When it comes to predicting the outcome of matches, do football pundits know their stuff? Three economists from UCC set about finding out...
“I’m going 2-2 Geoff”.
“I reckon Liverpool win this one 3-1”.
“City will be much too strong at home. 4-1 Geoff; Aguero to score first”.
Before the English Premier League came to an abrupt halt in mid-March, these were the type of forecasts we heard each weekend from the football pundits that intrigue and infuriate. Paul Merson, Matt Le Tissier, Michael Owen, Charlie Nicolas, Robbie Savage, Mark Lawrenson. The list goes on. None are shy of making predictions. That’s a pundit’s job after all. But just how good are they?
In 2014 we started putting them to the test. Week-by-week we collected the scoreline predictions of nine very well-known football pundits for the Premier League. All nine are former professional footballers, working for the main broadcasting companies that show live and highlight footage of Premier League games — Sky Sports, BT Sports, and BBC.
Following these pundits weekly continued for three years until May 2017. By then we had collected three full Premier League seasons and had 1,170 games to analyse. In total, we recorded nearly 3,500 pundit predictions. This was a sample size big enough to figure out if these ex-pros were actually better than the rest of us.
And what about the rest of us? We needed something to compare the pundits with.
For those unfamiliar with this website, it is a sports prediction platform that allows people all over the world to make predictions about sports events, in particular the Premier League. Using Superbru’s dataset we were able to match the pundit predictions between August 2014 and May 2017 with regular fans.
The website is popular. During the 2016/17 season, nearly 50,000 users were registered forecasters. In total, we had access to 32 million predictions made by normal football fans.
The acid test. How did the regular fans do when compared to our pundits when predicting match outcomes? Well, it turns out not great. In fact, the expert pundits are just that — expert. They do better than the rest of us. Of course, they should — after all, this is what they are paid to do.
The interesting thing, however, is that most previous research finds that pundits are no better than anyone else at predicting match outcomes. What makes our nine pundits better than the average fan is their superior ability in the prediction of draws. Players on the Superbru website are far less capable at predicting these outcomes.
About 25% of Premier League games end in a draw but fans normally don’t predict anything like 1 in 4. Neutral outcomes do not generally appeal to us. Where is the fun in a draw? Fans may like to pick a winner much more often than they should.
Our sample of nine experts actually outperforms bookmaker odds too. This means if a gambler blindly followed the pundits’ match predictions from 2014-2017 they would have reported a small profit.
Things start to unravel, however, when it comes to scoreline predictions. Neither pundits nor regular fans can beat the bookies when it comes to guessing the exact outcome.
This might be explained by a cognitive bias that many of us possess. When asked to predict the scoreline, both pundits and laypeople tend to overestimate the probability of goals and fans believe that some clubs, often the bigger ones, are more likely to be involved in high scoring matches.
This is maybe what we would like to see. Score lines like 2-1, 2-2, and 3-2 are a regular feature of the dataset. Sadly, football is much more mundane than this. The three most likely outcomes in any football game are 1-0, 1-1, and 0-1, in that order.
Maybe the most interesting scoreline that allows the bookies win out over the gambler is 0-0.
Approximately one in ten Premier League matches ends scoreless, though fans and pundits alike rarely predict stalemates. We tend to have a bias toward thinking that something rather than nothing will happen. It’s commonly referred to as wishful thinking.
And these findings are important. If the naïve prediction strategies extend to gambling behaviour, bettors can be nudged towards backing low probabilities outcomes.
When have you ever seen “Manchester United to win 1-0 and Rashford to score first — 8/1” advertised? Probably never. It is more likely to read “Man United to win 4-1 and Rashford to score first, 33–1”.
The probability of the second outcome is vanishingly small.
But a €10 bet could result in a return of €340 — a very attractive proposition to those seeking to win money.
In the scenario, Manchester United have to win. Next, The Red Devils have to score exactly four goals. And concede exactly one. And, to top it all off, Marcus Rashford has to score the first goal in the game.
While the odds might seem attractive, not even the expert pundits can stay ahead.
It is just too complex a task. Listening to their advice will help but the more complex it becomes — adding goals and goal scorers — the less likely you are to succeed.
Something to keep in mind when the Premier League returns next week.
- This work titled “Expert performance and crowd wisdom: Evidence from English Premier League predictions” is co-authored by David Butler, Robert Butler and John Eakins – Centre for Sports Economics and Law, CUBS, UCC, and is forthcoming in the European Journal of Operational Research.