Value betting: the only definition that means anything
A value bet is one where your probability estimate exceeds the de-vigged market probability by more than your own uncertainty. Everything else called value betting is a price you liked. How to compute it, and the four ways the calculation goes wrong.
A value bet exists when your estimated probability of an outcome is higher than the probability implied by the price after the bookmaker margin is removed. If you think an outcome is 40% likely and the de-vigged market says 35%, that is a 5-percentage-point edge. If you compare against the raw price instead, you will find edges that are entirely made of margin.
The calculation, in full
Take a three-way market: home 2.10, draw 3.40, away 3.80.
| Step | Home | Draw | Away |
|---|---|---|---|
| Raw implied probability | 47.6% | 29.4% | 26.3% |
| Sum | 103.3% | ||
| De-vigged (proportional) | 46.1% | 28.5% | 25.5% |
| Your estimate | 52.0% | 26.0% | 22.0% |
| Edge | +5.9 pp | −2.5 pp | −3.5 pp |
The home selection carries an edge of 5.9 percentage points. Against the raw price it would appear to be 4.4 points — smaller, because the raw price already includes margin working against you. Using raw prices in the other direction, which is more common, inflates edges instead.
edge (pp) = your probability − de-vigged market probability
EV per unit = (your probability × (odds − 1)) − (1 − your probability)The four ways it goes wrong
- Comparing against raw odds. Manufactures an apparent edge equal to roughly the margin, on every bet, in the flattering direction.
- Wrong normalisation target. Double chance markets sum to 2, not 1. Normalising them to 1 produced a fictional 41-percentage-point edge in our own engine on live prices — see double chance.
- Ignoring settlement rules. Pushes and half-settlements change the payoff. An EV formula with only two outcomes cannot price a quarter line.
- Treating your estimate as exact. A 2-point edge from a model whose typical error is 5 points is not an edge; it is noise with a sign.
How much edge is enough?
The honest answer depends on how good your estimate is, and most people have no measurement of that. In the absence of one, the practical threshold used by people who do measure it is well above zero — small computed edges are dominated by estimation error.
Our own engine answers this question differently, and the difference is worth naming because it cuts against the advice above. We do not use a minimum edge as a condition for publishing. Whether a signal goes out is decided by whether three independent analyses converge; the edge decides how large the position is, from a quarter of a unit to a full one. A negative edge produces the smallest position rather than silence, and the number is printed next to the signal. The reasoning is that a threshold on edge is a statement about our own precision, and we do not yet have the calibration record to justify a particular number. Every published signal carries the version of the rules that produced it, so a decision made last month stays explicable under this month's settings.
The test that catches most errors
Compute the edge on both sides of a two-way market. If both sides show positive edge, the calculation is wrong — one side must be at least as bad as fair. This single check catches de-vig mistakes, settlement mistakes and normalisation mistakes, and it costs nothing.
Check any of this against our record
Every signal CONSENSUS publishes carries the bookmaker odds fixed before the event starts and the settled result afterwards — including the drawdowns and the losing runs. The running total is on the front page and every entry is in the log.