Match Rating System

[AndrewKent]

Hi folks, Its me again, still on my quest to find learn how systems are developed. In a previous thread I asked about this system: http://www.football-data.co.uk/ratings.pdf well, I've got as far as capturing all the data, working out the probability of 1X2 based on this, AND have plotted all the graphs with best fit lines which give me the calculations I need. Question is, in the example, the author is looking at match ratings of "0" which gives the equation of "y = 1.56x + 46.47", 46.47 being the intercept point on the Home Wins graph, which is point "0". Now, if I wanted to have the same calculation but for a match rating of say, -5, would the 46.47 AND 1.56 have to change? Considering that a rating of -5 (and excuse the eyeball calculation here) is 38-39%? Therefore, the 1.56 (and R squared) calculations would be different. Any help? Andy

[relf]

Re: Match Rating System If the match rating is -5, you just need to substitute x=-5, 1.56 and 46.47 remains the same, that gives 38.67%. Which means the fair odds is 2.59. Anything above that is value. I did take a look at this before, but I realized it's a bit too simplistic to judge from the previous 6 matches. After all, most teams have VERY DIFFERENT home and away forms, and the ratings can vary quite a lot depending on whether the 6 games are made up of 3 home and 3 away, or 4 home 2 away, or 2 home 4 away, etc. But if I were to sort based on the previous 6 home results, or the previous 6 away results, then it's a bit too much to ask because there's only 19 home/away matches in a season.

[AndrewKent]

Re: Match Rating System Thanks relf, there was a flaw in my calculation, that's why it wasn't working but it's sorted now and it matches what you said. On the other point your raise about this being too simplistic, I understand. It's only one measure. If however this SAME approach was taken on other measures would it be more worthwhile? For example, a points system for last 6 games/Last 6 Home/Away games. or bringing Shots into the mix as well? So the approach remains the same, but the data changes? Andy

[Faxy Fowler]

Re: Match Rating System relf - agree with the above post. In my view its an interesting approach in the PDF but it misses the point in that with 47% of games being a home win, and 2.16 therefore being fair on average when the teams are even, you'll find teams with odds over 2.16 win far less, those under 2.16 win far more, so all the over 2.16 teams in isolation still won't win often enough compared to the under 2.16s to make it viable in my opinion. Equally if you go for those that on the face of it don't provide value, you'll get an excellent strike rate but the odds are so poor that the more rare losses hit harder. Andrew - I played the recommended system on the PDF using homes only for the 2008-09 season with a starting bank of £250 and a flat £3 per bet, using the top five English leagues and the best odds on offer with either Betfair (after commission) or Bet365. One or two erroneous bets aside, I ended up around £100 down over the season and only 1 selection over 5.00 won, which was Stoke against Arsenal. I had a lot of bad luck with late goals going against (most notably, a late Arsenal equaliser at Sunderland cost me around 24 quid). I did something like 750 bets over the course of the season and you will almost always find yourself betting on home outsiders. My bank was only ever ahead after week 2, but there was never any danger of losing the whole bank - it never got below £190 till the last 3 or 4 weeks of the season. The bank hit its lowest at the end of the season which ties with what many people say about season ending games being better avoided. I might have been up if I'd gone with favourites only or stuck to odds under 3, or used more money the lower the odds, but not by enough to make it worth the effort. Conference games seemed easily the most profitable but all the other divisions lost money. As I say, my rudimentary methods and the pain in the backside of having to put all the bets on got the better of me in the end!!! I looked after the season at using points instead of goal differences but it didnt look a whole lot more profitable on back testing. I don't know how a win 6 games ago is any more or less relevant than it was 7 games ago, yet it suddenly becomes irrelevant if we look at the last 6 games? I don't see the evidence of the last 6 games being relevant - or at least, being the most indicative. Choosing less games to derive odds means its more unreliable based on inconsistent opponents, choosing more means games from even longer ago are being assessed which are arguably even less relevant. I am personally doubtful as to the merits of shots on goal. Don't let me stop you thinking about it, but look at the shots on goal from the recent Man City-Everton game (SKY had it something like 24-4 to City) yet Everton won 2-1. Teams like Stoke get goals using different techniques to Arsenal, for example. The stats are evidently better than going for it blind but there's a bit of me suggesting the efficiency of football markets is extremely tough to work out. Better to go for the Asian Handicaps maybe where the bookie margin tends to be closer to 4% than 8%; at least its 4% you don't need to find. My new idea is to try and look at teams form over a longer period and assess whether they are presently in their best or worst form at that moment. For example, teams like Chelsea, Preston, Sheffield United and Lincoln are as bad as they've been for ages, so worth laying (Chelsea 1 win in 9 premiership games - would have got us something approaching 7 pts for 4/5 pts risk over that time), equally teams like Man City, Bournemouth, Norwich and Wycombe are so on top of their game they are worth backing. I'm going to give it a go before posting any selections and maybe backing them or opponents as appropriate in asian handicaps is a way forward.

[Donkson]

Re: Match Rating System also keep in mind that the data in the pdf is basically decade old.

[AndrewKent]

Re: Match Rating System Hi guys, thanks for the imput. I agree with you both that this measure on it's own isn't profitable and needs other things added to it. I'm going to tip things on it's head here and ask rather, is THIS system any good, what is in your opinion is a better set of measures? Again, it's all down to personal opinion and experience so I'm happy to pursue any measures people think are worthwhile. Cheers, Andy

[AndrewKent]

Re: Match Rating System Donkson, the data might be but I wondered if the calculations were still relevant. I have up to date data to analyse so that's not a problem.

[relf]

Re: Match Rating System

Thanks relf, there was a flaw in my calculation, that's why it wasn't working but it's sorted now and it matches what you said. On the other point your raise about this being too simplistic, I understand. It's only one measure. If however this SAME approach was taken on other measures would it be more worthwhile? For example, a points system for last 6 games/Last 6 Home/Away games. or bringing Shots into the mix as well? So the approach remains the same, but the data changes? Andy

It would definitely be useful if you add other factors into the research. My opinion is that purely using this simplistic match rating will get nowhere. If I still remember, I back-tested this on one of the recent seasons (can't remember which one), and it was a disaster. So hopefully, by adding some other elements, it improves the predictions/value/etc.

[AndrewKent]

Re: Match Rating System Yes but what other eliments wouldnyou consider to be more important?

[relf]

Re: Match Rating System Frankly, I don't know. Personally I don't think shots or possession are accurate in judging the team's strength. I guess, perhaps the more experienced punters can contribute to this discussion.

[Donkson]

Re: Match Rating System

Donkson' date=' the data might be but I wondered if the calculations were still relevant. I have up to date data to analyse so that's not a problem.[/quote'] I think its definitely relevant, form is always relevant. I think the idea is fairly sound but imo there are a few statistical things in the pdf (in terms of the regression and analysis) that should to be tweaked to give a more accurate and profitable prediction. One in particular is the idea of using all results to generate a formula to use for a fraction of the games, this is quite flawed imo.

[AndrewKent]

Re: Match Rating System Thanks for the input guys, Now, although I have had only a limited amount of experience with regression analysis in the past, is it worth continuing? I feel from looking at the example, and from what I've built, that it might give some good insights, if tweaked a bit more as you say. I intend to run some tester analysis using this system against the last 2,3,4 and 5 games and a final one looking at a season as a whole to see which test set turns up the best predictions. @ Donkson: What in your opinion would be a less flawed way to analyse the data? Obviously no system is perfect but can you make any suggestions? Andy

[Donkson]

Re: Match Rating System Well my main point is that the regression is done using the results from all games within whatever time period it is, then when they do the backtest, they use the same formula to test with different parameters such as -5 to 5 ratings only etc.. Where really it would make sense to only do a regression of the specific data that you want to test, games with a match rating of 19 or -12 don't have any statistical significance to games between -5 and 5. It doesn't necessarily mean that their results should be better or worse, but they would definitely be different.