On this page
- What live win probability measures
- A fictional match: the score, clock and cards change the question
- Why xG is not a win percentage
- How a model updates its estimate
- A probability estimate can be useful and still be wrong
- Model probability and odds-implied probability are different
- Frequently asked questions
- Sources and scope
A team can take no shots for several minutes and still see its win probability change. The clock is moving, the score may be changing, and a red card can change how the remaining play is expected to unfold. Live win probability estimates the final outcome from a particular match state. Expected goals, or xG, measures the scoring likelihood of shots. They answer different questions.
Before reading a live percentage, check which team it refers to, whether a draw is a separate outcome, when it was updated, and what counts as the end of the match. A forecast for normal time plus added time is different from a forecast of who advances after extra time or penalties.
What live win probability measures
For a match that finishes normally, a three-outcome forecast assigns probabilities to a home win, a draw and an away win. At one timestamp, with the same information and finishing rules, these outcomes are mutually exclusive and cover all the possibilities in that forecast. Their probabilities add to 100%. They are not three independent events.
For example, a fictional estimate of 45% home win, 30% draw and 25% away win describes uncertainty about the final result. It does not mean the home side has completed 45% of the work needed to win. Nor does it promise that the home side will win this particular match.
Opta describes live win probability as a forecast of win, draw and loss before and during a match. Its published explanation models future scoring rates and simulates the remaining play, using information such as team quality, the current score, time remaining, chance quality and red cards. These are model estimates, rather than measurements of a known underlying probability. Opta Analyst’s explanation describes the approach; the current MA21 developer documentation documents home, draw and away predictions and relevant match inputs.
A fictional match: the score, clock and cards change the question
Imagine a standard match scheduled for two 45-minute halves. In this example, the outcome is the result after normal time and any added time, assuming the match completes normally. Extra time, a penalty shootout, qualification, abandonment and a market being void are outside the example.
The percentages below are assigned for teaching. They are not calculated by a football model, taken from a live match or produced by Opta. Their purpose is to show how a complete distribution can change as the match state changes.
| Match state | Score, home–away | Home red cards | Home win | Draw | Away win |
|---|---|---|---|---|---|
| 20 minutes | 0–0 | 0 | 45% | 30% | 25% |
| 75 minutes | 0–0 | 0 | 35% | 45% | 20% |
| 76 minutes, home goal | 1–0 | 0 | 75% | 18% | 7% |
| 78 minutes, home red card | 1–0 | 1 | 60% | 25% | 15% |
At 75 minutes, there is less time left to break the 0–0 tie than at 20 minutes. The fictional draw probability is higher. That direction is plausible in this example, but it is not a rule that the draw probability must rise whenever time passes: the rest of the match state matters too.
The home goal changes the starting score for the remaining play. The away team now needs at least one goal to avoid losing. The later red card changes the home team’s circumstances while it is protecting the lead. In the assigned figures, the home-win estimate falls from 75% to 60%, a decline of 15 percentage points. This is not a universal red-card adjustment. A real model’s response depends on its inputs and assumptions, including the score and time left.
Added time matters because the amount of play remaining is part of the question. It is not the same thing as the separate extra-time periods a competition may use to resolve a tie. IFAB Law 7 sets out normal match duration and allowances for time lost; Law 10 covers the result and permitted procedures when competition rules require a winner.
Why xG is not a win percentage
A shot with 0.30 xG has an estimated 30% chance of becoming a goal under that shot model. A team’s accumulated xG adds the expected goals attached to its assessed shots. Neither number is the team’s probability of winning the match. Opta’s sports data glossary distinguishes the shot-scoring question from the win, draw and loss question.
This difference has two parts. First, an expected count of goals is a different quantity from the probability of a final result. Dividing one team’s xG by the two teams’ combined xG does not turn it into a valid win probability. Second, accumulated shot xG describes chances already recorded. A live outcome model must also consider what might happen in the remaining play.
Two teams can have equal accumulated xG while one leads 1–0. The same accumulated xG and score can also mean different things with ten minutes left and with an entire half left. A model may use chance quality so far as evidence about future scoring, but its predicted future scoring rates are not simply the past xG totals relabeled.
If neither team takes another assessed shot for a few minutes, their accumulated shot xG can stay unchanged while a live outcome forecast changes as time runs down. Conversely, a shot can add xG without changing the score. Read the guide to expected goals in football for shot values, totals and the limits of comparing providers.
How a model updates its estimate
A useful way to understand the process is to separate three steps: describe the current match state, estimate how the teams might score from here, and convert possible remaining scorelines into final outcomes. The published Opta explanation uses simulations for that last step. A goal or red card changes the state; a change in estimated remaining time changes the opportunity for future goals.
The details belong to the model and its version. Opta’s 2021 explainer describes the approach at that time; it is not a specification of every current product’s training window or coefficients. The current Opta Predictions overview describes forecasts using historical, seasonal and live information, but it does not let you reproduce a proprietary model or establish its accuracy from the product description alone.
Check the display’s timestamp as well as the match clock. A probability may reflect information received before an event you have already seen. A corrected event, a revised time estimate or a different provider’s inputs can also produce different numbers. A refresh rate is not a guarantee that every displayed percentage includes every event immediately.
Also check the outcome horizon directly. A data feed’s ability to carry extra-time or shootout events does not, by itself, establish whether a particular win-probability field covers normal time, extra time or eventual qualification. Use the provider’s definition for that display. The fictional table above has an explicit scope so its draw outcome remains unambiguous.
A probability estimate can be useful and still be wrong
A team assigned a 60% chance can lose. The estimate leaves 40% for the other outcomes combined; an upset alone does not show that the model was badly calibrated. Calibration asks whether events assigned similar probabilities occur at roughly those frequencies over a suitable collection of forecasts.
That requires many relevant predictions, known outcomes and a clear evaluation method. Repeated snapshots from one match are related observations, so they should not casually be treated as many independent matches. Research on in-game win probability in soccer illustrates the use of calibration evaluation for a particular academic model. Its results are not an accuracy audit of today’s Opta products.
The practical limits include the quality and timeliness of the data, the competitions represented in the model’s development, and how well its assumptions fit the match being watched. A percentage with a decimal place can still carry substantial uncertainty. General probability reasoning is covered in expected value and probability.
Model probability and odds-implied probability are different
A live model estimates an outcome from its data and assumptions. An odds-implied probability is derived from a quoted price. The price can incorporate a bookmaker’s margin and other pricing decisions; converting it to a percentage does not establish the true chance of winning.
Before comparing the two, match the teams, outcome, timestamp and finishing rules. A model’s normal-time home-win probability cannot be compared directly with a price for the home team to qualify after a shootout. Even with matching definitions, a difference between two estimates is not proof of a reliable advantage or a profitable trade. The odds and implied probability guide explains the conversion and why a complete set of quoted probabilities may exceed 100%.
Frequently asked questions
Does 60% win probability mean the team should win?
It means that winning is assigned 60% under that model, at that moment and for that defined outcome. The remaining 40% covers a draw or the opponent winning in a three-outcome forecast. It is not a guarantee or a prediction of the exact score.
Can win probability change without a shot or goal?
Yes. Less time remains to change the score, and other relevant information can change. Accumulated shot xG may stay the same while the live outcome estimate moves. The size and direction of a change depend on the match state and model.
Does a higher xG total mean a team is more likely to win?
Not on its own. You also need the actual score, the time left and the rest of the information used by the model. Accumulated xG is not a substitute for a complete outcome forecast, and its ratio is not a win percentage.
Do live win probabilities include extra time and penalties?
Check the definition of the particular display. A normal-time forecast can include a draw, while a forecast of which team advances answers a different question. In this guide’s fictional example, the horizon is normal time plus added time, excluding extra time and shootouts. A competition’s sporting rules and a betting service’s settlement terms also need to be checked separately.
Explore market questions
Read each market’s question, outcome rules and applicable service conditions before considering participation.
Sources and scope
Uses Opta’s published explanation and documentation for the provider’s approach, IFAB for match-duration distinctions, and a separate academic study for calibration concepts. The match snapshots are assigned teaching figures, not a model calculation or current match forecast. The example assumes normal completion after normal time plus added time.
- Opta Analyst: Explaining Live Win Probability (LWP)
- Stats Perform: Soccer Live Win Probability Feed (MA21)
- IFAB Law 7: The Duration of the Match
- IFAB Law 10: Determining the Outcome of a Match
- Stats Perform: Sports data glossary
- Stats Perform: Opta Predictions
- Robberechts, Van Haaren and Davis: A Bayesian Approach to In-Game Win Probability in Soccer
HypaForecast publishes educational guides and operates a prediction-market service. This guide does not establish that a football market or live forecast is available through the service, or that a reader is eligible to use it.
Share this guide