Play a quiz
Numbers game Sponsored post

Maths and football: putting Betegy's algorithm to the test

In 2012 a prediction service claimed it could read the weekend's results from data and team news. Four matches in England and Spain made a neat test.

First published 18 Nov ’12 3 min read Articles →
AICalculator, pencil and graph paper beside a pint on a pub table (picture made with AI)
Picture made with AI · About our images

Some things in football are certain. One team will win the Premier League, three will go down, and the Saturday night pundits will spend the season stating the obvious at length. Everything in between is guesswork, which is exactly why a company called Betegy thought there was a gap for mathematics.

In November 2012 the title race was a blur of sky blue, red and blue, and at the other end a cluster of sides were producing the kind of under-par numbers usually only celebrated on a golf course. Picking a sure thing from that fixture list was, as ever, not simple.

A formula for the weekend

The pitch was straightforward. Betegy ran an algorithm over the possible outcomes of each match, then adjusted those probabilities using news analysis: a star striker ruled out on Friday, a goalkeeper short of fitness, a manager rotating before a European tie. The service then labelled each forecast by confidence, from a "recommended" pick down to an "extreme" one, and suggested an exact score.

That last part is the bold bit. Predicting a home win is one thing. Naming the scoreline is where most supporters, fuelled by optimism and the memory of last week's surrendered three-goal lead, come unstuck.

The appeal for the ordinary fan is obvious. Nobody has time to read every injury bulletin in two countries. A model that does it in seconds, and is not swayed by the fact that its owner supports the away team, removes the two biggest weaknesses of the Saturday morning punter: laziness and loyalty.

Four matches, two leagues

The test weekend of mid-November 2012 used two Premier League games and two from La Liga, each with an obvious favourite on paper:

  • Aston Villa's trip to Manchester City, the champions, which looked doomed from the start for the visitors.
  • Reading at home to an Everton side in strong form, flagged by Betegy as an "extreme" call.
  • Barcelona against Real Zaragoza at the Camp Nou.
  • Real Madrid against Athletic Club, another fixture where the bookmakers saw only one winner.

Three of the four were labelled as recommended. The game at the Madejski Stadium was the outlier. Exact-score bets were placed on the recommended lines through a handful of bookmakers, as the service directed.

AISteep upper tier view of a floodlit Spanish stadium pitch at dusk (picture made with AI)
Picture made with AI · About our images

The maths came out ahead. Three of the four outcomes went as forecast, and at the Camp Nou the algorithm called the precise scoreline, enough to cover the losing ticket and leave the weekend's stake intact. Not a fortune. A neat proof of concept, though, from a model that had never stood on a terrace.

What the numbers missed, and what came next

The season itself showed both the strength and the limit of the approach. The favourites largely held: Manchester United won the 2012–13 Premier League in Sir Alex Ferguson's final season, and Barcelona took La Liga with 100 points. At the bottom, Reading went down alongside Queens Park Rangers and Wigan Athletic, who won the FA Cup in the same campaign. No algorithm built on form tables would have offered much on that last combination.

Football's appetite for data has grown enormously since then. Expected goals, once a niche blog metric, became a fixture of Match of the Day graphics, and clubs such as Brentford and Brighton built recruitment models that would have looked like science fiction on a 2012 fixture list. The basic idea Betegy sold, that a computer reading team news can beat a supporter reading the table, is now standard practice inside the game.

The mathematics behind services like this is older than it looks. In 1982 the statistician Mike Maher showed that the goals a side scores in a match follow a Poisson distribution quite closely, so an attack rating and a defence rating for each team are enough to put a probability on every scoreline. In 1997 Mark Dixon and Stuart Coles corrected the model for the low-scoring draws it underrated and gave recent matches more weight. Most exact-score forecasts still start from that pair of papers.

The fan, meanwhile, still has a hunch about Saturday. That has not changed at all, and it still loses to the spreadsheet more often than anyone admits in the pub.

Please gamble responsibly. GambleAware.org 18+

Questions and answers

How can a computer model predict football results?

A model of this kind starts by calculating the probability of each possible outcome, then adjusts it using team news, such as an injured striker, a goalkeeper short of fitness or a manager resting players before a European tie. It can process that information in seconds, and without any loyalty to one side.

Which matches were used to test the prediction model in November 2012?

Two Premier League and two La Liga fixtures: Manchester City against Aston Villa, Reading against Everton, Barcelona against Real Zaragoza and Real Madrid against Athletic Club. Three of the four forecasts proved right, and the model named the exact scoreline at the Camp Nou.

Who won the Premier League and La Liga in 2012–13?

Manchester United won the Premier League in Sir Alex Ferguson's final season, while Barcelona took La Liga with 100 points. At the bottom of the Premier League, Reading, Queens Park Rangers and Wigan Athletic were relegated, even though Wigan won the FA Cup that same season.

How has data changed football since 2012?

Enormously. Expected goals went from a niche blog metric to a regular feature of Match of the Day graphics, and clubs such as Brentford and Brighton built recruitment models around data. The idea that a computer reading team news can outperform a fan reading the league table is now standard inside the game.

Keep reading

Articles

Who would win the Premier League? The 2012–13 race in January

Seven points clear at the halfway mark, but haunted by the previous May. A look back at how the 2012–13 race was judged mid-season and what Sir Alex Ferguson's last campaign delivered.

Read on →Who would win the Premier League? The 2012–13 race in January
Articles

Top tips for the Premier League title: the 2011–12 race in March

Two Manchester clubs, two points apart, and Spurs clinging on in third. The prices of March 2012 told one story; the final day told another.

Read on →Top tips for the Premier League title: the 2011–12 race in March
Articles

The Manchester derby in statistics: cold hard facts of City v United

Before the FA Cup semi-final of April 2011, a derby preview in numbers: head-to-head records, cup meetings, record wins, top scorers and one very tough goalkeeper.

Read on →The Manchester derby in statistics: cold hard facts of City v United