Implied probability
implied = 1 / decimal odds
The probability a bookmaker's price corresponds to before margin is removed.
Example: odds 2.10 → 1 / 2.10 = 47.6%
Methodology
How the numbers are produced
Every figure on this platform comes from the pipeline below. Nothing is hand-picked, and the same rules are applied to every bout.
Warning: Limitations you should hold in mind
- MMA is high-variance: a well-calibrated 70% still loses roughly three times in ten.
- Historical data is incomplete for newer fighters, and coverage is disclosed per bout.
- A positive edge reflects a disagreement with the market, not a known mispricing.
- Outputs are analytical estimates, not betting advice, and not a profit forecast.
The pipeline
1. Collect historical data
Fighter records, bout outcomes, methods, rounds, physical attributes and activity are collected per athlete. Only information available before the bell is used for any given bout.
2. Build features
Differences between the two fighters — rating, experience, layoff, age, reach — are computed. When a feature cannot be computed, coverage drops and this is reported rather than silently imputed.
3. Estimate a probability
A calibrated classifier maps the feature set to a win probability for each fighter. Bookmaker odds are never a model input, which is what keeps the estimate independent.
4. Observe the market
Bookmaker decimal odds are recorded with a timestamp. Suspended, unopened or stale markets are marked as such and are not compared.
5. Remove the margin
Raw implied probabilities sum above 100%. Dividing by that overround gives no-vig probabilities that are directly comparable with the model.
6. Decide
Edge and expected value are computed per side. A value result requires both edge above the threshold and positive expected value; otherwise the result is no bet or unavailable.
Definitions
Overround
overround = implied A + implied B − 1
The bookmaker's built-in margin. A two-way market with no margin sums to exactly 1.
Example: 0.476 + 0.556 = 1.032 → 3.2% overround
No-vig probability
no-vig = implied / (implied A + implied B)
The margin-free market probability. This is the only market number compared with the model.
Example: 0.476 / 1.032 = 46.1%
Edge
edge = model probability − no-vig probability
Expressed in percentage points. Positive edge means the model rates the fighter higher than the margin-free market does.
Example: 0.520 − 0.461 = +5.9 pp
Expected value
EV = (model probability × (decimal odds − 1)) − (1 − model probability)
Expected return per one unit staked, under the model's own probability. It is an analytical measure, not a forecast of profit.
Example: (0.52 × 1.10) − 0.48 = +0.092 per unit
Decision policy
value if edge > threshold AND EV > 0, else no bet
If the model abstains, the market is missing, suspended or stale, or coverage is insufficient, the result is unavailable rather than a forced decision.
Example: threshold currently 3.0 pp
What each decision means
- Possible value
- The model's probability exceeds the no-vig market probability by more than the threshold, and expected value is positive. This is a disagreement worth recording — it is not a recommendation and it can be wrong.
- No bet
- The model and the market broadly agree, or the difference is inside the threshold. This is the most common outcome and the platform reports it plainly.
- Unavailable
- An input is missing or unusable: the model abstained, the market is closed, suspended or stale, or data coverage is too thin. No comparison is made.
How the model is judged
Accuracy alone is a poor measure for probabilistic forecasts. valUFCBet reports Brier score and log loss — both reward being well calibrated rather than merely confident — alongside a calibration table comparing predicted probability bands with observed win rates. Evaluation is run only on bouts the model had not seen during training.