Monday, 28 September 2009

Shaping Form


One of the problems I have with the Elo rating system is that winning is everything. Your points are at stake, and if you lose, you lose them. It doesn't matter who you lose to. If you're say Accrington Stanley and you lose to Manchester United your rating diminishes just the same as if you'd lost to Rochdale. If you win, then who you beat matters. It doesn't seem right that if you lose, the strength of your opponent doesn't matter.

The other thing I have been pondering for a couple of days, is weighting a teams 'performance' rating. This idea was triggered by Peter Nordsted's comment that he looks at how "teams have fared in recent matches against opposition of a similar quality". While this would add another level of complexity to an already complex enough equation, it did get me thinking that it would be relatively easy to adjust a team's performance rating to take into consideration the opponent's strength. Clearly, it is far better for a team to have 10 corners and 20 shots on-target away to Arsenal than at home to Darlington.

Back to the Elo ratings for a moment, and the best predictor right now is not the rating itself, but the relative difference in ratings for the two teams over the past six games. For example on Saturday Birmingham City played Bolton Wanderers. Over the previous six games, Birmingham had gone down by 123 points, whereas Bolton had improved by 3 points. Although both home and away selections are in profit, the system is especially accurate if the form team is at home with 22 winners from 30, but 4 of those losers were on opening day when the form dated back to last season, and should probably be ignored.

Odds-on favourites in the Premier League are also having a good run right now. Since August 24th, only 1 (Chelsea at Wigan) has failed to win. 22 winners from 23 selections. And still the draws are nowhere to be seen.

2 comments:

Anonymous said...

Cassini,

I really must find out how you do your Elo ratings! :)

Im not sure how a teams rating gain when they win is linked to the quality of opposition but a teams rating decline isnt related to the quality of opponent.

The way I understand Elo ratings is, whatever rating increase the winner receives, the losing teams rating declines. The extent of this is primarilly driven by the extent of the rating difference beforehand (I include margin of victory but thats beside the point in this case.) Therefore, if a team lost to Man Utd, their rating would fall much less then if they lost to Accrington. Im intrigued! :)

You make a good spot about the rating history. If you manage to keep a log of teams ratings as they progress, its much easier to see false spikes and "blips." Whilst this isnt fool-proof, it doesnt help you avoid some of the occassions where you might unwittingly overstake on a "perceived value" bet.

JPG

RobbieFowler said...

Hey nice blog, and some very interesting posts about the elo and form ratings. I will take a closer look.. I have added your blog to my blog list and would appreciate the same?

I started a sports betting blog a few weeks back, mainly focusing on football but other sports like boxing and tennis too..

check it out..

http://lumpon.blogspot.com/

good luck!