Saturday, November 28, 2009

EPL Mini Preview

Sorry for the lack of posts this week. I've been visiting family for Thanksgiving, a major holiday in the US. So as not to leave you completely hanging, here are the model predictions for the EPL this week. I should have a more-detailed preview for Arsenal - Chelsea and the Clasico tomorrow. I know it's late but perhaps you can read it the morning of the matches to get in the mood.


Blackburn - Stoke
Blackburn: 43.5%
Stoke: 27.4%
Draw: 29.1%

Fulham - Bolton
Fulham: 65.9%
Bolton: 11.7%
Draw: 22.4%

Man City - Hull City
Manchester: 75.6%
Hull: 7.1%
Draw: 17.3%

Portsmouth - Manchester United
Portsmouth: 22.5%
Man United: 47.9%

West Ham - Burnley
West Ham: 56.5%
Burnley: 18.2%
Draw: 25.2%

Wigan - Sunderland
Wigan: 29.8%
Sunderland: 42.1%
Draw: 28.1%

Aston Villa - Tottenham
Aston Villa: 39.8%
Tottenham: 32.3%
Draw: 27.9%


Wolves - Birmingham
Wolves: 36.9%
Birmingham: 31.9%
Draw: 31.1%

Everton - Liverpool
Everton: 26.1%
Liverpool: 47.5%
Draw: 26.4%

Arsenal - Chelsea
Arsenal: 49.1%
Chelsea: 25.7%
Draw: 25.2%


  1. With regards to Saturday's matches, the model doesn't seem to be accurate or even close. Makes me wonder if the matches are really 'predictable', such that it's better than an average's person guess. Just my thoughts.


  2. Having looked them over, it does seem a bit hit or miss. The most glaring match where the prediction feels off is Portsmouth - Man U. It certainly doesn't seem like Manchester United had less than a 50% chance of winning. I don't know what the actual percentage was but it should be higher. In contrast it appears the model may have overrated the away side in Wigan - Sunderland, though we shouldn't use the result to reach that conclusion since we're only talking about one match.

    I'm rather more optimistic. One reason I haven't been writing as much is that I've been working on testing a new model that incorporates various statistics like shots and corners. It seems to do much better and I'll probably feel confident enough to start using it next week. Using stats helps because there is a lot of luck when it comes to the score of a match. Using a lot of stats one can do a much better job of determining how good different teams are which helps the predictions a lot.

    Thanks for the comment.

  3. Based on the bookies' odds, I believe they rate Man Utd at about 70%? Pretty similar to Man City's chances.

    In an ideal (or near perfect) situation, I think stats would be pretty accurate, but I can't help but think that there's lots of external factors. Just to list a few (you get what I mean), poor refereeing decisions, reaction from players (referring to Wigan for bouncing back after their humiliating defeats), injuries to key players, or instances when managers rest players when playing weaker teams (complacency).

    Moreover, (correct me if I'm wrong), I learnt that stats and probability are based on long term. Thus I feel that it's difficult to predict the outcome of 1 match, as compared to the final league positions at the end of 1 season. That's what I think.


  4. The things you list are certainly factors that introduce randomness. However they don't mean that it's unpredictable in my view. In other words, those types of things are why Manchester United were, going by the bookies, 70% instead of 100%. I've been doing a lot of work because I think that it's possible to build a model to predict accurately that they are at 70%.

    As far as stats in general, the deal is that they converge in the long term. In terms of football, that means that they will be more accurate as the season goes on.