US-based video game developing giant EA Sports has now correctly predicted the winner of the FIFA World Cup the fourth time in a row. Coincidence or smart modelling?
The good folk over at EA Sports do it by running thousands of simulations of the tournament games on its uber popular FIFA gaming franchise. But they’re not the only ones predicting winners using simulation models. Banks, investment firms, think tanks, universities, and several others – organisations and individuals alike – consistently take a shot at setting up models to predict the winner of the World Cup – and mostly, fail.
But why exactly is forecasting football tournaments – or most other sporting tournaments for that matter – so notoriously difficult?
Let’s start from the right end of the forecasting spectrum.
EA Sports aren’t the only ones setting up a good track record in predicting the winner. Joachim Klement, a strategist from UK-based investment firm Liberum Capital, has now correctly predicted three of the last winners of the FIFA World Cup as well – the latest of the predictions proved correct only a week or so ago. In his note, Klement writes: ‘football matters – so much that academics have spent some of their time developing models to forecast the matches.”
To predict the winner, Klement set up a model with roots in a study conducted by the University of Nottingham quite some years ago – a study which showed that certain economic and climate variables may have a significant role to play in predicting the winners of international football matches.
Some of the factors he used in the model included GDP per capita, population size and demographics, temperature (the optimum temperature for football has been found to be 14°C), whether or not a country is a host nation etc. He writes, “our model uses all these socioeconomic variables to predict the outcome of the World Cup but also adds the current FIFA ranking points to indicate the strength of the current squad.”
Yet, modelling football tournaments remains a notoriously difficult task.
One aspect to consider is the fact that it is much easier to predict a single game than an entire tournament. The more the number of conditions are added to any conditional probability model, the lower the actual winner’s probability gets. In fact, landslide winners in these models usually end up with a probability of around the 20%-mark. In fact, 20% is a rather high number – that most organisations chose to attribute to Brazil this time.
Football, as a sport, can hardly be modelled in two dimensions – and models can hardly ever encapsulate all human and surrounding variables – an element referred to in common parlance as ‘luck’. Although upon selecting the criteria for his model, Klement did, in fact, try to model this very ‘luck’ factor as well:
“Using these variables, we can explain some 45% of the variation in success across nations in a World Cup. But this means that some 55% of the outcome is determined by luck. Thus, our model includes an element of chance when determining the outcome of matches between any two teams. Yes, a team can have a high chance of winning against another team, but upsets happen, and our model takes these effects into account.”
In spite of these adjustments, if you consider how wrong most of his predictions for the knockout stages are, it is close to astonishing that he got the winner right.
The one thing worth noting here, is the fact that the World Cup isn’t exactly something forecasters model for a living. As his note humorously outlines, it is often meant to satirise the hubris of economists who think they can ‘predict absolutely everything.’
This, coming from economists who have over the past year or so misjudged almost everything, makes for hilarious reading.
[Read Liberum Capital’s research note here]
Know more about the syllabus and placement record of our Top Ranked Data Science Course in Kolkata, Data Science course in Bangalore, Data Science course in Hyderabad, and Data Science course in Chennai.