Separate three layers
| Layer | Observable inputs | Appropriate conclusion |
|---|---|---|
| Match state | Clock, score, player count, period, restart and review status | What was known about the match |
| Forecast | Declared model, parameters, features and update timestamp | What one method estimated |
| Execution | Available prices and size, traded volume, status, request and match time | What could be transacted |
The Betfair historical feed specification documents publish time, market status, available-to-back and available-to-lay ladders, and traded volume for one exchange dataset. Those fields allow a price path to be reconstructed, but they do not identify the private reasons behind every order.
Elapsed time changes the remaining event
At 0-0 before kick-off, the market concerns the full qualifying period. At 0-0 late in the match, less playing time remains. A forecast should therefore condition on actual elapsed and remaining time rather than interpret the same score as the same state. The Bayesian in-play football model is one peer-reviewed example of updating outcome probabilities as match information arrives, with model assumptions and posterior checks that must be evaluated on the intended data.
Material events change inputs
A goal, sending-off, penalty decision, substitution, injury, or change in competition state can alter a model's inputs. The direction and size are not universal: score state, team strength, remaining time, player identity and model design matter. Under the IFAB VAR protocol, checks and reviews apply to specified match-changing incidents and the referee makes the final decision. A market can therefore suspend while the visible incident is unresolved.
Market mechanics affect executable price
On an exchange, the last traded price, best available price and volume-weighted execution price are different objects. Betfair's exchange rules explain matched and unmatched requests, market administration and matching for that product. A small trade at one price does not establish that a larger stake could have been filled there.
Worked decomposition
Consider an illustrative home-win price path:
| Time | Match state | Best displayed price | Available size | Status |
|---|---|---|---|---|
| 18:00 | 0-0, 11 v 11 | 2.40 | GBP 120 | Open |
| 18:15 | home goal shown | - | - | Suspended |
| 19:05 | goal confirmed | 1.62 | GBP 45 | Open |
| 19:20 | same score | 1.64 | GBP 300 | Open |
The table supports a narrow description: the market suspended around a shown goal and reopened with a shorter home-win price. It does not prove how much of the change came from the goal, elapsed time, order flow, model revision, or prior information. The numbers are invented to demonstrate the record.
Validation protocol
- Use one clock convention and UTC timestamps.
- Join event and price feeds without using corrections that arrived after the decision cutoff.
- Mark suspended intervals as unavailable rather than carrying the previous price forward.
- Compare the requested price with the volume-weighted accepted price.
- Predeclare the event window and model before reading the outcome.
- Test predictions on later matches and report calibration as well as discrimination.
- Retain no-cause as the conclusion when the data cannot distinguish explanations.
The Gambling Commission's in-play standard is a reminder that event information can reach participants at different times. A broadcast timestamp and a market publish timestamp should not be treated as simultaneous without evidence.
Next step
Use Odds Movement Explained for the next part of this topic.
Continue learning
- Next guide: In-Play Over/Under Goals
- Related guide: Late-Goal Statistics
Assumptions and limitations
The example is illustrative and is not a forecast or trading result. Exchange feed fields do not generalise automatically to sportsbook products. Public data cannot normally reveal operator parameters, private information, or participant intent, so causal language requires a separate identification design.
GambleAware
GamCare
Gambling Therapy
GAMSTOP