Following on from my most recent post, I continue looking back at the preseason projections that Aaron Nielsen and I made at the start of 2013-14 European season. I will add projected and actual league tables soon; I wanted to get my assessments written down now.
With the exception of newly promoted Grödig, our forecast model did a decent job of predicting places. Red Bull Salzburg were a near-consensus pick to win the title, and we did as well, so that achievement isn’t too earth-shattering. But point predictions were very poor — a points RMSE of 11.3 is very high for a ten-team competition. The reason is that we were way too optimistic about defensive performance. Every team in Austrian Bundesliga let in more goals than we expected, and only three sides (Admira, Rapid Wien, Red Bull) allowed less than 15 goals below expectations. Of course, this meant that there were a lot more goals than expected, 165 more (or 38% above our forecast). Of that number, 112 goals were distributed among three teams: Sturm Graz scored 21 more goals than expected, Grödig scored 44 more goals than expected, and Red Bull 47 more than expected. It didn’t make a difference to sides like Graz or Red Bull, but for a side tipped for relegation like Grödig, this jump in offensive performance vaulted them into a Europa League place.
Belgium (Pro League)
Belgium’s Pro League has a playoff system after 30 matches, so we only predicted the league during their regular season. The top six teams pass to the championship playoff, and we projected all of them pretty easily. For what it’s worth, we projected Anderlecht to finish first in the regular season, and they ended up winning the championship playoff. Oostende was a big surprise as they were projected to finish in the relegation playoff and ended up in the playoffs for the Europa League (they wouldn’t have been able to play as they were denied a license to play in UEFA competitions). Mechelen and Mons were the big negative surprises as they were expected to finish the regular season in mid-table yet sank to the lower reaches. Mons had a nightmare season and ended up getting relegated. Predicting goals scored seems to be a crapshoot, but the goals scored/allowed RMSEs were among the best of all the leagues we projected. And once again the predicted points were conservative for the top teams.
Belgium (Second Division)
This model’s performance shows how a pre-season projection is subject to events out of control of any forecast. Visé were projected to finish in first place and be promoted to Pro League. But off the field Visé’s foreign investor group failed to come up with promised funds and the club virtually collapsed. Players didn’t get paid and eventually were given special dispensation to find new clubs. Needless to say, Visé scored 31 fewer goals than expected, let in 37 more goals than expected, and scored 43 fewer points than expected on their way to finishing bottom on 13 points. If it wasn’t for that performance, the points RMSE of the model would have been below 10.
The PFG A has a playoff system in which the top seven enter a championship playoff and the bottom seven a relegation playoff. We projected the top seven teams with the exception of Lokomotiv Plovdiv and Cherno More Varna. We projected Ludogorets to finish first in the regular season, which they did on the way to winning the league in the playoff. The total number of goals that we predicted for the league was almost bang on (517 vs 515), which is probably down to dumb luck, but in general we’ve managed to predict that total to within 10% for all of our leagues. The distribution among teams was way off though, as most teams actually performed better in terms of defensive goal statistics than expected. Expected total points were way off as well, particularly for teams at the very top and bottom of the table.
Prva Liga was among our best performing models — only Rijeka and Slaven Belupo’s point projections were off by more than 10 points, and only Slaven and Zadar’s final positions were off by two places. These results happened despite being way off in our projected goal totals; there were 58 more goals scored in the league than we expected, and every team except RNK Split and Rijeka had a worse defense than we expected. I think what is going on is that there are many goals scored/allowed combinations that will yield the same number of expected points in a Pythagorean expectation. At any rate, a points RMSE of 7.4 is good for a ten-team league.
We projected a comfortable league title for Viktoria Plzeň, and they did play in line with our expectations, but Sparta Prague overachieved on both sides of the field to win the title going away. We predicted Teplice to be among the relegated teams, and they finished in fifth place – 15 points and ten positions better than expected. The biggest shock of the league had to be Sigma Olomouc, who we (and a few other analysts in Czech football) predicted to finish in a Europa place but in fact the club suffered a severe disorganization and dropped out of the top flight after 30 years. Most of the projections in the league were good, but the few bombs blew up the points and positions RMSEs.
We were all over the map with the Superliga expectations from the start. When I took a look at the table at the start of the season and saw that Nordsjælland and Esbjerg were near bottom, I knew it would be a long season. What’s funny is that the expected goals scored/allowed was quite good for almost all of the teams, but expected points were way off. And in a league of this size, being off by 10-12 points can be a difference between mid-table and champion (or relegation). Lots of near-misses in the table — FC Copenhagen (predicted to finish first) would have won had it not been for AaB’s surge, while FC Vestsjælland and SønderjyskE might have been relegated had it not been for AGF and Viborg’s implosion.
Below is a statistical summary for those leagues. I’ll compile this table into one large list when I finish the third part of this series.
|Total Goals||Pred Goals||Pct Change||RMSE Pts||RMSE GF||RMSE GA||RMSE Pos|