Value betting,the complete guide

In one sentence

A value bet is a bet whose odds pay more than the real risk: the probability that the outcome happens, multiplied by the odds, is greater than 1. Betting this way does not win every time; it wins on average, over a large number of bets, if — and only if — the estimated probability is right.

Everything else in this guide follows from that sentence: how to read odds, how to estimate a probability, how to measure the edge, and why it takes patience to know whether the edge really exists.

1. Odds are a probability

Decimal odds tell you what a 1-unit stake returns (stake included): at 2.50, you get 2.50 back if the bet wins. They also hide a probability, called the implied probability:

Odds do not tell you what will happen. They tell you what the bookmaker thinks will happen, at a given price — and that price includes its margin. More detail, with a converter: Odds and implied probability.

2. The bookmaker’s margin

Add up the implied probabilities of a match’s three outcomes: the total is more than 100%. The excess is the bookmaker’s margin, what it earns whatever the result.

Here the margin is 2.7%. To recover the “fair” probability the bookmaker has in mind, you remove the margin: spreading it proportionally, A goes from 40.0% to 38.9%, the draw to 28.6% and B to 32.4%. There are other ways to remove it, which give slightly different figures: see The bookmaker margin and Removing the margin.

3. What a value bet is

A bet has value when your estimate of the probability is higher than the one implied by the odds. The gap is measured by expected value (EV), the average profit per unit staked:

Take the Team A win at 2.50 again. If you estimate its chances at 45%:

• EV = 0.45 × 2.50 − 1 = +0.125, i.e. +12.5%: on average, each unit staked returns 12.5 cents of profit;
• with an estimate of 40% (the odds’ own), EV is 0: no edge;
• with 35%, EV is −12.5%: the bookmaker has the edge.

So everything rests on the estimated probability. Being five points wrong is enough to turn a +12.5% bet into a neutral one: a value bet is never better than the model that found it. That is why a gap that is too small is not worth playing — it is lost in the estimation error. The module What is a value bet? and the expected value calculator go through the reasoning.

4. Why you often lose, even with an edge

A +12.5% bet still loses 55 times out of 100. The edge only shows on average, and the average takes time to appear: that is variance.

Take a bettor with a genuine +5% edge, at average odds of 2.10. Here, by number of bets, is the range their measured yield falls in 95 times out of 100, and the probability that they are still losing money despite their edge:

Two practical consequences. First, a few weeks of results prove nothing, either way: at these odds, it takes around 1,700 bets for a +5% edge to stand clearly apart from luck. Second, losing streaks are normal: you need a stake and a bankroll that can absorb them. See Probability and variance, Streaks and drawdowns and Sample size, with a simulator.

5. How much to stake

Two approaches dominate.

Flat staking: the same stake on every bet, for example 1% of the bankroll. Simple, robust, and it lets variance play out without things getting out of hand.

The Kelly criterion: a stake proportional to the edge, which maximises long-run bankroll growth if the probability is exact:

In our example (45% at 2.50), Kelly says 0.125 / 1.50 = 8.3% of the bankroll. That is a lot: the slightest overestimate of the probability means staking too much, and losses pile up fast. In practice people play a fraction of Kelly — a quarter, i.e. 2.1% here — or flat stakes. See Bankroll basics, The Kelly criterion and Robust Kelly.

6. Where the probability comes from

The whole edge comes from an estimate that is more accurate than the market’s. It comes from a model: a goals model (such as Dixon and Coles’, described in our methodology), a model that accounts for players, or a model that combines several sources. A good model shows two things:

• it is calibrated: when it says 60%, the outcome happens about 60 times in 100 — see Model calibration;
• it is judged on bets published in advance, never on results reconstructed after the fact.

7. How to judge a value-betting service

Before following anyone, check four things:

1. Bets are published before kick-off, with the odds at that time, and are never changed afterwards.
2. Losses are published just like wins, from day one.
3. The number of bets is large enough for the yield to mean something (see the table in section 4).
4. The odds are actually available where you bet: odds found at a bookmaker that is not licensed in your country are no use to you.

Our own track record follows these rules:

From 4 July 2026 to 12 October 2026, 236 distinct value bets were published before kick-off and then settled, across 223 matches, at average odds of 2.13. Depending on the country and its bookmakers, the yield ranges from +4.0% (France) to +9.9% (Italy), losses included.

The detail, bet by bet and country by country, is on the track record. The statistical tests for judging a track record are in Testing performance statistically.

8. The limits

• Bookmakers limit winners. An account that wins regularly often sees its stakes capped: that is the operational reality of value betting.
• The market learns. Mispriced odds are often corrected within hours; a value bet has to be played quickly.
• Bettor biases are expensive: chasing losses, over-reading streaks, backing a favourite team. See Bettor cognitive biases.

Going further

Learning covers this whole guide in 18 free modules, with calculators and a quiz per level. Technical terms are defined in the value betting glossary.