Settings

Language, shortcuts & timezone

Language

Timezone

Time zone changes apply on this device. A new device may start at GMT+8; check match times before reading.

Skip to content
Referrals

Settings

Customize your preferences

How Margin-Removal Methods Change Football Probabilities

By Opta Tips Team Updated

Compare proportional and additive margin removal on the same fictional 1X2 book, check invalid outputs and test how the chosen method changes a model comparison.

Proportional and additive margin removal produce different home, draw and away percentages from fictional decimal odds of 1.80, 3.60 and 4.50.

Removing overround from football odds produces a probability estimate under a chosen margin-allocation rule. Two calculations can both turn the same complete 1X2 book into a total of 100% while assigning different probabilities to the home win, draw and away win. The method determines how much of the excess is removed from each outcome.

That choice matters when a reader compares a football model with a market-derived benchmark. A small apparent difference can change direction when the allocation rule changes. The practical task is to preserve the original prices, explain the transformation and check how sensitive the comparison is to the method.

The calculations below use fictional decimal odds and a single invented pre-match snapshot. They concern three mutually exclusive results for the same defined regulation-time period. They provide an arithmetic exercise, with no claim about a current fixture, a bookmaker’s actual pricing process or an OptaTips performance record.

Keep one complete book together

Start with home win at 1.80, draw at 3.60 and away win at 4.50. The reciprocal of each decimal price gives its raw implied figure. Our guide to soccer odds formats explains why the format needs confirming before taking a reciprocal.

Fictional outcomeDecimal oddsRaw implied figure
Home win1.8055.5556%
Draw3.6027.7778%
Away win4.5022.2222%
Total—105.5556%

The unrounded total is 19/18, approximately 1.055556. Its excess above one is 1/18, approximately 0.055556. Expressed on the percentage scale, the book exceeds 100% by about 5.5556 percentage points. The excess describes the quoted prices. The bookmaker’s realised financial result also depends on the stakes and outcomes.

All three prices need the same source, observation time and settlement scope. Combining the best home price at one operator with a draw price at another and an earlier away quote creates a different object. Preserve that distinction before choosing any adjustment. A clean calculation cannot reconstruct information omitted from the input record.

Proportional normalization preserves the raw ratios

Proportional normalization divides every raw implied figure by their total. Write r for an outcome’s inverse decimal price and S for the sum of all three inverse prices. The adjusted probability is r ÷ S. The implied package calls this its basic method and documents the formula explicitly. Introduction to the implied package.

For the home win, the calculation is (1 ÷ 1.80) ÷ (19/18), giving 10/19, or approximately 52.6316%. The draw becomes 5/19, approximately 26.3158%, and the away win becomes 4/19, approximately 21.0526%. The three exact fractions sum to one.

Each raw figure is multiplied by 18/19. The home-to-draw ratio therefore remains two to one, and the draw-to-away ratio remains five to four. This method removes the same relative proportion from every raw figure. In this example that relative reduction is about 5.2632%.

The reductions measured in percentage points differ. The home figure falls by approximately 2.9240 points, the draw by 1.4620 and the away win by 1.1696. A larger raw figure loses more points because all three are being scaled by the same multiplier.

The additive method removes equal probability amounts

The additive method takes the excess S − 1 and divides it equally between the number of outcomes. For this three-way book, subtract (S − 1) ÷ 3 from each raw figure. Clarke, Kovalchik and Ingram discuss additive and normalization methods among the alternatives for adjusting bookmaker odds. Adjusting Bookmaker’s Odds to Allow for Overround.

Here the excess is 1/18. Dividing it by three gives 1/54, approximately 0.018519. Each outcome loses about 1.8519 percentage points. The home probability becomes 29/54, the draw becomes 14/54 and the away probability becomes 11/54.

Those exact fractions also sum to one. Their percentage values are approximately 53.7037%, 25.9259% and 20.3704%. This is a complete distribution, yet its home estimate is more than one percentage point above the proportional estimate from the same prices.

The equal point deduction has a different relative effect across outcomes. It removes about 3.3333% of the home raw figure, 6.6667% of the draw figure and 8.3333% of the away figure. The longer-priced outcome gives up a larger share of its starting figure.

Read the difference between the two distributions

The comparison below rounds each result to two decimal places for display. All preceding calculations use the unrounded inputs.

Fictional outcomeRaw impliedProportionalAdditiveAdditive minus proportional
Home win55.56%52.63%53.70%+1.07 percentage points
Draw27.78%26.32%25.93%−0.39 percentage points
Away win22.22%21.05%20.37%−0.68 percentage points

Both adjusted columns total 100% at the displayed precision. Their disagreement concerns how that total is distributed. Proportional scaling preserves the raw ratios. Equal subtraction changes them. Identical inputs therefore support more than one margin-adjusted distribution once the allocation rule is allowed to vary.

That mathematical fact limits the meaning of a label such as fair probability. A calculator can apply its formula correctly while relying on an allocation assumption that fits a particular dataset poorly. Establishing forecasting quality requires dated observations and outcomes beyond this one illustrative book.

The range between these two home estimates, approximately 52.63% to 53.70%, describes sensitivity to two specified transformations. It has no stated statistical coverage. Another admissible allocation could fall outside it. A confidence interval or predictive uncertainty interval would require its own construction, assumptions and interpretation. Preserve the individual method outputs instead of labelling their minimum and maximum as the likely bounds of the match’s actual chance.

The original book supplies the total excess, while the allocation remains an additional assumption. Showing both methods makes that extra choice visible and provides a useful check on comparisons that depend on a small probability difference.

Check whether the resulting probabilities are valid

An additive deduction can exceed a very small raw implied figure. Consider a separate fictional book at decimal odds of 1.05, 10.00 and 100.00. The raw figures are approximately 0.952381, 0.100000 and 0.010000. Their total is approximately 1.062381.

Dividing the excess by three gives a common deduction of approximately 0.020794. Subtracting that from the final outcome’s raw 0.010000 produces approximately −0.010794, or −1.0794%. A negative value cannot represent that outcome’s probability. The package documentation explicitly identifies negative outputs as a possible problem with the additive method. Additive-method documentation.

Setting the negative value to zero would leave the remaining values summing above one. Dividing them by their new total could restore a 100% sum, but that extra clipping-and-normalization sequence introduces another transformation. Preserve the failed output and document any replacement method instead of presenting the repaired numbers under the original formula’s name.

For every converted book, check that the outcomes are complete, each adjusted probability lies between zero and one and the total agrees with one within the declared numerical tolerance. Rounding belongs after these checks. Early rounding can obscure a small invalid result or make two methods appear identical.

Test a model comparison under both assumptions

Assume a fictional football model gives the home win a 53% probability. Hold that model output fixed while comparing it with the two transformations of the original 1.80, 3.60 and 4.50 book.

Against the proportional benchmark of approximately 52.6316%, the model is higher by about 0.3684 percentage points. Against the additive benchmark of approximately 53.7037%, it is lower by about 0.7037 points. The model and the raw prices have stayed the same. The direction of the comparison changed with the margin-allocation rule.

These differences concern a market-derived benchmark. Evaluating the quoted payout requires the actual price and an independently justified probability estimate. Under the isolated full-win or full-loss assumptions, using the fictional model’s 0.53 with actual decimal odds of 1.80 gives 0.53 × 1.80 − 1 = −0.046, an assumed expected net result of −4.6% per unit before costs. Exceeding one transformed benchmark therefore does not establish a favourable expected return at the original quote.

Where a claimed model advantage is smaller than the change between reasonable benchmark methods, report that sensitivity. Choose and document the evaluation method before inspecting which version favours the model. This lets a later reviewer separate a consistent analysis from a method chosen after the comparison was observed.

Keep the transformation reproducible

A reusable record needs the original three prices, their format, source, capture time and market definition. Add the inverse prices, booksum, adjustment method, unrounded adjusted probabilities and display-rounding rule. For software, retain the package version and relevant options as well.

If the purpose is a closing-price review, preserve the method across the selected comparison window. The closing line value audit guide explains the separate requirements for defining the close, checking price availability and retaining missing observations. A margin adjustment supplies one component of that record.

Finally, keep coherence and forecasting performance as separate checks. The probability-consistency worksheet examines whether related football forecasts can fit together. A margin-removal calculation must satisfy its own probability constraints, while its usefulness as a benchmark remains an empirical question. Keeping both the inputs and the chosen method visible gives readers enough information to reproduce the result and assess its limits.

Reader discussion

Comments 0

No comments yet.

Leave a comment

Your email address will not be published. Required fields are marked *