Football Correct Score

[topman]

Anyone any thoughts on correct score predictions? I know there are numerous threads out there and I have spent a lot of time analysing data, past results etc.

The main problem I come up with is the nature of football i.e. the relatively small number of goals scored. When you average these out over a number of seasons it gets possibly less meanginfull.

What am basically doing is comparing (H)ome (G)oals (F)or with the (A) (Goals) (A)gainst and getting the average which will be my home team score. Doing the reverse for the away team, HGA - AGF and this will be my away teams score. But the amount of reuslts that come in at 1.344, 1.674 etc. etc.

So it boils down to what do you do with a figure like 1.344. Straightforward rounding will obviously give you 1.0 and technically that's correct. I have tried applying a 'weighting' to my rounding so that it only rounds up if the remainder is > .8 (1.79 would still be 1.0) but I feel like I'm 'dickying' the figures.

This is what I have come up with for the first weekend and a lot of them do seem reasonable predictions.

Arsenal 2 1 Aston Villa
Bolton 1 1 Tottenham
Everton 1 1 Watford
Newcastle 1 1 Wigan
Portsmouth 2 1 Blackburn
Reading 2 1 Middlesbrough
Sheff Utd 1 1 Liverpool
West Ham 1 1 Charlton
Chelsea 2 1 Man City
Man Utd 2 1 Fulham

Appreciate this is far from an exact science (and Arsenal will probably knock 6 past Villa) but would appreciate feedback. Was also contemplating doing them in doubles (10 matches = 45 bets). If I got 2 correct scores I'd almost even break, 3/4 scores would be potentially lucrative.

[mathare]

First thing that strikes me is that you have every team scoring and that won't happen. You'll also find that 2-1 and 1-1 appear a lot in your predictions if you go down this route.

A better option could be to look into the Poisson distribution, knowing the average number of goals each team scores home and away you can work out the probability of each side scoring 0,1,2,3,4 etc. Then either use the most probable number of goals by each side to give you your score prediction or work out the probability of each score line from the Poisson probabilities and bet only the value selections (those whose probability exceeds that suggested by the odds).

[topman]

Hi Mathaer, must say I'm not too familiar with the Poisson method (sounds like an ex-girlfriend's perfume) but I will do some research

I had explored the option of calculating the actual number of goals scored and using that figure (20 games 1 goal, 33 games 2 goals etc.). But would still be comparing home scored with away conceeded?

Still think I might come up with similar results but will carry on experminting (I do have 3 days after all!)

[mathare]

I had explored the option of calculating the actual number of goals scored and using that figure (20 games 1 goal, 33 games 2 goals etc.). But would still be comparing home scored with away conceeded?Up to you :) It's your system, you set the rules. Why not test a few systems out?


Still think I might come up with similar results but will carry on experminting (I do have 3 days after all!)You have all the time in the world. The Premiership season starts on Saturday - so what? There is nothing that says you have to start the system then. Better to spend the time testing and paper trading the system and getting it right than rushing to be ready for the start of a season. The game will be around long enough for you to make money with the right system

[sportingprofit]

You could deduct goals conceded, to get a better average.

[topman]

Another problem with looking at actual number of goals scored in games is sheer volume of data. It's obviously easy to see Chelsea's GF / GA for the last four years which I have done for all teams. But different kettle of fish to go through 4 years worth of results?

[mathare]

Another problem with looking at actual number of goals scored in games is sheer volume of data. It's obviously easy to see Chelsea's GF / GA for the last four years which I have done for all teams. But different kettle of fish to go through 4 years worth of results?Then don't test it. But don't expect to know whether the system is any good or not.

I don't really see it as much effort to be honest. It could be done easy enough with a spreadsheet.

I wouldn't put cash on any system I hadn't thoroughly tested so if you're not going to back-test this you should start by paper-trading

[topman]

No Mathare I didn't mean that - obviously you are correct and any system should be thoroughly tested. I'm just annoyed a bit with myself after collecting four years worth of data, analysing, testing etc. Now I think I really need to back and collect actual game results for four years. But probably worth doing.

Appreciate your feedback.