PDA

View Full Version : Football Predictions System



mathare
29th July 2003, 22:19
I have a football system that is shortly going to enter it's second season and I need some advice from you guys.

The system rates every match and gives a prediction as to what the result will be (home win, draw, away win). Every team has a number of points in the system. It compares the points for the home and away teams and based on the difference predicts the results. The higher the number, the more sure it is of a win. The nearer zero the more sure it is of the draw. I never back it on draws, too risky.

Last season I took the top 12-15 qualifiers and used them as tips on a football website, split into 3 from 5 perms and an 4-way accumulator. But that was last year.

Have you lot got any ideas on better ways to back this system?

Let me provide a few stats from the first season to give you some idea of how accurate it is. I'm not sure what stats will help you assess it so I am providing a whole range.

Overall strike rate: 42.77%

Strike rates based on ratings (the higher the rating the more sure the system is of the result)
Rating S/R Number in bin
0+ 42.77% 2726
100+ 48.06% 2041
200+ 51.80% 1413
300+ 54.52% 974
400+ 58.06% 670
500+ 61.02% 449
600+ 64.41% 295
700+ 66.00% 200
800+ 67.86% 140
900+ 71.74% 92
1000+ 74.24% 66

So over the season there were 66 games rated at over 1000 and I got 74.24% of them right. I used to use 500-550 as my cut-off to produce the top 12-15 per week.

Strike rate per division:
Premier: 49.47%
Division1: 39.67%
Division2: 42.03%
Division3: 37.32%
Conference: 43.07%
SPL: 53.51%

Any thoughts? If you need any more info let me know. I think there is something in this system. A strike-rate of over 42% in it's first year seemed pretty good to me bearing in mind that at the start all teams were considered equal, e.g. Man Utd had the same starting score as Sunderland.

John
29th July 2003, 23:34
I can't really think of anything to add other than to wish you good luck with it - I think what you're doing sounds good and you've said you've got high strike rates from it in the past.

You mentioned that backing the draw is too risky - it might pay off to have a look at last season's final league standings and also results trends, to find teams which draw a lot of their games in a certain area. For example there may have been teams which draw a lot of away games when they played teams slightly above/slightly below them at the time in the league.

If you can find trends like this and adapt them it could greatly improve your profits and reduce the risk factor of backing predicted draws. :wink

Scoop
30th July 2003, 14:19
Your system sound good. Do you have to adjust teams as per new players joining/leaving a club.

Also keeping a track of booking and sending offs of key players could help you. what i mean by this is Shay Given for Newcastle is a top class keeper if he was suspended or injuried would the team have the same rating. Same for if Robbie Kean is injuried who have spurs got to replace him.

mathare
30th July 2003, 21:56
I see exactly what you mean and the answer is no, the system does not have any dependence on players at all.

Each and every team was given a starting score of 1000. Both teams in a match contribute to a kitty for that game. Winner takes all, if it is a draw the kitty is split. So after a game the score for the teams involved change depending on the result and therefore what they get back from the match kitty. Very simple and once it is set up for a season needs very little input from the person running it, other than entering the results from each match.

A mate of mine came up with an interesting staking idea for me today though. It is far too complicated for me too explain here but it works a bit like dutching I suppose. I use three perms, each a 3 from 5 as before, but the stakes for each perm differ according to the scores my system produces for the teams in the perm. So if my first perm had 5 teams in whose results were all rated at over 1000 I would back those more heavily than a perm containing 5 teams whose results were all over 700. The odds for the 1000+ rated teams would be worse but the strike rate is better and so the stake is higher, so the returns for each perm should work out about equal.

Thoughts?

MattR
4th August 2003, 20:33
Sounds like its a good system then. That staking idea seems to make sense. After all those plus 1000 ones are the ones with the best hit rate.
How does it do with games that look tight on paper?

mathare
4th August 2003, 21:03
I'm quite pleased with it bearing in mind it is only a season old.

Games that look tight on paper - tricky one. there is no formula you can apply to say whether or not a game looks tricky on paper. It's a subjectove matter, hard to quantify.

Scoop
14th August 2003, 09:26
Hi mathare, will you be posting your predictions for saturday.

mathare
14th August 2003, 12:52
Hi mathare, will you be posting your predictions for saturday.

Unfortunately not. I'm going to Kenya tonight for a couple of weeks. I will try and get them posted for the last weekend in August but I am more likely to start posting the qualifiers in September so check back here then