect became stronger.

So the relationship looks less like millions of independent opinions averaging themselves into truth and more like a familiar financial-market structure:

Informed traders identify mispricing, trade against less-informed participants and gradually force prices toward fair value.

The crowd still matters.

Without it, there would be considerably less liquidity and fewer opportunities for informed traders to express their views.

But the paper suggests the crowd is more often the raw material for price discovery than its primary source.

 

These traders do not appear to be insiders

There is another obvious explanation.

Perhaps the profitable 3% simply know things everyone else does not.

Prediction markets can be unusually vulnerable to private information. A corporate employee may know earnings before release. Someone close to a political decision may know what is about to happen. A television producer may already know the winner of a recorded show.

The researchers found evidence that this happens.

But not enough to explain overall market accuracy.

They estimate that roughly 89% of the Polymarket markets in their sample were not naturally prone to insider trading, particularly sports, crypto prices and major elections.

Skilled traders still improved prices in those markets.

The researchers also constructed three different tests for suspicious insider-like accounts. They identified 967 accounts, but those traders accounted for only around 0.2% of total trading volume.

Where suspected insiders appeared, they could move prices dramatically.

But they appeared too infrequently to explain the broader efficiency of the market.

Only 11 of those 967 suspected insider accounts were also classified among the persistently skilled group.

The two populations were almost completely different.

 

So what are the best traders actually doing?

The answer is less mysterious than insider information.

The paper identifies three recurring sources of edge.

First, speed.

Skilled traders reacted faster to publicly available information.

Around scheduled FOMC decisions and corporate earnings announcements, their trading shifted in the direction of the new information faster and more reliably than other groups.

That suggests an advantage in monitoring, processing and execution rather than privileged access.

Second, arbitrage.

When logically related contracts temporarily produced inconsistent prices, skilled traders were overwhelmingly the group exploiting those discrepancies.

They effectively enforced the law of one price.

Third, behavioral mistakes.

Prediction markets exhibit the same favorite-longshot bias found across betting markets.

People tend to overpay for exciting, unlikely outcomes and underpay safer favorites.

The skilled traders systematically took the other side.

They sold overpriced long shots and bought underpriced favorites.

The crowd’s biases became their edge.

 

But that edge may be shrinking

This is where the market is beginning to change.

CNBC’s September 14 reporting highlighted the same research while examining the increasing professionalization of prediction markets.

Prediction-market platforms are courting Wall Street, bringing in deeper professional liquidity and more sophisticated competition. That should improve pricing, but it also makes it harder for traders to make money from the same inefficiencies.

When ten sophisticated traders notice the same arbitrage instead of one, the opportunity disappears faster.

Spreads compress.

Reaction times shorten.

Obvious mispricings vanish.

Theis Jensen, one of the paper’s authors, told CNBC that the share of traders with a persistent edge could eventually fall from around 3% to below 1% as competition increases. That is his projection, not a finding of the historical study.

Julie Hoover, an equity research analyst at Bank of America, made the same broader point: as markets become more efficient and spreads tighten, straightforward mispricing and arbitrage opportunities become harder to find.

There is an important counterpoint. 

Prediction markets contain an enormous number of small, specialized contracts.

A large quantitative fund may be exceptionally good at macroeconomic markets but have no particular advantage pricing a niche political race, regional weather event or obscure technology launch.

Large institutions also face capacity constraints. In thin markets, their own orders can move the price enough to erase the edge they are trying to capture.

That potentially leaves a long tail of opportunities for small specialists with deep domain knowledge, better models or faster execution.

 

Prediction markets are starting to look like real financial markets

Perhaps the most important implication of the research is not that prediction markets can be beaten.

It is that their internal structure increasingly resembles other sophisticated markets.

There are market makers supplying liquidity.

There are less-informed participants trading for entertainment, conviction or other reasons.

There are specialists exploiting behavioral biases.

There are arbitrageurs enforcing price consistency.

There are occasionally insiders.

And there is a small group of traders whose information processing and execution systematically moves prices closer to reality.

As more professional capital enters the market, the easiest opportunities should disappear.

That is bad news for anyone hoping prediction markets remain easy money.

But it may be very good news for prediction markets themselves.

More competition between skilled traders means better-calibrated prices. Better prices increase their usefulness not just for speculation, but potentially for forecasting, hedging and real-time economic information.

CNBC also pointed to Federal Reserve research showing that Kalshi’s macro contracts matched, and in some cases outperformed, conventional forecasting benchmarks, including the Bloomberg consensus for headline CPI.

The paradox is simple:

The better prediction markets become, the harder they become to beat.

And if the latest research is right, the next phase of the industry will depend increasingly on finding the people who can still do it.

 

Publishing notes

SEO title: Who Makes Prediction Markets Accurate? The Skilled 3%

Meta description: A study of $13.76 billion in Polymarket trades finds that just 3% of accounts drive much of prediction-market price discovery. Here is what separates skill from luck, and why those edges may be getting harder to find.

Suggested slug: /insights/who-makes-prediction-markets-accurate-skilled-traders

 

Sources

Gómez-Cram, Roberto; Guo, Yunhan; Jensen, Theis Ingerslev; Kung, Howard. “Prediction Market Accuracy: Crowd Wisdom or Informed Minority?” Working paper, June 25, 2026.

CNBC, September 14, 2026. “Prediction markets are becoming more professionalized, but also harder to beat.”

https://www.cnbc.com/2026/09/14/prediction-markets-efficient-win-lose-beat.html