World Cup Prediction Model Update - Taking Stock After the Final
In today’s final update, we take stock of our World Cup model’s performance through the competition.
We start by reviewing the first-round matches. Exhibit 1 plots the actual goal difference against the predicted goal difference in the 16 first-round matches. The correlation between the actual and predicted goal difference is 41%, which does not seem too bad given the inherent randomness of football. That said, the model only got 9 of the 16 second-round teams right, which seems quite disappointing.
Moving on to the knockout stage, Exhibit 2 shows the performance of our model in predicting which teams would advance to a particular stage of the tournament, up to and including the championship. We use three different timing conventions: 1) prior to the start of the tournament, 2) after the conclusion of the group stage, and 3) with predictions updated through the eve of every match. We view this performance as quite good. In particular, the fully updated version correctly predicted the results of all eight second-round matches, all four quarterfinals, one of two semifinals, and the final.
Combining the two stages, Exhibit 3 provides an overall sense of how well our model predicted the actual performance of all 36 teams in the competition. The chart compares the actual ranking of each team with the ranking predicted by our model at the start of the competition. (We assign equal ranks to teams that came fourth in the group stage, third in the group stage, were eliminated in the second round, or were eliminated in the quarter-final.) The overall correlation between the actual and predicted ranks is 48%. The chart also shows which teams outperformed our predictions (most notably Costa Rica, Algeria, Nigeria and Mexico) and which teams underperformed our predictions (most notably Spain, Italy, Portugal, and England).
As a concluding observation, we note that the model’s biggest single miss of the competition—and by one calculation the most surprising result in World Cup history—was the Brazil-Germany semifinal. The model predicted a 2-1 victory for Brazil, but the actual result was a 7-1 win for Germany. We regret the miss. But, speaking as Germans, we also note that there are more important things than being right.
Jan Hatzius and Jari Stehn
Goldman, Sachs & Co.
Legal and Certification Disclosures
We, Jan Hatzius and Jari Stehn, hereby certify that all of the views expressed in this report accurately reflect our personal views, which have not been influenced by considerations of the firm's business or client relationships.
For Reg AC certification, see above. For other important disclosures, go to www.gs.com/research/hedge.html