How Do Bookmakers Set Odds? The Balanced Book Is a Myth

TL;DR: Bookmakers do not set odds to balance the money on each side. Research on 19,770 wagers found that in the median game, around two thirds of the money landed on one side, and the operator left it there on purpose. The price is chosen to attract the side bettors already favour, then the margin is engineered upward through product mix. On an exchange, nobody chooses the price at all.

The version you were taught

Ask most people how a bookmaker works and you get the same answer. The bookmaker sets a line, money comes in on both sides, and the line moves until the two sides are even. Whoever wins, the operator collects a commission from the losers and walks away flat. No exposure, no opinion, no risk.

Every introductory guide to sports betting repeats some version of this. It is a comforting story, because it casts the bookmaker as a neutral clerk. It is also wrong, and the evidence has been sitting in the Economic Journal since 2004.

What 19,770 wagers showed

Steven Levitt got hold of something researchers almost never see: prices and quantities. An online sportsbook ran a high-stakes handicapping contest during the 2001 to 2002 NFL season, 285 entrants paying $250 each, five picks a week against the spread. Levitt could see every pick, the line it was made at, and the result. Nearly 20,000 wagers across 242 games.

If the balanced-book story were true, the money on each game would cluster near a 50-50 split. It did not come close. In only a fifth of games did the preferred side attract between 50% and 55% of the wagers. In the median game, close to two thirds of the money went one way. In roughly one game in ten, more than 80% landed on a single side.

That alone might mean the bookmaker was trying to balance and failing. Levitt tested that too. If the operator were doing its honest best, no observable feature of a game would predict which way the money leaned. In practice, whether a team was at home, which week of the season it was, and which club was involved all predicted the split, and together explained about half the variation. The imbalance was not noise the bookmaker could not see. It was a feature the bookmaker declined to remove.

The number the operator was protecting

Bettors like favourites. Across Levitt's sample, 60.6% of wagers went on the favourite. When the favourite was the away side, it climbed to 68.2%.

So the operator shaded the line. Over 21 seasons and 4,793 games, favourites covered the spread 48.2% of the time. Away favourites, the ones drawing more than two thirds of the money, covered 46.7%. The side attracting the most money was the side least likely to win.

Levitt then did the arithmetic. Under the standard vig, where you risk 110 to win 100, a bookmaker holding a genuinely balanced book earns 5.0% of money staked. Shading the line so bettors win 49.45% of their wagers instead of 50% lifts that to 6.16%. A 23% increase in gross profit, produced by a distortion small enough that no casual bettor would notice it.

The distortion has a ceiling, and the ceiling is arbitrage. Push the line far enough and a mechanical strategy of taking the unpopular side turns profitable. Levitt's conclusion was that operators had pushed right up to that boundary and stopped there.

What the research does not claim

Levitt found no evidence that any bettor in his sample beat the market over time. Success in the first half of the season predicted nothing about the second half. He was careful about the caveat, since a third of entrants dropped out and the attrition was not random. Later work by Krieger, Fodor and Stevenson, using a comparable contest across 2004 to 2010, found the profitability gain from strategic line-setting had shrunk considerably, and that favourites had stopped underperforming so reliably.

So the honest reading is narrower than "bookmakers always win." The finding is about who chooses the price and why. When one party sets the number unilaterally and has better information than the people trading against it, that number is a commercial decision before it is a forecast.

The margin is a product decision, and it has been rising

The 5% vig Levitt modelled belongs to a simpler era. Read what the largest operator in the world tells its own shareholders.

Flutter reports a metric called structural revenue margin: the share of money staked it expects to keep before results run hot or cold. In its US business that figure reached 14.2% across 2025 and 15.5% in the fourth quarter. Across its international business it sat at 16.6%. Management has told investors it is working toward a 16% structural hold over the longer term.

Flutter also says where the increase comes from. Not from sharper forecasting. From parlays and same-game parlays, where the margin compounds across every leg. Every product decision that nudges you from a single selection toward a six-leg builder raises the hold. The margin is not a fixed cost of doing business. It is a dial, and it has been turning one way.

The ceiling is your balance

Here is where the model runs into its own arithmetic, in the operator's own words.

In February 2026, Flutter told shareholders that US handle growth had come in behind expectations because of what it called an unfavourable recycling impact. Persistently high gross margins during the NFL season had reduced customer activity and betting volumes across the market. The company noted its own margin ran 470 basis points above the rest of the market in December, that it finished the season 100 basis points ahead of its expected margin, and that it saw higher churn as a result. Guidance for 2026 opened by attributing weak trading to the very high margins achieved in the closing weeks of the previous quarter.

Read that again, because it is the most honest description of the business anyone has published. Take too much and customers run out of money and stop. The operator's revenue depends on you losing at a rate slow enough that you can afford to keep going.

That is a hold-optimisation problem. It has nothing to do with forecasting a football match.

What price formation looks like with nobody in charge

An exchange removes the pricing decision from the operator.

On a sports prediction exchange, you post an order at the price you believe is right, and it either matches against someone else's order or it sits on the book. A contract trades between one cent and 99 cents, and that number is its implied probability. Nobody shades it, because there is nobody with a position to protect. The exchange takes a commission on matched volume and holds no view on the outcome.

Two consequences follow, and both are structural rather than promotional.

The price reflects what the marginal trader will pay right now. If it drifts away from consensus, the traders who disagree take the other side and pull it back. That is the same continuous double auction that runs equities, commodities and every liquid derivatives market.

And a trader who consistently prices better than the rest of the book is taking money from other traders, not from the operator. The exchange has no exposure to restrict. Levitt spotted this in a footnote in 2004: an early exchange-model platform was charging under 1% for matching buyers and sellers, against the traditional vig, and taking no position on outcomes.

What to do with this

Three things follow for anyone trading sports seriously.

Stop treating the posted price as a forecast. It carries information, since the people setting it are good at their job, but it also carries a deliberate lean toward whichever side the crowd prefers. The gap between those two things is where your work goes.

Check the margin before you check the price. Add up the implied probabilities across every outcome in a market. Anything over 100% is the operator's cut, and on a multi-leg product it will be far larger than you expect.

And know which question you are answering. On a sportsbook you are asking whether you can beat a number that was set by a specialist and shaded against the popular side. On an exchange you are asking whether your read is better than the person taking the other side of your order. Those are different problems, and only one of them has an operator in the middle with a reason to want you wrong.

FAQs

Do bookmakers balance their books? Mostly not. Analysis of nearly 20,000 wagers on NFL games found that in the median match, about two thirds of the money sat on one side, and observable factors such as home advantage predicted the imbalance. If operators wanted an even book, that information was available to them.

If they do not balance the book, how do bookmakers make money? By pricing so the popular side wins slightly less than half the time. Levitt calculated that shading the line to a 49.45% bettor win rate lifted expected gross profit from 5.0% to 6.16% of money staked, a 23% increase over balancing the book, at minimal risk across a full season.

What margin do bookmakers hold on sports? Flutter reported a structural revenue margin of 14.2% in its US business across 2025, 15.5% in the fourth quarter, and 16.6% internationally, with a stated longer-term target of 16%. The company attributes the increase to parlay and same-game parlay products rather than to better forecasting.

Why do favourites cover the spread less than half the time? Because bettors prefer them. Across 21 NFL seasons and 4,793 games, favourites covered 48.2% of the time, and away favourites 46.7%. Operators shade the line toward the side attracting the money, then stop just short of the point where taking the unpopular side becomes mechanically profitable.

How does a sports prediction exchange price differently? No one sets the price. Traders post buy and sell orders on a shared order book, and a contract trades between one cent and 99 cents, its implied probability. The operator earns a commission on matched volume and holds no position, so it has no reason to lean the price either way.