Favorite-longshot bias explained: why longshots cost more than they pay

Across 4.6 million Polymarket buys of every kind from July 15 to September 30, 2026, every 10¢ band from 50¢ to 90¢ won a little more often than its price said, and every band under 50¢ won a little less. That is the favorite-longshot bias, the oldest pattern in betting data, and on Polymarket no band missed its price by a full point.

Sort every bet at a venue by its odds and count how often each group won. In a fair market, 10% shots win 10% of the time and 80% shots win 80% of the time. In every racetrack and sportsbook dataset published since Richard Griffith's 1949 study, they don't: longshots win less often than their price says and favorites win more often.

The return is where it hurts. The house margin applies to everything, so both groups lose on average, but longshots lose far more. A typical racetrack finding is that favorites return about 95 cents on the dollar while horses at 100 to 1 return 20 or 30 cents. At a sportsbook the shape is gentler: heavy favorites come close to break-even before the vig, and big underdogs and long parlays lose several times the vig.

On Polymarket the tilt is real and too small to trade on its own. Only 3 of the 10 bands in the chart above have an interval that leaves out zero, all of them on the favorite side. The first chart on this page draws the sports data against a fair price; the sections below find the one place the tilt gets bigger.

Arithmetic first. A +900 underdog carries an implied probability of 10%. Say the bias is at work and it wins 7% of the time. 100 bets of $100 win $900 on 7 of them and lose $100 on the other 93: $6,300 won against $9,300 lost, minus 30 cents on the dollar. You thought you were paying 10¢ for a 10% chance. You were paying 10¢ for a 7% chance.

Now the favorite on the other side. If the underdog's true chance is 7%, the favorite's is 93%, and -900 charges 90% for it. Over 100 bets of $100, the 93 wins pay $11.11 each and the 7 losses cost $100 each: plus $333 on $10,000 staked, a 3.3% return before the vig. After a typical vig it's roughly break-even, which is what the studies find for heavy favorites.

The chart above is the same example with real money. Large Polymarket sports buys at 5 to 10¢ paid 7.62¢ on average and won 4.27% of the time: the example's 3-point miss, at a price where 3 points is close to half of what you paid for. Favorites at 90 to 98¢ missed by about half a point, and half a point of 93 is nothing.

Same mistake, different bill. Misjudging a probability costs you a share of your stake, and the smaller the probability, the bigger the share.

On Polymarket sports markets, the best-graded wallets put 29% of their large buys on favorites at 80¢ and up. The worst-graded put 19% there and 42% on coin flips between 40 and 60¢. Neither group bought many longshots: 2% and 3% of their buys were under 20¢.

Three explanations for the bias have held up. The most tested is that people overweight small probabilities: a 3% chance feels like more than 3%, so a payout at 33 to 1 feels like a better deal than it is. It shows up in lotteries, in insurance and in every experiment that asks people to price a small chance of a big outcome.

The second is a taste for the payout itself. A bettor with $20 and an evening to fill gets more from a 20-to-1 shot that might pay $400 than from a favorite that might pay $22, and accepts a worse price for it. Sportsbooks know this, which is why the bias is steepest in parlays and long underdogs, the products with the biggest multipliers.

The third is who bets what. Informed bettors find their value in favorites and mid-priced outcomes, where a small misprice on a large probability is worth trading, and casual bettors drift to longshots. So the longshot price is set by the less informed side, which is also why a bookmaker can shade it without losing sharp business.

The chart above backs half of the third explanation. The well-graded money leans to favorites, but the poorly graded money doesn't chase longshots; it crowds into coin flips. Large buyers of every grade leave the cheap end mostly alone. None of the three explanations needs a bookmaker, so a market with no house line should still show the bias, and the next sections test it.

One market can make a whole price bucket look like a bias. In the chart above, the 10-20¢ bucket's interval is 24.7 points wide, 4 times the next widest, and a single market did that. The next section names it.

The test itself is simple. Sort the bets by the price paid into buckets, 10¢ wide is standard, and compare two numbers in each: the average price paid and the share that won. The difference, in percentage points, is the bucket's edge. A fair market has every bucket near zero. The bias shows as negative edge at the cheap end and positive edge at the expensive end.

Report the dollar return too. Points understate the longshot problem: a 7¢ contract that wins 4% of the time is only 3 points mispriced, and it returns 57 cents on the dollar. Weighted by stake and held to settlement, the dollar return is the number you feel.

Then put an interval on each bucket and cluster it by event, not by bet. 500 bets on one longshot that comes in are one bet that came in. Count them as 500 independent results and you'll find a profitable longshot bucket that is really one tournament. You need thousands of bets and many events per bucket before the shape means anything.

Polymarket sports markets have no bookmaker. A contract pays $1 if the team wins, and the price is set by buyers and sellers on an order book. 0xinsider ran the test above on 416,548 large buys (the smallest about $3,000 until July 2, 2026 and $10,000 after) across every Polymarket sports market that settled yes or no between April 2 and September 11, 2026: soccer, esports, the NBA, tennis, baseball, hockey, cricket, MMA, the WNBA and golf.

Across the whole sample buyers paid 60.6¢ and won 60.9% of the time. Bucket by bucket, buys from 20¢ to 98¢ sat within two points of their price, with confidence intervals that include zero. Buyers at 80 to 90¢ paid 84.7¢ and won 84.3%. Buyers at 40 to 50¢ paid 45.4¢ and won 46.1%. The market was calibrated, which is a different result from every sportsbook study.

The 10 to 20¢ bucket shows a dollar return of plus 28.8%, and it is the clustering lesson from the last section. One market, Spain to win the 2026 World Cup, drew 490 large buys at an average of 15.2¢, and its buyers made $41.3 million when Spain won. The rest of the bucket, 1,576 markets, lost $14.1 million between them. The bucket's interval runs from minus 4.98 to plus 19.73 points, and the honest reading is zero. The full numbers are in the favorite-longshot bias study.

The one bucket where the textbook bias leans is the cheapest. Buys under 10¢ paid 6.50¢ on average and won 3.53% of the time: minus 2.97 points, with an interval from minus 5.22 to plus 0.81 that reaches zero. Those 2,040 buys put $28.3 million into contracts and got back $12.5 million. A dollar staked returned 44 cents.

In 5¢ steps the bottom is worse: buys between 2 and 5¢ paid 3.19¢ and won 1.35% of the time. From 10¢ upward the win rate catches the price.

These buys are the bias without the bookmaker, because nobody shaded these prices. Buyers staking about $3,000 or more chose to pay 6.5¢ for outcomes that arrived 1 time in 28, because 6.5¢ buys a 14-to-1 payout, and the payout is what they were buying. Overweighting small chances and liking big payouts explain it on their own.

At a sportsbook, favorites are the better of two bad bets: still negative after the vig, but far less negative than longshots. On Polymarket sports markets, favorites aren't a bad bet before fees at all. Buys at 80 to 90¢ returned plus 0.43% on the dollar and buys at 90 to 98¢ minus 0.49%. Buys at 70 to 80¢ returned plus 3.64% across 49,786 buys.

Split by sport, favorites at 70¢ and up were calibrated in soccer, tennis, esports, hockey and cricket and slightly negative in the NBA and baseball. MMA was the exception: fight favorites paid 80.4¢ and won 61.6% of the time, minus 19 points, and two UFC upsets carried most of the loss. MMA has 333 settled markets in the sample against 11,813 for soccer, so one card moves the whole sport. It is a lesson about small samples more than about fighters.

The fee does what the vig does, but smaller and visible. Polymarket's sports taker fee at 90¢ is 0.5% of the stake and at 95¢ is 0.25%; a resting limit order pays none. A 90-to-98¢ favorite that returns minus 0.5% before fees returns about minus 0.8% after them for a taker. A sportsbook favorite at the same probability carries the full 4.55% hold.

The calibration in sports is not a property of the platform. Run the same buckets over large buys on politics, crypto, geopolitics, culture, finance, world and business markets in the same window, and 6 of the 10 buckets miss their price by 3 points or more, in both directions. In sports, only the bucket Spain distorted did.

Those markets are thinner, settle over months rather than hours, and a surprising summer in politics moves every bucket at once, so the comparison says less about mispricing in the abstract than about where a large buyer's price has matched the result. In sports it has. The result is consistent with the third explanation above: sports markets on Polymarket are deep, trade constantly, and attract the informed side of the market on both ends of the price.

Don't buy the contract under 10¢ for the payout. Those buys were 0.3% of the money in this sample and lost more dollars than any other band: $15.8 million of the $28.3 million staked. The gain at 10 to 20¢ in the chart above is the Spain market, not a strategy. If you believe a 5¢ outcome has a 10% chance, that's a bet. If you want the 20-to-1, that's a lottery ticket at a bad price.

Stop expecting a favorite premium. At a sportsbook, backing heavy favorites is a way to lose less. On a Polymarket sports market the favorite is priced at its probability, so it's neither a way to lose less nor a way to win. Your edge has to be a probability the market hasn't priced yet, or a counterparty who is wrong. To judge the one price in front of you, use the five checks for a fair price.

And test a market before you trust a rule from another one. The bias is a fact about bettors: it appears where the people buying longshots outnumber the people selling them, and it fades where informed money prices both ends. A trader who brought the sportsbook shape to Polymarket sports would have spent April to September fading favorites that were fairly priced.

One of the 8,077 sports buys in 0xinsider's large-trade feed from September 20 to October 10, 2026 was under 10¢. The feed admits buys of $10,000 or more, ten times the study's floor, and at that size almost nobody buys the longshot. When someone does, it's rare enough to be worth a look.

The large-trade feed shows each of those buys with the buyer's grade beside it, live with Pro and 24 hours later without Pro. A wallet's trader profile shows its calibration edge, the gap between the prices it paid and how often it won, which is the test above applied to one wallet. The leaderboard ranks wallets by that record.

On each sports board, the soccer, NBA, UFC and the rest, the Profitable wallets block shows which side the profitable wallets are on for each game and how much they have on it, on a delay without Pro. If the well-graded wallets are on the 15¢ side, read the game again. If they're absent, the longshot price is what it says.