• Hi Guest, The software has been updated but I have not had a chance to tweak anything yet.
    It took longer than I had hoped, so I just turned it on and hope everything is OK
    If you spot anything that does not look rigfhyt then please let me know.
    Ark Royal
  • There seems to be a problem with some alerts not being emailed to members. I have told the hosts and they are investigating.
  • Hi Guest, If you are seeing that Lurker has appeared under your name then please take a look here to see why. AR

What are the best ratings for soccer ?

I 'm using the elo ratings from clubelo.com currently.
When I say "currently" I mean before the virus, which let's hope will go away soon and we are going to see football resuming.

I have constructed a probability scale using those.

From this forum thread:


I borrow something:


..... Here are this week's Elo ratings fully updated:-
ENGLISH PREMIER LEAGUE.
Liverpool 2046 v Watford 1623 423 1
Burnley 1679 v Newcastle U 1716 -37 X
Chelsea 1871 v Bournemouth 1678 193 1
Leicester C 1839 v Norwich C 1590 243 1
Sheffield U 1668 v Aston V 1631 35 X
Southampton T 1640 v West Ham U 1685 -45 X
Manchester U 1803 v Everton 1699 104 1
Wolverhampton W 1751 v Tottenham H 1839 -88 2
Arsenal 1809 v Manchester C 2000 -191 2
Crystal P 1743 v Brighton & H A 1647 96 1

The contest between Elo and Paul Steel Ratings was surprisingly won by the former 7/10 to 5/10 but there was some strange results around again.
Ark Royal's selections also beat Paul Steele ratings scoring 6/10 so well done there also. ;)
Using those elo values my probability calculator says:

LIVERPOOL - WATFORD ............................... 81.7% - 11.2% - 7.1%
BURNLEY - NEWCASTLE .............................. 38.3% - 37.6% - 34.1%
CHELSEA - BOURNEMOUTH ...................... 67.2% - 17.7% - 15.1%
LEICESTER - NORWICH ................................ 73.8% - 14.9% - 11.3%
SHEFFIELD UTD - ASTON VILLA ............... 51.7% - 23.4% - 24.9%
SOUTHAMPTON - WEST HAM ................. 36.9% - 28.0% - 35.1%
MAN UTD - EVERTON .................................. 58.9% - 20.9% - 20.2%
WOLVES - TOTTENHAM .............................. 33.4% - 28.7% - 37.9%
ARSENAL - MAN CITY .................................. 26.2% - 29.7% - 44.1%
CRYSTAL PALACE - BRIGHTON ................. 59.0% - 20.9% - 20.1%

The first number is the probability of the home win, the second number is the probability of the draw and the third number is the probability of the away win.
Probability is important because it tells me if a certain result is a value bet (or looks like a value bet).

So in the last match (Palace v. Brighton) it advises me to buy the home win if the price offered is 1.69 or higher (1.69 = 1/0.59), to buy the draw if the price offered is 4.78 or higher and to buy the away win if the price is 4.98 or higher.

Strictly speaking I also use the elo values of the teams before the last match to make an average, as this improves things a little, but for the
sake of simplicity I don't do that now.

Now my studies show that this elo-probability formula of mine has a predictivity value of 38.5%.
The predictivity is a benchmark I 'm using and it is a quantity that derives when we compare the elo probabilities with the results.
38.5% is not bad. I don't know if it beats the bookies but it's not bad, for the English premier.

Let me compute the (partial) predictivity from just those ten matches:
The matches were held in December 2019 and the results were:

1 - 1 - 2 - X - 1 - 2 - X - 2 - 2 - X

So predictivity = (0.817 x 0.383 x 0.151 x 0.149 x 0.517 x 0.351 x 0.209 x 0.379 x 0.441 x 0.209 ) ^ (1/10) = 0.314

(it's on the low side - Chelsea losing to Bournemouth and Leicester drawing against Norwich made it so)


The thing is if those are true probabilities, then the selection system should work.
That is if I don't find a good price then reject the match (or declare it a "no bet"), if I do find a good price then make the bet.

If however I am a loser in the long run -say after 3-4 months of betting- what does it mean ?
It means they are not "true" probabilities !
But I did all this work very carefully -or you did it if I teach you how- so what went wrong ?
Let me tell you what went wrong.
Nothing went wrong but the bookies were using better probabilities !
In this game the "true probability" is the one that achieves the higher score, in the sense of the long multiplication I did previously.

So the situation is what it is and I can't report on new matches because as I said we are in this stoppage period.
But the starting point was the ratings themselves.
Those ratings are computed and recomputed every week as we know by means of another standard formula, which you can see in the clubelo.com site. It's a theory that was first proposed by a Russian mathematician, Dr Arpad Elo.
There are other ratings used here and there.
For the world cup of 2014 I remember FIFA were using some other ratings (but when I asked them how those compare to Elos they replied that "they don't make comparisons").
It's a fact the the ratings makers tend to avoid producing benchmarks - the Elo sites too don't give benchmarks.
But from all those verious rating systems which ones do you think are the best is my question to the forum (so hopefully we make the score go higher) ?
 
Last edited:

Delboy99

Gelding
..... it all sounds so very technical that I thought that there was some attempt being made to reach a formula for stopping the coronavirus. Actually these ratings are from Club Elo and I would have thought by the very fact that the margin of difference created by each team rating would be enough to work out which team has the best chance of winning etc. There are loads of sites offering Elo ratings and most of these appear to be similar in the production of their predictions. For example look at the Liverpool v Watford match with a rating difference of 423, do we really need to work out probability as to who is actually supposed to win the match. Again the ratings between Burnley v Newcastle -37 actually shows just how close this match is likely to be. Interesting line of thinking though and I suppose something one could pursue if the inclination is strong enough. ;)
 
..... it all sounds so very technical that I thought that there was some attempt being made to reach a formula for stopping the coronavirus. Actually these ratings are from Club Elo and I would have thought by the very fact that the margin of difference created by each team rating would be enough to work out which team has the best chance of winning etc. There are loads of sites offering Elo ratings and most of these appear to be similar in the production of their predictions. For example look at the Liverpool v Watford match with a rating difference of 423, do we really need to work out probability as to who is actually supposed to win the match. Again the ratings between Burnley v Newcastle -37 actually shows just how close this match is likely to be. Interesting line of thinking though and I suppose something one could pursue if the inclination is strong enough. ;)
Well L'pool's difference is pretty telling but I always need the probability figure to work out the price, or the price I 'd like to see rather.
Also I have to measure success rate somehow..
Do you know rating systems besides clubelo I could calibrate ?

p.s. I know coronavirus too (!), if you give me some numerical data I can model the expansion rate.
A rough calculation I made shows that if the dangerous contacts we make with other people are limited to 1 in 72 hours or less, then
the epidemic dies (of course we don't want to make any dangerous contacts, but it was my test parameter, the frequency of dangerous contacts).
I also assume that my lifetime as a dangerous carrier is 5 days, after that I 'm sick and I 'm in bed but no longer dangerous to the community.
 
Here is a new thought.
Suppose we use Elo ratings or Steel's ratings or another system.
Suppose the team we follow is Crystal Palace.
Come next season in August, they play against Liverpool at Enfield and they win.
This will give Crystal Palace some 20 points.
The league matches continue and it is proven that Liverpool are no longer a force. They lose away, they lose at home and Klop faces the sack.
Should n't we then go back and snatch from Crystal Palace those 20 points ? Make them 5 say, retroactively.
This thing should be done allover the place in a kind of musical chairs algorithm.
Could it result in more accurate predictionwise ratings ?
 
Last edited:
To a first approximation every team is supposed to have their present elo at the beginning of the season and the subsequent elos are recomputed on that basis. So L' pool ends up with a different rating and C. Palace likewise.
Should n't this then predict the next results more accurately ?
 

Delboy99

Gelding
... What you say is incorrect. The Steele Power Ratings commence each season on a new mark of 10 points and I believe that they are not used as ratings until so many games have been played. As these games are played the ratings change and arrive at a point when we can start to use them. I would have to dig out the book to confirm just how many games this is. :)
 
... What you say is incorrect. The Steele Power Ratings commence each season on a new mark of 10 points and I believe that they are not used as ratings until so many games have been played. As these games are played the ratings change and arrive at a point when we can start to use them. I would have to dig out the book to confirm just how many games this is. :)

Ok so let's suppose Steel ratings are good for the second half of the season, after the new year say.
If they work then we win and we don't care about the first half of the season.

But I think I have a point here.
Even if you start all the teams from the same mark retrospecitve reevaluation may help.
My difficulty is that to do something and to check this with number crunching I need an entire database of elos, from the dawn of time if I could.
 

footysystems

Gelding
Why don't you just rate on current form scoring each win draw loss goals against goals for and so on much easy and clear..

Drag the data from soccerstats..
 
Why don't you just rate on current form scoring each win draw loss goals against goals for and so on much easy and clear..

Drag the data from soccerstats..
Give me an example of what you think.
My method is easy and clear and it's theoretical score is 38.5%. In the first post I used the elos from delboy's post and worked out the rest in one minute. The important think is the score -or the punter's bank balance if you like- but ease of use is also important of course.
 

footysystems

Gelding
Not on my PC if you look through football systems threads on the forum going back around 2014 my way examples are there...
 
Not on my PC if you look through football systems threads on the forum going back around 2014 my way examples are there...
I will search.
I mean to use my system from now on and see what happens.
To be sure I don't think it is any good for the matches that are being played now, Carribean, Siberia, Indochina ..., I mean later -after the virus age.
But I always declare "no bet" certain suspect matches, e.g. second leg matches when the affair has been settled from the first leg and the like.
 
Explain please ? Not sure what you mean ?
I have to check if any of those databases in 538 are available for download.
I mean what's their catch rate ?
The global minimum is the number 1/3 (we make entirely random guesses) and the global maximum is the number 1 (we know everything in advance). The maximum I ever recorded was 42% with Greek leagues of the 80s-90s. The Greek league in those years (and maybe even now) was easy in the sense of high preponderance of home wins - 55%.
 

footysystems

Gelding
This just using elo ? 42% strike rate ? This I take it's home draw away markets pre ko ?

There are many many more opportunities inplay and getting higher prices for teams that were odds on. Do you program ?? Python and so on ?
 
Top