Value Betting in Football: Evidence, Calculations and Limits

Start with the decision, not the result

A value decision needs four items recorded before kickoff: an estimated probability, the offered decimal price, the exact market and settlement terms, and the time the price was available. Expected value then combines the possible returns using their probabilities; OpenStax's expected-value treatment provides the underlying weighted-outcome principle.

For a one-unit back bet at decimal odds 2.20 with estimated win probability 0.48:

EV = (0.48 * 1.20) - (0.52 * 1) = 0.056 units

That is an illustrative model expectation of 5.6% of stake, not a promised return. If the defensible probability were 0.44 instead, EV = (0.44 * 1.20) - (0.56 * 1) = -0.032 units. A four-point probability change reverses the decision.

The evidence chain

1. Define the event. The model target and the bet must use the same settlement period and outcome.
2. Estimate probability. Freeze the model and information cutoff before observing the result.
3. Test probability quality. Compare later forecasts with outcomes using proper scores and calibration. scikit-learn's calibration guide explains why a 60% forecast should be evaluated across comparable predictions, not from one match.
4. Capture an executable price. Record operator, market, selection, timestamp, availability and accepted stake.
5. Apply costs and uncertainty. Include commission where relevant and test probability sensitivity.
6. Settle every qualifying decision. Preserve voids, rejected bets and missing prices rather than deleting inconvenient rows.

Market prices are useful but not truth

Odds can provide strong public probability forecasts, but the conversion method and source matter. A study covering 37 competitions found differences between normalization, Shin and regression methods, and found that exchange odds were not uniformly the most accurate source in smaller markets (Štrumbelj, 2014). Another 11-league study found different efficiency results by league and by whether mean or best available prices were used (Angelini and De Angelis, 2019). Neither study establishes a timeless rule for today's market.

A minimum decision record

Assumptions and limitations

This collection is educational and uses illustrative prices unless a dated dataset is cited. Positive expected value can coexist with long losing periods, and historical results do not guarantee future performance. Betting should use money that is affordable to lose; the NHS gambling-harm guide provides support routes and practical safeguards.