Analytics for sports markets: the numbers that matter and how to read them

Wallets graded S, A or B beat the price on Polymarket sports markets by 0.89 points in July 2026 and lost to it by 0.66 in August. In August the D and F wallets beat it by 1.03. Judge the grades by August alone and you'd have followed the worst wallets on the platform.

That is the problem analytics solves. A sports bet has two parts, a probability and a price, and it's good when the probability is higher than the price says. The result arrives later and carries most of the luck. The price you paid is known the moment you pay it.

So keep both numbers in view. Before the bet, check the price against your own probability. After it, grade yourself on the price you got, not on whether it won. A sportsbook writes the price as a line, -110 or +240; Polymarket writes it in cents, 52¢ or 29¢, and the cents are already the probability.

The 8 numbers below come in the order you meet them: implied probability, hold, expected value, closing line value, calibration, the win-rate trap, sample size and timing. Each is measured on 0xinsider's record of large Polymarket sports buys, 416,548 of them for calibration alone.

Every price is a probability in disguise. For negative American odds, divide the odds by the odds plus 100: -110 is 110 divided by 210, or 52.4%. For positive odds, divide 100 by the odds plus 100: +300 is 100 divided by 400, or 25%.

Decimal odds are simpler: one divided by the odds, so 1.91 is 52.4% and 4.00 is 25%. The odds calculator converts either format for every outcome of a game and removes the overround in one step.

A Polymarket price skips the conversion. A contract pays $1 if the outcome happens and nothing if it does not, so a price of 52¢ is a 52% implied probability.

The implied probability is the market's estimate, and the whole question of whether to bet is whether yours is different. Write your number down before you look at the price.

Add the implied probabilities of every outcome and a fair market sums to 100%. Anything above that is what the market keeps.

A standard sportsbook line of -110 on both sides of a spread sums to 104.8%, so the book's hold is 4.55% of every dollar wagered whichever side wins. A three-leg parlay paid at 6 to 1 on three -110 legs sums to 112.5%: the same bets, charged more.

On Polymarket there is no line to shade, so the hold is visible in two parts. The first is the spread between the best bid and the best ask. Across 0xinsider's sports trades from April to July 2026, the median spread at the moment of a large buy was 1¢ in every sport and in every timing bucket inside the final week before kickoff, so a buyer crossing the book at 50¢ pays about half a cent above the midpoint, 1% of the stake.

The second is the taker fee, which Polymarket sets per market from a category default: the sports rate is 0.05 times the share count times the price times one minus the price. At 50¢ that is 2.5% of the stake; at 90¢ it is 0.5%. NFL and college football games carry 0.03 and NBA games no fee, as of October 21, 2026, so read the rate on the market.

So a taker buying both sides of a Polymarket sports market at 50¢ pays about 103.5% of the payout, against 104.8% at a -110 sportsbook, and the top tier of Polymarket's taker rebate program returns half the fee. A resting limit order pays no fee and earns the spread instead, so a maker's hold is zero. The full schedule is in the Polymarket fees guide.

Know your hold at whatever venue you use. It is the edge you have to clear before you are ahead.

Expected value is what a bet returns per dollar, on average, if it were placed many times. For a binary contract it is the probability you believe divided by the price you pay, minus one.

Believe 55% and pay 50¢ and the expected value is plus 10%: a dollar staked returns $1.10 on average. Believe 55% and pay 60¢ and it is minus 8.3%.

Implied probability gives you the price side. Your model, your read, or the sharp money gives you the probability side. Hold tells you how far above zero the expected value has to sit, before fees, for you to be ahead.

A bet with positive expected value can lose, and a bet with negative expected value can win, and neither result tells you which kind of bet it was.

Refuse a bet whose expected value you cannot state. If you believe the home side is 55% and the price is 55¢, there is no bet, however strongly you feel about the team. If the price is 48¢ there is a bet whether or not you like the team.

Closing line value is the difference between the price you paid and the price the market settled on at kickoff. Buy at 48¢, watch the market close at 55¢, and you have 7¢ of closing line value whether or not the team wins.

The chart above is one real case. The Falcons cost 26.5¢ a week before they played the Packers on September 24, 2026, and 31.5¢ in the last hour before kickoff. A buyer a week out had 5¢ of closing line value before a snap, and the Falcons won. Closing line value is how professional bettors measure themselves week to week. Results take a season to mean anything. The price does not.

The closing price is the market's best estimate, with every lineup, injury and sharp opinion priced in. Buy below it consistently and you are buying outcomes for less than they are worth, which over enough bets turns into profit. Buy above it consistently and you are the bettor the professionals are trading against.

The formula, the sign convention and the fee adjustment are in the closing line value guide. Pick of the Day records the entry price and, where captured, the closing price for each pick, so its closing line value is public before the result is known.

Across 416,548 large Polymarket sports buys from April 2 to September 11, 2026 (the smallest about $3,000 until July 2, 2026 and $10,000 after), buyers paid 60.6¢ on average and won 60.9% of the time.

A market is calibrated when outcomes priced at 60¢ happen 60% of the time. Split into 10 price buckets, every bucket from 20¢ to 98¢ sat within two points of its price, with confidence intervals that include zero. The one lean was buys under 10¢, which won 3.53% of the time at an average price of 6.50¢ and, weighted by stake, returned 44 cents on the dollar. Its interval reaches zero too.

Check calibration first in any market you plan to bet in. If the prices are systematically wrong in one direction, that direction is a free edge.

On a Polymarket sports market they are not. The one lean is the longshot under a dime, and it is a lean, not an established result. There is no line to beat. Whatever edge you have has to come from knowing something the price does not, or from knowing who is on the other side. The full numbers are in the favorite-longshot bias study.

Win rate is the number everyone quotes and the one that says least. A trader who only backs 90¢ favorites wins 90% of the time and has demonstrated nothing, because the market already priced those outcomes at 90%. A trader who backs 30¢ underdogs and wins 35% of the time has beaten the market by 5 points and looks, on win rate alone, like a loser.

The chart above is the same 416,548 sports buys, with the average price paid in each bucket beside the share of buys that won. The two bars match all the way up. Win rate is the price you like to pay.

The gap between the two bars is the calibration edge. 0xinsider displays it on trader profiles alongside the historical grade and other performance metrics.

When someone shows you a record, ask for the average price. A 60% record at an average price of 55¢ is a 5-point edge. At 62¢ it is a 2-point loss.

A trader whose true win rate at even money is 55%, which is an excellent rate, will over 100 bets post anything from 45% to 65% about 19 times in 20. Over 500 bets the range is 50.6% to 59.4%. It takes roughly 2,500 bets before the measured rate sits within 2 points of the truth, and a 2-point edge is an excellent edge.

The binomial formula gives those ranges, and it is the reason results are a poor guide and prices a good one. Closing line value shows through in a few dozen bets because the closing price is a fact, not a sample. Win rate needs thousands.

Any tipster, tout or wallet with a 200-bet record is showing you a sample too small to distinguish skill from a good month.

Read a wallet's grade beside its resolved-market count. A short record leaves more uncertainty than a comparable long one. The confidence scoring guide explains sample size and confidence.

Buys placed 1 to 7 days before kickoff lost to the price by 2.19 points, and the confidence interval clears zero. From a day out through the final whistle every bucket was calibrated.

0xinsider timed 401,482 large Polymarket sports buys against kickoff, and half of them landed after the game had started. Early money was paying an average quoted spread of 21¢ a week out against 1¢ on the day, and it was wrong more often than the price said.

The same bet costs a different price at different times. A sportsbook line opens soft and tightens toward kickoff as money and information arrive. A Polymarket sports market does the same and then keeps trading through the game.

So your closing line value depends on when you bet, and the data says early is expensive. The timing study has the numbers, and the line movement guide covers how to read a moving price.

Across 161,128 sports buys from June to September 2026, each graded as the wallet stood when it traded, wallets graded S, A or B came in at +0.50 points against the price they paid and wallets graded D or F at -0.18, both within noise.

The gap is an in-play gap: before the game the cohorts were within a point of each other (-0.48 against +0.37), and in-play they were 2.45 points apart. The study corrected these figures on October 7, 2026; its first version used grades written after the trade and reported plus 1.25 and minus 1.21.

Every bet has a counterparty, and on Polymarket the counterparty has a public record. 0xinsider grades every wallet from S to F on its settled history, so a sports board shows which side the profitable wallets are on for a given game and how much they have on it, live with Pro and on a delay without Pro. No sportsbook shows its customers that. The full numbers are in the sports grade study.

Use it as a probability input, not as an instruction. A game where the S, A and B wallets are 70% on the home side at 52¢ is a game where the graded money agrees with your 55%. A game where they are on the other side is a game to look at again.

The Kelly criterion gives the stake that grows a bankroll fastest for a known edge: edge divided by the odds. For a 55% belief at a 50¢ price, that is 10% of the bankroll.

Nobody knows their edge to the point, so most professionals bet a quarter to a half of Kelly and accept slower growth for a smaller chance of ruin. An edge only pays if you are still betting when it arrives, and the sample-size math above says it arrives slowly.

The chart above runs that trade-off on our public Pick of the Day record: 443 picks that won 63.9% at an average 60.29¢, where Kelly says 6.4% a pick. Picks sharing a kickoff use the same prior balance; stakes are reduced equally if their sum exceeds it. At 6.4%, the bankroll reached 3.35 times the start by October 2, fell 67.7% by October 7 and finished 71.0% up. At 2% it finished 36.9% up, with a 27.6% deepest fall. Requesting 20% per pick lost the bankroll. Successive kickoff groups are assumed settled before the next, and each pick pays the taker fee its market carries on Polymarket's schedule read October 21, 2026: 0.03 on NFL and college football games, none on NBA games, 0.05 on the rest. Until that date the replay charged a uniform 0.05: Kelly said 6.2%, and that stake finished 68.6% up after a 66.5% fall.

Size by rule, never by feel, because a feeling is not a probability. And never let a bet's size depend on how the last one went. The Kelly criterion guide works through the formula with Polymarket prices.

Corrected on October 17, 2026: three Pick of the Day picks, from June 23, September 2 and October 1, 2026, were withdrawn from the record because they were published after their games had started. The figures here now leave them out. This section said 446 picks at an average 60.43¢ and a Kelly stake of 5.9%, which reached 3.11 times the start by October 1 and finished 61.3% up; at 2% the bankroll finished 33.9% up, with a 28.6% deepest fall.

F wallets placed more of the large sports buys in 0xinsider's feed than any other grade: 3,296 of 8,077 from September 20 to October 10, 2026, from 277 wallets. S wallets placed 1,714, from 62. The feed is mostly the best and the worst wallets on the platform. Each S wallet trades more often; there are more than 4 times as many F wallets.

Implied probability is the price on every sports board: the NFL, NBA, soccer, tennis, MLB, NHL, UFC and esports boards. Each board shows the moneyline, spread and total markets for each game, the kickoff countdown, and the live score once play starts.

Calibration edge and win rate sit side by side on every trader profile, with the realized P&L, the positions the wallet holds now, and the grade that combines them. The leaderboard ranks wallets by that record, and the trader grades guide covers what each letter means.

Who is on the other side is the Profitable wallets block on each game, and the large-trade feed, which shows every large buy with the buyer's grade beside it, live with Pro and 24 hours later without Pro. Timing is the timestamp on each alert. Closing line value is on Pick of the Day for each pick whose closing price was captured, and the parlay side of the market is the combos ledger.

So read the letter before the dollar figure. A large buy in the feed is more likely to come from an F wallet than from any other grade, and size alone tells you nothing about which kind it is.