The Edge Is in the Price, Not in Picking the Winner
Why a team that wins most of the time can still be a bad bet, how to strip the margin out of a price, and why even a careful estimate rarely proves an advantage.
Original educational writing based on the cited research. Grey-bordered notes report what studies found; gold-bordered notes are our interpretation. All prices and teams in examples are hypothetical.
Winners and prices are different questions
Most football conversations ask one question: who will win? A quantitative analyst asks a different one: is this price right? The two are easy to confuse, because the favourite usually wins. But a bet is a contract with a fixed payout, and whether that contract is worth taking depends on the relationship between the payout and the probability — not on which side is more likely.
Take a hypothetical favourite quoted at decimal odds of 1.40, and suppose it genuinely wins 70% of the time. It is the likeliest outcome by a distance. Now take a hypothetical underdog at 4.50 that genuinely wins 25% of the time. Most people would rather back the favourite. The arithmetic disagrees.
Expected value per unit staked: EV = p × o − 1| Bet | Odds | True probability | Break-even probability | EV per unit |
|---|---|---|---|---|
| Favourite | 1.40 | 70.0% | 71.4% | −2.0% |
| Underdog | 4.50 | 25.0% | 22.2% | +12.5% |
The favourite wins seven times in ten and still loses 2 pence per pound staked on average, because its price needs a 71.4% win rate just to break even. The underdog loses three times in four and still returns +12.5% on average, because its price only needs 22.2%. The edge — if there is one — lives in the gap between probability and price. Picking more winners is neither necessary nor sufficient.
Break-even probability and the bookmaker's margin
Every decimal price implies a break-even probability of 1/o. At 2.10 that is 47.6%; at 3.40 it is 29.4%; at 3.60 it is 27.8%. If those three prices are the home, draw and away options for one match, the implied probabilities add up to more than 100%:
1/2.10 + 1/3.40 + 1/3.60 = 0.4762 + 0.2941 + 0.2778 = 1.0481Another way to read the same number: if you could back all three outcomes in proportion to their implied probabilities, you would get back about 95.41% of what you staked whatever happened (1 ÷ 1.0481). That shortfall is the price of the market existing. It is also the first hurdle any claimed edge must clear: an estimate that merely matches the market's view loses the margin every time.
Removing the overround
The simplest way to turn 104.81% back into 100% is proportional normalisation: divide each implied probability by the book sum. That gives fair probabilities of 45.43% / 28.06% / 26.50% and a fair home price of 2.201. But proportional removal assumes the bookmaker spread its margin evenly across outcomes in proportion to their probability, and that assumption is questionable.
| Method | Home | Draw | Away |
|---|---|---|---|
| Proportional | 45.43% | 28.06% | 26.50% |
| Shin | 45.87% | 27.87% | 26.26% |
| Power | 46.02% | 27.80% | 26.19% |
Shin's method, derived from a model in which bookmakers protect themselves against better-informed bettors Shin (1992), and the power method both take relatively more margin from the long shots. In a comparison of conversion methods, Shin probabilities forecast outcomes more accurately than simple normalisation Štrumbelj (2014), and a later comparison found the power method comparable Clarke, Kovalchik & Ingram (2017). Here the methods disagree by up to about 0.6 percentage points on the favourite. That sounds small. It is not, when the edges you are looking for are only a few points wide.
Try it: change the draw price and watch how the methods diverge for long shots.
Lab 2 · Remove the margin
Synthetic · deterministic| Method | Home | Draw | Away |
|---|---|---|---|
| Raw implied (with margin) | 47.62% | 29.41% | 27.78% |
| Proportional | 45.43% (2.201) | 28.06% (3.563) | 26.50% (3.773) |
| Additive | 46.02% (2.173) | 27.81% (3.596) | 26.17% (3.820) |
| Power k = 1.0461 | 46.02% (2.173) | 27.80% (3.597) | 26.18% (3.819) |
| Shin z = 0.0241 | 45.87% (2.180) | 27.87% (3.588) | 26.26% (3.808) |
- Proportional: Divides every implied probability by the book sum. Simple, but assumes the margin is spread in proportion to probability.
- Additive: Removes an equal amount from every outcome. Easy to explain; it can push long shots below zero.
- Power: Raises every implied probability to a common power. Takes relatively more margin from long shots.
- Shin: Derived from Shin's insider-trading model. Takes more margin from long shots; performed well in Štrumbelj (2014).
Reading the result. The methods agree on the favourite more than on long shots. None of them recovers the “true” probability: they are different assumptions about how the bookmaker spread its margin. Differences of a point or two can decide whether a price looks like value, so state which method you used.
Educational calculator. Uses only the numbers you enter or seeded synthetic data — no live prices, no staking, no real money.
Two bookmakers, one match
Now add a second hypothetical bookmaker quoting 2.20 / 3.30 / 3.50 for the same match. Its book sums to 104.33%. Taking the best price for each outcome across both books — 2.20 home, 3.40 draw, 3.60 away — gives a combined book of 102.64%: still above 100%, but with roughly half the margin of either bookmaker alone. That is the mechanical benefit of comparing prices.
| Outcome | Book A | Book B | Consensus | Best price | EV at best price vs consensus |
|---|---|---|---|---|---|
| Home | 45.43% | 43.57% | 44.50% | 2.20 | −2.1% |
| Draw | 28.06% | 29.05% | 28.55% | 3.40 | −2.9% |
| Away | 26.50% | 27.39% | 26.94% | 3.60 | −3.0% |
Against the two-book consensus, even the best available price on every outcome has negative expected value. A higher number on one screen is not, by itself, value. Comparing prices reduces the margin you pay on bets you already have a reason to make; it does not create the reason. To claim value you still need either a better probability than the consensus or a price far enough from the consensus to survive the doubts discussed below.
Identifying an apparent edge
Suppose your own model rates the home side at 49%. Against the quoted 2.10, that implies EV = 0.49 × 2.10 − 1 = +2.9%. Against the fair consensus of 45.4%, you disagree with the market by roughly 3.6 points. Before celebrating, ask the question research keeps asking: why would you know better than the market?
Using the market against itself: consensus prices
One family of strategies does not try to out-forecast the market at all. It treats the average of many bookmakers as the fair price and looks for a single bookmaker whose price sits well above it. If the de-vigged consensus for the home win is 45.43% and one book offers 2.38, EV against the consensus is 0.4543 × 2.38 − 1 = +8.1% — on paper.
A widely discussed preprint tested this idea on historical football odds and in a short real-money trial, and reported positive results — along with bookmakers restricting the authors' accounts Kaunitz, Zhong & Kreiner (2017). Treat it carefully: it is not peer reviewed, the live sample was small, and historical outlier prices may have been stale, mistyped or available only for tiny stakes. The reported restrictions are themselves a finding: an edge that cannot be executed at scale is mostly theoretical.
Closing line value: a useful diagnostic, not a verdict
Prices move between opening and kickoff as information arrives. Many practitioners measure whether they systematically take prices better than the final, closing price — closing line value:
CLV = taken odds ÷ fair closing odds − 1The reasoning is intuitive: if closing prices are close to efficient, repeatedly beating them implies positive expected value. The sources reviewed in this library do not establish that CLV predicts long-run profit; this is not an exhaustive claim about all research. Closing prices are not always the best forecasts: one study of MLB line movement found that forecasts did not always improve towards game time and that line moves showed overreaction Simon (2024). CLV can be distorted by which de-vig method you use, by timing, and by measuring only the bets you remembered to record.
Price uncertainty: the estimate is a range
Your 49% is an estimate, with error. If an honest range is 45% to 53%, EV at 2.10 runs from 0.45 × 2.10 − 1 = −5.5% to 0.53 × 2.10 − 1 = +11.3%. Because EV is linear in probability, the whole story is in those two endpoints: the range straddles break-even (47.6%), so the bet is positive only if the optimistic half of your range is right.
Enter your own price and range. The lab reports EV at both ends and states whether the advantage survives the uncertainty.
Lab 1 · Fair odds and expected value
Synthetic · deterministicNo clear advantage. Your range straddles break-even. Whether the bet has value depends on which end of your estimate is right, so there is no clear advantage. A careful analyst treats this as a pass unless better evidence narrows the range.
Show data table
| True probability (%) | Expected ROI |
|---|---|
| 35% | -27% |
| 36% | -24% |
| 37% | -22% |
| 38% | -20% |
| 39% | -18% |
| 40% | -16% |
| 41% | -14% |
| 42% | -12% |
| 43% | -10% |
| 44% | -8% |
| 45% | -5% |
| 46% | -3% |
| 47% | -1% |
| 48% | 1% |
| 49% | 3% |
| 50% | 5% |
| 51% | 7% |
| 52% | 9% |
| 53% | 11% |
| 54% | 13% |
| 55% | 16% |
| 56% | 18% |
| 57% | 20% |
| 58% | 22% |
| 59% | 24% |
| 60% | 26% |
| 61% | 28% |
| 62% | 30% |
| 63% | 32% |
Educational calculator. Uses only the numbers you enter or seeded synthetic data — no live prices, no staking, no real money.
Theoretical EV versus realised results
Even if +2.9% were exactly right, results would not show it quickly. One unit staked at 2.10 with a 49% chance has a standard deviation of about 1.05 units per bet. After 100 bets the standard error of your average return is about 10.5%; after 1,000 bets it is still about 3.3% — larger than the edge itself. Roughly 5,000 bets are needed before a true +2.9% edge sits about two standard errors from zero.
That is why a good month proves almost nothing, and why a study that reports a profitable season should be read alongside evidence that single-season inefficiencies often fail to persist Winkelmann et al. (2024).
From a quoted price to an executable one
A quote on a screen is not a bet you can necessarily make. Prices go stale, move when you submit, carry stake limits and can be cancelled. Suppose that +8.1% outlier at 2.38 could only be half-filled and executed 3% worse than quoted. Expected value per intended unit falls to about +2.4% — and that is before account restrictions, which can end a strategy entirely.
- Timestamp every price you analyse; a quote captured after information arrived is not one you could have taken.
- Model the fill: maximum stakes, partial acceptance and rejections.
- Model slippage: the price you actually get, not the one you saw.
- Count every bet, including the ones that were refused.
In WCWins Value 7, a pick's price is locked at submission and the stake is free virtual Credits — a deliberate simplification for learning. Real markets are less forgiving.
What the evidence supports, and what it does not
Check your understanding
Practice only: saved on this device, no XP and no account record.
References
- Raymond D. Sauer (1998). The Economics of Wagering Markets.Journal of Economic Literature 36(4), 2021–2064Wagering prices are, to a first approximation, efficient forecasts, with documented exceptions.Original source (opens in a new tab)
- Erik Štrumbelj (2014). On determining probability forecasts from betting odds.International Journal of Forecasting 30(4), 934–943Margin-removal method affects forecast accuracy; Shin outperformed basic normalisation in the samples studied.Original source (doi:10.1016/j.ijforecast.2014.02.008) (opens in a new tab)
- Lisandro Kaunitz, Shenjun Zhong, Javier Kreiner (2017). Beating the bookies with their own numbers – and how the online sports betting market is rigged.arXiv:1710.02824 (statistics, applications) · preprint, not peer reviewedA consensus-outlier strategy and the practical problem of account restrictions (preprint).Original source (opens in a new tab)
- Jay Simon (2024). Inefficient Forecasts at the Sportsbook: An Analysis of Real-Time Betting Line Movement.Management Science 70(12), 8583–8611Line movement can overreact and closing forecasts are not always the most accurate.Original source (doi:10.1287/mnsc.2022.00456) (opens in a new tab)
- David Winkelmann, Marius Ötting, Christian Deutscher, Tomasz Makarewicz (2024). Are Betting Markets Inefficient? Evidence From Simulations and Real Data.Journal of Sports Economics 25(1), 54–97Single-season inefficiencies often do not persist across seasons and leagues.Original source (doi:10.1177/15270025231204997) (opens in a new tab)
Further reading
- Shin (1992): Prices of State Contingent Claims with Insider Traders, and the Favourite-Longshot BiasThe insider-trading pricing model behind the Shin method.
- Clarke, Kovalchik & Ingram (2017): Adjusting Bookmaker's Odds to Allow for OverroundA comparison of additive, normalisation, Shin and power adjustments.
- Hegarty & Whelan (2026): Market structure and prices in online betting markets: theory and evidenceHow market competition shapes the favourite–longshot pattern in football odds.