Pi rating and xG Elo explained: how to read football ratings

Sign in to Match Terminal and open a fixture with ratings data. The Pre-match model outlook is free for signed-in users. It puts two ratings beside each team, Pi and xG Elo, then gives home, draw and away chances and an expected goal margin. Each number answers a slightly different question about the match.

What xG means in football

Expected goals, or xG, estimates the chance that a shot becomes a goal. A 0.20 xG shot would be scored about twice in ten similar attempts. Add up a team's shots and you get its match xG. The calculation considers details such as distance, angle and how the chance was created. Providers can give the same shot different values because their models use different data. Hudl StatsBomb's xG guide explains the shot-level calculation.

On our fixture page, xG Elo is a team-strength rating. It is not a forecast of how many expected goals the team will create in this match. Past xG helps update the rating after some matches, as explained below.

How our Pi rating works

Our Pi-style model keeps separate home and away ratings for each team. The number beside a club is the rating for its venue in this fixture. A higher number means greater strength in that setting, so the same club may show a different Pi rating on its next away trip.

After a match, Pi compares the goal difference with its pre-match expectation. When a team's goal difference beats that expectation, its rating rises. When it falls short, the rating drops. The ratings carry on across seasons.

Take a match where Pi expects the home team to finish 0.73 goals ahead. A 1-0 home win is only 0.27 goals above that expectation, so the update is small. A 3-0 win is 2.27 goals above it and produces a larger rise. A 0-0 draw falls 0.73 goals short, so the home team's rating drops even though it did not lose. The model responds to the gap between expectation and result, not simply the W, D or L in the form guide. This is why a comfortable win can tell it more than a narrow one.

The home and away ratings learn from one another. A better-than-expected home result mainly changes that team's home rating, but its away rating gets a smaller nudge in the same direction. An away fixture can therefore tell the model something about the team you'll face at home next month, while still keeping the venues distinct.

The panel's Pi expected goal margin combines the two venue ratings with a home advantage allowance. It is shown from the home team's perspective. At +0.73 goals, the model expects the home side to finish about 0.73 goals ahead on average; a negative value favours the away side. This is an average across possible outcomes. It does not specify a final score or a home-win percentage.

Constantinou and Fenton's original pi-rating research used score differences to rate football teams. Match Terminal uses its own approximation of that approach, so its numbers will differ from the authors' published ratings. Our Pi model also has a fixed home advantage term when it estimates the margin for a fixture. Separate home and away team ratings still capture differences in how each club performs at those venues.

Why a new rating needs matches

Our Pi ratings start at zero for an unseen team. Zero is the model's starting point, not evidence that the team is average. Every result moves it towards an estimate supported by match history. Early numbers can therefore change sharply as the model learns about a newly seen side.

The rating-development figure from Constantinou and Fenton illustrates this cold-start problem in their original model. Six English Premier League teams began at zero in the first match of 2007/08. The authors wrote that roughly two seasons of results, about 76 matches per team, might be enough for acceptable estimates under their chosen settings; teams further from the average, such as Chelsea and Manchester United in that example, could need another season. Those are findings about their historical experiment, not a fixed waiting period for Match Terminal's ratings. They explain why a team's first few values deserve more caution than a rating built from years of results.

How xG Elo works

Elo starts teams on a common points scale and adjusts their ratings after completed matches. The change depends on how a team performed against the pre-match expectation. In our model, a club rated 1,589 is rated more strongly than one at 1,543. The 46-point gap has to go through the model before it becomes a win chance; it does not mean 46%.

When match xG is available for both teams, our update blends an xG-based performance score (70%) with the result (30%). If either team's xG is missing, it uses the result alone. A team's rating history can therefore include matches without xG. This is Match Terminal's xG-adjusted Elo approximation, not an official Opta rating.

That blend matters when the score and the chances tell different stories. Imagine two evenly rated teams and a 0-1 home defeat in which the home side created 2.0 xG to the visitor's 0.5. The loss pulls the rating down, while the stronger xG performance pulls it up. In our model, the combined signal can even lift the home team's rating despite the defeat. If xG was unavailable for either team, the same result would be judged from the score alone.

The panel's home, draw and away percentages come from xG Elo. The model compares the teams' ratings, accounts for home advantage and assigns a draw chance. The three percentages add up to about 100%, allowing for rounding. They are estimates, not betting odds or promises about the result.

There is no direct conversion from a Pi number to the three percentages shown on our page. That matters when comparing models elsewhere: a researcher can build match probabilities from Pi ratings, but those probabilities will depend on the extra method used to turn ratings into forecasts. The penaltyblog Pi-rating example did exactly that. On its chosen Dutch league test, its Pi setup recorded a ranked probability score of 0.199 against 0.204 for its Elo setup, where lower is better. These figures describe that author's models and test data. They do not measure the accuracy of Match Terminal's Pi or xG Elo ratings.

Reading the panel on a fixture

These illustrative values show how the pieces fit together:

The panel also shows a 21.0% draw chance and a +0.73 Pi expected goal margin. Both ratings favour the home side. A home win is the most likely single result under xG Elo, but the draw and away win together still account for almost half the probability. The Pi margin describes an average lead. It does not predict a 1-0 score or an exact one-goal win.

Compare teams within each column. Pi and Elo use different scales, so 1.51 Pi cannot be compared directly with 1,589 Elo. If they disagree about which team has the edge, look at the recent games and team statistics. Pi responds to score margins and venue; xG Elo can also respond to chances created in past matches.

What the ratings miss

The ratings are saved before kick-off, using earlier matches. They cannot account for a starting line-up announced later or an injury just before the game. A new or lightly observed team also has less match history behind its rating. Treat the displayed percentages as estimates even when they appear to one decimal place.

To see the free ratings, sign in and open an eligible Match Terminal fixture. Read them alongside the line-ups and recent matches before making a prediction.

Statistics and predictions on this site are informational only. Gamble responsibly, never stake what you can't afford to lose, and if betting stops being fun, BeGambleAware has free, confidential help. 18+