Bookmaker margin is often described through the overround in quoted odds. Removing that excess does not reveal a verified chance of winning: the result depends on how the adjustment allocates margin across outcomes. Two calculation rules can turn the same prices into different probability estimates.
This is the limitation behind a calculator's "fair odds" or "no-vig probability" result. A total of 100% shows that its numbers form a probability distribution under the chosen assumptions. It does not show that the distribution predicts the event accurately.
The implied package's first-party documentation describes several adjustment methods with different assumptions. The comparison below uses its basic proportional and additive rules. It is an explanation of those rules, not a recommendation to bet or a claim that either method recovers the actual chances.
First separate the price from the adjustment
For decimal odds D, the raw implied percentage is 100 ÷ D. Smarkets' probability explanation gives that conversion. It expresses a quoted price numerically; it does not supply an independent observation of the outcome's probability.
For a complete market, add the reciprocal prices. The total is the book percentage; its excess above 100% is the overround. Smarkets' margin guide explains summing the prices across the possible outcomes. Our odds-format guide covers the conversion and payout notation in detail.
All examples here are hypothetical two-outcome markets. Outcomes A and B are mutually exclusive and exhaust the possibilities, both decimal prices come from the same snapshot, and fees, taxes, promotions and settlement adjustments are excluded. These are not live event quotes or suggested stakes.
At 1.91 on each side, each raw implied percentage is approximately 52.3560%. The total is approximately 104.7120%, giving an overround of approximately 4.7120 percentage points. At 2.00 on each side, the total is 100% and the overround is zero. Those calculations still do not establish the actual chance of A or B.
Two rules for the same quoted prices
Use fractions rather than percentages in the formulas: q_i = 1 ÷ D_i, and B = the sum of all q_i. For these single-winner markets, the target total after adjustment is 1.
The proportional rule divides each raw implied value by the full book total: p_i = q_i ÷ B. It removes the same proportion from each value. Divide by B, not by the excess B − 1.
The additive rule subtracts the same absolute amount from each raw value: p_i = q_i − ((B − 1) ÷ n), where n is the number of outcomes. It distributes the excess equally in probability units. That is a different assumption from proportional scaling.
Both formulas appear in the documentation. The package manual also makes the choice of method explicit rather than treating adjustment as a unique conversion.
| Hypothetical prices A / B | Book total / overround | Proportional estimate for A | Additive estimate for A |
|---|---|---|---|
| 1.91 / 1.91 | 104.7120% / 4.7120 pp | 50.0000% | 50.0000% |
| 2.00 / 2.00 | 100.0000% / 0.0000 pp | 50.0000% | 50.0000% |
| 1.50 / 2.80 | 102.3810% / 2.3810 pp | 65.1163% | 65.4762% |
The table is calculated from the stated formulas, with results rounded to four decimal places. Here pp means percentage points. B's adjusted estimate is the complement of A's in each row. No row is a measured forecast or a bookmaker's disclosed internal model.
Why a 100% total does not settle the question
In the symmetric 1.91/1.91 example, both rules return 50% for each outcome. They agree because equal inputs receive equal treatment. That agreement is not additional evidence that the real event is evenly balanced.
The asymmetric example makes the assumption visible. Raw implied values from 1.50 and 2.80 are approximately 66.6667% and 35.7143%. Proportional scaling produces 65.1163% for A; subtracting equal amounts produces 65.4762%. Both distributions sum to 100%, yet they disagree about A.
The arithmetic cannot choose which allocation describes the event better. That requires evidence beyond the two prices and a declared adjustment rule. Renaming either column "true probability" would conceal that missing evidence.
There is also a validity limit. The CRAN documentation shows that the additive method can produce negative adjusted probabilities for some inputs, particularly low-probability outcomes in larger markets. Our table does not establish that the additive rule works for every price set. Other methods have their own assumptions and numerical constraints; an unfamiliar calculator should explain its method and how it handles invalid results.
Keep margin, settlement and charges apart
Overround is computed from prices. Realised sportsbook hold is computed from settled wagering and payouts, while accounting profit includes further costs. Our hold and theoretical-margin explainer separates those measures. A price-derived 4.7120-point overround is not a guaranteed 4.7120% profit on money wagered.
The examples also exclude exchange commission and other charges. A comparison that ignores a relevant charge can misstate the effective return. Check the applicable provider's terms; a historical fee example in an educational article is not a current rate for every account or country. The exchange-versus-sportsbook guide explains the different counterparties and liability structures.
A later quote is another snapshot, not a validation of the earlier adjusted probabilities. Keep all prices from one market and time together. Nor should the single-winner formulas be applied unchanged to a market where several selections can win: the CRAN documentation explicitly provides a different target total for multiple-winning-outcome markets.
When reading a no-vig result, ask which prices were used, whether the outcome set is complete, which method removed the margin and which charges were excluded. Treat the output as an estimate under that method. An internally consistent calculation is useful to understand, but it does not make gambling safe, predict the winner or establish a profitable betting opportunity.