How Do Sportsbooks Set Odds?

A sportsbook's odds are an opinion about probability with a fee baked in. Somebody — an in-house trader, a model, or a data supplier the book pays — estimates how likely each outcome is. The book converts that estimate into a price, then shades the price so that both sides together imply more than 100%. That excess is the margin, and it is how the book gets paid whether or not its opinion was right.

Most US sportsbooks do not build that opinion from scratch for every game. A small number of books and suppliers set a price that the rest of the market treats as the reference, and the retail books follow it within minutes, adjusting for their own customers. The skill in bookmaking is less about the first number than about how the book manages the money that arrives after it.

Who actually makes the first number?

Three kinds of operation produce prices, and a given book may use more than one.

In-house trading teams. The largest operators employ traders and quantitative modellers who build power ratings, simulate games and price markets directly. These books can hang a line before anyone else and live with the consequences.

Data and odds suppliers. Many books license prices. Sportradar, a supplier that files annual reports with the SEC, describes its Managed Trading Services in its fiscal-2025 annual report (Form 20-F) as "a sophisticated trading, risk, and liability management solution" that embeds its odds services. The filing spells out the mechanics: client books forward their customers' bet slips to Sportradar, which can accept or decline each one, and on the slips it accepts it takes a share of the client's revenue, or of its loss. So on some sites the price you see — and the decision whether your bet is taken at all — is partly made by a vendor you have never heard of. The same filing says the tools let operators differentiate their odds and manage liabilities according to "rules and thresholds that they control," which is why two books running the same feed can still show different numbers.

Market-making books. A few books aim to be the most accurate price in the market rather than the most profitable per bet. They run thin margins, take large wagers, and treat winning customers as a source of information. Pinnacle is the best-known example. Its own site sells "low margins" and says it welcomes bets that are part of an arbitrage strategy because it is confident in its ability to "price markets correctly." A book that invites arbitrageurs is telling you its price is meant to be the reference, not the target.

Why do most books' odds look the same?

Because most of them are copying. Tobias Moskowitz's 2021 study in the Journal of Finance, covering 117,442 betting contracts on 59,592 NFL, NBA, MLB and NHL games, names Pinnacle, 5Dimes and BetCRIS as the "market setting" books of his sample period and notes that other books "move on air": when a market-setter moves, the rest follow without having taken significant bets themselves. In his data, opening lines across books were nearly identical — they differed less than 1% of the time.

That was true a generation earlier as well. Steven Levitt's 2004 paper in The Economic Journal observed that individual bookmakers were not really setting prices but "following the lead of a handful of influential odds makers" paid by the large Las Vegas casinos.

Copying is rational. A retail book's edge is marketing, promotions and a customer base that bets for entertainment. Its trading desk is a cost centre. Borrowing a sharp reference price and adding its own margin is cheaper than employing enough modellers to price 15 leagues independently, and safer than being the one book with a stale number.

How is the margin added?

Start with a fair coin-flip game: each side wins 50% of the time. Fair American odds on both sides are +100. The book instead posts -110 on both sides.

Convert -110 to implied probability: 110 ÷ (110 + 100) = 52.38%. Both sides together imply 52.38% + 52.38% = 104.76%. The 4.76 points over 100 is the overround.

What the book expects to keep is slightly smaller than that. If it takes $110 on each side, it holds $220 and pays the winner $210 (stake plus $100). It keeps $10 of $220 staked, or 4.55%. The general formula is hold = 1 − 1 ÷ (sum of implied probabilities) = 1 − 1 ÷ 1.0476 = 4.55%.

It is also why a bettor at -110 needs to win 52.4% of the time just to break even — the figure Levitt uses for the standard 110-to-win-100 bet.

The margin does not have to be split evenly. If the book's number says one side is a 60% shot and the other 40%, fair odds are -150 and +150. The book might post -165 / +140:

• -165 implies 165 ÷ 265 = 62.26%
• +140 implies 100 ÷ 240 = 41.67%
• Total 103.93%, so hold = 1 − 1 ÷ 1.0393 = 3.78%

Books routinely load more of the margin onto the side they expect the public to bet. That is a pricing decision, not arithmetic, and it is invisible unless you strip the margin out yourself. The Academy's guide to removing the vig from a price shows the methods.

What happens after the line opens?

The opening number is a draft. From there, three inputs revise it.

Bets. Every accepted wager tells the book something, and not all wagers say the same thing. A book tracks who is betting: a limit bet from an account with a long record of beating closing prices is treated as information and the line moves; the same amount from a recreational account may move nothing. Books that license trading from a supplier get this filtering from the vendor's accept/decline system as well as their own.

News. Injuries, confirmed lineups, weather and suspensions change the true probabilities. The price moves whether or not anyone has bet yet. The next guide in this series, why betting lines move, walks through a move from start to finish.

Other books. A retail book watching a market-maker move by half a point will usually follow within minutes, because leaving a stale price up invites anyone faster to bet into it.

Those moves were historically small and infrequent. Levitt's casino-line data, published in 2004, show the posted NFL spread changed on average 1.4 times in the five days before kickoff, and 85% of those changes were the minimum half point. Today's online markets update far more often, but the principle is the same: most of the price is fixed at the open, and later moves are refinements.

Do books try to get equal money on both sides?

Less than the folklore says. The textbook model has the book balancing its liabilities so it keeps the margin whatever the result. Levitt tested that with roughly 20,000 NFL picks placed by 285 bettors in a high-stakes handicapping contest at an online book during the 2001–02 season, and found the opposite: in the median game almost two-thirds of the bets were on one side, and the book's spreads showed no sign of being set to even that out. He concluded that bookmakers deliberately take positions — pricing favourites slightly too high because customers prefer them — and that doing so raised gross margins by 20–30% over a balancing strategy.

A modern retail book with millions of customers and hundreds of games a week can carry exposure on individual games, because the results average out across the volume. What it cannot afford is being wrong in a way sharp bettors can repeat. That is the real reason lines move: not to balance the book, but to stop feeding informed money a bad price.

What this means for the person placing the bet

Three consequences follow.

First, the book's price already contains an opinion and a fee. To win over time you need to be right more often than the price implies by more than the fee. Our Academy explainer on implied probability covers the conversion.

Second, the most accurate number in the market is not necessarily the one on your phone. It is the market-maker's price, especially close to game time. That is the subject of opening vs closing lines.

Third, retail books copy the reference price, but they do not all copy it at the same speed or add the same margin. The gaps between them are small and short-lived, and the books know it — which is why accounts that consistently take those gaps tend to get their limits cut.