A value bet is one whose expected return is positive at the available price, using the true probabilities and settlement payoffs. In practice those probabilities are unknown, so a claimed value bet is an estimate that can be wrong.
Finding a likely winner is a different task. A strong favourite can be overpriced in probability terms and offer a poor return for its risk. Start with the exact market, assess the evidence and then compare the actual odds.
The basic value calculation
For a standard win-or-lose cash bet before fees, expected net return per £1 equals p × d − 1, where p is the event probability and d is decimal odds. Break-even probability is 1 ÷ d.
With a hypothetical 54% chance and odds of 2.00, estimated expected return is 0.54 × 2.00 − 1 = 8%. This is an average implied by the assumptions, not a promised return on the next bet or the next hundred bets.
If formats are unfamiliar, read how football odds and implied probabilities work before applying the calculation.
Build a probability estimate for the market
Assess relative team strength, venue, opponent quality and current availability. For a goals market, scoring and concession patterns are relevant; for a first-half market, full-match averages alone are insufficient.
A recent 8-in-10 hit rate is an observation, not automatically an 80% forecast. Matches vary in opposition and circumstances. Estimates should account for sample size, changing personnel and how well the method has worked on previously unseen cases.
Stress-test the apparent advantage
At a hypothetical price of 1.90, break-even probability is 52.63%. Suppose your central estimate is 55%, but reasonable input changes produce estimates from 50% to 58%.
| Assumed chance | Expected net return per £1 | Interpretation |
|---|---|---|
| 50% | −£0.050 | Negative under this assumption |
| 55% | +£0.045 | Positive central estimate |
| 58% | +£0.102 | Positive optimistic estimate |
This range is a teaching sensitivity test, not a calculated confidence interval. It shows why the estimated 4.5% advantage should not be presented as established fact. There is no universal edge threshold that protects every model from error.
Removing margin is not forecasting
Normalising all implied probabilities in a market so they total 100% can provide a useful market benchmark. It does not supply an independent estimate of the fixture, and the chosen margin-removal method makes assumptions about how pricing distortion is distributed.
Comparing a bookmaker’s price with another number derived solely from the same prices does not, by itself, demonstrate an exploitable advantage.
Refunds and partial settlements need a wider formula
For a draw-no-bet example, let win, draw and loss probabilities be 45%, 30% and 25%, with a hypothetical price of 1.80. Net payoffs per £1 are +£0.80, £0 and −£1. Expected net return is (0.45 × 0.80) + (0.30 × 0) − (0.25 × 1) = £0.11.
Using only the unconditional win probability and the binary shortcut would give the wrong result because a draw refunds the stake. Asian quarter lines require the same payoff-by-outcome approach, including half wins and half losses.
Keep a record before claiming success
Save the forecast, uncertainty, market, available price, timestamp and result. Compare like-for-like prices when reviewing market movement. A better closing price can provide context, but does not certify a profitable method or remove selection bias.
Use the PBT calculator guide to check arithmetic. Use Situational Match Analysis to examine contextual reasoning, retaining any research-only labels. Then apply spending and exposure limits; a positive estimate does not justify unaffordable stakes.
Frequently asked questions
Does value mean long odds?
No. Value depends on probability relative to price, at short or long odds.
Can a value selection lose?
Yes. Even a correctly priced advantage includes losing outcomes and runs.
Is my FIS score a probability?
Do not assume that. A research grade cannot be substituted for a market-specific probability estimate.
Does a larger sample prove my edge?
It provides more evidence only if the data and evaluation are appropriate. Dependence, bias and changing methods still matter.
18+. No guaranteed returns. Gambling support.