Current navigation level
Previous navigation level
Next navigation level
Current navigation level

Soccer Analytics - Forecast Quality

On this page we check the quality of previous forecasts. To this end, we compare our forecasts to forecasts from other external sources.

The main objective of the model is to forecast the final table. It is not expected that the final table can be predicted exactly. Every season has its own surprises and disappointments. To check how good our forecast was last season, we compare our forecast with the tips from many other forecasts collected on the BStat homepage. All of these predict the ranks of the final table. We calculate the rank correlation coefficient with the real final table and the respective forecasts. The closer a forecast was to the truth, the higher this value. Our new forecast method was beaten by three other forecasts in the 2017/18 season. Guessing the exact position is often a matter of luck, especially in a narrow league like the Bundesliga. Therefore, we additionally look at the forecast quality for expected points using the root mean squared error. This quality measure is smaller the closer the forecast is to the actual points. In this category, our new forecast outperforms all others.

  Rank correlation
Root mean
squared error
Season 2017/18
SEW new
0.64 8.9*
SEW old
0.47 10.8 0.60 -
Goalimpact 0.71 9.0
Fupro 0.63 11.5
0.43 16.8
Kick-Forum 0.61 9.4
Spiegel Online
0.75* -
FiveThirtyEight 0.63 9.9
 Euro Club Index
0.61 9.0 0.63 12.1
General-Anzeiger 0.74  - 0.85 16.4
Club Elo
0.62 -
 * indicates best prediction performance.

In the 2016/17 season, our forecast was only beaten by Spiegel Online in predicting the rankings.

Season 2016/17
Quality measure SEW Soccer Analytics F.A.Z. Spiegel Online
Rank correlation coefficient
 0.47 0.44 0.29 0.50* 0.42
RMSE 10.14* 10.87 - - 10.83
 * indicates best prediction performance.

In season 2015/16 our predictions were at least as good as all considered competitors.

Season 2015/16
Quality measure SEW Soccer Analytics F.A.Z. Spiegel Online
Rank correlation coefficient
 0.64* 0.64* 0.57 0.60
RMSE 8.57* 8.76 - -
 * indicates best prediction performance.