No-vig probability is a benchmark, not your edge

A price answers what a transaction costs. A probability estimate answers how often an outcome should happen. Confusing the two can make an ordinary market quote look like a betting advantage. No-vig normalization is useful for comparison, but it does not turn a bookmaker's prices into ground truth.

Start with the payout, not the label

Decimal odds include the returned stake. At decimal odds 2.10, a winning $100 stake returns $210: $100 of principal and $110 of profit. For an ordinary win-or-lose wager with no fees or refunds, the break-even probability is 1 divided by decimal odds. Here it is 1 / 2.10 = 47.619%.

For positive American odds +A, decimal odds are 1 + A / 100 and implied probability is 100 / (A + 100). For negative American odds −A, where A is the positive magnitude, decimal odds are 1 + 100 / A and implied probability is A / (A + 100). Thus −120 implies 120 / 220 = 54.545%, while +110 implies 100 / 210 = 47.619%.

A worked two-outcome no-vig example

Suppose the two mutually exclusive, exhaustive outcomes of the same market are priced at −120 and +110 at the same time. Their raw implied probabilities sum to 102.164%, rather than 100%. The excess of about 2.164 percentage points is the overround. It is not a promise that the bookmaker earns that percentage of every stake.

Proportional no-vig probability divides each raw implied probability by their sum: 54.545 / 102.164 ≈ 53.390%, and 47.619 / 102.164 ≈ 46.610%. Calculate with full precision and round only for display. Normalization assumes the margin is allocated proportionally. Other methods allocate it differently and can produce different estimates, especially for longshots.

The inputs must describe the same event, line, settlement rules and observation time. Do not combine opposite sides from unrelated handicap lines, omit the draw from a three-way market, or quietly treat a push as a loss. A win/loss formula may need to be conditional on no push, or expanded to include refund outcomes.

Expected value requires a separate probability assumption

For stake S, decimal odds d and estimated win probability p, expected net profit is S × [p × (d − 1) − (1 − p)], equivalently S × (p × d − 1). This is a probability-weighted average over possible outcomes, not a forecast of the next result.

At +110, assume your independently justified probability for outcome B is 50%. A $100 stake wins $110 or loses $100. Expected net profit is 0.50 × $110 − 0.50 × $100 = $5, a 5% expected return on stake before costs. If the true probability is only 47%, the same calculation becomes 0.47 × $110 − 0.53 × $100 = −$1.30.

Using the market's own normalized 46.610% instead gives about −$2.12 per $100 at +110. Removing the overround did not create an edge. To claim a positive estimate, you need a reason to believe the outcome probability exceeds the actual offered price's 47.619% break-even threshold, not merely a different way to rewrite that price.

Keep uncertainty visible

• Record the quote source and timestamp. A favorable historical quote may no longer be executable.
• Label the probability as a model estimate or user assumption. A market-derived benchmark is not independent evidence.
• Recalculate across a plausible probability range. An edge that disappears after a one-point change may be smaller than your estimation error.
• Include fees and the actual achievable payout. See fees and execution risk before treating gross EV as net EV.

Repeated positive estimated EV can still produce substantial losses, and correlated bets do not behave like independent trials. Stake sizing cannot fix an inaccurate probability estimate. Risk only money you can afford to lose; this example is education, not a recommendation to wager.

From calculation to a documented assumption

Open the analysis workspace (signup required) to work with your inputs. For evidence behind a probability estimate, start with weighted models and honest backtesting. Keep the offered odds, assumed probability and costs separately identifiable so another calculation can reproduce your result.