Friday 12 July 2013

Polish Prediction Engine

A little busy today to answer all of Danny's questions, so time only for a lazy post courtesy of the Wall Street Journal, although I have to say not up to that reputable publications normal standards by a long way:
In-Depth Analysis? Complete Bollocks more like
WARSAW — If someone could predict with 90% accuracy the outcome of a soccer game, you might be tempted to put a bet on that match.
Betegy, a startup here in the Polish capital, is claiming just that on some games in major European and global leagues.
CEO and founder Alex Kornilov, a Ukrainian who came to Poland to study, said the company has just signed a deal with a U.S. broadcaster to provide forecasts for its soccer service. The details will be made public shortly. The company is working on a prediction engine that will be able to forecast the score.
At the heart of the company is an algorithm that takes in a huge range of factors. These include not just past performance, but things such as whether it is the manager’s birthday, a significant home match, even the weather. “Imagine that, say, Manchester United is playing Liverpool,” said Mr. Kornilov. “We know that Manchester United has more players who can strike from long distance. Usually in poor weather the manager will tell players to shoot from further out.
Knowing these small things allows us to tweak the analysis in favor of Manchester United.”
Last year, the company was analyzing games in just five leagues–France, Germany, Italy, Spain, and the U.K.–but has now grown that to 25, including Japan and Latin America.
The company came about almost by accident. Although based in Warsaw, Mr. Kornilov had a software firm in Ukraine. One of his developers had, as a hobby, worked on as soccer-prediction engine. He approached Mr. Kornilov.
“I didn’t believe it, so first I turned him down. But he came back. So I said ‘OK, you are so insistent. I will open an account with a book maker to bet on the Bundesliga [the main German league].’”
He put in €50 ($64.60) and bet according to the predictions the engine produced. “When that went to €300 really quickly I realized ‘this is it’. We spent months testing it to prove it. It really worked.”
Not every game can be predicted, and different leagues command different levels of confidence. The U.K. Premier league has high levels, since there are a few strong clubs and that gives the league stability, Mr. Kornilov said. “On games that we can call, for some matches we can predict with around 92% to 93% accuracy. For others it is around 55% to 60%.” Given there are three possible outcomes of a match, picking at random would give a 33% chance of success.
The same stability was true of the German, Spanish and Italian leagues, all of which had predictable “super clubs.” But he said there were some leagues where the success rate falls off.
“The French league–totally unpredictable.”
The company is working on tournaments as well as leagues. In the recent Confederation Cup, Betegy correctly predicted Brazil’s victory, although Spain was widely tipped. The company didn’t quite get the score right (it predicted 2-1; the result was 3-0).
The company is considering branching out into other team sports such as basketball and American football. But it’s not for every sport, even ones that attract large amounts of betting.
“There is no one in Ukraine who understands cricket. I don’t even understand the rules,” he said.
I shall be adjusting my Elo ratings to account for the five success factors above. I presume Jose Mourinho's birthday is December 25th - any other manager's special days known? This is important. But this was the best line:
Do they need to win because they are at home for the 100th match?
There may well be no one in the Ukraine who understands cricket, and there also may well be no one in the world who understands what the hell that sentence is all about. And if Manchester United play Liverpool in 'poor weather' in the "UK" Premier League, well my money is now on United - with all those players who can strike from long distance. Price not an issue - I'm all in. It really works.

Full Disclosure - I have no affiliation with Betegy. Nor do I want one.


Unknown said...

"I shall be adjusting my Elo ratings to account for the five success factors above."

"Not before time!" is the cry doubtless issued from the hordes of your suffering subscribers.

The article above is a bit incoherent and quite a lot is lost in translation I feel. However team motivation, weather and players are important factors and I expect they are using weightings too so why not give these guys a chance?

This article does highlight the problems of Elo ratings in that in this day and age they do seem woefully simplistic. It feels a bit like going off to a track day in a Rover SD1 3500: While it may have been the mutt's nuts in the late 1970's it sadly won't be competitive today.

The problem with these ratings systems is that unless you have the skills required to build a powerful model you are always going to lose out to those who can. In 2013 if you are not factoring in weather, team motivation, distance travelled, the WAGS and does the ref like a bit of bunga bunga, chances are you are behind the curve and conkers will be done.

Cassini isn't it time you faced the truth and admitted Elo ratings are far too simplistic and stuck in a 1970's timewarp?

Anonymous said...

ELO were a shite band