Talkbet posted an article last month reviewing his first complete season for his Powerstats app, which appears to be somewhat similar, if a little more sophisticated, to my Elo ratings. He commented that different leagues have their own personalities, and having recently completed the time consuming task of catching up the results to the end of the season after my data loss in February, I can confirm Talkbet’s observations.
I maintained ratings for the top five divisions in England, the four Scottish leagues, as well as the top leagues in Italy, Spain, Germany and France. For cup and European matches, points moved from league to league. The starting points were necessarily somewhat arbitrary, but were based on the UEFA coefficients, which seemed as good a starting point as any.
I also started recording the Under / Over 2.5 goal markets midway through the season to see if there was any evident trend.
It’s been a while since I wrote about the ratings, so for any newcomers the story is this. I keep a spreadsheet (backed up on a frequent basis now) which contains the current ratings for all teams in the aforementioned leagues. For each match I enter the ratings for each team and a formula calculates the margin of victory required for the two teams’ ratings to stay closest to their current values. It’s a work in progress, but in certain areas, for example predicting the draw in the Premier League, it performed remarkably well.
The results are promising; no Holy Grail for sure, but something solid to work on. The next few posts will reveal some of the successes and failures of the ratings, overall and by league, and also those teams whose ratings did not reflect their final league positions.
No comments:
Post a Comment