Goalimpact (or its predecessor) was born in the year 2004. Jörg Seidel, the founder, was invited by his investment banker colleagues at a former company to bet on the 2004 European Championships. Jörg was not really interested in soccer (at the time, that is - now he supports Werder Bremen), but he was good at building models, mainly due to his academic background in finance and physics, plus his ability to code. Thus, he ended up creating an algorithm for the tournament that calculated probable game outcomes to fill out his betting sheet.
Initially, the algorithm was based on evaluating team strength - a sort of Elo Chess Rating for soccer. The model rated Greece as quite strong, something that no one else expected. Greece of course went on to win that tournament, and Jörg won his bet. Inspired by this success, he continued to utilise the model when betting on soccer matches. However, after the summer transfer period ended, he noticed that his model didn’t work as well anymore, mainly due to the fact that roster changes had occurred, and a team rating didn’t reflect this. He therefore changed the algorithm to rate players instead of teams.
This approach happened to work much better, as it can still estimate the strength of a team even after players change clubs. Needless to say, Jörg has won on almost every other tournament he has bet on.
This is how Goalimpact started.
How does Goalimpact work?
The algorithm measures how a player influences the goal difference per minute of their team. It only focuses on the scores, rather than the stats, of players. The reason for this top-down approach is that the game of soccer is complex. What do we mean by that? Complex systems imply that every action is connected on the field, creating something of an intricate web. Therefore, it is difficult to pin down which actions ultimately contribute to the success of a team, because everything that occurs on the pitch is intertwined.
The idea is similar to the +/- score in basketball.
However, a soccer game has an objective purpose: you play to win. As Johan Cruyff noted: “To win you have to score one more goal than your opponent.” The aim of a soccer game is to score goals and avoid conceding goals. In essence, soccer is all about a positive goal difference.
Therefore, one can define a good player in an objective manner:
A player is good if their team scores more goals and concedes less whenever they are playing - no matter why or how.
The algorithm measures the influence a single player can have on goal difference. Thus, Goalimpact has an objective definition of a “good player” underpinning its model.
The idea is similar to the +/- score in basketball, with the main difference being that with the +/- score, the opponent’s strength is not taken into account. Consequently, a high +/- score does not necessarily mean that a player is good, it could actually mean that they have only played against bad teams or players. Goalimpact solves this challenge by taking opponent strength into consideration. Ultimately, this provides Goalimpact with a relative value, rather than an absolute score, and allows users to compare every player across the world.
How does the algorithm learn?
The algorithm only requires match data, such as starting XI, goal minutes, player exchanges, and players’ dates of birth as inputs. Using these figures, the algorithm can calculate player quality, starting with 1,000 minutes of playing time. The model also takes exhaustion levels, red cards, and home field advantage into account. Since such data is readily accessible, Goalimpact has over 1,000,000 players in their database from all over the world. Using a player’s date of birth, it also corrects for the relative age bias, and allows users to predict the future potential of a player (Goalimpact Peak).
Goalimpact's algorithm takes team strength into account.
Let’s say Manchester City play against Manchester United. City score the 1-0 lead at minute 30. At half-time, United switch Casemiro for Donny van de Beek. United equalise at minute 60, and 1-1 remains the final score. Assuming equal team strength in this very simple scenario, only two player ratings would be adapted, since the final score was 1-1 and thus most players had a goal difference of zero. Casemiro’s rating would be increased, and van de Beek would be downgraded by the algorithm because their unique scores were 1-0 and 0-1, respectively.
One important thing to note: this assumes it is the first game played between the teams, and all players are equal in performance. In all subsequent games, the Goalimpact algorithm will take team strength into account - that is to say, if Manchester City play against Plymouth Argyle, we would expect City to win with maybe a goal difference of three. If City only win 1-0, the Plymouth players would have their ratings increased despite losing because they lost by fewer goals than expected.
However, one game is not enough to determine the rating with confidence, since one game can be quite random after all. With additional data, the Goalimpact algorithm can adapt, and only player quality will remain.
Anyway, enough dry theory – how about an example?
Case study: the discovery of Alphonso Davies
One success story for Goalimpact was Alphonso Davies’ rating. He played in Canada, and had a market value of less than €500,000 when Goalimpact began rating his potential as world-class.
The Goalimpact graph of Alphonso Davies.
The blue line details his career average Goalimpact rating over time,
while the red-dotted line is the career Goalimpact predicted rating based on average player development.
In general, a Goalimpact rating above 140 indicates top five European League potential. For comparison, the best players in the world have a Goalimpact rating of over 170. The players with the highest current Goalimpact rating are Thomas Müller, Sadio Mané, and Dani Carvajal.
How Goalimpact can aid bettors
With Goalimpact’s objective and bias-free approach, it provides a prediction model that is close to reality.
The sum of players’ Goalimpact ratings on the field is an estimate of the team’s total strength. Thus, the expected goal difference between two competing teams can be derived from the teams’ strength, and using the goal distribution of a game, one can obtain the probabilities for potential results.
Goalimpact therefore offers probabilities and outcomes for matches, and can be used as an additional source of information for betting. Bettors can use such information to compare differences between Pinnacle’s betting odds and markets with Goalimpact’s metrics.
Get the latest soccer odds on all of this season’s matches from the English Premier League and more at Pinnacle. You can keep up to date with Goalimpact via Twitter here, and find their comprehensive database here.