Julio del Corral Cuervo: Euro 2024: France is more likely to win and here we explain how we know it

The Euro Cup began on Friday, June 14. The question that hovers over the football environment is the usual one: what are the possibilities of each of the competing teams? Hence, more and more newspapers, companies and individuals show their predictions before the event begins.

Some of these predictions are updated as the championship progresses. I am going to explain the different methodologies used, while commenting on some predictions.

Spoiler: France is the most likely to win. Specifically, @Futbometrix1, an account by the prestigious economist Daniel Paserman, gives him a 16.5% chance. On the podium of predictions are followed England (13.7%) and Spain, with a probability of 11.7% of winning the European Championship.

There is no prediction without probability

Last week the XIV Ibero-American Congress of Sports Economics was held, organized by the Spanish Society of Sports Economics. In my presentation I stated that “predictions are either probabilities or they are not predictions.”

In the case of the Euro Cup, any prediction must make clear the probability of a team being champion. For example, Spain has a 10% chance of becoming champion. On the other hand, saying that a team will or will not win the competition is not valid in predictive terms. For example, “Spain is not going to win the Euro Cup” is not a prediction because the statement is not based on a probability calculation.

It is also important to understand the meaning of probability. If someone says that Spain has a 10% chance of winning the European Championship, it means that, if 100 European Championships were played under these same conditions, Spain would win ten. What it does not mean is that this prediction is appropriate if Spain does not win. As there are obviously not going to be 100 Euro Cups held, it is not an immediate task to define who or which model has predicted better.

For some time now, Daniele Paserman’s X @Futbolmetrix1 account has been running a contest to see who can best predict a tournament like the Euro Cup. Those interested send the probabilities of reaching each of the rounds for all the teams before any competition begins. In the case of the Euro 2024 in Germany, more than 30 have been received.

To determine which prediction is the winner, @Futbometrix1 uses three different statistical methods: Brier Score, Log Score and RPS. In all three models, the lower its value, the better the prediction.

Spain was the winner in the 2023 Women’s World Cup. The winning prediction was made with the average of several betting houses. These predicted better than up to 20 predictions based on different models. This result, which may seem paradoxical, is relatively common.

It is not enough to do many simulations

Below I am going to explain how predictions can be made for a tournament with the characteristics of the Euro Cup.

Most predictions are made by simulating the tournament through a high number of repetitions. How much is high? They generally range from 1,000 to 200,000 simulations, like those used by the newspaper El País for this Euro Cup. Calculating predictions by simulating the tournament is as simple as seeing the percentage of simulations in which a team reaches a certain round. El País says that the probability of Spain winning the Euro Cup is 11%, which implies that Spain was the winner in about 22,000 simulations of the 200,000 carried out.

However, the key to this methodology is not so much the number of simulations. In my experience, you can get close to the final probabilities of the model after about 200 or 300 simulations. The key lies in how to assign the probabilities (win, draw, loss) for each match. Here two notable alternatives emerge: past results or quality of the players.

The results recorded in matches held in the past are quantified by Elo rating, which is a modification of the famous chess rating, while others follow the FIFA or UEFA rating.

To measure the quality of the players, the easiest way is to add the transfer value provided by the Transfermarkt website. This provides a hypothetical transfer value for all players in professional leagues. To do this, it is based on collective assessment under the assumption that the many are more intelligent than the few.

Toni Kross is worth more than what is said

However, using the Transfermarkt value as a measure of a footballer’s quality has the problem that said value depends largely on age, and not performance. For example, Toni Kross, Real Madrid’s undisputed starting midfielder in recent seasons, is worth ten million euros. Exactly the same one assigned to Nico Paz, a 19-year-old Real Madrid midfielder who has played 19 minutes in LaLiga 2023-2024.

To achieve a more realistic valuation, the Transfermarkt value should be adjusted based on age.

Another way to measure the quality of players is through individual statistics, which can be obtained through artificial intelligence models. Obviously, these sources can be integrated through different combinations and scales. For example, to measure the quality of a team, El País integrates the Elo ranking and the Transfermarkt value.

Predictions are updated with results

Some forecasters not only release their pre-competition predictions but also update them based on the results already made. In this case, a decision that the models have to make is whether they update the quality of the teams as the competition progresses or stay with the quality established pre-tournament. A particularly relevant event occurs when surprise teams reach the final rounds, as was the case of Morocco in the 2022 World Cup in Qatar, reaching the semifinals.

From my Aristotelian vision, which places virtue at a midpoint, I believe that a good model should have taken into account that the Morocco team in the semifinals was a much better team than the team that started the World Cup. But one should not update its quality so much and infer that it was one of the four best teams in the tournament.

Once the probabilities are determined, a key data for the accuracy of a model, it is relatively easy to carry out the simulations, as it is enough to tell the computer program to choose the result of a match.

For their part, betting houses establish their initial probability by some means similar to those described above. But once a bet is opened, it is updated. Both due to new information that may appear, such as an injury to a relevant player, and due to the bets received.

Suppose I play a tennis match against Carlos Alcaraz, and that, by mistake, the odds establish me as the favorite (player with the lowest odds). Bettors will go en masse to bet on Alcaraz. This will make the bookmaker reconsider if they modify their odds. In such a way that the normal thing is that the final quota considers that it is half impossible or completely impossible for him to beat Alcaraz in a tennis match. Logically, the same thing happens with other bets such as the Euro Cup winner.

The odds for Euro 2024

To show an example of the predictions, I am going to use the average probability based on the 30 participants in the contest established by @Futbolmetrix1 of reaching each of the rounds and winning. The table appears ordered according to the probability of winning the Euro Cup.

And yes, the favorite team is France, with a probability of winning of 16.5%, followed by England with 13.7%. Spain is third with a probability of 11.7% of winning the Euro Cup. So, congratulations to the French fans, although their probabilistic advantage is not that great. We Spaniards entrust ourselves to that 11.7%.

This article has been published in ‘The Conversation’.