A value bet is a bet whose odds, as offered by the bookmaker, are higher than the fair odds, that is 1 divided by the real probability of the outcome. Its expected value is positive: EV = p × c − 1, with p the probability and c the odds. There is value when p × c exceeds 1. Guide published on 05/09/2026.
p is the probability that the outcome happens, between 0 and 1. c is the decimal odds, what €1 staked returns, stake included. EV is the expected gain per euro staked: positive, the bet pays on average over time; negative, it costs. The fair odds, those that cancel the expectation, are 1 / p.
You estimate a player has a 60% chance of winning: p = 0.60, fair odds = 1 / 0.60 = 1.67.
The point everyone misses: being the favourite and having value are two different things. A favourite at 1.60 who wins 60% of the time loses money over time. An underdog at 3.50 who wins 30% of the time makes money (0.30 × 3.50 − 1 = +0.05).
The odds are on the screen; the probability, nobody knows. All of value betting rests on the quality of that estimate. The market gives one by default: 1 / c, corrected for the bookmaker's margin. How to read odds and remove the margin. Finding value means estimating better than the market on some matches, which is rare and can be measured.
On our corpus, the market favourite wins 68.8% of the 220,859 settled singles matches since 2021. Our model, version 1.1.0, picks the right winner in 67.39% of cases on 59,960 ATP matches, against 67.07% for the bookmaker's favourite on the same matches, and agrees with the market 96.46% of the time. In other words: nobody beats the market as a whole. Value lies in the few percent of disagreement, and in the calibration of probabilities on those matches.
On our published engines, 86,303 matches were analysed and 990 passed the positive expectation filter, 1.15%. The rest is discarded: no volume for volume's sake. On the seasons alone that the model had not seen in training (2023 to 2026), the series of selected bets gives a ROI of 5.27% on 564 flat-stake bets, with 67.4% of bets won. This is a simulation, none of these bets was placed, and the gap in real conditions is usually a few points. The full measurement, version by version.
A bet whose offered odds are higher than the fair odds, that is 1 divided by the real probability of the outcome. Its expected value is then positive.
EV = p × c − 1, where p is the estimated probability and c the decimal odds. There is value when p × c exceeds 1. Example: a 60% chance at odds of 1.80 gives 0.60 × 1.80 − 1 = +0.08, an 8% expected value.
No. A value bet can lose, and often does: at odds of 1.80 it loses about four times in ten. Value is judged over hundreds of bets, never on one.
Few. On our published engines, 1.15% of analysed matches pass the filter, about two in a hundred.
See also: measuring the accuracy of a prediction, the full methodology.