townterrier Posted August 27, 2007 Share Posted August 27, 2007 I have recenty read a Kevin Pullein piece on ratings in which he states: "The Elo rating system lends itself very well to the purpose of sports betting in that it can be used not only to determine the relative strength but also to calculate probabilities and thus odds" I have created my Elo ratings system however I am now left with a table that can tell me Everton are so many points better than Bolton but how can I calculate odds from this as Pullein suggests I should be able to do with Elo ratings? Thanks for your help. Quote Link to comment Share on other sites More sharing options...
GaF Posted August 27, 2007 Share Posted August 27, 2007 Re: Converting Elo ratings into odds Do you have (or can you get) historical ELO's at a point in time for old games? I'll assume you can...... I'd be tempted to graph the difference in the ELO between the 2 teams (being positive if the Home team is favourite, and negative if the away team has a better rating) with the percentage of the time there is a home win/Away Win/draw (3 seperate graphs) for each "difference" and see if there is a pretty clear "line of best fit" (from which I could have the percentage chance and thus calculate the odds). Quote Link to comment Share on other sites More sharing options...
townterrier Posted August 27, 2007 Author Share Posted August 27, 2007 Re: Converting Elo ratings into odds Ah so there isn't a set way to do it? I assumed there was from Pulleins wording. Quote Link to comment Share on other sites More sharing options...
GaF Posted August 27, 2007 Share Posted August 27, 2007 Re: Converting Elo ratings into odds My assumption is that this is just an "arbitrary" rating (Can ELO be calculated a range of ways, or is it always one set formula? :unsure) based on previous results - a certain amount of empirical data would therefore be needed to project probabilities... I would suspect that an ELO rating in different sports, even where the value and formula is the same, would equate to different probabilities of the higher rating prevailing.... Quote Link to comment Share on other sites More sharing options...
GaF Posted August 27, 2007 Share Posted August 27, 2007 Re: Converting Elo ratings into odds I'm sure Datapunter and/or slapdash (and others) will be along shortly with far more informed answers :ok Quote Link to comment Share on other sites More sharing options...
townterrier Posted August 27, 2007 Author Share Posted August 27, 2007 Re: Converting Elo ratings into odds Ok mate thanks for the replies. Quote Link to comment Share on other sites More sharing options...
slapdash Posted August 27, 2007 Share Posted August 27, 2007 Re: Converting Elo ratings into odds The way Elo rating systems work is to adjust the ratings after each game according to the difference between the "actual score" and "expected score", so if you know how to update the ratings then you should be able to deduce how to calculate the "expected score". In a game without draws, the expected score (assuming the rating system works on the basis of 1 for a win and 0 for a loss) is exectly the probability of winning. If draws are possible, it's going to be more complicated. Quote Link to comment Share on other sites More sharing options...
merlin Posted August 27, 2007 Share Posted August 27, 2007 Re: Converting Elo ratings into odds .... I believe Elo ratings are worked out as a percentage of the pot. The home side contribute 7% of their total points to the pot, and the away team contribute 5% of theirs.If either team wins, they get the whole pot added to their points,while the loser loses whatever they contributed. If draw the pot is halved.I think this is correct from memory.;) Quote Link to comment Share on other sites More sharing options...
townterrier Posted August 27, 2007 Author Share Posted August 27, 2007 Re: Converting Elo ratings into odds I fully understand how Elo ratings work, how to create them etc. it is more how to convert the results into percentages and therefore odds that i am interested in. Quote Link to comment Share on other sites More sharing options...
London_Calling Posted August 27, 2007 Share Posted August 27, 2007 Re: Converting Elo ratings into odds Fiddle with this, and please report back with an opinion: A -= The lowest nummber in your range (league) B = the highest number in your range C = B-A The maths: your team rating - A / C That should give you relative percent values for your teams. Quote Link to comment Share on other sites More sharing options...
slapdash Posted August 28, 2007 Share Posted August 28, 2007 Re: Converting Elo ratings into odds I fully understand how Elo ratings work' date=' how to create them etc. it is more how to convert the results into percentages and therefore odds that i am interested in.[/quote'] The Elo rating systems I know more or less require you to calculate the "expected score" in order to update the ratings after a game: the change in ratings is a function of the differrence between "expected score" and "actual score". I suspect this is what Pullein is referring to. Maybe if you gave some more details about the particular system you use ...? Quote Link to comment Share on other sites More sharing options...
London_Calling Posted August 29, 2007 Share Posted August 29, 2007 Re: Converting Elo ratings into odds The Elo rating systems I know more or less require you to calculate the "expected score" in order to update the ratings after a game: the change in ratings is a function of the differrence between "expected score" and "actual score". I suspect this is what Pullein is referring to. Maybe if you gave some more details about the particular system you use ...? You're normally quite precise with language so can I ask if you meant "expected score" or 'expected outcome' (H/D/A)? Fwiw, if you have Home and Away ELO ratings, the week-on-week changes would seem to be a pretty sensible indicator of Form (a not insignificant issue by itself) - assuming you manipulate that derived data effectively. Would folks agree with that? Quote Link to comment Share on other sites More sharing options...
townterrier Posted August 29, 2007 Author Share Posted August 29, 2007 Re: Converting Elo ratings into odds London Calling, i'll post my findings of you're above suggestion asap, just hadn't had time yet. My ratings offer the expected outcome rather than the expected score. They take into account the date, venue, score and quality of opposition. Quote Link to comment Share on other sites More sharing options...
slapdash Posted August 29, 2007 Share Posted August 29, 2007 Re: Converting Elo ratings into odds You're normally quite precise with language so can I ask if you meant "expected score" or 'expected outcome' (H/D/A)? I meant "expected score", but didn't mean "score" as in "3-2 to Man. Utd"., I meant a number that summarizes the outcome. When Elo ratings are used in games like chess, the "actual score" is 1 for a win and 1/2 for a draw. Your "expected score" based on the current ratings of you and your opponent will be some number between 0 and 1, equal to the probability (according to the ratings) of a win for you plus half the probability of a draw. Quote Link to comment Share on other sites More sharing options...
asc Posted August 29, 2007 Share Posted August 29, 2007 Re: Converting Elo ratings into odds First of all, you should calculate the Elo ratings for as much as premier league matches as you can. With these historical results you can use a technique called Regression Analysis to calculate the "fair" odds. I did this "excercise" for all premier league matches for the periode 1993-2006 and the equation for P(home win) turned out to be: y=3.1632E-04x+4.7589E-01 (with x = the difference in elo rating between home and away team). Example: Manchest City - Aston Villa (28-04-2007) City: 817 points Villa: 873 points Diff: -56 points P(home win)=3.1632E-04*(-56)+4.7589E-01=0.4582 Odd(home win)=2.18 This is very close to the odds that were offered by the bookies, and I am quite sure that bookmakers also use Elo Ratings to calculate their odds. More info on regression analysis can be found in Joe Buchdahl's excellent book "Fixed Odds Sports Betting" Hope this info is useful to you.... Quote Link to comment Share on other sites More sharing options...
Chris3b Posted August 30, 2007 Share Posted August 30, 2007 Re: Converting Elo ratings into odds Hi Everybody, firstly, excuse me for my poor english. I'm French and new on this nice english betting forum. I think Elo rating and regression is a nice way to find value. The message of asc is very interesting and call reflexion. To Asc, and others expert in mathematics Can you give me some help about this formula ? y=3.1632E-04x+4.7589E-01 1) It is a formula regression ? 2) How did you find 3.1632E and 4.7589E 3) What is E ? (sorry for my poor mathematics) 4) How did you find 04 ? 5) What is the optimum elo rating for historical results ? 1 season of Home and away ? 2 seasons , 10 season ? 6) Asc, Have you backtested this formula for english premier league ? have you good results ? Thnak you very much for your help :ok Christophe Quote Link to comment Share on other sites More sharing options...
asc Posted August 30, 2007 Share Posted August 30, 2007 Re: Converting Elo ratings into odds To Asc, and others expert in mathematics Can you give me some help about this formula ? y=3.1632E-04x+4.7589E-01 1) It is a formula regression ? 2) How did you find 3.1632E and 4.7589E 3) What is E ? (sorry for my poor mathematics) 4) How did you find 04 ? 5) What is the optimum elo rating for historical results ? 1 season of Home and away ? 2 seasons , 10 season ? 6) Asc, Have you backtested this formula for english premier league ? have you good results ? Christophe (1) It's a formula that is the result of regression analysis. The formula can be determined by Excel, by putting your data in a graph (with x=elo_difference and y=number of home wins percentage). After adding a trendline, Excel will calculate the formula for the trend line. (2) Excel found it for me, see (1) . Excel can also determine the regression coefficient which is a measure for how "reliable" the formula is. For this formula (UK home wins) the regression coefficient was 0.88 which is very good. More on regression analysis can be found here: http://en.wikipedia.org/wiki/Regression_analysis (3)(4) E-04 is a scientific notation. For example: 5.5*E-4 Is the same as 5.5/10000=0.00055. So in 'normal notation' the formula is: y=0.00031632x+0.47589 (5) You need as much results as possible, if you have too few results, there is too much 'noise' and the regression coefficient (=reliability) will decrease. I calculated this formula for the period 1993-2006 (4045 matches). The first 10 competition of rounds for each season are not taken into account because every team starts at 1000 points so it takes some matches to get every team in the right position. (6) I used this system to find value bets for home wins (8% yield), but I was not able to backtest it for several seasons , because i dont have all these odds in my database. If you want to dive into this, you really have to buy Joe Buchdahl's book, because this technique is explained very well in it. Good luck:ok Quote Link to comment Share on other sites More sharing options...
Chris3b Posted August 30, 2007 Share Posted August 30, 2007 Re: Converting Elo ratings into odds Thank you very much asc for your answers. it's more clear for me. I'd like buy joe Buchdal's book, but i think i'll be out with english technicals terms. Why these sort of book is not traducted in french ? perhaps not enough french punters...:( Quote Link to comment Share on other sites More sharing options...
muppet77 Posted August 31, 2007 Share Posted August 31, 2007 Re: Converting Elo ratings into odds have you considered running the seasons into one another, thus eliminating the need to start again at 1000 points? perhaps give newly promoted teams the average points of those relegated? Quote Link to comment Share on other sites More sharing options...
trollerboy Posted August 31, 2007 Share Posted August 31, 2007 Re: Converting Elo ratings into odds @asc good work. i´ve also read buchdahls book and know about those models. what did you use: just the results or also shots on target/corners data? did you do that for leagues throughout europe? did you do that also for away wins? for you home wins: can you tell me the yield if you only had taken value bets with odds >1.9 ? Quote Link to comment Share on other sites More sharing options...
Dumyat Posted September 1, 2007 Share Posted September 1, 2007 Re: Converting Elo ratings into odds Asc: how did the system work in practice ie 2006-2007 season - did u actually achieve 8% yield. Quote Link to comment Share on other sites More sharing options...
longfoot Posted September 6, 2007 Share Posted September 6, 2007 Re: Converting Elo ratings into odds Sounds very interesting. I made calculations in Spain Primera 5054 matches and got following result: Quote Link to comment Share on other sites More sharing options...
asc Posted September 6, 2007 Share Posted September 6, 2007 Quote Link to comment Share on other sites More sharing options...
asc Posted September 6, 2007 Share Posted September 6, 2007 Re: Converting Elo ratings into odds Sounds very interesting. I made calculations in Spain Primera 5054 matches and got following result: Quite close to the equation that I calculated, based on 'only' 2872 matches (1993-2005, first 10 matches where skipped). y = 0.00025461x + 0.48437 with R^2=0.83062. liisy, did you also skip the first 10 matches and how many seasons did you use to calculate the data? Quote Link to comment Share on other sites More sharing options...
cavello Posted September 6, 2007 Share Posted September 6, 2007 Re: Converting Elo ratings into odds Because the ratings grow during the season I have rejected them as a way of assessing "true odds" An advantage of 500 points in November is not equivalent to an advantage of 500 points in April. Quote Link to comment Share on other sites More sharing options...
longfoot Posted September 7, 2007 Share Posted September 7, 2007 Re: Converting Elo ratings into odds liisy, did you also skip the first 10 matches and how many seasons did you use to calculate the data? No, I transferred ratings from season to season and promoted teams got relegetad teams ELO. I started from 93/94, but in regression analysis I eliminated that season in full. Quote Link to comment Share on other sites More sharing options...
longfoot Posted September 7, 2007 Share Posted September 7, 2007 Quote Link to comment Share on other sites More sharing options...
noolnool Posted September 11, 2007 Share Posted September 11, 2007 Re: Converting Elo ratings into odds How do you calculate these ratings? I mean, is there some kind of software to do it for you? It seems kind of a big task to calculate the ELO ratings for 1000's of matches manually. At least 2 weeks of work. Quote Link to comment Share on other sites More sharing options...
asc Posted September 11, 2007 Share Posted September 11, 2007 Quote Link to comment Share on other sites More sharing options...
asc Posted September 11, 2007 Share Posted September 11, 2007 Re: Converting Elo ratings into odds How do you calculate these ratings? I mean, is there some kind of software to do it for you? It seems kind of a big task to calculate the ELO ratings for 1000's of matches manually. At least 2 weeks of work. I have an MS Access database with the match results of all major european leagues since 1993/1994. I wrote code in VBA (Visual basic for applications) to calculate the elo ratings. Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.