Using Betting Exchanges as a Probability Benchmark

Choose the price you mean

In Great Britain an exchange is licensed as a remote betting intermediary: it brings betting parties together without having liability for their bets, so the prices it shows are offers from other customers. An exchange market can expose best available back offers, best available lay offers, last traded price, traded volume and deeper ladder levels. Betfair's getting-started guide explains backing, laying and matching, while its historical data specification defines timestamped best-price, traded-price, volume and market-status fields.

Declare one field and a timestamp rule before comparing it with a model or sportsbook.

Illustrative spread calculation

Suppose best back is 2.40 and best lay is 2.46.

• Back raw implied probability = 1 / 2.40 = 0.41667, or 41.667%.
• Lay raw reciprocal = 1 / 2.46 = 0.40650, or 40.650%.
• Reciprocal spread = 41.667% - 40.650% = 1.017 percentage points.
• Decimal midpoint = (2.40 + 2.46) / 2 = 2.43.

The midpoint is descriptive, not guaranteed executable. A benchmark record should retain both sides and their available sizes.

Include commission in a decision

Exchange commission may be charged on net market winnings and varies by account or market; Betfair's commission documentation provides one operator example, and any operator can change its terms. Expected value weights each payoff by its probability, as in OpenStax's definition. For a back bet at 2.40 with 5% illustrative commission on positive winnings, net win profit = (2.40 - 1) * (1 - 0.05) = 1.33 units. If p = 0.43, net EV = (0.43 * 1.33) - (0.57 * 1) = 0.0019 units.

Ignoring commission would give EV = (0.43 * 1.40) - (0.57 * 1) = 0.032 units. The cost changes the interpretation materially.

Exchanges are not uniformly the best forecast

A multi-sport study found exchange-derived probabilities were not always the most accurate source, especially in smaller markets, and that conversion method mattered (Štrumbelj, 2014). An earlier study of matched UK horse-racing data found significantly lower market biases on betting exchanges than in traditional betting media (Smith, Paton and Vaughan Williams, 2006); that is evidence from another sport and period, not a ranking for any football market. Treat an exchange as a named market benchmark that must be validated for the relevant sport, market, liquidity and timestamp.

Minimum benchmark record

• Market and selection IDs.
• Regulation or settlement definition.
• Published and retrieval timestamps.
• Best back and lay prices and sizes.
• Last traded price and its time, if used.
• In-play, suspended and complete flags.
• Commission assumption and account basis.
• Method for multi-outcome normalization.

Diagnose a thin exchange market

Suppose best back is 2.40 for only £3, best lay is 2.60 for £4, and the last trade at 2.42 was several hours earlier. A midpoint would conceal a wide spread, tiny executable size, and stale trade. Record the market as thin rather than converting 2.50 into a precise consensus probability.

Betfair's historical feed specification separates best available levels, last traded price, traded volume, published time, and market status. Use those fields to define minimum freshness and depth requirements before analysis. If the requirements fail, retain the row with a quality flag and exclude it under a rule fixed before outcomes are inspected.

Continue learning

• Next guide: Using Pinnacle as a Reference Price
• Related guide: Using xG in Value Betting

Assumptions and limitations

The prices and 5% commission are illustrative. Actual commission and access vary. An exchange quote does not remove model uncertainty or guarantee execution. The exchange odds guide covers mechanics; this page owns probability-benchmark construction.