Elo rating system (my thoughts)

[muppet77]

Re: Elo rating system (my thoughts) Yes afford but that is a model formula to convert the difference in rating to price.

[salford_student]

Re: Elo rating system (my thoughts)

Yes afford but that is a model formula to convert the difference in rating to price.

I dont understand ? you can only ever generate probs/odds by looking at the differrence in ratings as it all relative

[Machine]

Re: Elo rating system (my thoughts) Muppet is correct That only gives you the probability of rating a beating rating b simply converting that that probability into odds does not work. Lets say a ratings difference of 100 gives him a probability of 50% = 2.00 If you look at all the variables which make up the model with a difference of 100 you may find that of that 50% ---- 50% under 2.00 are winners and 50% over 2 are losers Therefore simply backing above 2.00 because you think you have "value" is in effect working the opposite way and taking you into negative equity , thats when they fall into the "in the long run" mode and keep backing the losers through the false belief they are getting value

[muppet77]

Re: Elo rating system (my thoughts) i meant that a difference in rating of 100 means.... what? how did you get the probabilty from this? plug it into an equation? what's this? merely a simplification of the bigger picture - an approximation or model. there is no golden rule that says "all 100 rating diffs will have probability equal to 50%". you are using a model (the ELO) and then another model (pricing model). this is where the weakness lies. if your ratings are ACTUAL values that you van use then that applies directly to spreads, rather than to bookies. eg if you ELO goals scored team A has a rating of 2.30 and B 0.70, you can use this info to play the spreads rather than saying team A has a rating of 2507 and B 1589.

[Machine]

Re: Elo rating system (my thoughts) You asking me the question or ELO If its me I suggest you re read my answer to SS

[Machine]

Re: Elo rating system (my thoughts) Lets try and answer your questions anyway

i meant that a difference in rating of 100 means.... what? how did you get the probabilty from this?

As ELO has stated this is a home draw away model and has 3 differing models for home dray away then it is natural to presume that he has plotted his x axis as difference in ratings against y axis %age of home wins and like wise for draw and away.

there is no golden rule that says "all 100 rating diffs will have probability equal to 50%".

See above, 50% was an example, but a 100 rating will always have the same Px if the above model is used.

eg if you ELO goals scored team A has a rating of 2.30 and B 0.70, you can use this info to play the spreads rather than saying team A has a rating of 2507 and B 1589.

totaly irrelevent as he is using H/D/A

[salford_student]

Re: Elo rating system (my thoughts)

i meant that a difference in rating of 100 means.... what? how did you get the probabilty from this? plug it into an equation? what's this? merely a simplification of the bigger picture - an approximation or model. there is no golden rule that says "all 100 rating diffs will have probability equal to 50%". you are using a model (the ELO) and then another model (pricing model). this is where the weakness lies. if your ratings are ACTUAL values that you van use then that applies directly to spreads, rather than to bookies. eg if you ELO goals scored team A has a rating of 2.30 and B 0.70, you can use this info to play the spreads rather than saying team A has a rating of 2507 and B 1589.

in the ELO system, you define what rating differnce will generate a given probability. you then plug this into the formula and then whenever you use the formula in future it will generate probabilities in accordance with this. so whenever you use the ELO system, a given rating difference will always give the same probabaility. One of the basic tenets of the ELO is that there is a one to one relationship between the rating difference and the probability.In other words rating difference to probs is set by definition. as to how accurate these are is a different matter, as that comes down to if the ratings generated are accurate. I'm not sure what a pricing model has to do with things, fair enough if your setting prices (i.e. a bookmaker or trader on BF) but even then it's a pretty trivial matter to apply a bit of overround and bias. If your looking for value bets all you need is the prob, e.g. probs*price, if thats greater than 1, then your bet is +EV and you bet. there's even enough information there to shove into the kelly equation to work out your stakes. Regarding spreads, what you've posted just wouldn't work for lot's reason, main being the ELO ratings are not meant to reflect real world values.

[salford_student]

Re: Elo rating system (my thoughts)

Muppet is correct That only gives you the probability of rating a beating rating b simply converting that that probability into odds does not work. Lets say a ratings difference of 100 gives him a probability of 50% = 2.00 If you look at all the variables which make up the model with a difference of 100 you may find that of that 50% ---- 50% under 2.00 are winners and 50% over 2 are losers Therefore simply backing above 2.00 because you think you have "value" is in effect working the opposite way and taking you into negative equity , thats when they fall into the "in the long run" mode and keep backing the losers through the false belief they are getting value

Lets say a ratings difference of 100 gives him a probability of 50% = 2.00 If you look at all the variables which make up the model with a difference of 100 you may find that of that 50% ---- 50% under 2.00 are winners and 50% over 2 are losers if something has a 50% chance of winning and then only wins 50% of the time, that proves the model is working! if something has a 50% chance of winning and you get odds of 2.1, then ev=0.5*2.1=1.05 i.e. a EV of +5% How is that taking you into negative territory ?

[Machine]

Re: Elo rating system (my thoughts) Becauase the variables on which you have taken you %ages of wins for the model can be negative ie you can have 50 winners @ 1.5 and 50 losers @ 2.1 = 50% probability of a win so you back all the ones @ 2.1+ which are rated @ 100 and voila big loss Boom value theory out the window . Your ELO model only shows how many won or lost not at what price they won or lost @

[muppet77]

Re: Elo rating system (my thoughts) As soon as you plot rating difference against win percent, drawing a line of best fit and generating an equation to produce a price you are using an ESTIMATION.

[salford_student]

Re: Elo rating system (my thoughts)

As soon as you plot rating difference against win percent' date=' drawing a line of best fit and generating an equation to produce a price you are using an ESTIMATION.[/quote'] But all models are ESTIMATION, the question is wether it's a good estimation or a bad estimation. take your corners model it produces good estimation, would you throw it away because it is an estimation. we will never know what the true probability of any event is especially with sports as the event only ever take place once. so the search is to find models which will estimate the true probability. this estimation will always have an error, the key is to make models that reduce this error as much as possible, and then to seek out bets that where your margin greater than your error.

[salford_student]

Re: Elo rating system (my thoughts)

Becauase the variables on which you have taken you %ages of wins for the model can be negative ie you can have 50 winners @ 1.5 and 50 losers @ 2.1 = 50% probability of a win so you back all the ones @ 2.1+ which are rated @ 100 and voila big loss Boom value theory out the window . Your ELO model only shows how many won or lost not at what price they won or lost @

if you predict something that is a 50% chance why would anyone back it at 1.5 ?

[muppet77]

Re: Elo rating system (my thoughts) salford, my point is that some strategies do use estimates, fine, they have to be. However when you have an estimate of an estimate you lose some accuracy, do you not agree?

[muppet77]

Re: Elo rating system (my thoughts)

Regarding spreads, what you've posted just wouldn't work for lot's reason, main being the ELO ratings are not meant to reflect real world values.

..sure they can - you adapt them so that they take into account the real world difference, rather than having a randomly chosen "5% points kitty" that is taken by the winner. a long read, but have a look at this thread on doing exactly that. http://www.punterslounge.com/forum/f21/football-rating-system-discussions-ideas-12655/index27.html this is what ratings should be IMO, missing out the stage of converting a rating into a probabilty/price.

[a1ehouse]

Re: Elo rating system (my thoughts) I spent a bit of time developing an Elo Access database to work on finding value. I had a good strike rate but the odds on offer were pretty poor. Had long winning streaks and then a couple of big losses would wipe profit out. I also worked up prices for win lose and draw and had really good fits for my models. Then my mrs fell ill and I've had to put a lot of this on hold. Was building a pretty decent database too and will fire it up again as the season starts. I do have a lot of faith in the ratings and included home advantages, goal scored etc to refine the model. However as with all systems and me old mukker muppet will confirm, it's all about finding value. Would post a link to the thread but the forum is playing up and not displaying it.

[Machine]

Re: Elo rating system (my thoughts)

if you predict something that is a 50% chance why would anyone back it at 1.5 ?

You are not getting it are you You are not backing at 1.5 You are falsely presuming that the data that you put into your model for the X axis @ 100 and Y axis @ 50% are all above evens,therefore backing at 2.1 will give you a value bet if you bet above it. THEY ARE NOT. Lets simplify it. The league is a sort of ratings table agreed. The home team win rating in this table has never fallen below 41% or 2.44 agreed so by your reckoning if we back anything above its chance of winning we have value agreed So we will only back at 2.5 + giving an edge Now lets see what happens:

year	Back+	HW%	edge	edge%
2001	2.50+	2.15	1.16	16%
2002	2.50+	2.13	1.17	17%
2003	2.50+	2.37	1.06	6%
2004	2.50+	2.21	1.13	13%

	games	games	Hwin	eligible	bet	profit 
year	tested	won	%	games	return	loss
2001	2035	945	46.44%	450	388.9	-61.1
2002	1688	791	46.86%	397	362	-35
2003	2108	891	42.27%	420	357.21	-62.79
2004	1753	794	45.29%	410	368.92	-41.08

Shock horror whats happened, all negative , it cant be its a "value" bet your backing at odds higher than its chance of winning. :loon The same principle will apply to an Elo model. ;)

[Machine]

Re: Elo rating system (my thoughts) PS Start involving kelly and you'll lose a damn sight more :( If your are in the "in the long run" camp; years 2005 to 2009 show similar negative results. So 9 years is quite a long run for a trial I would imagine. :ok sits back and waits for the " Heresy ";" He's a witch "," burn him "," Van Der Wheil lives "; value punters onslaught ;)

[Shinigami]

[Datapunter]

Re: Elo rating system (my thoughts) A simply hanging would suffice me thinks... ;) I've not looked at ELO in detail so may be mistaken here but from what i remember isn't it so that for each match the rating of the winning team increases and the losing team decreases. And the amount with which that happens is dependend on the relative strenght rating of the team at that point in time. So if a high ranked team beats a low ranked team they get a few more points. But if a low ranked team beat a high ranked team they get a whole lot of more points. This would, to some extend, effectively incorporate a difference in odds and it is not simply a matter that each team for each match is considered 50/50. Their relative rating at the start of the match puts them in weak/strong starting position. As a side question, is it possible to incorporate the odds in the ELO rating and by doing so not predict the relative strenght but rather try to predict the chance of a team performing better than what their odds would indicate ? Probably neccesary to include some other statistics that just the final result to get something accurate enough to be usefull. I don't see this as a single reason for making a bet or not, more of an indicator in a series of indicators adding up to a decision to bet or not. ( and i'm including live trading strategies in the term "bet" )

[Lardonio]

Re: Elo rating system (my thoughts)

PS sits back and waits for the " Heresy ";" He's a witch "," burn him "," Van Der Wheil lives "; value punters onslaught ;)

I'd rather say your example is irrellevant to most people. What you are describing is not value betting, as can be clearly seen by the numbers, and it would take a very inexperienced gambler to expect value after such a bad analysis.

[Machine]

Re: Elo rating system (my thoughts)

I'd rather say your example is irrellevant to most people. What you are describing is not value betting' date=' as can be clearly seen by the numbers, and it would take a very inexperienced gambler to expect value after such a bad analysis.[/quote'] Like Ive said before you show me the proof instead of blowing hot air then I will take you seriously. At least I am showing why simply betting above a percentage of wins will not work in practice. Please show me how the numbers are wrong and prove your point, (which I doubt you can). I will discus anything with anybody as long as they talk reasonable sense but the garbage I have got back from you is not even worth my typing time. I do not suffer fools so I will say again Either PUT UP OR SHUT UP. :ok

[Machine]

Re: Elo rating system (my thoughts) Sorry about the slow reply Datapunter but internet has been on and off all weekend (damn NTL/Virgin)

as a side question, is it possible to incorporate the odds in the ELO rating and by doing so not predict the relative strenght but rather try to predict the chance of a team performing better than what their odds would indicate ? Probably neccesary to include some other statistics that just the final result to get something accurate enough to be usefull.

its a long time since I tried ELO ratings,Im using RPI at the moment(similar idea to ELO) I have never tried incorporating the odds directly into the ELO model, but I did once plot an odds v ratings scenario from the Elo ratings. Used the outcome on unused data and a live test which was profitable but nothing to shout about most probably just variance.

[muppet77]

Re: Elo rating system (my thoughts) I once used odds as the metric in a rating system just to see what it looked like. I posted the results in a thread that can be found at the link that I posted earlier. It may take a bit of finding, but if anyone wants to paste an accurate link to the specific post then feel free. I never used them it was merely Out of curiosity.

[Machine]

Re: Elo rating system (my thoughts) Interesting read Muppet did you mean this thread HERE or mister onemores results which I see return negative using ratings

[muppet77]

Re: Elo rating system (my thoughts) Yes that's it, although not sure which post number it is.

[Shinigami]

Re: Elo rating system (my thoughts) the problem with ratings is the assumption that "strength" is transitive - the assumption that if A beats B, & B beats C, then A can beat C. however, although Scissors beat Paper, & Paper beats Rock, it doesn't translate to the conclusion that Scissors will beat Rock. the nature of "strength" can be non-transitive in the sense that different teams have their own unique "Achilles Heel", so if A beats B, it only means that A has exploited B's weakness. When B beats C, B also may have exploited C's weakness. But this does not necessarily mean that A will exploit C's weakness as well. i think this is what ratings fail to accomplish.

[Machine]

Re: Elo rating system (my thoughts)

the nature of "strength" can be non-transitive in the sense that different teams have their own unique "Achilles Heel", so if A beats B, it only means that A has exploited B's weakness. When B beats C, B also may have exploited C's weakness. But this does not necessarily mean that A will exploit C's weakness as well. i think this is what ratings fail to accomplish.

ELO is just a gloryfied league table in effect, all be it a more accurate view of how teams performances are due to the weighting nature of the ratings. It would be interesting if you could run say 3 years ELO ratings through a neural net algorithm incorporating an odds algorithm to find the optimum weighting and then test on unused data. Alas that is way above my programming skills.:(

[Shinigami]

Re: Elo rating system (my thoughts)

ELO is just a gloryfied league table in effect, all be it a more accurate view of how teams performances are due to the weighting nature of the ratings. It would be interesting if you could run say 3 years ELO ratings through a neural net algorithm incorporating an odds algorithm to find the optimum weighting and then test on unused data. Alas that is way above my programming skills.:(

i think NN algorithmm that runs on 3-years ELO or any other variables won't bet much useful , in the sense that a lot of things change in 3 years. The current squad is much different from what it is 3 years ago. Even our own biological cells die & regenerate every 30 days, so effectively you & i are almost entirely not the same person as we were 30 days ago. There's a limit to the effectiveness of a NN algorithm. Nevertheless, as educated ppl, we'd rather lose intelligently than to win ignorantly.

[Machine]

Re: Elo rating system (my thoughts)

Nevertheless, as educated ppl, we'd rather lose intelligently than to win ignorantly.

Agreed :ok But the idea of NN is that it becomes dynamic once the foundations have been built and will change the algorithm as and when the new data is input, surely this must be better than a static algorithm This is what Im currently trying to build myself , but as programming is only a hobby the end product could be some time off, but Ill get there eventually I hope.(it may be a waste of time but you never know until you try). :hope

[Lardonio]

Re: Elo rating system (my thoughts)

Like Ive said before you show me the proof instead of blowing hot air then I will take you seriously. At least I am showing why simply betting above a percentage of wins will not work in practice. Please show me how the numbers are wrong and prove your point, (which I doubt you can). I will discus anything with anybody as long as they talk reasonable sense but the garbage I have got back from you is not even worth my typing time. I do not suffer fools so I will say again Either PUT UP OR SHUT UP. :ok

You're such a nice person to ask me so kindly with a funny smile, thumbs up and all, but unfortunately I can't put up. I don't follow one set method than can be proofed, I just profit from betting - that's what finding value is about. So I can't put up, and I'll shut up whenever I feel like it, and write something whenever I feel like that. I can't really help, nor care, that a few people are either to ignorant, to rude or to stupid to understand the sense in what I'm writing.