Today the Bundesliga season 2012/2013 ended. It was a record breaking season of Bayern Munich, but besides this it also featured many other surprises. Especially the qualification to the Europa League by SC Freiburg and Eintracht Frankfurt was expected by virtually nobody. Both teams were expected to be in big danger of relegation. So let's see how the Goalimpact fared in predicting this outcome. Since my algorithm had some updates since the season started, I'll use the values based on the new algorithm and not the old published ones. (Intriguingly, the old algorithm predicted the standings better, so no cherry picking here. Still I'm confident that the new version is better.)
We see a positive correlation between the expected rank, based on the Goalimpact in September 2012, and the actual rank at the end of the season. Football is, probably more than any other sports, subject to the influence of luck. Therefore we should not expect the correlation to be one, but we see some big differences in the ranking.
SC Freiburg
The biggest surprise, at least if or expectations were based on the Goalimpact, was the qualification of Freiburg to the Europa League. Goalimpact was here in-line with the expectations of football pundits. It predicted Freiburg to be lucky if it would stay in the league with an expected rank of 17. Arguably, the incredible good over-performance of Freiburg is to large part the success of its trainer Christian Streich. He let Freiburg play an attractive, pressing reliant football. Due to this performance, the Goalimpact increased over the course of the season, albeit moderately to 103.8. So the Europa League participant may struggle against relegation in Bundesliga next season.
SpVgg Greuther Fürth
For whatever reason, the Goalimpact of Greuther Fürth was rather good at the beginning of the season. From all newly promoted teams, Fürth had the highest score and thus according to Goalimpact the best chances to avoid relegation with an expected rank of 9. However, the season was terrible and ended in the relegation on a distant last rank. In the course of the season, the Goalimpact dropped strongly to 101.9.
VfB Stuttgart
Stuttgart was the third biggest surprise. Another underachiever. According to Goalimpact it may have had chances to qualify for the Champions League on an expected fourth rank, but it turned out to be a disappointing 12th rank. The Goalimpact dropped 4.5 points to 109.7. As reader +Thomas Mach pointed out in a comment, Stuttgart plays the cup final and played Europa League until March. The team may not have had enough depth to compete in all three competitions.
Eintracht Frankfurt
Frankfurt's Goalimpact was quite high already at the begin of the season considering they were newly promoted to Bundesliga. They beat the expectation of rank 12 by six and qualified to Europa League. Goalimpact attributes this partly to luck and actually decreased their value from 107 to 104.7.
VfL Wolfsburg
The car city is the last surprise of this season. The underachieved by five ranks being 11th instead of the expected 6th. Still, the Goalimpact raised from 112.6 to 116.2. That may partly because of good transfers in the winter break.
All other teams are within three ranks of their expectation. Even the bad performance of Werder Bremen was expected, indicating that Thomas Schaaf was less guilty for the awful season than many claim. However, as a trainer he also didn't manage to raise the Goalimpact of his players. It actually dropped from 107 to 102.5. I'd recommend Bremen to buy some better players or face another season in fear of relegation. FC Augsburg unexpectedly managed to avoid the relegation, but only by a narrow margin. I'm personally glad they managed, because they play unconventional attractive football.
![]()
The regression shows that Goalimpact explains as much as 69% of the variation of in the goal differences. Stuttgart, Freiburg and Fürth are also outliers in this analysis, but Frankfurt and Wolfsburg are actually not that much off the regression line. Those two were lucky (unlucky) to get many (few) points out of their goals.
I'm very content with the prediction quality as this is an out-of-sample test. If we use the Goalimpact at the end of the season, and thus conduct the test in-sample, the R² is 82%. If we explain points, rather than goal difference, the R² is 65% out-of-sample and 78% in-sample. Which is 10%-points better than 68% Martin Eastwood found for the TSR in an in-sample analysis of the Premier League.
The regression also provides us also with a model how to translate the Goalimpact into an expected goal difference. The formula in the chart indicates, that every extra point in Goalimpact raises the expected goal difference by 2. A Bundesliga team with a Goalimpact of 111.6 has an expected goal difference of zero. This is about the hurdle you need to take to qualify for Europa League. To qualify for Champions League, the manager should aim for a team with a Goalimpact of 120 or more. In other words, Everton would have chances to play Champions League if they would just play Bundesliga instead of Premier League.
Overview
So here is the final table including the Goalimpact predictions.| No. | Team | Points | Goals | Goal Diff | Goalimpact Sep12 | Exp. Rank | |
| 1 | Bayern München | 91 | 98 | 18 | 80 | 144,0 | 1 |
| 2 | Borussia Dortmund | 66 | 81 | 42 | 39 | 127,6 | 2 |
| 3 | Bayer Leverkusen | 65 | 65 | 39 | 26 | 113,6 | 5 |
| 4 | FC Schalke 04 | 55 | 58 | 50 | 8 | 122,5 | 3 |
| 5 | SC Freiburg | 51 | 45 | 40 | 5 | 100,8 | 17 |
| 6 | Eintracht Frankfurt | 51 | 49 | 46 | 3 | 107,0 | 12 |
| 7 | Hamburger SV | 48 | 42 | 53 | -11 | 110,8 | 7 |
| 8 | Bor. Mönchengladbach | 47 | 45 | 49 | -4 | 108,6 | 8 |
| 9 | Hannover 96 | 45 | 60 | 62 | -2 | 108,1 | 10 |
| 10 | 1. FC Nürnberg | 44 | 39 | 47 | -8 | 107,0 | 13 |
| 11 | VfL Wolfsburg | 43 | 47 | 52 | -5 | 112,6 | 6 |
| 12 | VfB Stuttgart | 43 | 37 | 55 | -18 | 114,2 | 4 |
| 13 | 1. FSV Mainz 05 | 42 | 42 | 44 | -2 | 106,8 | 14 |
| 14 | Werder Bremen | 34 | 50 | 66 | -16 | 107,0 | 11 |
| 15 | FC Augsburg | 33 | 33 | 51 | -18 | 98,4 | 18 |
| 16 | 1899 Hoffenheim | 31 | 42 | 67 | -25 | 106,3 | 15 |
| 17 | Fortuna Düsseldorf | 30 | 39 | 57 | -18 | 106,2 | 16 |
| 18 | SpVgg Greuther Fürth | 21 | 26 | 60 | -34 | 108,2 | 9 |
We see a positive correlation between the expected rank, based on the Goalimpact in September 2012, and the actual rank at the end of the season. Football is, probably more than any other sports, subject to the influence of luck. Therefore we should not expect the correlation to be one, but we see some big differences in the ranking.
SC Freiburg
The biggest surprise, at least if or expectations were based on the Goalimpact, was the qualification of Freiburg to the Europa League. Goalimpact was here in-line with the expectations of football pundits. It predicted Freiburg to be lucky if it would stay in the league with an expected rank of 17. Arguably, the incredible good over-performance of Freiburg is to large part the success of its trainer Christian Streich. He let Freiburg play an attractive, pressing reliant football. Due to this performance, the Goalimpact increased over the course of the season, albeit moderately to 103.8. So the Europa League participant may struggle against relegation in Bundesliga next season.
SpVgg Greuther Fürth
For whatever reason, the Goalimpact of Greuther Fürth was rather good at the beginning of the season. From all newly promoted teams, Fürth had the highest score and thus according to Goalimpact the best chances to avoid relegation with an expected rank of 9. However, the season was terrible and ended in the relegation on a distant last rank. In the course of the season, the Goalimpact dropped strongly to 101.9.
VfB Stuttgart
Stuttgart was the third biggest surprise. Another underachiever. According to Goalimpact it may have had chances to qualify for the Champions League on an expected fourth rank, but it turned out to be a disappointing 12th rank. The Goalimpact dropped 4.5 points to 109.7. As reader +Thomas Mach pointed out in a comment, Stuttgart plays the cup final and played Europa League until March. The team may not have had enough depth to compete in all three competitions.
Eintracht Frankfurt
Frankfurt's Goalimpact was quite high already at the begin of the season considering they were newly promoted to Bundesliga. They beat the expectation of rank 12 by six and qualified to Europa League. Goalimpact attributes this partly to luck and actually decreased their value from 107 to 104.7.
VfL Wolfsburg
The car city is the last surprise of this season. The underachieved by five ranks being 11th instead of the expected 6th. Still, the Goalimpact raised from 112.6 to 116.2. That may partly because of good transfers in the winter break.
All other teams are within three ranks of their expectation. Even the bad performance of Werder Bremen was expected, indicating that Thomas Schaaf was less guilty for the awful season than many claim. However, as a trainer he also didn't manage to raise the Goalimpact of his players. It actually dropped from 107 to 102.5. I'd recommend Bremen to buy some better players or face another season in fear of relegation. FC Augsburg unexpectedly managed to avoid the relegation, but only by a narrow margin. I'm personally glad they managed, because they play unconventional attractive football.
Backtesting
Goalimpact actually doesn't predict the league points but goal differences. The correlation between those two numbers is higher, but if to have a mathematically view on the Goalimpacts performance, we need to compare it with the goal differences. An easy way of doing this, is a linear regression that explains the actually goal difference in the season with the Goalimpact at the start of the season.
The regression shows that Goalimpact explains as much as 69% of the variation of in the goal differences. Stuttgart, Freiburg and Fürth are also outliers in this analysis, but Frankfurt and Wolfsburg are actually not that much off the regression line. Those two were lucky (unlucky) to get many (few) points out of their goals.
I'm very content with the prediction quality as this is an out-of-sample test. If we use the Goalimpact at the end of the season, and thus conduct the test in-sample, the R² is 82%. If we explain points, rather than goal difference, the R² is 65% out-of-sample and 78% in-sample. Which is 10%-points better than 68% Martin Eastwood found for the TSR in an in-sample analysis of the Premier League.
The regression also provides us also with a model how to translate the Goalimpact into an expected goal difference. The formula in the chart indicates, that every extra point in Goalimpact raises the expected goal difference by 2. A Bundesliga team with a Goalimpact of 111.6 has an expected goal difference of zero. This is about the hurdle you need to take to qualify for Europa League. To qualify for Champions League, the manager should aim for a team with a Goalimpact of 120 or more. In other words, Everton would have chances to play Champions League if they would just play Bundesliga instead of Premier League.