Methodology
Removing the Vig: Four Methods Compared
There is no single 'fair' price behind a margin — there are several de-vig methods, and they disagree most exactly where it matters, on longshots.
Key takeaways
- There is no single fair line behind a margin — only a fair line under an assumption about where the margin sits.
- Proportional taxes every outcome equally in relative terms; the other three do not.
- All four agree on a near-even two-way market and disagree most on the longest price.
- The disagreement is largest exactly where a wrong fair line creates the biggest phantom edge.
- Pick a method for a stated reason, then keep it fixed — switching methods mid-analysis manufactures edges.
Why de-vigging is not one operation
To recover a fair line you have to decide how the margin was distributed across outcomes. The proportional method assumes it was applied evenly in probability terms. That is simple but tends to over-tax favourites and under-tax longshots.
Novus Odds implements four alternatives so you can compare them: proportional, power, Shin, and odds-ratio — each solving for a fair set of probabilities that sums to one, but under different assumptions about where the margin lives.
It is worth being blunt about what this means. De-vigging is not a measurement. It is an inversion of a model of bookmaker behaviour, and the model is unobservable. Two analysts looking at identical prices will derive different fair probabilities and both will be internally consistent. The choice of method is an assumption you are making, and it should be stated rather than defaulted into.
Power, Shin and odds-ratio
We corrected all three during our math review: power now actually solves its exponent, Shin uses the standard closed form, and odds-ratio does a proper solve rather than an ad-hoc shrink.
Each carries a story about the market. Proportional says the book applied a flat percentage load to every outcome. Power says the load scales with the probability in a smooth nonlinear way. Shin says a known fraction of the money comes from bettors with private information, and prices are set to survive them — which mechanically pushes probability toward favourites. Odds-ratio says the market's odds are right up to a single multiplicative distortion.
None of these stories is verifiable from prices alone, which is precisely why four methods exist rather than one.
- Power — raises each probability to a solved exponent so the set sums to one; bends the favourite-longshot curve.
- Shin — models a share of informed money and shifts probability toward favourites; the classic favourite-longshot correction.
- Odds-ratio — solves a single odds-ratio so the fair set sums to one; a middle-ground 'wisdom of the crowd' method.
One board, four answers
Take a lopsided three-way board with raw implied probabilities of 0.70, 0.22 and 0.14 — a total of 1.06, so 6% overround to remove.
Proportional divides through by 1.06 and returns roughly 0.660, 0.208 and 0.132. Power solves an exponent slightly above one and returns something near 0.672, 0.203 and 0.125 — it has taken more from the longshot and less from the favourite. Shin, modelling informed money, pushes further in the same direction, landing near 0.677, 0.201 and 0.122. Odds-ratio typically sits between proportional and power.
The favourite moves by under two points across all four methods. The longshot moves from 0.132 to 0.122 — a relative difference of about 8%. If your model says that outcome is worth 0.128, then proportional tells you the price is a fraction too short and Shin tells you it is a genuine edge. Same board, same model, opposite conclusion, and the only thing that changed was an unstated assumption.
Where the methods disagree
On a near-even two-way market the four methods barely differ. On a lopsided field they diverge most on the longshot, which is exactly the price where a wrong fair line produces the biggest phantom edge. Comparing methods on the same board is the quickest way to see why the choice matters.
There is a structural reason the disagreement concentrates there. Every method has to remove the same total probability mass, and they differ only in how they allocate it. Allocation differences are small relative to a 0.70 favourite and large relative to a 0.03 longshot — so the same absolute disagreement is a rounding error at one end of the board and a decisive difference at the other.
The practical rule that falls out: choose a method for a reason, write the reason down, and hold it fixed across an analysis. An edge that only appears when you switch de-vig methods is an artefact of the switch, not a property of the market. The Model Comparison lab exists partly to make that failure mode obvious — run the same tape through all four and watch which conclusions survive.
Frequently asked
Which de-vig method is correct?
None of them is correct in an absolute sense. Each inverts a different assumption about how a book distributed its margin, and that assumption is not observable from prices. Choose one for a stated reason and apply it consistently.
What does the Shin method assume?
That a fraction of the money wagered comes from bettors with private information, and that prices are set to remain viable against them. Correcting for that fraction shifts fair probability toward favourites, which is why Shin is the classic favourite-longshot correction.
Why do the methods agree on even markets?
Because every method must remove the same total probability mass, and on a symmetric two-way board there is essentially one way to split it. Differences in allocation only become visible when outcomes have very unequal probabilities.
Can switching methods create a fake edge?
Yes, and it is a common mistake. Longshot fair probabilities can differ by several percent relative across methods, which is enough to flip a marginal call. Any edge that depends on which de-vig you chose is an artefact of that choice.
Try it yourself
Everything in this article is something you can run and tweak in the lab — with your own settings and a reproducible seed.
Open the lab →