1.72 million addresses are only 3.14% “real winners”. Predict how the market's money will be taken by a few people?
Translation: Wenser
Original title: 10 Truths of Predicting the Market: Only 3.14% of Polymarket's 1.72 million addresses are “real winners”
Editor's note:Predictive market platforms such as Polymarket and Kalshi have always defined themselves as a “concentrated expression of group intelligence” to distinguish themselves from quiz platforms and raise valuations by emphasizing this narrative. However, a recent paper from the London Business School and Yale University found that after disassembling data on the Polymarket chain, less than 4% of addresses drove price changes and actual profits were quite impressive, while the remaining 97% of addresses were mostly “runners”, and over 67% of people lost money. Considering that the number of Polymarket user addresses has already exceeded 2.43 million, the paper's research data may be lagging behind, but the phenomena behind its findings are still worth pondering.
Fact 1: The accuracy of predicting the market has nothing to do with “group intelligence”; it is determined by a minority of 3.14%
This is the central conclusion of the entire paper, and it is also a direct challenge to the industry narrative.
Previously, many industry representatives were proud of this: Kalshi CEO Tarek Mansour said that predicting the market “uses group intelligence,” and Polymarket CEO Shayne Coplan also promoted many times that “financial interests can gather information more effectively than experts.” Robinhood CEOVlad TenevCall it “capitalism's quest for truth.” However, research data tells us that out of 1.72 million Polymarket accounts, only about 54,000 accounts (3.14%) were identified as “skill winners” (Odily Planet Daily note: the paper summarizes this type of person as professional players who can both predict and absorb information on an average basis, and respond efficiently when news appears).
The main driving force for predicting price discoveries in the market is this minority, not the crowd hiding behind “group intelligence” most of the time.
Truth 2: Whether you make money or lose money is probably luck. 67% of participants are essentially “philanthropists”
In this paper, Roberto Gómez-Cram and others used a set of sign-randomization statistics to divide all traders' accounts into four categories: skill winners (3.14%), luck winners (29.0%), luck losers (61.4%), and skill losers (6.4%).
The most counterintuitive number is — the lucky winners account for nearly 30%. They make money, but their transactions don't contribute to price discovery; the statistical level is no different from random coin toss.
In other words, making money in the forecasting market and “having the ability to predict the future” are two different things; the loser group, which accounts for about 67%, bears all the losses, essentially paying for the information advantage of a few people.
Fact 3: Top players in the profit list, 88% rely on luck to make money
Of the top 54,000 traders ranked by actual profit on Polymarket, only 12% were also identified as “skill winners” by statistical methods.
In other words, the vast majority of big winners with large profits on the rankings depended on the luck of one or two big gamblers.
A typical example is the account@majorexploiter——One weekend in early 2026, the account invested $4.5 million in three sporting events, making a profit of over $3.6 million.
The return on this type of concentrated betting is extremely unsustainable, with 60% of “lucky winners” becoming losers in out-of-sample testing.
Fact 4: The effectiveness of market forecasting skills far exceeds that of the traditional fund industry
The paper researchers randomly divided the betting events into training sets and test sets to verify outside of the sample.
The results showed that 44% of the accounts identified as “skilled players” in the training set were still identified as “skilled users” in the test set; as a comparison, the US actively managed mutual fund conducted the same test, and the skill effectiveness was only 10%.
Looking at it the other way around, “anti-skill” (continuous loss) is also highly consistent: 51% of the “skill losers” concentrated on training continued to maintain their status as losers in the test set, while the US Mutual Fund's figure rose to 20%.
The final conclusion is,Predict that the best players in the market are real experts, and chives are also real chives.
Fact 5: Skill-based winner orders are highly correlated with the final outcome
Based on the constructed order imbalance formula, the researchers found that for every 1% increase in the net buying index (OIB) of skilled winners, the next price increases by about 2 basis points, and the probability of the final event occurring increases by about 8 basis points, which is extremely statistically significant (t values are 12.71 and 9.51, respectively).
However, the order flow for lucky winners was not significant on either indicator (t values were only 1.47 and 1.49).
In other words, even though the profit is positive, the lucky winner's trading operations are not informative — a conclusion that is very solid from a data perspective.
According to research, the phenomenon can be seen is that in a market with a “yes” settlement result, skill winners are net buyers; in a market where the settlement result is “no”, they are net sellers; they continue to open positions in the direction of the final result. Market makers are mostly net sellers in “yes” markets, and are mostly net buyers in “no” markets, consistent with their role of adapting to directional order flows and earning price spreads rather than establishing insider orders.
Fact 6: Skilled traders are the only group that makes prices more accurate
Based on the premise that “some transactions actually drive the price to the final result,” the researchers constructed a “price discovery contribution index” to measure whether the price is closer or farther from the accuracy of the final result within each time window.
It was found that only when skill winners increased their share of trading volume, the betting event could significantly reduce pricing errors (coefficient - 5.00, t value - 5.54).
Looking at the other three groups — lucky winners, lucky losers, and skill losers — instead diverts the price from the final result — in reality, most people only make noise at the transaction level, and this effect is increasing as the market approaches settlement. The contribution factor for skill winners swelled to -9.61 during the final 20% of the betting event.
Fact 7: Skill winners are the only “News Trading” players
In order to minimize errors due to news transmission time, the researchers selected FOMC interest rate decisions and corporate earnings announcements, as research samples (Daily Daily Note: The former is the core of monetary policy expectations; the latter is the core of understanding company fundamentals).
Research data showed that only skill winners had a significant shift in order flows “in an unanticipated direction” within the short-term window after the news was announced.
In FOMC betting events, for every 1% increase in the unexpected direction, the net purchase volume of the corresponding skill winner increased by about 5% (t=3.94); since the FOMC accident direction was small (maximum about 6 percentage points), the reverse buying margin was large. For financial announcements, every 1% increase in the unexpected direction correlates with an increase of about 17 basis points in net purchases for corresponding skill winners (t= 2.62). In contrast, all other groups did not respond unanimously to the news, and some even operated in the opposite direction.
According to research data, market makers on Polymarket only account for 0.1% of the total number of accounts (about 1,660), yet participate in an average of 942 betting markets, making an average profit of $11,832 per account.
Furthermore, their order flow can predict price changes in the short term (because they are constantly “taking over” orders), but the predictive effect on the final event outcome is negative (figure 3 above: coefficient -5.69, t=-10.30).
This means that they took sales orders from insider traders for a short period of time, but the orders were “cut” by insiders for a long period of time, mainly making money by relying on trading price differences rather than directional judgments.
Fact 9: Insider trading only affects the outcome of a few incidents
Considering that insider trading is difficult to avoid in predicting the market, the study also conducted a data analysis on the impact of insider trading on price discovery. (Daily Planet Daily Note: The study uses two major criteria to flag suspicious transactions. The first is timing, that is, accounts that were opened shortly before a specific event, such as 7 days, and stopped trading after the event was settled; the second is strength of belief, that is, accounts that concentrate activity in a single event contract and hold abnormally large positions with a trading volume of at least $1,000 and profits of at least $1,000. (Accounts that meet both conditions are classified as insider traders.)
Among them, the paper identified about 1,950 suspected insider trading accounts using the two dimensions of “account time characteristics+position concentration”, and this batch of addresses made a profit of 15,000 US dollars per person.
It is worth noting that orders from these accounts are extremely accurate in predicting the price and results of some events (the final result prediction coefficient is 94.63, 12 times that of skill winners), but they are concentrated on only a few events, and their contribution to predicting the overall price discovery of the market is not obvious.
It is worth mentioning that the study disassembled the predictive market case of the “US military raid on Maduro” in detail: the three accounts placed bets a few days before the operation, concentrated on buying events with a probability of only 10%, and ended up making a total profit of more than 630,000 US dollars — one of the account holders was later charged by the CFTC as an active US military member.For details, please read“It took 4 months for Polymarket to catch the leaker of the military operation, but at a cost...”
Fact 10: Predicting the extremely unequal distribution of transactions in the market is comparable to the law of power
As of December 2025, Polymarket's trading volume increased from $3.3 million in December 2023 to $1.98 billion, more than 600 times in two years; during the same period, the number of monthly active accounts soared from 1600 to over 519,000.
The operating data is impressive, but the truth behind the data is even more counterintuitive — the average trading volume of the median active account on Polymarket was only $72, while the average trading volume of the top 1% of head accounts was $74,000, a difference of more than 1000 times.
In terms of overall trading volume, as of December 2025, the total trading volume of Polymarket was 13.76 billion US dollars, 1.72 million accounts, but the two groups of lucky losers and skill losers accounted for 67% of the accounts, contributed 39% of the trading volume, and borne 100% of the losses.
Undoubtedly, this is not a fair market where “everyone is equal and the wisdom of the group is concentrated,” but rather a zero-sum game ecosystem where a few people set the facts, the majority provide capital, and bear losses.
Twitter:https://twitter.com/BitpushNewsCN
Compare the TG exchange group:https://t.me/BitPushCommunity
Compare TG subscriptions:https://t.me/bitpush
