math · Updated 21 Jul 2026
Fair Odds: What the Bookmaker's Margin Hides
Bookmaker odds are not probabilities — they are probabilities plus a fee. Until the margin is stripped out, every price overstates every outcome, and any model-vs-market comparison is rigged.
Finding the hidden fee
Take a real-looking 1X2 book: 2.00 / 3.40 / 4.20. Convert each price to an implied probability (1 ÷ odds):
- Home: 1 ÷ 2.00 = 50.0%
- Draw: 1 ÷ 3.40 = 29.4%
- Away: 1 ÷ 4.20 = 23.8%
Total: 103.2%. Reality only has 100% to give — the extra 3.2 points are the margin (also called the overround or vig). It's how the bookmaker earns regardless of the result: whichever outcome you back, you pay a price implying slightly more probability than the market truly assigns.
Typical football books run 5–10% of margin; lower-tier leagues run fatter ones.
Removing it — and why the obvious method isn't enough
The obvious fix is to scale everything down proportionally so it sums to 100%. That's better than nothing, but it has a known bias: the margin isn't spread evenly across outcomes in real books. Bookmakers load relatively more of it onto longshots (the famous favourite–longshot bias), so proportional scaling leaves outsiders looking more likely than they are.
The power method fixes this: instead of dividing, it raises every implied probability to a common exponent chosen so the set sums to exactly 100%. Mathematically, that takes proportionally more off the longshots — matching how the margin is actually loaded.
For the book above, the power method gives roughly 48.9% / 28.3% / 22.8% — the market's honest opinion. Note the home side: the raw number said 50%, the fair one says 48.9%. Small, but it's precisely the range where "value" is usually claimed.
We tested proportional scaling against the power method on 43,000+ real closing books graded against actual results. Power won on every slice — most visibly on longshots, where raw 1 ÷ odds said 15.4%, proportional said 14.3%, and the power method's 12.5% matched the actual win rate of 12.5%. That validation is why every market % on Tofiko uses it.
Why this matters for everything downstream
- Comparing a model to the market only means something against fair probabilities. Compare against raw 1 ÷ odds and every model looks pessimistic about favourites and every "gap" is shifted by the margin's share.
- The fee is the hurdle. To profit long-term you don't need to beat the market's opinion — you need to beat it by more than the margin. That's why most bets lose even when the bettor's judgement is decent (see Expected Value).
- Honest bookkeeping. Our Performance page grades our models against the fair closing book — the hardest honest benchmark we can construct. Grading against raw prices would flatter us; we'd rather lose to the real number than beat a fake one.
Related
Frequently asked questions
What is the bookmaker margin (overround)?
Convert all three 1X2 odds to probabilities (1 ÷ odds) and add them up. The total lands above 100% — typically 105–110%. That excess is the margin: the bookmaker's built-in fee, baked into every price.
How do I calculate fair odds from bookmaker odds?
You must remove the margin from the whole set of outcomes at once. The simple way scales all implied probabilities to sum to 100%; the better way (the power method) solves for an exponent that does it — which distributes the margin more realistically between favourites and longshots. Fair odds are then 1 ÷ fair probability.
Why does the naive 1 ÷ odds method mislead?
Because it quietly includes the margin: odds of 2.00 read as '50%' when the market's true opinion is roughly 47-48%. Every outcome looks a few points more likely than the market actually believes — and any 'edge' computed against those inflated numbers is distorted.
Are the market percentages on Tofiko fair or raw?
Fair. Every market probability displayed on Tofiko has the bookmaker margin removed with the power method, which we validated on 43,000+ closing books. Where a corrupted price feed makes that impossible, the raw fallback is labelled.