Using data from all FIFA World Cup competitions that took place between 1994 and 2014, a step logit model is estimated to forecast the likelihood of success of each team in each tournament. The model correctly identiﬁes the winner in 5 out of the 6 tournaments, and among many variables considered, hosting the tournament, the team’s quality, and the stage of the generational cycle of the players are found to be the key contributors to its forecasting performance. Considering only the information available at the date preceding each of the last two World Cups, we can perform a more ambitious test of the model’s ability to forecast the winner at future tournaments. Our results indicated that Spain would win in 2010 and Germany in 2014, as they did. The model also predicted that Argentina would come in second in the last World Cup, thus getting it right for 3 out of the last 4 ﬁnalists in total. The model’s success rate is still remarkable when predicting which teams would reach the semi-ﬁnals (two in 2010, and three in 2014).
We make use of the model in two diﬀerent regards. First, the ability of FIFA’s rankings to serve as a measure of teams’ relative strength is evaluated and rejected. Speciﬁc suggestions on how to improve their current ranking methodology are presented. Second, using bookmakers odds and model probabilities, we show that a sophisticated bettor could (consistently) make a proﬁt. Therefore, our results hint at the possibility of deviations from eﬃciency in the large World Cup betting market.