Giraldi system - mainly based on ELO

[giraldi]

Last season I started here a topic for my system. After 3 very good month, 1 month pretty weak, I lost 15 (fifteen) games in a row and I (very wrong) quit. To lost 15 is a performance. I dare everybody to repeat this. It is obvious that was not system fault – just hazard. Before this season start I would like to present on detail my new system. I have to say that all is automated, all the coefficients and parameters may be adjusted and even test all the system against past 2-3 seasons. Also try to see the essential ideas and not my poor English Let-s say we have TeamA vs TeamB Our objective is to approximate the probability for the following events: PwA = Probability for TeamA to win the game PwB = Probability for TeamB to win the game Pd = Probability for the draw event Po = Probability to be scored more than 2 goals (over 2.5) Pu = Probability to be scored less than 3 goals (under 2.5) The story is the same for each probability so I will consider only PwA – the probability the game to be won by TeamA According to my system PwA=Ch*Ph+Cf*Pf+CePe Where Ph = Probability considering overall team’s results and value (let-s say historical value). Pf = Probability considering recent form of the teams. Pe = Probability considering recent events (missing players, …etc) All three probabilities are affected by Ch , Cf and Ce coefficients. For my system default values are Ch – 75% (0.75 ) and Cf – 25% ( 0.25 ). They can be adjusted as I wish Unfortunately I have not the power and resources to know or to calculate Pe so I will consider Ce=0. In my opinion Ce is not bigger than 5% (0.05) but on the other hand can make the difference. The good news – we have the power to bet on the events we chose. So the easier way to avoid Pe (the bookmakers are to good here) is to bet on “flat games” . I call “ flat ” a game where are not missing players, not much injuries, overall nothing special happened 1. Probability considering overall team’s results and value - Ph In my opinion this is most important. Maybe you have noticed – many times good teams have long bad series of results (or maybe not as good as we expect) but the odds are still very low. I agree with that because Real is still Real even though they lost last 2 games and so on. In my opinion the recent form or another punctual events are just particularity of the team value. In order to find the current overall value of a specific team, I gave, starting with the season 2004/2005, 2000 points for each first league team, 1800 for each second league team …and so on. After each game I recalculate the ELO coefficient taking into account the team ELO, the opponent ELO, goals difference and of course who is the home team. I have used a classic formula for football to do that. Finally, at this time every team has an ELO coefficient. Of course there are many variation over time, depending to results, ..but overall the current ELO reflects very good team position comparing with the other teams. Now is very easy: If TeamA has ELOA and TeamB has ELOB. I just count all the games (last 5 years) where home team had ELOA plus and minus an ELO coefficient and away team had ELOB plus and minus an ELO coefficient. The current ELO coefficient is 50 (may be adjusted). In this moment we have an answer for the question: How many games were won by the home team with ELOA against away teams with ELOB? Of course could be a problem if the number of games is too small to be relevant. Current I consider this minimum number 25 games (may be adjusted). If the number of games is smaller I will consider Ph =0 and all the probability is given by Pf . At this moment the average number of games is 70-100. There are games with more than 200 similar games in the past and in my opinion this is very relevant. On the other hand you will be surprised how close is this probability (in the case of many games) to bookmakers probability So at this moment we have Ph., as I said affected with Ch (0.75) coefficient

1. Probability considering recent form of the teams – Pf

Please allow me to be short, first of all to keep a little bit from my system only for me. Just in case. J Recent form, Pf is calculated based on the last 10 games (adjustable). I consider first the average number of goals scored by the home teams and the away teams per game on last 2 seasons (adjustable) Than I calculate how much has been better the home team attack/defence on the last 10 games comparing with the average league team. The same way for the away team. Considering ELO coefficients I calculate how much scored / received goals a team with similar ELO coefficients comparing with average league team. Now I have a picture, before the game, what is the recent form for the attack and defense for both team and the number of goals to be scored by each team. This is enough to calculate the probabilities using Poisson formula. Finally I use the formula PwA=Ch*Ph+Cf*Pf+CePe And I have final probability for PwA The same way for the other probabilities Of course this is just a game inside another game. It is very hard to beat bookmakers but I am sure that a perfect system exists somewhere. Even with this perfect system the chances to became rich are very low because we can not afford to play all the events at disposal in order to get advantage as THEY do with us

[zbrochu]

Re: Giraldi system - mainly based on ELO Hi Giraldi, I am very glad to see you are back. I really like your approach and If I were not 110% busy with my own stuff I would love to dig into this nad maybe have some ideas how to help you. If you have any question, you have my email I will be happy to help. As far as ratings are concerned, one word of caution. It is very tricky to evaluate newly promoted and relegated teams. You assumed premier league teams has 2000 starting points and teams from one division lower have 1800. It sound reasonable but it would be good to do some research, try different values to see what is best number for this. You can also for safety just ignore promoted teams in your ratings or just observe how what are you predictions for games involving these teams. If you for example find that you hava value bet on them frequenty or very rarely, it means you either overvalue them or undervalue them. In order to have best chance at testing your method, I would recommend many low-margin bookmakers like PINNACLE, SBO, etc. It is hard enough to beat the market using this arsenal of portfolio and I think it is almost impossible to beat big leagues in big sports using one bookmaker's account. If you have 55% hit rate, this means it is very good results because it would mean +8% yield with portfolio of respectable bookmakers I mentioned, while using only one bookmaker with higher margin like Bet365, would change your yield to +5% and variance would be much bigger and streaks of 15 lost units would be more likely and this could be unneccessary disouraging for you. Good luck!