1. Convert the quoted price
| Format | Raw implied probability |
|---|---|
| Decimal d | 1 / d x 100 |
| Fractional a/b | b / (a + b) x 100 |
| American +A | 100 / (A + 100) x 100 |
| American -A | A / (A + 100) x 100 |
Examples:
- Decimal 2.50: 1 / 2.50 x 100 = 40.00%.
- Fractional 3/2: 2 / (3 + 2) x 100 = 40.00%.
- American +150: 100 / (150 + 100) x 100 = 40.00%.
The three displays encode the same payoff.
2. Calculate the whole market
Suppose a three-way market is priced at 2.00, 3.60 and 4.20:
- 1 / 2.00 x 100 = 50.00%.
- 1 / 3.60 x 100 = 27.78%.
- 1 / 4.20 x 100 = 23.81%.
- Total = 50.00% + 27.78% + 23.81% = 101.59%.
The 1.59 percentage points above 100 are the displayed overround. They are not a forecast of the operator's realised profit.
3. State the de-margin method
Proportional normalization divides each raw share by the total:
| Selection | Raw share | Proportional share |
|---|---|---|
| Home | 50.00% | 50.00 / 101.59 x 100 = 49.22% |
| Draw | 27.78% | 27.78 / 101.59 x 100 = 27.34% |
| Away | 23.81% | 23.81 / 101.59 x 100 = 23.44% |
The normalized shares sum to 100%, subject to rounding. They remain one convention for allocating margin. They do not reveal a unique true distribution.
Probability assessment comes afterwards
A model can be compared with the accepted price only after its probabilities have been evaluated. scikit-learn's calibration guidance describes how predicted probabilities are checked against observed frequencies. One winning or losing result cannot validate a 55% estimate.
Separate three different percentages
Readers often encounter three percentages that answer different questions. A raw implied probability is the reciprocal of one quoted price. A de-margined share is produced by applying a named method to all prices in a complete market. A model probability is generated independently from data and assumptions. Only the first can be calculated from one price alone; OpenStax provides the underlying probability terminology.
For a decimal quote of 1.80, the raw reciprocal is 55.56%. If the complete market totals 106%, proportional normalization would reduce every raw share by the same ratio. That adjusted share is a description of one margin-removal convention, not evidence that the outcome will occur at that frequency. A separate forecast must be evaluated on future observations; scikit-learn's calibration guidance explains the relationship between predicted probabilities and observed frequencies.
Verification checklist
- Use the accepted or executable price, not a stale screenshot.
- Include all mutually exclusive selections before calculating overround.
- State whether a reported percentage is raw, normalized or modelled.
- Preserve the original timestamp and market definition.
- Treat a percentage difference as an estimate-dependent comparison, not proof of value.
If a result cannot be reproduced from the recorded prices and stated method, it is not ready to support a decision. The formula is simple; the evidence boundary is the important part.
Related resources
Use overround for market comparison and true odds versus bookmaker odds for terminology.
Continue learning
- Next guide: How to Investigate an Outlier Odds Price
- Related guide: Odds Conversion Guide
- Definition: Implied Probability
Assumptions and limitations
All percentages use the displayed price before commission, deductions or boosts. Proportional normalization is transparent but not uniquely correct. Market definitions, settlement rules and available prices must match before comparing probabilities.
GambleAware
GamCare
Gambling Therapy
GAMSTOP