Positive Expected Value Betting, Explained for 2026

Positive expected value (+EV) betting is the practice of placing a wager only when the offered payout exceeds what the outcome's true probability justifies. If an outcome wins 42% of the time and a sportsbook prices it as though it wins 40% of the time, the bettor who takes that price earns money on average, whatever any single bet does. The method reduces to three moves: estimate the true probability, compare it to the price on the screen, and bet only when the comparison favors you.

Each of those moves hides real work. The comparison is arithmetic. The estimate is the discipline, and the gap between bettors who profit and bettors who generate spreadsheets full of imaginary edge lives almost entirely inside it.

What does the EV formula compute?

Expected value is the probability-weighted average of a bet's outcomes. A standard single bet has two: it wins and pays a profit, or it loses and costs the stake.

EV = p * profit - (1 - p) * stake

p is the true probability of winning. profit is what the book pays on a win, excluding the returned stake. stake is the amount risked. A positive result means the bet makes money on average. Dividing that result by the stake expresses the same number as an edge percentage, which is how practitioners quote it.

Worked through at American odds of +150 with a fair win probability of 42%:

Two details in that table do most of the explanatory work. Break-even probability is a pure function of the price: at +150 the bet breaks even at exactly 40%, because 100 / (100 + 150) = 0.40. Any true probability above that threshold makes the price +EV, and anything below it makes the price -EV. The second detail is how thin the margins run. Two percentage points of probability produced a 5% edge here, and 5% counts as large. Most opportunities surfaced in liquid markets carry between roughly 1% and 4% of stake, and they decay within minutes as books correct.

Where does the true probability come from?

The formula consumes a number nobody actually knows. Everything rides on how p gets produced.

Model-based estimates come from a bettor's own projection system: player-level simulations, power ratings, injury-adjusted priors. Building a model that beats the market on major US sports is a research program measured in years, and most attempts fail. Market-based estimates take a different route. Trust the sharpest available market, strip out its margin, and treat the result as fair. Nearly all practical +EV betting runs on this second approach.

Stripping the margin is called devigging. A sportsbook's two-sided prices sum to more than 100% in implied probability, and the excess is the vig, the book's structural margin. Take a market posted at -110 on both sides. Each -110 price implies 52.38%, the pair sums to 104.76%, and the multiplicative devig divides each side by that total to recover 50% fair on each. If another book hangs +105 on one of those sides, the fair 50% against a 48.78% implied price yields EV = 0.50 * 105 - 0.50 * 100 = +$2.50 per $100 staked, a 2.5% edge.

Which market earns the role of truth-teller matters as much as the method. Industry practice anchors on sharp, high-limit books, with Pinnacle as the default and consensus blends across several sharp books as the more robust alternative. Our field guide to closing line value covers the anchor choices and the multiplicative, additive, and Shin devig methods in detail. The estimate degrades at the edges of coverage. Sharp anchors thin out on player props and small college markets, devig methods disagree most on longshot prices, and a consensus dominated by one slow book inherits that book's blind spots. An error of a single percentage point in p flips the sign on most real opportunities, which is why the quality of the estimate decides everything downstream.

How does +EV relate to closing line value?

+EV is a prediction made at placement. Closing line value is the audit that arrives when the market closes, and it runs entirely on public data.

A simple mechanism joins the two metrics. Sharp closing lines are the market's best public estimate of true probability, so a bettor whose fair-value process is accurate keeps taking prices the close later vindicates. Consistent positive CLV across a few hundred placements is the strongest publicly checkable evidence that a +EV process is real, which is why professionals grade themselves on CLV first and on profit second. Wins and losses stay noisy for thousands of bets; price quality registers on every single one.

The reverse direction is the useful diagnostic. A process that surfaces bets which then settle with negative CLV is producing probability estimates the market keeps overruling, and the market is right far more often than any private model. Treat a bet log with a healthy computed edge and a flat CLV column as an alarm, since one of the two numbers is lying and it is rarely the closing line.

Why is +EV a long-run claim?

Per-bet variance dwarfs per-bet edge. In the worked example the expected profit is $5.00, and the standard deviation of a single bet's result is roughly $123: the wager either wins $150 or loses $100, and no averaging has happened yet. The edge is about one twenty-fifth of one standard deviation. At that ratio, 50 bets reveal almost nothing, and a bettor with a genuine 5% edge will live through losing stretches of 100 or more bets without the underlying math changing at all. The average asserts itself over hundreds to thousands of placements. Bettors who abandon a sound process during an ordinary downswing convert real edge into realized loss.

Stake sizing is what lets a bankroll survive long enough for the averaging to happen. Kelly (1956), in "A New Interpretation of Information Rate," derived the fixed fraction of bankroll that maximizes long-run growth: f = (bp - q) / b, where b is the net odds received (1.5 for a +150 price), p is the win probability, and q = 1 - p. The worked example gives f = (1.5 * 0.42 - 0.58) / 1.5, about 3.3% of bankroll. Full Kelly assumes p is exactly right. Since p is an estimate, an optimistic error means systematic overbetting, and overbetting damages a bankroll far faster than the mirror-image underbet slows its growth. The standard professional response is fractional Kelly: stake a quarter to a half of the computed fraction and accept slower compounding in exchange for surviving your own estimation error.

How does a real +EV workflow run?

Every serious operation runs the same five-step loop, whether by hand or through software.

1. Find the price. Scan the boards of every accessible book for current prices on every market of interest. Breadth is the point, because the mispriced number sits at whichever book has been slowest to move.

2. Estimate fair value. Devig the sharp anchor or the sharp consensus for the same market at the same moment. The estimate must be current. A fair value computed five minutes ago describes a market that no longer exists.

3. Compare. Convert the found price to implied probability and set it against the fair number. The gap, with the vig removed from the comparison, is the candidate edge.

4. Size. Apply fractional Kelly to the edge, the odds, and the current bankroll. Cap stakes where book limits or account-preservation concerns bind first.

5. Record. Log the placement price, the fair value at placement, and the stake, then capture the closing price when it arrives. The CLV column of that log is the process grading itself.

Run manually, the loop takes minutes per bet, and minutes is longer than most edges survive in liquid markets. Software fills that gap: live odds feeds across dozens of books, continuous devigging against sharp anchors, and surfaced opportunities ranked by edge, delivered inside the decay window. We wrote about the engineering behind that scanning layer in an inside look at how CLV.gg is built.

Where do +EV bettors lose?

Four failure modes account for most of the gap between computed edge and realized profit.

Bad probability estimates. The formula outputs whatever the estimate implies. Thin anchor coverage on props, a devig method mismatched to longshot prices, or a consensus leaning on one slow book all produce fair values that are wrong by more than the edges being chased. The symptom is a bet log full of "+EV" placements with a flat or negative CLV column.

Stale lines. Staleness cuts in two directions. When the data feed lags the book, the screen shows edges that are already gone, and the bet slip comes back repriced or rejected. When the book itself lags the market, the posted price is real +EV for as long as it survives, and every scanner in the market is racing toward the same number, so fills come back partial and the accounts that hit stale lines repeatedly are the first ones flagged. A workflow that ignores price timestamps cannot tell the two cases apart.

Limits and restriction. Recreational-facing sportsbooks limit or close accounts that win. A bettor running a genuine +EV process is, by construction, the account the risk desk is hunting for. Long-run profitability therefore depends on operational choices: spreading volume across books, preserving account life, and treating access to bettable prices as the scarce resource it is. The math survives restriction; the volume that monetizes the math often does not.

Ignoring the vig. Comparing raw implied probabilities across books makes negative-EV bets look positive. A -105 price beats a rival's -110, but against a fair value of 50% derived from a devigged market, -105 still implies 51.22% and still loses money on average. Every comparison has to run against a devigged fair number. Beating another book's marked-up price proves nothing.

The formula in this post fits on an index card, and it has been public knowledge for as long as bookmakers have posted odds. What separates the small population of bettors who profit from it comes down to three unglamorous capabilities: probability estimates good enough to survive the vig, the discipline to size and log every placement, and infrastructure fast enough to reach prices before they correct. The first is a modeling problem. The second is a habit. The third is why this category of betting has turned into a software business.

CLV.gg is the live sports-betting intelligence software we build at ixprt. It scans sportsbook prices in real time and surfaces +EV opportunities, arbitrage, middles, low-hold markets, steam moves, and per-bet CLV, so the workflow described above runs continuously instead of by hand.