Mouseycat Posted May 4, 2016 Share Posted May 4, 2016 Hi everyone, I would be grateful to receive any insights from anyone who creates his/her own elo rating formula for football. Rnew = Rold + K(1-E) 1. How do you determine the value of K. As far as I know, in chess they reduce the k value gradually for some reason that is inapplicable in football. Should K be valued based on a certain ratio between K and the base rating? 2. Elo rating calculates probability of gaining/losing points, aka win/lose, did anyone try to modify it to accommodate draws as well? 3. Did you have a separate elo rating based on home/away results? 4. Can elo rating system be applied on determining team to score first goal? Thank you in advance! Quote Link to comment Share on other sites More sharing options...
Sir Puntalot Posted May 4, 2016 Share Posted May 4, 2016 You can see mine at the bottom of https://www.punterslounge.com/ Explanation here https://www.punterslounge.com/football-ratings I weight mine slightly differently for home and away, but not too much. It's obviously slightly harder to win away than it is at home. Quote Link to comment Share on other sites More sharing options...
Mouseycat Posted May 4, 2016 Author Share Posted May 4, 2016 6 minutes ago, Paul Ross said: You can see mine at the bottom of https://www.punterslounge.com/ Explanation here https://www.punterslounge.com/football-ratings I weight mine slightly differently for home and away, but not too much. It's obviously slightly harder to win away than it is at home. Hi, I assume you have devised some sort of method (based on the difference +32) to calculate the probability of draw in this example? Example 3 Leeds 67 v 35 Leicester +32 2.50 3.25 2.50 Quote Link to comment Share on other sites More sharing options...
Sir Puntalot Posted May 4, 2016 Share Posted May 4, 2016 18 minutes ago, Mouseycat said: Hi, I assume you have devised some sort of method (based on the difference +32) to calculate the probability of draw in this example? Example 3 Leeds 67 v 35 Leicester +32 2.50 3.25 2.50 Nope, +32 is basically home rating minus away rating, 67 - 35 = +32. Quote Link to comment Share on other sites More sharing options...
Sir Puntalot Posted May 4, 2016 Share Posted May 4, 2016 Oh hang on, you mean the closer the rating the more chance of a draw? I would have to say yes, but draws are very hard to predict. I prefer to seek out good value in the odds. For example: If an away team was only -5.00 behind the home team, yet was 5/1 or more, I'd be keen on X2 and maybe even the full 5/1. Had a similar one in France earlier this season, Nantes at 5.75, they won 4-1 too I think! Nothing is the bible though, it can only be a guide. Quote Link to comment Share on other sites More sharing options...
Pawn_pt Posted May 4, 2016 Share Posted May 4, 2016 Hi, 1. How do you determine the value of K. As far as I know, in chess they reduce the k value gradually for some reason that is inapplicable in football. Should K be valued based on a certain ratio between K and the base rating? I've tried to predict every single result for 60+ leagues based on elo, and adjusted my K in order to increase the prediction accuracy. K also scales with the goal difference of each game. My K is 25 and it scales according to the following table: GD K Scaling 0 2,000 1 1,000 2 1,500 3 2,308 4 2,462 5 2,615 6 2,769 7 2,923 8 3,077 9 3,231 10 3,385 11 3,538 12 3,692 13 3,846 14 4,000 2. Elo rating calculates probability of gaining/losing points, aka win/lose, did anyone try to modify it to accommodate draws as well? I did it. =MIN((1/(1+10^((Elo Away Team - Elo Home Team)/400))+Home Factor);1)^1,75 (PREDICTION FOR HOME WIN PROBABILITY) =MAX((1/(1+10^((Elo Home Team - Elo Away Team)/400))-Home Factor);0)^1,75 (PREDICTION FOR AWAY WIN PROBABILITY) Draw = 1 - HOME WIN PROB - AWAY WIN PROB With this formulas i had the following averages for 1 / X / 2 = 45,2% / 25,7% / 29,1% 3. Did you have a separate elo rating based on home/away results? No. But I have two Elos, one for national league / cup games and another for european league games. 4. Can elo rating system be applied on determining team to score first goal? Never tried it. For goals predictions I use the Adjusted Poisson formulas. Cheers, Quote Link to comment Share on other sites More sharing options...
Mouseycat Posted May 5, 2016 Author Share Posted May 5, 2016 On 5/4/2016, 5:31:41, Paul Ross said: Oh hang on, you mean the closer the rating the more chance of a draw? I would have to say yes, but draws are very hard to predict. I prefer to seek out good value in the odds. For example: If an away team was only -5.00 behind the home team, yet was 5/1 or more, I'd be keen on X2 and maybe even the full 5/1. Had a similar one in France earlier this season, Nantes at 5.75, they won 4-1 too I think! Nothing is the bible though, it can only be a guide. Hey! yeah was asking about it. I agree with your reasoning, even though it's really hard to find something like that unless you only consider the past 8 games like you mentioned. Thank you for your input! Quote Link to comment Share on other sites More sharing options...
Mouseycat Posted May 5, 2016 Author Share Posted May 5, 2016 17 hours ago, Pawn_pt said: Hi, 1. How do you determine the value of K. As far as I know, in chess they reduce the k value gradually for some reason that is inapplicable in football. Should K be valued based on a certain ratio between K and the base rating? I've tried to predict every single result for 60+ leagues based on elo, and adjusted my K in order to increase the prediction accuracy. K also scales with the goal difference of each game. My K is 25 and it scales according to the following table: GD K Scaling 0 2,000 1 1,000 2 1,500 3 2,308 4 2,462 5 2,615 6 2,769 7 2,923 8 3,077 9 3,231 10 3,385 11 3,538 12 3,692 13 3,846 14 4,000 2. Elo rating calculates probability of gaining/losing points, aka win/lose, did anyone try to modify it to accommodate draws as well? I did it. =MIN((1/(1+10^((Elo Away Team - Elo Home Team)/400))+Home Factor);1)^1,75 (PREDICTION FOR HOME WIN PROBABILITY) =MAX((1/(1+10^((Elo Home Team - Elo Away Team)/400))-Home Factor);0)^1,75 (PREDICTION FOR AWAY WIN PROBABILITY) Draw = 1 - HOME WIN PROB - AWAY WIN PROB With this formulas i had the following averages for 1 / X / 2 = 45,2% / 25,7% / 29,1% 3. Did you have a separate elo rating based on home/away results? No. But I have two Elos, one for national league / cup games and another for european league games. 4. Can elo rating system be applied on determining team to score first goal? Never tried it. For goals predictions I use the Adjusted Poisson formulas. Cheers, hey thanks for that load of information. Let me thank you first while I take some time to analyse this. (Y) Quote Link to comment Share on other sites More sharing options...
Mouseycat Posted May 5, 2016 Author Share Posted May 5, 2016 17 hours ago, Pawn_pt said: Hi, 1. How do you determine the value of K. As far as I know, in chess they reduce the k value gradually for some reason that is inapplicable in football. Should K be valued based on a certain ratio between K and the base rating? I've tried to predict every single result for 60+ leagues based on elo, and adjusted my K in order to increase the prediction accuracy. K also scales with the goal difference of each game. My K is 25 and it scales according to the following table: GD K Scaling 0 2,000 1 1,000 2 1,500 3 2,308 4 2,462 5 2,615 6 2,769 7 2,923 8 3,077 9 3,231 10 3,385 11 3,538 12 3,692 13 3,846 14 4,000 2. Elo rating calculates probability of gaining/losing points, aka win/lose, did anyone try to modify it to accommodate draws as well? I did it. =MIN((1/(1+10^((Elo Away Team - Elo Home Team)/400))+Home Factor);1)^1,75 (PREDICTION FOR HOME WIN PROBABILITY) =MAX((1/(1+10^((Elo Home Team - Elo Away Team)/400))-Home Factor);0)^1,75 (PREDICTION FOR AWAY WIN PROBABILITY) Draw = 1 - HOME WIN PROB - AWAY WIN PROB With this formulas i had the following averages for 1 / X / 2 = 45,2% / 25,7% / 29,1% 3. Did you have a separate elo rating based on home/away results? No. But I have two Elos, one for national league / cup games and another for european league games. 4. Can elo rating system be applied on determining team to score first goal? Never tried it. For goals predictions I use the Adjusted Poisson formulas. Cheers, Hey I didn't get the bold portions of your content. Isn't (1/(1+10^((Elo Away Team - Elo Home Team)/400))+Home Factor);1) the formula for home win probability? Is 1,75 supposed to be 1.75 or something? Thanks! Quote Link to comment Share on other sites More sharing options...
Pawn_pt Posted May 5, 2016 Share Posted May 5, 2016 The ^ 1,75 is a way to reduce the Win / Loss probability achieved from the regular Elo Formula in order to also extract a Draw probability. Quote Link to comment Share on other sites More sharing options...
Mouseycat Posted May 5, 2016 Author Share Posted May 5, 2016 41 minutes ago, Pawn_pt said: The ^ 1,75 is a way to reduce the Win / Loss probability achieved from the regular Elo Formula in order to also extract a Draw probability. The excel formula doesn't work though. Could you help to run me through this again. =MIN((1/(1+10^((Elo Away Team - Elo Home Team)/400))+Home Factor);1)^1,75 (PREDICTION FOR HOME WIN PROBABILITY) 1. What is the purpose of the min and max values here? 2. If (1/(1+10^((Elo Away Team - Elo Home Team)/400) computes the probability of home win, is the value similar to the (PREDICTION FOR HOME WIN PROBABILITY) ? If not, how do we compute (PREDICTION FOR HOME WIN PROBABILITY) ? 3. Is this middle part );1)^1,75 supposed to fit in an excel formula, don't really understand this part As far I know, ^ is the power of something, so I am not really sure about this. Sorry for so many questions once again. Quote Link to comment Share on other sites More sharing options...
Pawn_pt Posted May 6, 2016 Share Posted May 6, 2016 How does it not work ? Please see the attached file. Book1.xlsx Quote Link to comment Share on other sites More sharing options...
Mouseycat Posted May 6, 2016 Author Share Posted May 6, 2016 1 hour ago, Pawn_pt said: How does it not work ? Please see the attached file. Book1.xlsx hey it works now. thanks a lot! this will help greatly in the developing of my work. Quote Link to comment Share on other sites More sharing options...
Pawn_pt Posted May 6, 2016 Share Posted May 6, 2016 You can check some formulas and hipothesis about elo used to predict soccer results in a website named clubelo. Google it and read his articles / explanations. It helped me out! Quote Link to comment Share on other sites More sharing options...
Mouseycat Posted May 6, 2016 Author Share Posted May 6, 2016 1 hour ago, Pawn_pt said: You can check some formulas and hipothesis about elo used to predict soccer results in a website named clubelo. Google it and read his articles / explanations. It helped me out! Quote Link to comment Share on other sites More sharing options...
world Posted May 6, 2016 Share Posted May 6, 2016 1 hour ago, Pawn_pt said: You can check some formulas and hipothesis about elo used to predict soccer results in a website named clubelo. Google it and read his articles / explanations. It helped me out! Hi, I've never used elo ratings before but am planning to. For example if I want to use the ratings from here: http://sinceawin.com/data/elo/league/div/e0 and use your formula/excel how do I do that? What's the Elo Base there (you've 2000), how to determine Home Factor (you've 5%) for certain league? Thanks a lot! Quote Link to comment Share on other sites More sharing options...
Pawn_pt Posted May 6, 2016 Share Posted May 6, 2016 I do not have the answer to your questions! I've used 2.000 as using 400 was showing a Winning % of 99% and i wanted to show some more common and acceptable % for 1 / X / 2. I've studied the home win percentage for each league, and I can say the home factor is different according to each league. On the last 5 years i have the following average % for each result: DIV 1 X 2 ESP1 48,5 28,4 23,1 GER1 46,3 29,5 24,2 ENG1 46,0 31,2 22,8 ITA1 46,0 30,3 23,7 PRT1 44,3 30,3 25,4 RUS1 44,3 30,7 25,0 NED1 47,8 29,7 22,4 BRA1 49,7 33,5 16,8 BEL1 47,5 30,3 22,3 FRA1 44,9 32,8 22,3 UKR1 44,1 28,8 27,2 CZE1 49,2 30,9 19,9 GRE1 50,2 32,2 17,6 TUR1 45,9 30,7 23,4 AUT1 42,8 31,5 25,6 NOR1 48,8 29,8 21,5 SUI1 44,2 29,7 26,2 DEN1 41,4 31,0 27,5 SWE1 45,5 30,9 23,6 BLR1 44,5 31,6 23,9 ROM1 45,0 31,0 24,0 HUN1 46,9 30,2 23,0 ISL1 42,3 28,1 29,5 NIR1 41,9 26,3 31,8 FIN1 45,8 31,3 22,9 IRL1 41,3 29,0 29,7 POL1 45,9 31,7 22,4 SVK1 48,6 34,0 17,5 EST1 42,0 24,3 33,7 BUL1 50,2 28,3 21,5 SRB1 47,0 30,3 22,7 SCO1 41,8 28,6 29,6 LTU1 42,9 25,7 31,4 CYP1 43,8 28,9 27,3 SVN1 44,2 29,2 26,6 WAL1 44,4 27,8 27,9 MOL1 44,3 27,9 27,8 CRO1 48,5 31,3 20,2 GER2 43,1 32,2 24,7 ENG2 42,9 30,6 26,5 ESP2 46,4 31,2 22,4 ITA2 42,0 34,6 23,4 FRA2 45,1 35,5 19,4 BEL2 43,7 28,8 27,5 NED2 46,3 28,5 25,2 SWE2 44,6 29,9 25,5 BRA2 51,2 30,5 18,3 AUT2 45,0 28,8 26,3 PRT2 45,7 35,0 19,4 RUS2 44,8 31,8 23,4 POL2 42,8 31,3 25,9 UKR2 47,7 29,1 23,3 FIN2 45,1 28,3 26,7 CZE2 49,3 29,8 20,9 NOR2 47,0 28,0 25,0 SUI2 42,6 28,1 29,4 DEN2 42,4 30,0 27,6 SCO2 41,2 29,8 29,0 GER3 44,8 32,4 22,8 ENG3 42,7 30,3 27,0 SCO3 44,4 27,2 28,4 ENG4 41,2 31,5 27,3 ENG5 43,7 30,7 25,6 Considering that the result 1 happens in average on 45% of the matches, the home factor in Greece, Brazil and Bulgaria is higher than for example in Denmark or Ireland. In my model, I am currently using a general home factor of 2,75% and a elo base of 2.250. But I've tried many other values and picked the ones I believe worked better for me. I do not have all the matches of each team, and its quite hard to build such model (in excel) with all the games of the last for example 50 years. For that reason, i've designed my model to reset elos yearly, in the beginning of each year, based on the last 5y performances weighted by time. I can match the booking house odds, but to beat them... is another story And unfortunately, that's hard as hell, at least for me Cheers, Quote Link to comment Share on other sites More sharing options...
matrix1974 Posted February 27, 2018 Share Posted February 27, 2018 Hi All, The whole topic is here https://en.wikipedia.org/wiki/World_Football_Elo_Ratings This formula is designed to produce We value. We = 1/(10^(-dr/400) + 1) But why does not it produce red values there What do you say in this regard ? Examples The same examples have been used on the FIFA World Rankings for a fair comparison. Some actual examples should help to make the methods of calculation clear. In this instance it is assumed that three teams of different strengths are involved in a small friendly tournament on neutral territory. Before the tournament the three teams have the following point totals. Team Points A 630 B 500 C 480 Thus, team A is by some distance the highest ranked of the three: The following table shows the points allocations based on three possible outcomes of the match between the strongest team A, and the somewhat weaker team B: Example 1 Team A versus Team B (Team A stronger than Team B) Team A Team B Team A Team B Team A Team B Score 3 : 1 1 : 3 2 : 2 K{\displaystyle K} 20 20 20 20 20 20 G{\displaystyle G} 1.5 1.5 1.5 1.5 1 1 W{\displaystyle W} 1 0 0 1 0.5 0.5 We{\displaystyle W_{e}} 0.679 0.321 0.679 0.321 0.679 0.321 Total (P) +9.63 -9.63 -20.37 +20.37 -3.58 +3.58 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.