Four years ago for Brazil´s FIFA’s World Cup, I developed a forecasting model that correctly predicted Germany’s tournament win.
At that moment, the Wall Street Journal featured Palisade’s World Cup 2014 simulation model in their article, “The Journal’s Prediction”. Reporter Matthew Futterman consulted me on our “brilliantly logical prediction model” that picked who would win–with home-field advantage, and without.
As Futterman put it,
“Some 50,000 iterations later, the team with the highest probability to win the World Cup was Germany at 19.9%. Spain, the defending champion, was second at 16.1%, followed by Argentina and Switzerland at 6.1% and Brazil at 6%.
In other words, Germany is probably the better team, but Brazil is still the favorite. It all comes down to how much playing in front of a Brazilian crowd will motivate the hosts. We probably didn’t need an algorithm to know that.”
Come now Russia 2018.
Our model has been updated with the latest information and arranged groups and games for Russia 2018.
This @RISK Excel model was built to forecast the probabilities of each one of the 32 national teams to win the soccer’s World Cup in Russia 2018.
The model considered information from FIFA (www.fifa.com), the global soccer organization. FIFA maintains rankings for its more than 200 national teams. Using games and rankings data from the past eight years (2011-2018), including games still to be held until May 31st. 2018, the model calculates the probability of winning the World Cup.
The first stage for building this model consists in calculating the historic odds of winning, losing or getting a draw between ranked teams. By classifying teams in eleven equally weighted-probability bins, at any given time it is possible to calculate such probabilities. For example, a team on the highest bin (11) would have a larger probability of beating a team in, say, an intermediate fourth bin. In this case, historic information yields that the highest ranked team would have a 86% chance of beating an intermediate team, a 7% chance of losing and also a 7% chance of getting a draw. That is part of the beauty and mystery of soccer – the emotion of an underdog of, from time to time, beating the favorite.
Once these probability tables were calculated, the model considered all the first 48 games. The 32 teams are allocated on eight groups of four teams each. Within each group, all four teams play against each other. The top two teams from each group advance to a group of 16 teams. A win at this level accounts for 3 points, a defeat for 0 points. Assimetry is given by the fact that a draw accounts for only one point to each one of the teams. An astounding amount of classifying results are possible. Even though teams may tie in terms of points on their quest to classifying for a second round, two cascading rules untie equally pointing teams depending on the net difference of goals scored/received, and also depending on the specific result of two equally pointing/net scoring teams. If the tie persists, tossing a coin will determine which team advances. All these rules were incorporated into a Monte Carlo simulation model that ran tens of thousands of possible iterations, solving for example, for ties in points/net scores and also “tossing coins” for highly unlikely draws. At the core, the model also considered historic goal scoring records for winning, drawing and losing teams. Obviously, teams that win a match will score more goals than teams that either draw or lose.
After the sixteen teams are classified, they will run on a single match eighths-final game, leaving eight, then four and then two teams for a final game. These additional 16 games (including a game to determine third place among the two losers of semifinals) were also simulated. After running fifty thousand iterations, the model will probabilistically channel each team into an eventual tournament win and will calculate its odds. Depending on certain ranking assumptions, probability calculations may vary somehow significantly. However, what was considered a rather robust approach to consider historic and current rankings yielded the following results.
Our simulated results are posted as follows. Brazil has the largest probability of winning the tournament even though it is not ranked first according to FIFA.
This and many other intermediate probabilistic results may be generated for the tournament using this Monte Carlo simulation method. For example, you could fetch information on what are the odds of your own national team of winning each stage.
We want to transform this whole model into an Android/Apple app and we are currently in active search of investors/developers/marketers for it.