How Value Bets Work
Every set of odds carries an implied probability. Decimal odds of 2.00 imply a 50% chance (1 divided by 2.00). Odds of 4.00 imply a 25% chance. The formula is straightforward:
Implied probability = 1 / decimal odds
A positive estimate occurs when p is above the raw break-even probability. At odds of 3.00, break-even is 1 / 3.00 = 33.33%. If a pre-event model estimates 42%, the expected net return per unit is (0.42 x 3.00) - 1 = 0.26, or 26% before fees, using the standard expected-value calculation.
The probability difference is 8.67 percentage points. The 26% expected-return figure is not the same unit. The edge definition explains this distinction in more detail.
A Practical Football Example
Assume an illustrative model assigns a selection p = 52% and a 10 pound bet can be accepted at 2.20. The modelled outcomes are:
| Outcome | Probability | Net result | Probability-weighted result |
|---|---|---|---|
| Win | 0.52 | +12.00 pounds | +6.24 pounds |
| Loss | 0.48 | -10.00 pounds | -4.80 pounds |
| Expected net result | +1.44 pounds |
The same result follows from 10 x ((0.52 x 2.20) - 1) = 1.44. This is the ordinary weighted-outcome definition of expected value. It verifies the arithmetic, not the 52% input.
Value Is Conditional
The example is positive only under its stated probability, price, stake, and fee assumptions. A 45% estimate at odds of 2.20 has expected net return (0.45 x 2.20) - 1 = -1%; a small probability error can therefore change the sign. For a market with several outcomes, remove the overround before treating market probabilities as a comparison baseline.
Even when the true expected value is positive, a finite series can lose. Conversely, a weak process can win over a finite series. Realised return and forecast quality should be reported separately.
How to Estimate True Probability
No method reveals true probability directly. A credible estimate should preserve time order as documented by TimeSeriesSplit and test forecast probabilities using calibration diagnostics. It should:
- define the outcome and settlement period precisely;
- use information available at the decision timestamp;
- preserve time order between model training and testing;
- beat relevant simple and market baselines on unseen events;
- be checked for calibration and probability score;
- include every decision generated by the rule; and
- record the price and stake actually accepted.
TimeSeriesSplit documents why training on future observations creates an invalid time-ordered test. Comparing prices can improve execution for an already selected outcome, but an outlying price does not prove the outcome itself has value.
The Relationship Between Value and Profit
An outcome estimated at 55% can lose. That is ordinary uncertainty, not evidence by itself that the estimate was wrong. Repeated forecasts need a proper evaluation rather than explanations written after each result; scikit-learn's calibration guidance explains how predicted probabilities are compared with observed frequencies across groups of forecasts.
This is why every probability estimate should be logged and evaluated out of sample. There is no universal bet count at which an assessment becomes reliable: precision depends on the forecast distribution, edge size, dependence between bets, and evaluation method. Calibration plots and uncertainty intervals reveal more than a fixed threshold; scikit-learn's calibration guidance describes the comparison between predicted probabilities and observed event frequencies.
Past performance does not guarantee future results. A process should be revised when new, properly separated evidence contradicts it; "trust the process" is not a substitute for validation.
18+. Gambling involves risk. For support, visit begambleaware.org. Please gamble responsibly.
Related resources
Continue with Expected Value: What It Means in Betting for the next part of this topic, or return to Betting Glossary: Every Betting Term Explained in Plain English to compare the other guides in this collection.
Continue learning
- Next guide: Void Bet
- Related guide: Yield
- Go deeper: What Is Value Betting? Probability, Price and Uncertainty
Assumptions and limitations
A value conclusion is conditional on a probability estimate, an executable net price, and matching settlement rules. The arithmetic can show model-market disagreement but cannot prove that the model is calibrated or that a profit will be realised. OpenStax expected value supports the calculation; future validation must use data unavailable when the estimate was made.
GambleAware
GamCare
Gambling Therapy
GAMSTOP