Expected Value in Betting

The essentials

Expected value, usually shortened to EV, is the probability-weighted average result of a decision if the same conditions could be repeated many times. For a simple win-or-lose bet, it combines the estimated chance of winning with the net profit if it wins and the loss if it loses.

Positive EV does not mean the next bet will win. Negative EV does not mean the next bet will lose. The calculation depends on an uncertain probability estimate, and real results can sit far from the average for a long time.

The expected-value formula

For a two-outcome settled bet:

EV = (probability of winning × net profit if successful) - (probability of losing × stake lost)

Using decimal odds and a one-unit stake, the same calculation can be written:

EV per unit = (estimated win probability × decimal odds) - 1

Multiply by 100 to express the result as a theoretical percentage of stake. The probability must be written as a decimal, so 45% becomes 0.45.

Calculate every possible net outcome

The formula must use net profit, not total return. At decimal odds of 2.30, a £20 winner returns £46, of which £26 is profit and £20 is the returned stake. If it loses, the net result is minus £20.

The expanded calculation is (0.45 × £26) + (0.55 × -£20) = £11.70 - £11 = £0.70. That equals 3.5% of the £20 stake.

Markets with more than two settlement outcomes need each branch included. An each-way bet can win both parts, place only, lose both, dead heat, be void or face a deduction. A cash-out decision introduces another known result rather than the original binary pay-off. Simplifying these structures into one win probability can materially distort EV.

Fair odds and the break-even probability

Fair decimal odds are the reciprocal of an estimated probability:

Fair odds = 1 ÷ estimated probability

An estimated 45% probability corresponds to fair odds of 2.222. A price above that figure produces positive theoretical EV under the estimate. A lower price produces negative theoretical EV.

The market-implied break-even probability is 1 ÷ decimal odds. Odds of 2.30 require about 43.48%. The difference between an estimated 45% and the required 43.48% creates the apparent edge.

The hard part is estimating probability

Odds and pay-offs are observable. The true probability of a future sporting event is not. A model or judgement can omit team news, surface conditions, tactical changes, participant dependence or selection bias. Even a well-calibrated method produces uncertain estimates.

Do not treat the market percentage as automatically true. Bookmaker prices include margin and can reflect risk management as well as information. Removing the margin can produce a useful market baseline, but it does not prove an independent edge.

A probability estimate needs a stated method, data available before the event and testing on unseen outcomes. If the probability was adjusted after learning the result, it cannot support the original EV claim.

Use sensitivity analysis instead of one confident number

Small probability changes can reverse the answer. At odds of 2.30, an estimate of 45% gives theoretical EV of 3.5%. At 43%, the same price gives -1.1%. At 41%, it gives -5.7%.

Calculate a plausible range rather than reporting only the preferred estimate. Ask what happens if the model is two percentage points too high, the available price is shorter, or a charge applies. A decision that looks positive only under the most optimistic input has little margin for error.

Bookmaker margin and market EV

In a complete market, implied probabilities often add to more than 100%. That excess is the overround, a visible measure of price margin rather than a guaranteed operator profit on every event. If you use raw implied probabilities as your forecasts, every outcome can appear less attractive than a margin-free version.

One simple proportional method divides each implied probability by the total market percentage. More advanced methods distribute margin differently. The choice matters, especially when outcomes have very different prices. State the method rather than presenting a de-margined probability as objective truth.

Compare equivalent markets at the realistically obtainable price. Better prices improve EV mechanically, but only if stake size, terms and settlement rules also match.

Include commission, deductions and practical costs

Exchange commission can reduce the net profit on winning markets. Racing deductions, dead heats, taxes in other jurisdictions and subscription costs for data or tips can also change the practical result. Use the current terms that apply to the specific account and market.

Positive EV can lose repeatedly

EV describes an average over a hypothetical long run. Variance describes how widely actual outcomes can move around it. A 45% selection is expected to lose more often than it wins, even if a sufficiently large price makes it positive EV. Losing runs are therefore compatible with the original estimate.

After 20 independent bets, a profitable method can show a loss. After hundreds, results can still be affected by clustering, changing market conditions and probability errors. Independence itself is often questionable because several bets can depend on the same team, weather or information source.

Never increase stakes merely because results are “due” to return to expectation. The next event does not compensate for the previous sequence.

Observed profit is not the same as expected value

A winning record estimates realised performance. It does not reveal the pre-event probabilities by itself. A lucky sequence of negative-EV bets can make money, while positive-EV decisions can lose. The larger and cleaner the sample, the more information it provides, but sample size cannot repair biased data or unavailable prices.

Record the probability estimate and its timestamp before the event, along with the requested and accepted odds. This allows a later review to separate forecasting, price acquisition and outcome variance.

Expected value in accumulators

If legs are independent and each probability estimate is credible, multiply the probabilities to estimate the chance that every leg wins. Multiply decimal prices to calculate the combined return. The accumulator EV per unit is combined probability × combined decimal odds - 1.

Correlation breaks the simple independence assumption. A team winning and its striker scoring, or rain affecting several selections at one meeting, can be positively or negatively related. Bet-builder prices may already account for this. Multiplying standalone prices or probabilities can give a false answer.

EV for a lay position

A lay bet has a small potential win and potentially larger liability. Include both. If you lay £10 at decimal odds of 4.00, liability is £30. If you estimate the laid outcome has a 20% chance of winning, the lay wins £10 with 80% probability and loses £30 with 20% probability before commission.

EV = (0.80 × £10) - (0.20 × £30) = £2. This is theoretical positive EV under the estimate. If the true chance is 25%, EV is zero before commission. The four-point movement from 20% to 24% reduces EV sharply.

What closing prices can and cannot tell you

A consistently better price than a mature closing market can be useful evidence that selections were obtained efficiently. It is not proof of profit, and the closing market is not a perfect statement of truth. Limits, timing, market depth and the method used to remove margin all matter.

Record whether the headline price was realistically available for the full stake. A tiny amount at a favourable exchange quote does not justify marking the whole bet at that level.

EV when there are several settlement outcomes

Expected value is the sum of probability × net result across every mutually exclusive outcome. Suppose a £10 market can return £35 total with 25% probability, return the £10 stake with 10% probability and lose the £10 with 65% probability. The net results are £25, £0 and -£10.

EV = (0.25 × £25) + (0.10 × £0) + (0.65 × -£10) = £6.25 - £6.50 = -£0.25. The bet has theoretical EV of -2.5% under those probabilities.

This method is useful for place terms, dead-heat scenarios or a market with a partial refund, but only when the branches are complete. Outcomes must not overlap, and their probabilities must add to 100%.

Test whether probability estimates are calibrated

A set of 60% forecasts is calibrated if events in that group occur about 60% of the time over an appropriate sample. Calibration does not mean every group is profitable because the prices may be too short. It tests whether probability language corresponds with observed frequency.

Group pre-event forecasts into sensible bands, keep the bands fixed and compare predicted with observed rates. Very narrow bands can contain too little data, while broad bands can hide differences. Also inspect by sport, market and time period, but avoid searching many slices and reporting only the most flattering one.

A calibrated model can still lack discrimination, use stale inputs or fail after market conditions change. Re-test prospectively and preserve versions so later changes do not rewrite the historical method.

Set a decision threshold above zero

A calculated edge of 0.2% is easily overwhelmed by probability error, a small price movement or a cost omitted from the model. Some research processes therefore require a buffer before considering a price. The threshold is not a universal recommendation and does not make the remaining bets safe.

Define the threshold before reviewing outcomes. It might reflect model error observed on unseen data, typical slippage and commission. Apply the same rule to passes as to bets. Lowering the threshold because a match is televised or because the day has produced no selections is not evidence-led.

Record near misses. They help show whether a later rule change is genuine research or an attempt to capture remembered winners.

A repeatable EV assessment

1. Define the exact market and settlement rules.
2. Record the available price and amount before the event.
3. Estimate probabilities using a stated method and pre-event data.
4. Check that all outcomes are mutually exclusive and complete.
5. Calculate fair odds and the break-even probability.
6. Include margin, commission, deductions and realistic matching.
7. Run sensitivity tests for probability and price error.
8. Check correlations and total exposure.
9. Save the decision whether you bet or pass.
10. Review a complete sample without rewriting old estimates.

Common expected-value mistakes

• Using total return as profit.
• Using a bookmaker price as the true probability without addressing margin.
• Choosing a probability because it makes the bet positive.
• Ignoring commission, deductions or partial matching.
• Assuming accumulator legs are independent.
• Calling one winning bet proof of positive EV.
• Changing the model after results and reporting the revised backtest only.
• Staking more to recover losses from a supposedly positive-EV method.

Common questions

Does positive EV guarantee profit?

No. It is a probability-weighted estimate, not a promise about one bet or a finite sequence.

Can odds alone reveal EV?

No. Odds provide the pay-off and break-even probability. You still need an independent and uncertain probability estimate plus costs and rules.

Is the highest price always the best EV?

For the same outcome, rules and probability, a higher obtainable price improves EV. Different terms, limits, commission or settlement conditions can invalidate a simple comparison.

Use this information responsibly

Betting always carries a risk of loss. A calculation, budget or record can make decisions more transparent, but it cannot make gambling safe or produce certain returns. Never use money needed for housing, bills, food, debt repayments, savings or other priorities.

If gambling is affecting your finances, relationships, work, sleep or wellbeing, stop and seek support. The National Gambling Helpline is available free at all times on 0808 8020 133.