Expected Value (EV) in Sports Betting: Formula and Examples

EV answers one question: if you made this exact bet 1,000 times, what would your average profit or loss per bet be? It is not a prediction of what happens in the next 90 minutes. A striker with a 0.71 xG shot can miss 9 times out of 10 and still be the correct process - the outcome and the quality of the decision are separate things. EV betting applies the same logic to wagering: you are grading the decision, not the result. The formula is EV = (Probability of Winning x Amount Won per Bet) - (Probability of Losing x Amount Staked). If you back a team at decimal odds of 2.10 with a $100 stake, and your model puts their true win probability at 52%, you win $110 profit 52% of the time and lose $100 the other 48% of the time. Run those numbers and you get a positive figure, meaning the bet is theoretically profitable across a large enough sample - even though it fails to hit nearly half the time. The entire discipline of value betting rests on one skill: producing a probability estimate that is more accurate than the one baked into the bookmaker's price. Get that estimate right and EV becomes a mechanical calculation. Get it wrong and the formula just confirms your bias with false precision.

Written in full, the formula is: EV = (True Probability x Net Odds) - (1 - True Probability), where Net Odds equals decimal odds minus 1 (the profit portion, excluding the returned stake). Multiply the result by your stake to get EV in currency terms, or leave it as a decimal to express EV as a percentage of stake. To use the formula you need two inputs the bookmaker gives you for free and one input you have to generate yourself. The bookmaker gives you decimal odds, which convert to implied probability via 1 / decimal odds. A price of 1.83 implies a 54.6% win probability. The bookmaker also embeds margin into that number - the true fair price, once you strip out the overround, is usually a percentage point or two lower. The input you must generate is your own probability estimate, built from a model, historical base rates, or line movement analysis. This is the hard part and the reason most bettors never get past break-even. Anyone can plug numbers into the EV formula. Few can produce a probability estimate accurate enough to beat a 54.6% implied price consistently across a full season.

Take a Premier League match where the bookmaker prices the home win at 1.83 on the standard match-result market. Implied probability from that price is 1 / 1.83 = 54.6%. Suppose your model, built on expected goals differential, home advantage weighting, and recent squad rotation, puts the true win probability at 61%. Plug that into the formula: EV = (0.61 x 0.83) - (0.39 x 1.00) = 0.506 - 0.39 = 0.116. Expressed as a percentage of stake, that is an 11.6% EV - nearly double our 6% threshold. On a $100 stake, that translates to an average expected profit of $11.60 per bet if the same scenario repeated indefinitely. Now compare that to a market where your edge is thinner. Say you like an Asian Handicap -0.5 on the same team at 1.95 (implied probability 51.3%), and your model gives them a 54% cover probability. EV = (0.54 x 0.95) - (0.46 x 1.00) = 0.513 - 0.46 = 0.053, or 5.3% EV. That falls below our 6% cutoff, detailed further on /methodology/, and under our framework it does not clear the bar for a recommended stake - even though it is technically positive.

Bookmakers do not publish true probabilities. A three-way match-result market with prices of 2.10, 3.40, and 3.75 sums to implied probabilities of 47.6%, 29.4%, and 26.7% - a total of 103.7%. That extra 3.7% is the bookmaker's overround, and on some markets, particularly obscure handicaps or lower-league totals, that margin runs as high as 8-9%. To find fair odds you have to devig the market - proportionally scaling down each implied probability so the total equals 100%. Using the example above, dividing each probability by 1.037 gives fair probabilities of roughly 45.9%, 28.4%, and 25.7%. Only after this step can you compare your model's output against a genuinely fair benchmark rather than a margin-inflated one. Sharp books like Pinnacle run margins closer to 2-3% on major markets, which makes their closing lines the closest publicly available proxy to true probability. Comparing your pre-match price against the closing line - a technique called closing line value (CLV) - is one of the fastest ways to sanity-check whether your probability model has any real predictive skill, independent of short-term results.

Not every positive number is worth acting on. A 1.5% EV bet is mathematically profitable in theory, but it sits inside the margin of error of almost every probability model - a rounding difference in your input assumptions can flip it negative. This is why we apply a 6% minimum edge threshold before a bet gets flagged as actionable, a standard explained in full on /methodology/. The 6% line exists because model error is not zero. If your true probability estimate has even a 2-3 percentage point margin of error - which is realistic for most public models - a bet showing 2% EV could genuinely be sitting anywhere between -1% and +5%. A bet showing 9% EV has enough buffer that even a meaningful model error still leaves it in profitable territory. This is also why volume and selectivity matter more than most bettors assume. Across 200 tracked selections, filtering for 6%+ EV rather than any positive number typically cuts bet volume by 60-70%, but the resulting subset shows measurably tighter variance around its expected return. Fewer bets, better filtered, outperform more bets loosely filtered - almost every time we have tested it.

Traditional bookmakers build their margin into every price you see, typically 6-9% on standard markets. Betting exchanges strip that structure out entirely. On the Betfair Exchange, you are trading against other bettors rather than against a bookmaker's book, and typical margins on liquid markets run closer to 2-5% before commission - a meaningfully smaller drag on your long-term EV. Exchanges also unlock lay betting, which matters directly for EV calculations. If your model says a team's true win probability is 30% but the exchange back price implies only 24%, you can lay that outcome instead of trying to find value on the other side of a fixed-odds market. This doubles your opportunity set: you are no longer restricted to betting on outcomes you think will happen, you can also bet against outcomes the market is overpricing. The lower margin structure on the Betfair Exchange (https://www.betfair.com/exchange/plus/?PLACEHOLDER) compounds meaningfully over a large sample. A bettor placing 500 wagers a season at an average 7% EV loses roughly 2-3 percentage points of that edge to a standard bookmaker's built-in margin before variance even enters the picture. On an exchange, with the true market price closer to fair value, more of that calculated edge survives contact with the actual settlement.

The most frequent error is using the bookmaker's implied probability as if it were the true probability, then wondering why long-term results disappoint. If you take every bet where the implied probability looks slightly generous without devigging the market first, you are not finding EV, you are just finding bookmaker margin working against you at a slower rate. The second error is overconfidence in a probability model that has never been backtested against closing lines. A model that shows an average 9% EV across 50 selections but has never been checked against actual closing-line movement is unverified. Compare your model's implied probabilities to the closing price on those same 50 events - if your numbers consistently sit closer to the opening line than the closing line, your model is not adding information, it is just early. The third error is stake sizing disconnected from the EV figure itself. Betting a flat $100 on a 4% EV selection and a $100 on a 14% EV selection ignores the fact that the second bet deserves a materially larger allocation under something like the Kelly criterion. Most EV bettors under-size their best opportunities and over-size their marginal ones, which flattens the entire benefit of doing the calculation in the first place.

Positive EV does not mean you win. A bet with a 61% true win probability still loses 39% of the time, and in a run of 20 such bets, losing 10 or more is well within normal statistical variance - a binomial distribution with p=0.61 and n=20 puts a 10-loss result inside one standard deviation. Anyone judging a betting strategy off 20 results is measuring noise, not skill. The sample size required to distinguish a genuine 6% edge from random variance is larger than most bettors expect. With typical odds around 1.90-2.00, you generally need 300-500 settled bets before the observed return on investment converges meaningfully toward the calculated EV. Below that volume, a run of bad variance can make a legitimately +8% EV strategy look like a losing one, and a lucky streak can make a break-even strategy look like a winner. This is precisely why the 6% threshold matters more at scale than it does on any single bet slip. A 6% edge over 50 bets can still show a loss. A 6% edge over 500 bets, at reasonable stake consistency, converges toward its expected return with much tighter variance. EV betting is a volume game played over a long horizon, not a prediction tool for Saturday's fixture list.