Expected value: the average result of a bet

Expected value is what a bet is worth on average if you could repeat it many times. It combines the price and the true chance of winning into one number.

The formula

For a bet that either wins or loses, expected value (EV) is the probability of winning times the profit, minus the probability of losing times the stake.

EV = p × profit − (1 − p) × stake

With decimal odds and a stake of 1 this simplifies to a neat form: EV per unit = p × odds − 1. If the result is negative the bet loses money on average. If it is zero the bet is fair. If it is positive the bet gains on average, as long as your value of p is right.

Example 1: a coin flip at 1.95

A fair coin has p = 0.5. Stake 10 at 1.95: a win makes 9.50 profit, a loss costs 10.

EV = 0.5 × 9.50 − 0.5 × 10 = -0.25, or -2.5% of the stake. The fair price for a coin is 2.00; any lower price gives a negative EV.

Example 2: a single-zero roulette wheel

A single-zero wheel has 37 pockets. A bet on one number pays 35 to 1, so EV per unit = (1/37) × 35 − (36/37) × 1 = −1/37 = -2.70%.

A bet on red covers 18 pockets and pays 1 to 1: EV = (18/37) × 1 − (19/37) × 1 = −1/37 = -2.70%. The two bets feel very different but have the same expected loss per unit. On a double-zero wheel with 38 pockets, the same bets have an EV of −2/38 = -5.26%.

Example 3: a dice game

Suppose a game pays 4 to 1 if a fair die shows a six. EV = (1/6) × 4 − (5/6) × 1 = −1/6 = -16.67%. The fair payout would be 5 to 1, because five losing faces have to be balanced by one winning face.

How EV changes with the price

Hold the true chance fixed and vary the price. The table assumes the event really happens 40% of the time. Its fair decimal price is 1 ÷ 0.4 = 2.50.

Expected value if the true chance really is 40%
Decimal oddsImplied probabilityEV per 1 stakedEV as % of stake
2.2045.45%-0.120-12.0%
2.5040.00%0.0000.0%
2.8035.71%0.12012.0%
3.0033.33%0.20020.0%
3.2031.25%0.28028.0%

Below 2.50 the EV is negative; above it, positive. The break-even point is exactly where the implied probability equals the true chance.

The catch: you never know p exactly

In a game with dice, cards or a wheel, p comes from counting outcomes, so EV can be worked out exactly. For sports and most other events, p is an estimate, and the EV you calculate is only as good as that estimate. If you think an event is 40% likely but it is really 33%, a bet you believe is worth +10% may in fact be worth less than zero.

That is why the margin matters so much (see overround and margin). The price already includes a cost, so your estimate has to be better than the market's by more than that cost before EV turns positive.

EV is an average, not a promise

A bet with negative EV can win today, and a bet with positive EV can lose ten times in a row. EV describes the centre of the results over many repeats; it says nothing about how far a short run can stray from that centre. The variance guide measures that spread, and the long-run guide shows how quickly results settle towards the EV.