KEY TAKEAWAYS
Sportsbooks set odds by estimating how likely each outcome is, then adjusting the price so the book earns a margin. The estimate typically comes from models and human traders. The margin is built in: the prices on all sides of a market imply probabilities that add up to more than 100%. After the odds go live, the book can move them as bets arrive.
This matters because a price is not a forecast on its own. It is a forecast, a profit margin and a risk decision combined, which is why two sportsbooks can quote slightly different numbers for the same game.
This guide explains how a price is built, how to read the margin inside it, how risk is managed after opening, and what standards say about changing and logging odds. It describes the system, not how to bet, and it names no sportsbooks.
How Do Sportsbooks Set Odds?
A sportsbook typically builds a price in three layers. First, a probability estimate for each outcome, produced by statistical models and reviewed by traders. Second, a margin added on top. Third, a risk layer that sets limits and watches exposure. The exact methods are commercially sensitive, so published descriptions are general.
A few terms help. The line or price is the odds offered on an outcome. A trader is the person (or system) responsible for a market. Handle is the total amount wagered. Liability is what the book would pay out on a given result. The margin, also called the vig or the overround, is the built-in profit.
Academic work describes the key restriction clearly. In a published paper on bookmaker odds, Dominic Cortis shows that odds set by bookmakers should have implied probabilities that add up to at least one, otherwise a bettor could find a guaranteed profit across the outcomes. That is why a fair price of 2.00 on each side of an even market is shaded to something lower, such as 1.90.
Under the Hood: Implied Probability, Margin and Liability
Implied probability converts a price into the chance it represents. For decimal odds it is one divided by the odds. For American odds on a favorite (a minus number) it is the odds without the sign divided by that number plus 100.
favorite: implied probability = |odds| ÷ (|odds| + 100)
where:
decimal odds = the total return per unit staked; |odds| = the American odds without the minus sign.
The overround is the sum of the implied probabilities minus one. Researchers describe it as the sum of the inverses of the decimal odds, and bettors commonly use it to estimate a bookmaker’s margin. A standard US spread is priced at −110 on each side. Each side implies 110 ÷ 210 = 52.38%, so the two sides sum to 104.76% and the overround is 4.76%. If the book takes equal money on both sides, it keeps 1 − 1 ÷ 1.0476 = 4.55% of the amount wagered.
The same arithmetic works for three outcomes. Cortis gives a published football example with decimal odds of 1.86, 3.75 and 4.05. Their inverses sum to 1.0512, an overround of about 5.12%.
Liability is where risk management begins. Suppose, in a fictional market, a book offers 1.90 on both sides and takes $60,000 on side A and $40,000 on side B. If A wins it pays $114,000 against $100,000 taken, a $14,000 loss. If B wins it pays $76,000, a $24,000 profit. Had the money been $50,000 each, it would pay $95,000 either way and keep $5,000.
Cortis notes that bookmakers may adjust odds according to the volume of wagers on different outcomes. Researchers Tadgh Hegarty and Karl Whelan add a caution about reading margins: the simple overround calculation is exact only if the margin is spread evenly, and evidence of favorite-longshot bias shows that long shots typically carry larger losses than favorites.
Walkthrough: The Life of One Pre-Game Market
This is a hypothetical walkthrough with fictional numbers. In a typical sportsbook, one market runs in seven steps:
- Data in: the system receives schedules, results and statistics from its data sources. GLI-33 asks that statistics and line data come from a source the regulator allows and that the operator review their accuracy and timeliness.
- Opening price: a model estimates probabilities, a trader reviews them, and the margin is added. A 50-50 game might open at 1.90 on each side.
- Publish and limit: the price goes live with betting limits that cap the exposure on any one wager.
- Money arrives: bets are accepted and liability by outcome is tracked, for example $60,000 against $40,000.
- Adjust: a trader or automated rule may shorten the heavily backed side to 1.85 and lengthen the other to 1.95.
- Confirm: if a player’s price changes before acceptance, GLI-33 asks for a notification so the player can confirm the new price, unless the player has opted in to auto-accept changes where the regulator permits it.
- Suspend and settle: the market is typically suspended at the start of the event, with the time and reason logged as GLI-33 requires, then settled from a confirmed result.
Variations: Bet Types and How the Margin Changes
The margin looks different depending on the market. The table below uses fictional but typical prices and the arithmetic above. For a parlay, each leg’s payout ratio multiplies: two legs at 1.90 give 0.95 × 0.95 = 0.9025, so the book keeps 1 − 0.9025 = 9.75%, assuming independent legs priced with equal margins.
| Market | Example prices | Overround | Share kept if balanced |
|---|---|---|---|
| Two-way spread | −110 / −110 | 4.76% | 4.55% |
| Two-way price | 1.90 / 1.90 | 5.26% | 5.00% |
| Three-way match result | 1.86 / 3.75 / 4.05 (published example) | 5.12% | 4.87% |
| Two-leg parlay of two 1.90 legs | 1.90 × 1.90 = 3.61 against a fair 4.00 | n/a | 9.75% |
Cortis notes that bookmakers can increase profitability by offering more multiples, also called accumulators. The table shows why: margins compound across legs. In-play markets add timing and data delay to these questions, which is covered separately.
Myths, Limits and Misconceptions
Myth: the odds are the true probability. A price includes a margin, so the implied probabilities exceed 100%. Removing the margin gives only an estimate, and Hegarty and Whelan show the simple method can be off when margins differ by outcome.
Myth: a book always tries to balance the money. Balance is one tool. Cortis shows that keeping wagers in ratio with implied probabilities locks in profit, but also that a bookmaker does not need the exact probability to expect a positive profit, so some exposure can be a deliberate choice.
Myth: a move in the price predicts the result. A price changes because of data, money and risk. It does not reveal an outcome, and the same limit applies to models in general, as explained in why AI cannot predict random gambling outcomes.
Limit: this article gives no picks. It explains how prices are built, not which side to take.
Key Facts and What to Know
The essentials are listed below, with steps a player can legitimately take.
- A price is an estimate, a margin and a risk decision combined.
- The overround equals the sum of the implied probabilities minus one.
- Margins on multi-leg bets compound, so the book’s share rises with the number of legs.
- Under GLI-33 a platform asks the player to confirm a changed price, unless the player has opted in to auto-accept, where allowed.
- Check the price and the confirmation screen before accepting a bet, and read the house rules for how voids and cancellations are handled.
- If a bet is cancelled or settled unexpectedly, keep the bet ID and contact the platform’s support team.
- The GLI-33 requirements cited here come from version 1.1; GLI issued a version 2.0 draft for public comment on June 30, 2026, so details may change.
Related Concepts and Next Steps
For prices that move while the game is under way, read how live betting works behind the scenes. The systems that take bets and hold balances are covered in how an online gambling platform works. A logical next concept is how a cash-out value is calculated from a bet’s current estimated probability. Readers interested in modeling can also follow how data science is applied to gambling and where its limits lie, from probability estimates to risk limits and market surveillance.
Frequently Asked Questions
Sources & References
- Dominic Cortis — Expected Values and Variances in Bookmaker Payouts: A Theoretical Approach Towards Setting Limits on Odds, Journal of Prediction Markets (2015)
- Tadgh Hegarty and Karl T. Whelan — Estimating expected loss rates in betting markets: theory and evidence, Applied Economics (2025)
- Gaming Laboratories International — GLI-33: Standards for Event Wagering Systems, Version 1.1
- Gaming Laboratories International — Executive Summary of Changes for GLI-33 v2.0 Public Comment Draft (issued June 30, 2026)


