Samstag, 23. August 2014

Borussia Dortmund vs Bayer Leverkusen: Goalimpact of Lineups

Dortmund has the slightly better starting XI and due to the home field advantage are expected to win. Betfair favors Dortmund slightly more.

Borussia Dortmund: 51.9%
Draw: 28.0%
Bayer Leverkusen: 20.1%

Borussia Dortmund

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Marco Reus114.5116.225.2Deutschland27121167
Miloš Jojic109.6119.922.3Serbien [U21]734733
Erik Durm104.3115.222.3Deutschland [U21]1209308
Sebastian Kehl104.1142.734.5Deutschland43435004
Mitchell Langerak108.8124.625.9575153
Pierre-Emerick Aubameyang109.0110.925.1Gabun23616667
Sokratis Papastathopoulos105.9106.326.2Griechenland24920504
Ciro Immobile105.4108.724.4Italien1419801
Matthias Ginter107.0127.620.5Deutschland [U21]13011781
Lukasz Piszczek130.1136.929.2Polen26621319
Henrikh Mkhitaryan128.6129.625.5Armenien22418865
Average111.6121.725.6
Bench
Average0.00.00.0


Bayer Leverkusen

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Hakan Çalhanoglu97.7118.520.5Türkei13510865
Gonzalo Castro122.3125.527.2Deutschland [U21]36529722
Emir Spahic87.6121.733.9Bosnien-Herzegowina24922434
Bernd Leno106.6153.822.4Deutschland [U21]20419034
Karim Bellarabi104.6108.124.3Deutschland [U21]1248615
Stefan Kießling121.3133.130.5Deutschland40630012
Sebastian Boenisch112.1116.327.5Polen17513347
Tin Jedvaj90.3128.918.7Kroatien [U21]221883
Heung-Min Son96.6108.022.1Südkorea15510400
Simon Rolfes125.1152.132.5Deutschland48840314
Ömer Toprak119.7121.725.0Türkei22319068
Average107.6126.225.9
Bench
Average0.00.00.0



Hertha BSC vs Werder Bremen: Goalimpact of Lineups

Based on the starting XI, Hertha BSC is a 57.0% favorite to win this match. Draw probability is 26.3%, and Werder Bremen might be lucky to get three points with a 16.7% chance. Thus Goalimpact is more optimistic about Berlin's chances than the betting markets are.

Hertha BSC

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Ronny111.3117.228.316610330
Hajime Hosogai95.2101.028.2Japan17814389
Peter Niemeyer90.1103.430.7Deutschland [U21]30822787
Thomas Kraft104.4119.526.022520785
Genki Haraguchi95.1102.223.317112437
Peter Pekarík107.3112.327.8Slowakei19416291
Julian Schieber119.7120.825.5Deutschland [U21]20210938
Sebastian Langkamp94.896.326.6Deutschland [U21]18015831
Johnny Heitinga139.4152.930.8Niederlande44837423
Roy Beerens106.7108.526.6Niederlande Olymp.31023540
Nico Schulz95.7111.421.3Deutschland [U21]1198159
Average105.4113.226.8
Bench
Jens Hegeler109.3110.826.522114245
Sami Allagui96.7102.628.2Tunesien22015214
Rune Jarstein103.8107.729.8Norwegen19918383
Sandro Wagner96.698.526.7Deutschland [U21]22211627
John Anthony Brooks109.2123.821.5USA1068773
Marcel Ndjeng102.6128.132.330723211
Änis Ben-Hatira103.6103.726.0Tunesien19111656
Average103.1110.727.3


Werder Bremen

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Zlatko Junuzovic93.095.426.8Österreich34529159
Cédric Makiadi87.398.530.4DR Kongo30823742
Clemens Fritz89.6122.133.7Deutschland41334372
Assani Lukimya99.6105.828.524119838
Izet Hajrovic97.0104.923.0Bosnien-Herzegowina1167117
Santiago García93.493.626.11028780
Eljero Elia90.094.227.5Niederlande29819341
Franco Di Santo85.987.425.3Argentinien1769534
Luca Caldirola93.9100.023.5Italien [U21]12010735
Felix Kroos101.9108.423.416811315
Raphael Wolf81.895.926.216515275
Average92.1100.626.8
Bench
Davie Selke92.9121.919.5Deutschland [U19]715057
Gálvez99.6101.425.2746359
Nils Petersen101.9102.725.7Deutschland [U19]22314514
Marnon Busch94.3122.019.7665694
Richard Strebinger94.1152.921.5Österreich [U19]837599
Ludovic Obraniak98.2105.529.8Polen36125422
Fin Bartels91.395.527.526120034
Average96.0114.624.1



Freitag, 22. August 2014

Bayern München vs VfL Wolfsburg: Goalimpact of Lineups

The odds based on starting XI show Bayern as clear favorite. But Goalimpact is slightly more skeptical than the betting markets. This is at least partly due to the low ratings of Bernat and Gaudino.

Bayern München: 66.3% (Home)
Draw: 22.3%
VfL Wolfsburg: 11.5%

Bayern München

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Manuel Neuer191.4197.128.3Deutschland41038408
Dante133.6147.830.8Brasilien31328034
Juan Bernat105.1120.321.4Spanien [U21]855752
Philipp Lahm195.9209.530.8Deutschland58552427
David Alaba131.6142.922.1Österreich24419998
Holger Badstuber128.3130.924.8Deutschland23921340
Gianluca Gaudino67.0114.717.8271931
Mario Götze147.7158.722.2Deutschland20514057
Robert Lewandowski144.0144.125.9Polen28222524
Arjen Robben153.4165.330.5Niederlande49936884
Thomas Müller173.5175.924.9Deutschland35728296
Average142.9155.225.4
Bench
Pepe Reina171.2173.331.9Spanien62658177
Xherdan Shaqiri130.1138.822.8Schweiz21213884
Lucas Scholl65.0110.318.1241395
Sebastian Rode106.9111.823.8Deutschland [U21]14812119
Pierre-Emile Højbjerg105.1140.119.0Dänemark [U21]705158
Claudio Pizarro103.9155.435.8Peru61044418
Julian Green101.9135.119.2USA564300
Average112.0137.824.4


VfL Wolfsburg

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Max Grün111.4118.327.311610755
Sebastian Jung123.1127.024.1Deutschland [U21]24321798
Naldo116.0139.431.9Brasilien32929657
Robin Knoche119.6130.522.2Deutschland [U21]15913773
Ricardo Rodríguez105.7117.721.9Schweiz14512533
Vieirinha106.3112.528.5Portugal23216692
Aaron Hunt98.5103.927.9Deutschland [U21]33023326
Kevin De Bruyne107.3114.823.1Belgien18313863
Luiz Gustavo137.6140.527.0Brasilien25219979
Josuha Guilavogui109.4114.123.8Frankreich1309474
Ivica Olic100.4142.934.9Kroatien35323729
Average112.3123.826.6
Bench
Patrick Drewes105.9164.421.5797378
Marcel Schäfer100.1108.930.2Deutschland36130493
Timm Klose107.0107.726.3Schweiz14112040
Daniel Caligiuri111.4112.926.619713197
Junior Malanda98.3122.919.9Belgien [U21]786345
Maximilian Arnold112.7135.420.2Deutschland [U21]1099092
Bas Dost103.3105.125.2Niederlande [U21]18413690
Average105.5122.524.3





Season Prediction

In lack of time, we just publish the pre-season estimates as plain table. For those leagues that started already, the few games so far have been ignored.

Premier League

Rank Team Goalimpact Peak GI Age
1 Manchester City 135.8 146.3 28.8
2 Chelsea FC 133.4 140.5 27.0
3 Arsenal FC 124.6 132.4 25.9
4 Manchester United 123.3 131.2 27.2
5 Tottenham Hotspur 119.6 125.2 26.3
6 Liverpool FC 119.1 129.8 26.1
7 Everton FC 112.7 123.3 27.4
8 Southampton FC 109.2 115.0 26.5
9 Swansea City 107.2 113.6 26.1
10 Newcastle United 106.4 113.5 25.1
11 Leicester City 106.0 115.1 25.9
12 Aston Villa 105.9 110.8 25.4
13 West Bromwich Albion 105.6 113.1 26.5
14 Hull City 104.6 110.7 25.9
15 Sunderland AFC 104.5 110.8 26.3
16 Stoke City 103.8 108.6 26.0
17 West Ham United 103.6 111.4 26.7
18 Queens Park Rangers 103.3 112.9 27.4
19 Burnley FC 102.9 108.4 25.4
20 Crystal Palace 101.7 108.7 27.6

Bundesliga

Rank Team Goalimpact Peak GI Age Bwin Rank TM.de ClubElo EuroClubIndex
1 Bayern München 149.9 157.9 26.8 1 1 1 1
2 Borussia Dortmund 127.8 130.9 26.4 2 2 2 2
3 FC Schalke 04 118.3 127.3 25.5 3 3 3 3
4 VfL Wolfsburg 117.2 124.0 25.3 5 4 5 5
5 VfB Stuttgart 112.6 120.1 26.1 8 8 15 15
6 Bayer Leverkusen 112.0 124.0 25.4 4 5 4 4
7 Hamburger SV 111.7 118.4 25.9 9 9 17 16
8 1899 Hoffenheim 111.7 118.5 25.6 7 6 9 14
9 Bor. Mönchengladbach 110.9 120.4 24.1 6 7 6 6
10 Hertha BSC 110.8 118.2 26.3 14 12 13 14
11 1. FC Köln 108.2 119.8 24.2 13 16 16 17
12 SC Paderborn 07 107.8 116.7 24.6 18 18 18 18
13 1. FSV Mainz 05 107.7 119.2 24.8 11 13 8 7
14 SC Freiburg 106.6 116.1 24.1 17 15 11 12
15 Hannover 96 106.5 113.6 25.7 10 11 10 8
16 FC Augsburg 105.6 113.7 24.6 12 17 7 9
17 Eintracht Frankfurt 103.3 112.0 25.3 16 10 12 10
18 Werder Bremen 100.1 112.0 24.3 15 14 14 13

Dienstag, 19. August 2014

Supercopa 2014: Real Madrid - Atlético Madrid: Goalimpact of Linups

Real Madrid would be the favorite even if it would field their worst eleven players. Not sure how to value the home field advantage in this on. Here are the odds based on the starting XIs depending on where you put the HFA.

Real Madrid: 71.2% (Home), draw: 19.8%, Atletico Madrid: 9.1%

Real Madrid: 40.5%, draw: 30.5%, Atletico Madrid: 29.0% (Home)

Real Madrid: 56.1%, draw: 26.6%, Atletico Madrid: 17.3% (Neutral Ground)


Real Madrid

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Iker Casillas195.1196.233.2Spanien75470056
Pepe136.5156.131.4Portugal33329545
Sergio Ramos168.0173.928.3Spanien50745096
Marcelo140.8141.526.3Brasilien29424541
Dani Carvajal127.8137.422.6Spanien [U21]917590
Toni Kroos135.6138.624.6Deutschland30521932
Gareth Bale125.4127.425.1Wales32527481
Xabi Alonso170.3198.032.7Spanien63051740
Luka Modric126.0132.528.9Kroatien33227256
Cristiano Ronaldo206.5213.529.5Portugal62051827
Karim Benzema154.4156.226.6Frankreich41127102
Average153.3161.028.1
Bench
Keylor Navas101.9108.327.7Costa Rica15214106
Raphaël Varane118.9135.021.3Frankreich1119205
Fábio Coentrão119.8121.026.4Portugal23117835
Arbeloa127.3147.831.5Spanien33828846
Ángel Di María142.9144.326.5Argentinien38125142
Isco117.4128.022.3Spanien16512106
James Rodríguez125.1132.823.1Kolumbien20414874
Average121.9131.025.5


Atletico Madrid

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Moyá93.596.930.3Spanien [U21]22220276
Diego Godín121.5127.628.4Uruguay34230803
Siqueira93.199.028.318413725
Juanfran118.1125.229.6Spanien36527603
Miranda133.3140.629.9Brasilien32729603
Mario Suárez114.6118.727.4Spanien22216131
Koke120.9130.522.6Spanien19312908
Raúl García118.8124.628.1Spanien [U21]40827165
Gabi113.2129.631.138630311
Saúl92.4119.619.7Spanien [U21]665056
Mario Mandžukic124.8130.728.2Kroatien27221298
Average113.1122.127.6
Bench
Jan Oblak91.4149.021.6Slowenien787190
Cristian Ansaldi106.2111.527.820317104
José Giménez90.4119.519.5Uruguay312689
Tiago105.2135.733.3Portugal36925949
Cristian Rodríguez101.2107.628.8Uruguay32718513
Raúl Jiménez109.8116.923.3Mexiko1168501
Antoine Griezmann106.0112.623.3Frankreich22517061
Average101.5121.825.4


Samstag, 16. August 2014

Skill and Luck in Bundesliga

Comparing the Bundesliga outcome with Goalimpact pre-season prediction, we find a close alignment. As Goalimpact is a pure player rating system that omits team factors, this indicates that the player quality is the dominating factor in team performance. It explains nearly 80% of the variance between the teams' goal differentials leaving little to explain for other factors and chance.

Ranking

The Bundesliga season 2013/2014 finished. Let's check how well the different pre-season predictions worked apart from the easy prediction of Bayern winning the title. To recall, these were the predictions (sorted by Goalimpact). Final ranks within one rank range around prediction are marked green.

No. Team Goalimpact Points Goal Diff Bwin Rank ClubElo Euro Club
Index
tm.de Last Year Actual Final
Rank
1 Bayern München 139,8 84,7 +64,8 1 1 1 1 1 1
2 Borussia Dortmund 119,8 60,2 +23,1 2 2 2 2 2 2
3 FC Schalke 04 119,0 59,2 +21,3 3 4 4 3 4 3
4 Bayer Leverkusen 113,8 52,9 +10,6 4 3 3 4 3 4
5 VfL Wolfsburg 112,3 50,9 +7,3 5 7 8 5 11 5
6 VfB Stuttgart 107,5 45,0 -2,8 6 13 7 7 12 15
7 Hannover 96 106,1 43,4 -5,6 10 8 6 10 9 10
8 1. FSV Mainz 05 105,7 42,9 -6,4 13 11 11 16 13 7
9 Bor. Mönchengladbach 105,6 42,7 -6,7 6 6 5 8 8 6
10 Hertha BSC 105,4 42,5 -7,1 12 14 13 15
17
11
11 1899 Hoffenheim 105,3 42,4 -7,3 13 16 16 12 16 9
12 Eintracht Braunschweig 105,0 42,0 -7,9 18 18 18 18
18
18
13 SC Freiburg 104,6 41,5 -8,8 13 5 9 13 5 14
14 Hamburger SV 103,6 40,3 -10,8 8 10 10 6
7
16
15 1. FC Nürnberg 103,5 40,2 -11,0 16 9 12 14 10 17
16 Werder Bremen 101,2 37,4 -15,8 11 17 15 11 14 12
17 Eintracht Frankfurt 100,7 36,8 -16,8 9 12 14 9 6 13
18 FC Augsburg 99,2 35,0 -19,9 17 15 17 17 15 8

The yellow column in the following table shows the rank correlation of the final table with the predicted table by the different algorithms or sources.

Goalimpact Bwin Rank ClubElo Euro Club
Index
tm.de Last Year current
Goalimpact 100% 78% 69% 83% 70% 50% 71%
Bwin Rank 100% 75% 87% 97% 75% 69%
ClubElo 100% 92% 75% 91% 64%
Euro Club Index 100% 84% 82% 65%
tm.de 100% 80% 61%
Last Year 100% 50%
current 100%

Obviously, all algorithms seem to do something right as all of them are better than just assuming last year's table. Each of the predictions placed some teams too high or too low, but overall the betting markets and Goalimpact have a slight edge to Euro Club Index and ClubElo. Transfermarkt performed a bit worse than the later two. Despite the fact that the difference to the betting market is certainly not statistically significant and, hence, pure random, we are proud to have obtained the first spot.

Let's see how the predictions worked out. Not only Bayern München, but also with Borussia Dortmund, Schalke, Leverkusen and Wolfsburg the entire Top-5 was predicted correctly by Goalimpact and the betting markets. Other metrics were more skeptical on Wolfsburg's chances, but their 5th place wasn't also clear cut to be fair.

All in all, 8 out of the 18 teams ended up right on the predicted rank or just one rank off. Given that Goalimpact is a player rating system (as opposed to team rating system) this may indicate that the success of a team is to large extend driven by the quality of its players. With that insight in our minds, any outliers may indicate towards a strong team factor. Augsburg was seen as the weakest team by Goalimpact, but they ended up on rank 8. We tend to conclude that this is achieved by a good tactical concept and a clear match plan that covered up the players' weaknesses. On the other hand, all team based ratings but ClubElo had them listed them as relegation candidate, too, and they should, in principle, also cover team factors.

Goal Difference

Actually, Goalimpact doesn't predict just the order of teams. It outright predicts the goal difference of the teams. If we compare the actual outcome of the Bundesliga goal differences for all teams with the predicted values, we see that 11 out of 18 values were realized within the range of ten goals around the predicted goal differentials. This is a good result indicating that Goalimpact has a lot of predictive value and that is also well calibrated. The regression line looks like this.


We find a regression line with a slope close to 1 and a R² of 77%. This is excellent for an out of sample result. Notice that the high R² is strongly driven by the distance that Bayern Munich at the far right hand side had from the rest of the teams. Without the Bayern Munich dot, the R² drops to 60%. But even this distance has been predicted well. Goalimpact had a pre-season expectation of Bayern Munich to finish with a goal differential of +65 and it turned out to be +71. Whilst it was a rather easy bet to predict Bayern Munich to finish first by some margin, to predict the correct goal difference and hence the distance of the victory correctly is considerably more difficult.

Despite the good results at the upper part of the table, there were some strong outliers at the left hand side of the chart. Especially Augsburg and Mönchengladbach performed much better than expected outperforming expectations by whopping 20, respectively 23 goals. Braunschweig (-23) and Nürnberg (-22), on the other hand, performed much worse than predicted.

Braunschweig was newly promoted to Bundesliga and was seen by all other metrics as a sure bet to get relegated again. Goalimpact was more optimistic and predicted a goal difference of -8. It turned out to be -25, thus -17 worse than expected. The other metrics got this one better. This raises the question if Goalimpact maybe systematically overrates teams in lower leagues. To check this, we look at the other promoted team Hertha BSC Berlin. They were predicted to have a goal difference of -7. Here Goalimpact was much more optimistic than the other metrics. Hertha, however, performed according to this optimistic prediction and ended up with a positive differential of -8 and thus only one goal worse than predicted. The total underperformance of all promoted teams was hence -24. It is too little data to make definitive statements on over- or undervaluation of those teams, but it is a difference to be watched..

Summary

Both in ordinal and cardinal predictions Goalimpact performed very good. As the prediction was based solely on aggregated player ratings, this indicates that team performance is very strong determined by the player quality. Out-of-sample, this factor alone explains 78% of all variation in the teams' goal differentials. Other factors, such as luck and tactics, explain together only the remaining 22% over the course of a whole season. Given the fact that the Goalimpact rating system is not perfect and a better system might even explain a larger part of the variance, player quality is possibly even more relevant than implied by the results presented here.

Updated: Do to an error in the excel sheet we compared the goal differentials of the 31st matched day instead of the 34th and last. These numbers and findings have been updated.

Sonntag, 20. Juli 2014

RB Salzburg - Rapid Wien 6:1

Red Bull Salzburg takes the pole position in the Austrian Bundesliga right from the start and, frankly, is unlikely to ever lose it again. Even the second team in the league, Rapid Wien, is no match to Salzburg's exquisite offensive and defensive power that would without problem be very competitive in the German Bundesliga, too.

The odds based on starting XI were RB Salzburg: 59.4%, draw: 25.3%, Rapid Wien: 15.3%. It would be even stronger biased towards Salzburg if the best players would form the starting XI. However, Adolf Hütter chose to play with Schiemer that turned out to be the weak link in the defense. He was replaced by Ramalho in minute 55.

That some of the Salzburg players aren't even playing for their respective national teams seems odd and indicates a bias of football pundits against the Austrian league. Due to some technical reasons, the league may be a bit overrated in our algorithm a few points, but even if you'd subtract 10 points from all Schwegler, you'd still expect him to play for Switzerland.

RB Salzburg

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Péter Gulácsi119.9119.924.2Ungarn [U21]1079664
Christian Schwegler138.7145.130.127224169
Franz Schiemer111.5116.928.3Österreich26221968
Andreas Ulmer140.9146.428.7Österreich25422594
Martin Hinteregger132.9149.721.8Österreich [U21]15913243
Stefan Ilsanker134.4137.325.1Österreich [U21]23118078
Christoph Leitgeb122.1127.829.3Österreich31225285
Kevin Kampl139.2146.923.8Slowenien20617407
Sadio Mané128.2142.622.3Senegal1108739
Jonathan Soriano129.4135.028.820113914
Alan133.4136.725.01609158
Average130.1136.826.1
Bench
Peter Ankersen108.3115.823.8Dänemark816917
André Ramalho138.1151.822.31109886
Marcel Sabitzer96.1124.920.3Österreich15811560
Alexander Walke105.7105.731.1Deutschland [U20]32029642
Naby Keita81.2120.019.4231917
Valentino Lazaro75.7128.918.3Österreich [U17]271391
Massimo Bruno104.5129.420.8Belgien [U21]946219
Average101.4125.222.3


Rapid Wien

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Ján Novota110.7110.730.6686320
Thomas Schrammel115.9118.126.820918424
Mario Sonnleitner114.1118.727.8Österreich [U21]33430329
Christopher Dibon121.2129.223.7Österreich [U21]19016972
Maximilian Hofmann109.3133.320.9524430
Thanos Petsos108.0118.623.1Griechenland [U21]15310968
Stefan Schwab94.5102.123.8Österreich [U21]17513638
Louis Schaub103.7141.119.5Österreich [U21]1026562
Steffen Hofmann111.4143.333.844037150
Lukas Grozurek116.2129.222.5Österreich [U21]1679168
Deni Alar104.5109.524.4Österreich [U21]19811359
Average110.0123.125.2
Bench
Philipp Schobesberger109.2137.820.3947119
Dominik Wydra102.9131.720.3Österreich [U19]936865
Robert Beric111.2121.923.0Slowenien [U21]1088402
Marko Maric90.390.318.5Kroatien [U17]242185
Brian Behrendt117.2129.522.712610159
Mario Pavelic104.7129.620.8Österreich [U19]725840
Stefan Stangl95.5107.822.713911819
Average104.4121.221.2