Converting probability estimates into prices, applying margin correctly, and recovering true probability from a market.
In this lesson:
- Convert freely between odds formats and implied probability
- Calculate overround for markets of any size and express margin per outcome
- Remove margin from a set of prices to recover estimated true probability, using both proportional and power methods
- Explain favourite-longshot bias and how margin is deliberately distributed unevenly across a market
Every price contains two things
The central analytical move in sportsbook pricing is separating a price into its two components: an estimate of probability, and a margin applied to that estimate.
A compiler arrives at a view that a team has a 45% chance of winning. That view might come from a model, from expertise, from the market, or from a combination. The fair price corresponding to 45% is decimal odds of 2.222, since 1 divided by 0.45 gives 2.222. A book offering that price expects to break even over time.
The sportsbook does not offer 2.222. It offers something shorter, perhaps 2.10, which implies a probability of 47.62%. The difference between the genuine estimate and the offered implication is the margin.
Everything in this lesson follows from that separation. Analysing a market means asking what probabilities are being implied, how much margin has been added, and how that margin has been distributed.
The conversions
Working fluently between formats is a basic requirement, so the conversions are worth stating precisely.
Decimal to implied probability: divide 1 by the decimal odds. Odds of 2.50 imply 1 ÷ 2.50 = 0.40, or 40%.
Implied probability to decimal: divide 1 by the probability expressed as a decimal. A 40% chance gives 1 ÷ 0.40 = 2.50.
Fractional to decimal: divide the numerator by the denominator and add 1. Odds of 7/2 give 3.5 + 1 = 4.50.
American to decimal: for a positive figure, divide by 100 and add 1, so +250 gives 3.50. For a negative figure, divide 100 by the absolute value and add 1, so -150 gives 1.667.
The habit worth building is converting to implied probability immediately whenever assessing a price, because probability is the only representation in which prices across a market can be meaningfully summed and compared.
Overround
Take every outcome in a market, convert each price to implied probability, and sum them. In a market with no margin, the sum would be exactly 100%, since one of the outcomes must occur. In any real market it exceeds 100%, and the excess is the overround.
Consider a football match priced at 2.10, 3.40 and 4.00 for home, draw and away. The implied probabilities are 47.62%, 29.41% and 25.00%, summing to 102.03%. The overround is therefore just over 2%, which is a tight market, typical of a major fixture where competition is intense.
Now consider a first goalscorer market with forty players priced. Each carries some margin, and the accumulated total might produce an overround of 130% or considerably more. The same principle applies; the arithmetic simply produces a much larger figure because there are many more outcomes each contributing.
This is the mechanical reason exotic markets are more profitable than headline ones, and it explains a great deal about how sportsbooks construct their product offering.
Overround is not the same as margin on turnover
A distinction that is frequently muddled and matters commercially.
Overround expresses how much the implied probabilities exceed certainty. Margin as a percentage of turnover expresses what proportion of money staked the book expects to retain. They are related but not identical, and the second is always smaller than the first.
For a market with overround of 102.03%, the expected retention on turnover is approximately the overround excess divided by the overround itself, so 2.03 ÷ 102.03, giving roughly 1.99%. For a market with 130% overround, the expected retention is 30 ÷ 130, approximately 23%.
The distinction matters because turnover-based margin is what appears in financial reporting and what determines actual revenue. Quoting overround as though it were margin on turnover overstates profitability, modestly at low overrounds and substantially at high ones.
A further caution: expected retention assumes money is distributed across outcomes in proportion to their implied probabilities. Real money is not distributed that way, which means realised margin differs from theoretical margin even before results are considered. This is examined properly in the liability lesson.
Distributing margin across a market
Margin is rarely applied evenly, and the pattern of its distribution is one of the more interesting aspects of pricing.
The empirically observed pattern, documented across many sports and markets, is favourite-longshot bias: outcomes at long odds carry proportionally more margin than outcomes at short odds. A heavy favourite might be priced with very little margin while a rank outsider in the same market is priced with a great deal.
Two explanations are usually offered and both have merit.
The first is pricing uncertainty. Estimating whether something has a 2% or a 3% chance is proportionally far harder than estimating whether something has a 60% or a 61% chance, even though the absolute difference is smaller. A one-point error at long odds is a fifty percent error in relative terms. Wider margin at long odds compensates for that uncertainty.
The second is customer behaviour. Customers demonstrably accept worse value on outcomes offering large returns, because the appeal of a large potential payout outweighs sensitivity to price. Sportsbooks price accordingly, applying margin where it will be tolerated and competing hardest where customers actually compare.
There is a third, more prosaic consideration. Applying uniform margin to a long list of outcomes produces prices at the extreme tail that are commercially awkward, and rounding conventions at long odds are coarse, so precision is limited regardless.
Recovering true probability
A common analytical task is working backwards: given a set of offered prices, what probabilities did the compiler actually believe?
The simplest method is proportional removal. Divide each implied probability by the overround. In the football example, 47.62 ÷ 1.0203 gives 46.67%, 29.41 ÷ 1.0203 gives 28.83%, and 25.00 ÷ 1.0203 gives 24.50%. These now sum to 100%.
Proportional removal is convenient and, given favourite-longshot bias, systematically wrong. It assumes margin was applied evenly, which the previous section established it generally was not. The consequence is that proportional removal understates the true probability of favourites and overstates the true probability of longshots.
The power method addresses this by assuming margin was applied multiplicatively in a way that affects long odds more than short ones. Each implied probability is raised to a power, and the power is solved for such that the adjusted probabilities sum to exactly 100%. This produces estimates that better match observed outcomes, particularly in markets with a wide spread of prices.
Other approaches exist, including additive methods and techniques fitted to observed data for a particular sport and market type. The practical point for anyone doing this analysis is that no method is exactly right, that the choice of method materially changes the answer in markets with long-priced outcomes, and that stating which method was used is part of stating the result honestly.
Margin levels in practice
Typical margin varies systematically, and the variation follows a clear commercial logic.
Major football match odds carry very thin margin, often in low single digits, because these markets are heavily compared, heavily bet and intensely competitive.
Major league team sports in the largest markets are similarly tight.
Secondary competitions and lower leagues carry wider margin, reflecting both lower comparison pressure and genuinely greater pricing uncertainty.
Player and prop markets carry considerably more, since they are harder to compare directly and harder to price.
Exotic and multi-outcome markets carry the most, for the accumulative reason described above.
Accumulators compound margin across legs, which is why they are the highest-margin product in the sportsbook and why they receive so much promotional attention.
The strategic decision an operator makes is where on this spectrum to sit. A book competing on price in headline markets accepts thin margin there to attract customers, then relies on those customers also betting products where margin is healthier. A book that prices tightly across everything holds better margin per bet and takes considerably less volume.
Prices as a market rather than an oracle
A closing conceptual point that matters more as the course progresses.
It is tempting to treat sportsbook prices as forecasts, and to evaluate them by whether they predicted results. That is not quite what they are. A price is the number at which the sportsbook is willing to accept money on both sides while retaining margin, and it responds to where money goes as much as to the compiler's genuine belief.
When heavy money arrives on one outcome, the price shortens. Sometimes that reflects genuine information the market has and the compiler did not, in which case the movement is the price becoming more accurate. Sometimes it reflects popular sentiment, in which case the movement is the book protecting itself against imbalanced liability while the underlying probability has not changed at all.
Distinguishing these two situations is the essential judgement in trading, and it is why the identity of the customer placing a bet matters as much as the amount. The next lessons deal with what happens once money arrives and the book has to decide what its position actually means.
Correlation and why combinations are hard
A pricing problem that deserves early introduction, because it underlies the most commercially important product in the modern sportsbook.
Standard accumulator pricing multiplies the decimal odds of each leg together. Four legs at 2.00 produce a combined price of 16.00. This is correct only if the legs are genuinely independent, meaning the outcome of one tells you nothing about the others.
Legs drawn from different matches are approximately independent, and multiplication is a reasonable approximation. Legs drawn from the same match are not.
Consider a bet combining a team to win with its striker to score. These are strongly positively correlated: the team is considerably more likely to win in matches where its striker scores. Multiplying the two prices as though independent produces a combined price far longer than the true combined probability warrants, which means offering substantial value to the customer.
The reverse also occurs. Combining a team to win with the match finishing under a low goals total involves outcomes that pull against each other, and naive multiplication produces a price shorter than fair.
This is precisely why bet builders and same-game combination products require dedicated pricing rather than simple multiplication. The approaches used include simulation, where a model generates many synthetic versions of the match and the combined outcome frequency is counted directly; correlation matrices estimated from historical data; and conditional pricing, where each leg is priced given the assumed outcome of the preceding ones.
Two commercial consequences follow. First, operators that price correlated combinations well can offer these products confidently at healthy margin, and those that do not either price defensively, losing competitiveness, or price naively and lose money to customers who understand the mathematics. Second, correlated combinations concentrate risk, since a bet builder that lands typically requires several related things to have gone the same way, and many customers will hold similar combinations on the same popular scenario.
Reference prices and the market consensus
A practical note on where compilers actually anchor.
Very few sportsbooks price entirely independently. Most reference the wider market: prices offered by competitors, prices available on betting exchanges, and in some cases prices from sources regarded as particularly sharp. The exchange price in a liquid market is a genuine market-derived probability estimate, produced by participants risking money on both sides, and it is frequently a better estimate than any individual book's model.
This produces a recognisable dynamic. A small number of sharp operators and exchanges effectively set the market. Others price close to that consensus, differentiating on margin, product and promotion rather than on probability assessment.
The strategic question for any operator is whether to be a price maker or a price taker. Making prices requires genuine modelling capability and accepts that sharp customers will attack any error. Taking prices is safer, cheaper and offers no edge, which means competing entirely on other dimensions.
Most operators are price takers on the majority of their catalogue and price makers on a narrow set of markets where they have concentrated capability, which is a sensible allocation of a scarce resource.
Rounding, display and the practical constraints
A set of small practical matters that shape real prices more than theory suggests.
Rounding conventions limit precision. Fractional odds in particular move in coarse steps at longer prices, and a compiler wanting to offer a price between two available fractions must choose one. Decimal display allows finer granularity, which is one reason it has become standard in most markets.
Price ladders define the permitted increments a book will display. These exist for consistency and to prevent absurd precision, and they mean the offered price is frequently a rounded version of the calculated one. Across a market, the direction of that rounding is not neutral: rounding consistently towards the book adds margin, and rounding consistently away from it gives margin up.
Minimum and maximum prices cap the range offered. Very short prices below a threshold are often not displayed at all, since the return barely exceeds the stake and the market carries administrative cost for negligible turnover. Very long prices are capped because pricing precision at those levels is illusory.
Currency and stake granularity affect what customers can actually bet, and interact with maximum payout limits, which cap the total a single bet can return regardless of the odds and stake. Payout caps are a genuine risk control and a frequent source of customer complaint when they are not clearly disclosed.
Display consistency across web, app and any partner channels matters more than it appears, since a price shown differently in two places is a dispute waiting to happen.
None of this is intellectually interesting and all of it produces real money. Margin leaks through rounding conventions applied inconsistently, and errors here are systematic rather than random, which means they compound rather than cancel.
A worked margin analysis
To consolidate, work through a market end to end.
A tennis match is offered at 1.44 and 2.75. The implied probabilities are 69.44% and 36.36%, summing to 105.80%. The overround is therefore 5.80%, and expected retention on turnover is 5.80 ÷ 105.80, approximately 5.48%.
Removing margin proportionally gives true estimates of 65.63% and 34.37%, corresponding to fair prices of 1.52 and 2.91. The margin taken on the favourite is the gap between 1.52 and 1.44; on the underdog it is the gap between 2.91 and 2.75.
Expressed as a proportion of the fair price, the favourite is priced roughly 5.3% short of fair and the underdog roughly 5.5% short. This market therefore applies margin close to evenly, which is typical of a two-way market where both prices are relatively short.
Now compare a market where the prices are 1.05 and 12.00. Implied probabilities are 95.24% and 8.33%, summing to 103.57%. Proportional removal gives 91.96% and 8.04%, fair prices of 1.087 and 12.43. The favourite is priced 3.4% short of fair; the underdog 3.5% short.
Proportional removal always produces this even result by construction, which is precisely its weakness. Applying the power method to the same prices typically shows the underdog carrying materially more margin than the favourite, matching what is observed when these markets are compared against actual outcomes over large samples.
The practical takeaway is that any statement about how margin is distributed across a market depends entirely on the removal method used to make it, and analyses that do not state their method are not saying anything checkable.
Key takeaways
- Every price is a probability estimate with margin applied, and separating those two components is the foundation of all pricing analysis.
- Overround is the sum of implied probabilities above 100%, and it grows with the number of outcomes, which is why exotic markets carry far more margin than binary ones.
- Margin as a percentage of turnover is not the same as overround, and confusing them overstates profitability.
- Margin is rarely distributed evenly. Longshots typically carry disproportionately more, which reflects both risk management and observed betting behaviour.
- Recovering true probability from offered prices requires assumptions about how margin was applied, and the proportional method is convenient rather than accurate.