When large sports bets land on Polymarket: early money loses, and graded wallets bet live

If you bet a Polymarket sports game days before it starts, you are betting with the money that lost. Little of the money in 401,482 large sports buys arrived more than a day before kickoff, as the chart below shows, and buys one to seven days early lost to the price by 2.19 points. From a day out through the final whistle, the market was calibrated.

The wallets 0xinsider grades S, A or B place half their sports buys in-play. In the first hour of play they came in at +2.23 points and D and F wallets at -1.70, each graded as the wallet stood at the trade. So early is the timing to avoid, and in-play is where a winning record shows.

This study first ran on September 12. It scored every buy on a market that settled void, or that Polymarket could not resolve, as a loss, and those buys sat mostly in the early buckets and in tennis. A void market pays half a dollar a share on either side, so they were neither wins nor losses. Re-run on October 10, 2026 with them left out, buys one to seven days early go from minus 3.21 points to minus 2.19, and every early figure shrinks, as the chart below shows.

The September rule on the same data still lands near the published figures one to seven days out and in tennis, so the void markets made most of the change, not markets that settled since. Early money still loses, by less, and the six-to-24-hour bucket no longer leans against the buyer. Everything after this section is the corrected study.

The sample is every buy 0xinsider recorded as a large trade on a Polymarket sports market from April 2 to September 11, 2026, on a market that settled yes or no after the trade, at 2 to 98 cents, where Polymarket published a start time for the game. That is 401,482 buys by 13,295 wallets across 33,643 settled markets, carrying $7.94 billion, and the chart below shows which sports it comes from.

Each buy is placed in one of eight buckets by hours to kickoff: more than seven days out, one to seven days, six to 24 hours, one to six hours, the last hour, and then the first hour of play, the second hour, and anything later. Kickoff is the provider's scheduled start; a delayed start is not modeled, so a buy at scheduled kickoff plus ten minutes during a rain delay counts as in-play.

It is the calibration study's sample less the 15,000 buys on markets with no kickoff on record. Edge and dollar return are the same measures as the calibration study: win rate minus price paid, in points, and stake-weighted return held to settlement before fees. Intervals resample markets, 5,000 draws.

48.6% of the buys and 47.7% of the dollars landed after the game had started. The last hour before kickoff took 22.8% of the buys and the first hour of play 22.4%, so four in ten large sports bets on Polymarket land in the 120 minutes around kickoff. Only 12.6% land more than six hours before the game. A tenth of one percent land more than a week out.

The bets get bigger as kickoff approaches. Buys placed a week or more out averaged $10,799. One to seven days out, $15,830. Six to 24 hours, $17,295. One to six hours, $19,374. The last hour, $22,375. In-play the average drops back to $17,376 in the first hour and climbs to $23,274 for buys more than two hours into play, the largest of any bucket, on games that have gone long or gone live in a second market.

A sportsbook sees the same curve: handle is thin when the line opens and heaviest in the hour before the game. Polymarket's curve keeps going after kickoff.

Buys placed one to seven days before kickoff won 57.4% of the time at an average price of 59.6 cents: minus 2.19 points, and the 95% interval, minus 4.01 to minus 0.43, clears zero. The 515 buys placed more than a week out came in at minus 2.03 points, on an interval, minus 7.26 to plus 2.78, that includes zero on a small sample. Buys placed six to 24 hours out came in at minus 0.54, with an interval from minus 1.48 to plus 0.40.

From a day out the loss disappears. One to six hours out, minus 0.09. The last hour, plus 0.29. The first hour of play, plus 0.45. The second hour, plus 0.56. Later than that, plus 0.39. Every one of those intervals straddles zero. From a day before kickoff to the final whistle, a large buyer on a Polymarket sports market paid what the result was worth.

Early money loses on a thin book: the mean quoted spread on buys placed more than a week out was 21 cents, against about one cent on the day. The price has not yet absorbed the lineup, the weather, or the market's own second opinion. And the buyer who is willing to put $16,000 on a game five days out is not a random draw from all traders. In this sample that buyer was wrong more often than the price said.

Before kickoff most large buys are near a coin flip. In the last hour before the game 55.1% of buys were between 40 and 60 cents and 9.4% were at 80 cents or more. In the first hour of play the 40-to-60 share falls to 35.9% and the 80-plus share rises to 21.6%. In the second hour a third of all large buys are at 80 cents or more: a team is ahead, and the money is on it.

Those live favorites calibrate. In-play buys at 70 cents and up paid 84.1 cents and won 84.3% of the time, plus 0.21 points and plus 1.44% on the dollar across 83,215 buys. Pre-kickoff buys at 70 and up paid 81.6 and won 81.0%, minus 0.57. In-play buys between 30 and 70 cents came in at plus 0.67 and pre-kickoff buys in the same band at plus 0.04.

In-play longshots are the one live band that lost money. Buys under 30 cents during play won 23.0% at 22.2 cents, an edge of plus 0.76 points, and still returned minus 8.0% on the dollar, because the stakes that lost were larger than the stakes that won. A team that is behind is cheap for a reason.

Cricket is bet during the match: 89.5% of its large buys landed in-play. Tennis, 73.9%. MMA, 73.5%. Esports, 65.8%. At the other end, soccer is a pre-kickoff sport at 36.1% in-play, hockey 31.1%, the WNBA 27.2%. The NBA is split, 45.5%, and baseball 38.8%.

The split changes the sport's calibration. Tennis buys before the first serve came in at minus 0.83 points across 12,333 buys, and tennis buys during the match came in at plus 0.57 across 34,946.

MMA buys before the first bell came in at minus 8.91 points on 1,372 buys, most of it two upsets, and in-play at minus 2.73. Esports pre-kickoff minus 0.21, in-play plus 0.68. The NBA minus 0.09 and minus 0.02. Soccer, the one sport positive in both phases, plus 0.39 and plus 0.74.

In every sport with both phases, the in-play figure is better for the buyer than the pre-kickoff figure. The live market, with the game in front of everyone, has been the fairer one.

From June 1, when grade coverage widened, each buy also carries the grade its wallet held when it traded: the last ranking 0xinsider wrote for the wallet before the buy. Split by cohort, the timing mix is different.

Wallets graded S, A or B placed 51.4% of their sports buys in-play, 45.3% on the day before kickoff, and 3.3% more than a day out. Wallets graded D or F placed 37.7% in-play, 58.3% on the day before kickoff, and 4.0% more than a day out. C wallets sit between at 44.6% in-play; wallets with no grade yet at 41.7%.

The poorly graded wallets mostly buy before the game; the well-graded wallets place half their buys live. Neither cohort buys early: more than a day out is under 5% of every cohort's buys.

Corrected October 7, 2026. This section first reported S, A and B wallets at plus 3.07 points in the first hour of play and D and F at minus 3.03. Those figures took each buy's grade from a ranking row 0xinsider can rewrite after the trade, so some buys carried a grade decided by their own result. Every figure below grades the buy as the wallet stood when it traded, over the markets settled by October 7, 2026. The gap narrowed and still holds.

Before kickoff, the grade does not separate anyone. In the last hour before the game S, A and B wallets came in at -1.08 points and D and F at +0.32. One to six hours out, -0.26 against -0.24. Six to 24 hours out, -0.19 against +1.11. Every pre-kickoff gap is under a point and a half, and every interval includes zero.

In the first hour of play, S, A and B wallets came in at +2.23 points and +6.63% on the dollar across 12,540 buys, with a 95% interval from +0.10 to +4.35. D and F wallets came in at -1.70 points and -6.48% on the dollar across 8,839, interval -3.52 to +0.12.

The two cohorts were 3.93 points apart on the same games in the same hour, and the interval on that gap, +1.06 to +6.80, clears zero. Later than two hours into play, +1.54 against -1.15. The second hour is the narrow one, -0.15 against -0.58.

This is the mechanism behind the sports grade study, which finds the cohorts 0.85 points apart before kickoff, the wrong way round, and 2.45 apart in-play. The first hour of a game is when the most information arrives fastest: a lineup, an injury, a red card, a first-inning collapse. A wallet that reads it right, or reads it first, is paid for the read. A wallet that reacts to the score is paying the wallet that saw it coming.

The alert pipeline stored the quoted bid and ask on 60% of this study's buys from April 2 to July 17, 2026. Later trades come from on-chain fills, which carry no order book, so the cost of crossing the book can be measured on the quoted buys only. The median quoted spread was one cent in every sport and in every timing bucket but the earliest, where it was 2 cents. The table has the mean for each bucket.

By sport, before kickoff, the median was one cent everywhere and the 90th percentile was one cent in soccer, the NBA, baseball, and hockey and two cents in tennis, esports, MMA, and the WNBA. So a market order on an even game pays half a cent of spread plus the taker fee its market carries, which together cost less than a sportsbook's standard -110 line, as the chart below shows. NFL and college football games carry 0.03 and NBA games no fee, as of October 21, 2026, so those cost less still. A limit order that rests on the book and fills pays neither.

The early book is the exception that explains the early loss. A buyer a week out pays a 21-cent spread on average, on a book few traders have priced yet.

There is no price path here. This study times buys against kickoff and scores them against the result; it does not follow one market's price from listing to close. Polymarket's hourly price history is thin in the hours before kickoff for most markets in this window, so an open-to-close movement study would be describing the markets that happen to have snapshots.

Kickoff is the scheduled start on the market. In-play buckets are by clock, not game state: a buy 40 minutes into a soccer match and a buy 40 minutes into a tennis match are both first-hour buys. Edge is measured against settlement, so a live buyer who sold at the next goal is scored on a result they did not wait for. Fees and slippage are ignored.

The sample is large buys 0xinsider recorded, and the floor moved: about $3,000 until July 2, 2026 and $10,000 after, so most of the sample sits under $10,000, as the chart below shows. It is not every fill on the platform, and a market can be calibrated for large buyers while retail buyers pay a different price. The grade splits start June 1, grade each buy as the wallet stood at the trade, and are described in the sports grade study.

Ask a casual trader how they did and they will tell you their record. Ask a professional and they will tell you their closing line value: the difference between the price they got and the price the market closed at. Timing is the whole of that number. Bet early and you get a soft line and a wide spread. Bet at the close and you get the market's best estimate and nothing more.

Polymarket sports markets add a third phase. The book does not close at kickoff. It trades through the game, and by the second half the price on a team that is ahead looks nothing like the price before the game. Whether that live price is fair, and whether the people hitting it are the same people who bet before kickoff, is what this study measures.

The NRFI study shows the early half of that on one market. The same no-run bet placed a day before first pitch won when the price then rose and lost when it fell, while bought at first pitch the two came out within two points of each other. The chart below draws it.

The pattern held into the fall, a little less live. In the 30 days from September 11 to October 10, 2026, the NFL, whose season began as this study's window closed, was the biggest sport on 0xinsider's large-trade feed, and tennis was still bet mostly live and soccer mostly before kickoff, as the chart below shows.

Each game on a 0xinsider sports board (the soccer, NBA and tennis boards, and the rest) shows the kickoff countdown, the live score once play starts, and the current price. The board also shows which side the profitable wallets hold as the game goes on, live on Pro and delayed or locked without it. In-play is where this study found their edge.

The large-trade feed timestamps recorded large trades admitted by the feed rules with the wallet's grade beside it, live on Pro and 24 hours behind without it. Watching the live feed during a game is watching the first-hour money this study describes. An alert notifies you in the app and by email when a tracked wallet trades a game you pick, or when a trade matches your filters: 1 alert with an account, 50 on Pro.

Pick of the Day publishes up to twenty-five picks a day with the price each was frozen at, and its list of past picks shows each settled pick's closing line value where a closing price exists. The closing line value guide explains the measure, and the line movement guide covers how to read a moving price before and during a game.

The page draws on dated, read-only runs against the production database, all public in the 0xinsider research repository under polymarket-sports-markets. The sample, the timing buckets, the sport and price splits and their intervals come from the run on a copy of the database at 12:26 UTC on October 10, 2026: timing-2026-10-10.sql, its output and its market-level export, from which bootstrap.py recomputes the intervals without a database. The grade split is the run of October 7, 2026, described below. The stake floor by month and the feed's last 30 days come from the read of the replica at 13:22 UTC on October 11, 2026, in timing-charts-2026-10-11.sql and its output. The closing-line chart is the NRFI study's run at 22:28 UTC on September 13, 2026, in the files whose names start nrfi. The September 12 run, timing.sql and its output, stays beside them as the record of the first version.

The cost-of-a-bet chart is arithmetic, not a query: the 0.05 taker rate most sports game markets carry on Polymarket's schedule, read October 21, 2026, and this study's one-cent median spread in the last hour before kickoff.

The universe is the calibration study's universe, every large-trade alert for a Polymarket buy on a sports-category market from April 2, 2026 with a yes-or-no outcome settled after the trade and a price between 2 and 98 cents, joined to the market's game start time. Hours to kickoff is the start time minus the trade time. The window starts on April 2 because the alert table's outcome index was defective before that day; the calibration study documents it.

The grade split is a later run: on October 7, 2026 it was re-measured with each buy graded as the wallet stood at the trade, on a copy of the production database, in the files whose names start grade-at-trade-2026-10-07 in the same folder. Its intervals group buys by market, by wallet and by game, and show the widest.

Expect your counts to grow as open markets settle. The bucket shares should hold to a tenth of a point, as they did between the September and October runs in the chart below, and the edges should stay within their intervals.