Devig Methods Compared: Multiplicative, Additive, Power, and Shin

Devig Methods Compared: Multiplicative, Additive, Power, and Shin

Why devig methods disagree

A sportsbook market contains margin, or overround. Removing that margin requires an assumption about where the bookmaker placed it. The familiar no-vig calculation divides each implied probability by the market total, but uneven markets can produce meaningfully different estimates under other methods.

Use the 2-way no-vig calculator for spreads and totals, or the 3-way no-vig calculator for soccer 1X2 and other three-outcome markets. Both calculators use the multiplicative method; this guide shows what the other methods would change.

Multiplicative normalization

The multiplicative method divides every implied probability by their sum. If three prices imply 50%, 30%, and 25%, their 105% total becomes 47.62%, 28.57%, and 23.81%. It is fast, transparent, and always returns positive probabilities, but assumes every outcome carries margin in proportion to its implied probability.

Additive method

The additive method subtracts an equal share of the overround from every outcome. In a market with a five-point overround and three outcomes, it subtracts 1.67 percentage points from each. This can be intuitive for balanced markets, but it may return a negative probability for a sufficiently long-priced outcome.

Power method

The power method raises implied probabilities to a shared exponent chosen so the adjusted values sum to one. This redistributes margin nonlinearly and can account for favorite–longshot bias. Clarke, Kovalchik, and Ingram found the power method compared favorably with other adjustments in their analysis of bookmaker odds (research paper).

Shin method

The Shin method models the possibility that some bettors possess inside information, then estimates fair probabilities from the resulting market imbalance. It is more complex and its assumptions do not fit every market. Štrumbelj found Shin probabilities outperformed basic normalization in the betting data studied (International Journal of Forecasting).

One market, four answers

Take a soccer 1X2 market priced Home -150, Draw +280, Away +450. The implied probabilities are 60.00%, 26.32%, and 18.18%, which add up to 104.50%, so the book is holding a 4.50% overround. Each method removes that same 4.50%, but each puts it in a different place:

The worked lines behind each row:

• Multiplicative: 18.18% ÷ 1.045 = 17.40% for the away side. Every outcome shrinks by the same 4.3% of its own size, so the longshot keeps the most probability of any method.
• Additive: 4.50 points ÷ 3 outcomes = 1.50 points off each side, so the away side drops from 18.18% to 16.68%. Removing equal points is a much bigger cut, in relative terms, for a longshot than for a favorite.
• Power: the exponent that brings the total back to exactly 100% is about 1.048. Raising 0.1818 to that power gives 16.76%. Because every probability sits below 1, a small exponent takes the most, in relative terms, from the smallest numbers.
• Shin: solving for the insider share gives z ≈ 2.3%, and the away side comes out at 16.87%. Shin lands between multiplicative and additive here. In a two-way market, Shin's method gives the same fair probabilities as the additive method, so it only diverges in markets with three or more outcomes.

In raw percentage points the favorite moves more: 1.13 points separate the home estimates, against 0.72 points for the away side. Relative to its size, though, the longshot moves about twice as much. 0.72 points is roughly 4% of a 17% probability, while 1.13 points is about 2% of a 58% one. In price terms the away side's fair price runs from +475 to +499, and that is where a bet's edge is decided. This is favorite–longshot bias at work: the methods mostly disagree about how much of the margin sits on the unlikely outcomes.

A conservative worst-case estimate

“Worst case” is not a fifth devig formula. For the outcome you may bet, calculate several defensible fair probabilities and use the lowest estimate when testing expected value. If a wager is still positive EV under that conservative input, the conclusion depends less on one margin assumption. It remains an estimate—not a guarantee of true probability.

Suppose another book offers the away side above at +500 (a 6.00 decimal price). The expected value per unit staked is fair probability × 6.00 − 1:

• Multiplicative, 17.40%: +4.4% EV
• Shin, 16.87%: +1.2% EV
• Power, 16.76%: +0.6% EV
• Additive, 16.68%: +0.1% EV

Under the multiplicative estimate this looks like a clear edge. Under the most conservative estimate it is roughly break-even. The price is still a small positive-EV bet on every method, but the size of the edge depends almost entirely on the devig assumption, which argues for a small stake, not a confident one.

Practical workflow

1. Convert every offered price to implied probability.
2. Check that all outcomes come from the same market and timestamp.
3. Run multiplicative normalization first as a clear baseline.
4. Compare additive, power, or Shin outputs when the market is lopsided.
5. Use a conservative estimate in the EV calculator, then record the exact price taken.

Return to the sports betting learning path for the full sequence from odds through model evaluation.