Expected Value: What It Means in Betting

Expected value is the mean of a probability distribution. It can be calculated without literally repeating a bet, and one observed result need not equal the mean (OpenStax expected value).

The EV Formula

The expected value of a bet is calculated using a simple formula:

EV = (Probability of winning x Net profit) - (Probability of losing x Stake)

For a conventional binary bet with decimal odds and a full-stake loss as the only losing outcome, the same calculation can be written:

EV = (True probability x Decimal odds x Stake) - Stake

Refunds, dead heats, commission, tax where applicable, and partial wins require additional outcome terms. Do not force a multi-outcome market into the binary shortcut.

A Worked Example

Suppose a model assigns a 62% win probability to an illustrative selection and the accepted decimal price is 1.80.

Using a 10 pound stake:

• Profit if you win: 10 x 1.80 - 10 = 8 pounds
• EV = (0.62 x 8) - (0.38 x 10)
• EV = 4.96 - 3.80
• EV = +1.16

The estimated EV is +1.16 pounds per 10 pounds staked under those inputs, or +11.6%. Multiplying by 100 gives an expected total of 116 pounds only for 100 bets with the same valid probability, payoff, and stake assumptions. Actual results can be much higher or lower.

Positive EV vs Negative EV

A positive estimated EV (+EV) means the chosen probabilities and net payoffs produce a positive weighted mean. It does not establish that the input probability is true.

A negative estimated EV (-EV) means the same calculation produces a negative weighted mean. A market overround can make the quoted outcome set collectively exceed 100% in raw implied probability, but it does not ensure operator profit on every result or prove that every individual selection has negative EV.

For example, in a coin toss, fair odds would be 2.00 on each side. But a bookmaker might offer 1.91 on both heads and tails. At true odds of 50%, the EV of a 10 pound bet at 1.91 is:

• EV = (0.50 x 9.10) - (0.50 x 10) = 4.55 - 5.00 = -0.45

Under the stated fair-coin probability, the expected net result is -45 pence per 10 pound stake. It is a probability-weighted mean, not the outcome of each toss (OpenStax expected value).

What the Calculation Requires

1. A complete outcome set. Include every way the bet can win, lose, refund, or settle partially.

2. Net payoffs. Use the accepted odds and deduct commission or other applicable costs.

3. A defensible probability distribution. Probabilities must sum to one, and a forecasting method should be evaluated on unseen, time-appropriate data (OpenStax probability; scikit-learn probability calibration).

4. Uncertainty analysis. Recalculate with less favourable plausible probabilities. Calibration evidence can expose systematic differences between predicted probabilities and observed frequencies (scikit-learn probability calibration).

Calibration checks whether events assigned a probability occur at a corresponding frequency across a suitable sample; it does not make any one forecast certain (scikit-learn probability calibration).

EV and Sample Size

No fixed count guarantees convergence. Repeated independent observations from a stable distribution can make the sample average more informative, but betting probabilities, prices, stakes, and methods can change. Correlation and selection bias reduce the value of a simple bet count.

A losing result does not disprove a positive EV estimate, and a winning result does not prove one. The estimate should be challenged with later data, calibration, and a preserved record.

The Relationship Between EV and Other Metrics

Value betting compares an estimated probability with a price. Yield records realised profit relative to stakes. Closing line value compares prices at two times. None independently proves the others, although together they can provide different diagnostic evidence.

18+. Gambling involves risk. For support, visit begambleaware.org. Please gamble responsibly.

Continue learning

• Next guide: Fractional Odds
• Related guide: Free Bet Meaning
• Go deeper: Expected Value in Betting

Assumptions and limitations

The worked EV is conditional on every stated probability and net payoff. It does not validate the probability estimate, guarantee a finite-sample profit, or establish that opportunities are independent and stable. OpenStax supports the expected-value definition; fees, limits, voids, and taxes must be included when they affect the real payoff.