Cashing out on Polymarket sports: 38,601 large sells, checked against the final result

Large Polymarket sports sellers got 64.1 cents on average for outcomes that went on to win 64.9% of the time. Across 38,601 sells from April 2 to September 13, 2026, that is 0.87 of a point short of fair, on an interval from minus 2.15 to plus 0.33 that includes zero. Large buys in the same sports and window landed within a point of fair too; their market IDs were not restricted to the markets sold.

There was no measurable hidden margin in this sample, before explicit fees. Selling is the buying decision run backward: sell when the bid after any taker fee is above the probability you believe, and hold when it is below. A resting limit sell that fills as a maker pays no taker fee. The exits that cost real money were a handful of World Cup futures, and one of them, Spain, decides a whole price band on its own.

The first version of this study, on September 13, scored a sell on a market that settled void, or that Polymarket could not resolve, as if the side sold had lost. That credited the seller with a perfect exit. A void market refunds both sides, so those sells were neither good nor bad.

In the October 11 re-run, 476 sells on 174 markets fall in that group. The 104 on markets that could not be resolved went at 87.8 cents on average, so the first run counted each as a good exit worth 87.8 points.

Re-run on October 10, 2026 without them, the overall exit edge fell by 0.66 of a point, and selling a heavy favorite before kickoff, which had cleared zero as a good exit, moved back inside the noise. The September query on the same data gives minus 0.15 points overall, so the void markets made the change, not the newer sells. Exits are still fair on average, and everything below is the corrected study.

"Large" means what 0xinsider's large-trade feed recorded, and that changed mid-study. Until July 2, 2026 the feed kept sports fills of about $3,000 and up (the smallest sell here is $2,957.75); after that, $10,000 and up. The sample is every sell it recorded on a Polymarket sports market from April 2 to September 13, 2026, at 2 to 98 cents, on a market that settled yes or no after the sale.

That is 7,544 wallets on 8,928 markets, $781.8 million received, $20,253 a sell on average, and 67.8% of the sells landed after the start.

The measure is the exit edge: the exit price minus the share of sold outcomes that went on to win, in percentage points. Positive means you got more than the outcome turned out to be worth; negative means you let go of something worth more than you were paid. It ignores what the seller first paid, which this data cannot see.

Intervals resample markets rather than sells, because two hundred sells on one market are one result. Every sale is scored at its own price, before fees.

Pick any exit price from 20 cents up and the outcomes sold won about as often as the price said. Every one of those bands has an interval that includes zero, and none is wider than eight points. Under 10 cents, 706 sells came in at plus 2.43 points, a touch above fair and also inside the noise.

The band that looks like a disaster is 10 to 20 cents: 1,575 sells at 15.2 cents whose outcomes won 25.7% of the time, minus 10.44 points. Its interval runs from minus 34 to plus 7, because a few World Cup futures carry the whole band.

Spain's sellers let go of 18.2 million shares for $2.8 million. Spain won the World Cup, and those shares paid $18.2 million. Nine of the ten markets with the most large sells under 20 cents were 2026 World Cup winner markets, the tenth was the San Antonio Spurs to win the 2026 NBA Finals, and Spain was the only one of the ten that won.

Take Spain out and the 2,045 other sells under 20 cents, on 600 markets, went at 12.1 cents on outcomes that won 9.5% of the time: plus 2.57 points, a figure this study does not bootstrap. Without Spain, selling a long shot was a good exit.

That is the favorite-longshot bias seen from the other side. The calibration study found buyers under 10 cents getting back 44 cents on the dollar, so the seller of a 12-cent ticket is usually on the right side. Usually is the word. The one you sell will sometimes win, as its price said it could, and one Spain outweighed nine good exits.

Dumping an underdog before the game was the bad pre-game exit. 144 sells under 20 cents went at 14.1 cents on outcomes that won 26.4% of the time, minus 12.29 points, on an interval from minus 23.80 to minus 1.60 that clears zero. The average seller let a long shot go at 14 cents that went on to win about one game in four.

Heavy favorites leaned the other way: 835 sells at 87.2 cents let go of outcomes that won 81.4% of the time, plus 5.81 points, on an interval from minus 0.28 to plus 12.45 that reaches zero. Between 20 and 80 cents, where 5,960 of the pre-game sells sit, exits were fair.

The underdog result rests on 96 markets, so read it as the place a pre-game exit deserves a second look, not as a rule. Futures, the tournament and season markets where the World Cup sells sit, have no kickoff time on record and are in neither phase.

The classic cash out, taking the money on a position that is winning, was fair. 9,991 sells at 80 cents and up during play went at 89.2 cents on outcomes that won 89.5% of the time, minus 0.33 points, on an interval from minus 1.88 to plus 1.26. No in-play price band has an interval that clears zero.

The comeback a live seller fears came a little more often than the price said: 735 sells under 20 cents went at 13.7 cents on sides that came back 18.2% of the time, minus 4.58 points, on an interval that just includes zero. If the bid beats your own read of the game, take it; the price is not taxing you for locking in.

Corrected October 7, 2026. This section first graded each sell with a ranking row 0xinsider can rewrite after the trade, and reported S, A and B sellers at -2.31 points and D and F sellers at +0.18. The figures below grade each sell as the wallet stood when it traded, over the markets settled by October 7, 2026. The finding holds: the grade does not separate exits.

From June 1, each sell carries the grade its wallet held when it traded. Wallets graded S, A or B came in at -2.10 points across 2,574 sells, interval -5.04 to +0.84, and wallets graded D or F at -0.15 across 4,888.

Split by exit price, S, A and B sellers came in at -1.79 points between 20 and 80 cents, interval -5.51 to +1.92, and at -2.59 at 80 cents and up. D and F sellers came in at +1.05 and -0.38. No cohort's interval excludes zero.

That fits what the grade measures. The sharp money study found graded and poorly graded wallets separating only on the buys they make in play. What a good trader does well shows in what they buy and when. Their exits are rebalancing, locking in, or making room for the next position, and on Polymarket those exits happen at fair prices whoever makes them.

Cricket is the one sport where sellers left money on the table: 1,566 sells at 66.8 cents let go of outcomes that won 70.6% of the time, minus 3.73 points, on an interval from minus 6.71 to minus 0.34. 96% of them landed during play.

Treat it as a lean. With seven sports, the chance that at least one interval misses zero by luck alone is about 30%, and cricket's misses by 0.34 of a point. Soccer's minus 1.61 is mostly Spain: its 236 sells under 20 cents alone account for minus 1.33 of it. Tennis, where 89% of exits land mid-match, sits 1.4 points under fair on an interval that reaches plus 0.13.

September's 280 sells gave up 8.64 points, on an interval from minus 13.86 to minus 3.22, and tennis did it. 170 of them were tennis sells, 94.1% during the match, at 70.2 cents on players who won 80.0% of the time: minus 9.82 points.

September is thin. It runs only to the 13th, it holds only sells of $10,000 and up, the feed's floor since July 2, 2026, and 102 of its 189 markets were tennis. Two weeks of large tennis exits under the price is worth watching as the sample grows; it is too short to trade on yet.

The market does not charge you to leave, so the only question is whether the bid is above or below your own probability. If you think a 60-cent position wins 70% of the time, holding is worth 10 cents a share more than selling; if you think it wins 50%, selling is worth 10 more. Across 38,601 exits the bid was right to within a point, so a seller's edge, like a buyer's, has to come from knowing more than the price.

Two places in this data deserve a second look before you sell. A cheap underdog in a single game, dumped before kickoff, sold for about half its win rate, so check that the price and not your patience is telling you to go. A futures long shot you no longer believe in is usually worth selling, because long shots are the most overpriced thing on the board. The World Cup shows the one you sell will sometimes win.

It does not know what the seller paid. A sell at 60 cents might lock in a profit bought at 40 or cut a loss bought at 80, and the data sees only the exit. The exit edge answers whether the exit price was fair, not whether the round trip made money.

It does not know why a wallet sold. Some sells are hedges, some are market makers flattening inventory, some are wallets moving money to a different outcome in the same match. The fairness of the price does not depend on the reason, but the grade split mixes all of them.

It cannot tell a taker sell from a maker fill, and it scores every sale before fees. And the thin cuts are wide: a few hundred markets give an interval several times the width of the whole sample's, and under 20 cents the result moves with a single tournament.

A sportsbook cash out is an offer. The game moved, and the book offers to settle your bet early for an amount it computes, with no way for you to see how that amount relates to the bet's value.

On Polymarket there is no offer. Your position is shares that each pay $1 if the outcome happens, and cashing out means selling them on the order book to whoever is bidding, or resting a limit sell at a price you choose. The price is the market's and every sale is public, which is what makes the question in this study measurable.

What you pay is the cost of the trade. A taker crosses the spread, 1 cent at the median when large sports buys filled, from a week before kickoff to the end of play, and pays the taker fee the market carries, taken from the sale in USDC: on most sports about 0.5% of the proceeds at 90 cents and 2.5% at 50 cents. NFL and college football games carry 0.03 and NBA games no fee, as of October 21, 2026. A resting limit sell pays no fee.

Sells are rare on the feed, and tennis is where they cluster: 187 of the 495 large sports sells from September 11 to October 10, 2026 were tennis, 94.7% of those mid-match. If you want to watch traders cash out, watch tennis.

The large-trade feed lists recorded taker fills of $10,000 or more that meet its market-volume-share floor, marked BUY or SELL, with the price and the grade of the wallet behind it, live on Pro and 24 hours behind without it. Cards on the sports boards list the top five holders on each side of a game, with their grades, for every visitor; a game's event page adds its Top holders tab on Pro. A wallet's profile lists its open positions beside its record.

The cash-out guide works the sell-or-hold decision through partial exits and limit sells. A run of large sells on one side of a live game is a reason to check the price, not proof the sellers know something: on this study's evidence, they sold at fair prices.

Every number outside the grade split and the chart cuts comes from one run of a set of read-only queries on a copy of the production database at 12:32 UTC on October 10, 2026; the September 13 run stays beside it as the record of the first version. The cuts the section charts added (sport and month intervals, the World Cup holds, the size floor, void sells and the 30-day feed) come from a second run at 13:33 UTC on October 11, 2026.

The query text, the raw output, the market-level exports and the bootstrap are public in the 0xinsider research repository, under polymarket-sports-markets, in the files whose names carry the words cash-out. The exports are committed, so the intervals reproduce with the bootstrap alone and no database.

The universe is every large-trade alert for a Polymarket sell on a sports-category market from April 2, 2026, settled yes or no after the sale, at 2 to 98 cents; void and unresolved markets are left out. The query output reports the opposite sign from this page, the value given up, which is the sold outcome's win rate minus the exit price; every exit edge here is that number negated.

The grade split is a later run: on October 7, 2026 it was re-measured with each sell graded as the wallet stood at the trade (the latest daily ranking dated on or before the trade day and last written at or before it), in the files whose names start grade-at-trade-2026-10-07 in the same folder, with intervals that group trades by market, by wallet and by game and show the widest.

Expect the counts to grow as markets settle. A day after the main run, the same query found 18 more sells, all on one June soccer market the first run did not have, and the overall exit edge barely moved.