allen29 Posted December 28, 2015 Share Posted December 28, 2015 (edited) Hello. I'm really excited to finally start my own thread after working on developing my betting model. I aim to constantly post my predictions and picks. From time to time, I will also monitor the accuracy of my picks compared to the market So, here are some brief details about my model and picks. About the model(s) Why logistic picks? Because the model that generates the probabilities for the outcomes is a multinomial logistic regression. It combines Elo ratings with with my own ratings developed with data from whoscored and data about injuries and suspensions. The sample used was last seasons from 2009 to 2015 for all covered leagues. There are two models that I use: the main difference between them is that the second one uses data about injuries and suspensions from whoscored. Because of that, the "Preview" page of the game needs to be available before generating predictions. Therefore, any prediction that is more than two days into the future is generated without injuries and suspensions data and from my tests they have lower accuracy (as measured by AUC metric). Bet selection I will post my model predictions and my picks. I will try not to have a rigid approach, say bet when the odds value is above X, I will also study a bit the games before placing bets. From my cross-validation tests, my optimal results where when probability difference between was at least 4-5% or the expected value of the bet was at least 10-15%. Usually, the second is my threshold, but I could make some exceptions from time to time. Leagues covered For now, I cover the top 5 European leagues and Championship. In the future I would like to add more leagues though. Odds The market odds are from Pinnacle as they are very easy to get from their API. However, I will try to look for higher odds at other bookmakers before placing the bets So, these are my predictions for the upcoming fixtures. Also, my first bets. Staking plan Not 100% sure, in principle level stakes. I might do exceptions for long shot games as they pose the greatest threat to the bank. Edited December 28, 2015 by allen29 Quote Link to comment Share on other sites More sharing options...
jtw1 Posted December 28, 2015 Share Posted December 28, 2015 Good luck Quote Link to comment Share on other sites More sharing options...
allen29 Posted December 28, 2015 Author Share Posted December 28, 2015 Thank you! The preview page is now available for Spain and some Championship games. I added a new column that specifies is the probabilities were calculated with or without the injuries and suspensions data. Here are some updates. An interesting thing that I noticed is that odds tend to drop on events that have "high value". For example, Valencia to beat Villareal was 5.14 2 two days ago and now it is at aprox. 4.6. I think it is key to bet some time in advance if the odds are really good. One thing that I plan for the future is to collect data about moving odds and see if I can predict their movement few days in advance. The difference between my prediction and the bookmaker would be one interesting variable I think. Anyway, here are my updated predictions. I've been somewhat cautious with long-shots and went for DC or handicap instead. Quote Link to comment Share on other sites More sharing options...
allen29 Posted December 28, 2015 Author Share Posted December 28, 2015 Short update: The bet type from Sunderland - Liverpool is wrong. It is a home bet, not an away bet. Quote Link to comment Share on other sites More sharing options...
evil_inside Posted December 28, 2015 Share Posted December 28, 2015 Could you elaborate on the features you used and on the performance of your model on the test data ? Thanks. Quote Link to comment Share on other sites More sharing options...
allen29 Posted December 29, 2015 Author Share Posted December 29, 2015 (edited) I can talk briefly about the model, but in case it will perform well I would like the knowledge to stay in-house. One argument is Elo difference. The other arguments are what I call Strengths for each teams. They are computed using statistics from whoscored. The stats that are included in this index were determined after assessing their predictive performance. They are weighted accordingly to their predictive ability. For the second model the strengths are adjusted based on the contribution of players that are expected to be on the pitch towards team productivity (goals,shots, key passes) and average minutes played by starting 11. There is also a dummy variable for a particular league. (can you guess which one?). The models for championship were built using only data from championship as this league is not comparable with the big 5 IMO. I have less confidence in it because it was built on about 10% of the data the other two models covering big 5 were built. The results looked promising, so I decided to give it a go. The models will be constantly updating with fixtures that have been played this season and for which I have the stats needed. Here are some histograms with the results of both models broken down by: Overall (H+A), Home and Away Bets. There were no odds restrictions. 200 simulations with train: 75% of the data, test - 25%. Method used was repeated cross-validation. The rule for placing the bet was: Expected Value >= 0.15. Of course you can simulate thousands of times by trying out different thresholds and minimum odds combinations but I think this just increases complexity. Odds used were maximum odds provided by football-data and were downgraded by subtracting 1.5% from their underlying probability. I think this is at least similar to what Pinnacle offers, if not to cautious because I like to hunt days in advance for best odds. AUC for home wins (left) and away wins (right). This is relative because I used odds from football-data, I don't know if they are closing odds or not. I suspect closing odds to be a bit more accurate. Edited December 29, 2015 by allen29 Quote Link to comment Share on other sites More sharing options...
evil_inside Posted December 29, 2015 Share Posted December 29, 2015 I am currently using Python and scikit-learn for my systems. I have also tried Weka, Orange and Rapid Miner. XGBoost and deep learning seem an interesting next step. I believe the odds in Football-data are something in between the starting and the closing odds. Although, I don't think odds move so much anyway. Oddsportal has a very good archive of odds. How about the fluctuations of the capital ? Quote Link to comment Share on other sites More sharing options...
allen29 Posted December 30, 2015 Author Share Posted December 30, 2015 Update: I think I will only post my picks from now on as all predictions cover to much games and it would make huge posts. @evil_inside: I haven't looked that much into fluctuations of capital and money management. Will do that when I have more time at my disposal, I will try to see how I can minimize the losses. There were some fluctuations, I think largely from either long streak of longshot losses or to frequent wins which could spike the returns. I will try to avoid bets with probability less than 10-15%, probably will go for DC or handicap instead. Quote Link to comment Share on other sites More sharing options...
allen29 Posted December 30, 2015 Author Share Posted December 30, 2015 Forgot this one. Quote Link to comment Share on other sites More sharing options...
thecurlyone1 Posted December 30, 2015 Share Posted December 30, 2015 What's the bet ?? Quote Link to comment Share on other sites More sharing options...
allen29 Posted December 31, 2015 Author Share Posted December 31, 2015 It is a bet on Malaga to win Away, my mistake. Today was a terrible day, let's see if Garry Neville can help us tomorrow. Brighton - Wolves was removed as there was only 3% value after preview page was up. Solly March is injured and seems to be an important player. Quote Link to comment Share on other sites More sharing options...
allen29 Posted December 31, 2015 Author Share Posted December 31, 2015 Sorry, my mistake, I meant away bet on Celta. Celta to defeat Malaga. Sorry, I need more sleep. Quote Link to comment Share on other sites More sharing options...
Xtc12 Posted January 1, 2016 Share Posted January 1, 2016 Hi Allen29 have you backtested this in anyway and if so what were the results. Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 1, 2016 Author Share Posted January 1, 2016 Another loss yesterday. Can it get any worse than this? Surely it can, but let's see. @Xtc12: Yes, I did, I have a post on that with the results on 29th of Dec. Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 2, 2016 Author Share Posted January 2, 2016 Again, not the best day. Bets against Liverpool are my favorites. As a Liverpool fan, if they win, I'm happy. If they loose, at least I have consolation in winning my bet. Dissapointed by Leicester result, with the missed penalty. Celta in serious drop of form, how the hell did they managed to be 4th in the league? Their odds went up to 3.94 at kick-off, no idea why, no major changes in their starting 11. Nolito was a known absence. Quote Link to comment Share on other sites More sharing options...
Matthew Posted January 4, 2016 Share Posted January 4, 2016 With regard to your interest in dropping odds, and I can;t find the post where I originally made this available but you can find a slug of data for hourly obtained odds in via this web link: www.soccersystembuilder.com/hr_by_hr_odds.zip On 28/12/2015, 15:13:29, allen29 said: Thank you! The preview page is now available for Spain and some Championship games. I added a new column that specifies is the probabilities were calculated with or without the injuries and suspensions data. Here are some updates. An interesting thing that I noticed is that odds tend to drop on events that have "high value". For example, Valencia to beat Villareal was 5.14 2 two days ago and now it is at aprox. 4.6. I think it is key to bet some time in advance if the odds are really good. One thing that I plan for the future is to collect data about moving odds and see if I can predict their movement few days in advance. The difference between my prediction and the bookmaker would be one interesting variable I think. Anyway, here are my updated predictions. I've been somewhat cautious with long-shots and went for DC or handicap instead. Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 4, 2016 Author Share Posted January 4, 2016 A dreadful start. So far you would have gotten nice profit by betting against my picks. Regression to the mean, where are you hiding? Quote Link to comment Share on other sites More sharing options...
Rabadzija Posted January 4, 2016 Share Posted January 4, 2016 I already betting ageinst your picks.Model is perfect.Keep going. allen29 1 Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 4, 2016 Author Share Posted January 4, 2016 That's great, @Rabadzija, I have more suggestions for you then. Doria, Inter and Roma, all away wins. Rabadzija 1 Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 6, 2016 Author Share Posted January 6, 2016 Genoa close to coming back from 0-3 at half time, losing "derby della Lanterna" in style and helping me continue my miserable loss streak. Decided to bet early for games these weekend as I see some interesting odds. Possibly 1-2 more coming on Thursday or Friday. Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 7, 2016 Author Share Posted January 7, 2016 Oh, boy that makes it 7 consecutive losses. I'm on a roll here! Nice to see the odds of Chievo vs a Roma plagued by suspensions and injuries drop to where my model predicted. Similar drop for Empoli vs Inter. Better to take a value bet and lose than not to take any.... Anyway, added more games for the weekend. let's see if the tides will turn Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 11, 2016 Author Share Posted January 11, 2016 Another poor performance this weekend. Bit unlucky when you consider that Lorient doped a 2 goal advantage. Currently Aways are causing the biggest Loss at -9.28 pts, DC and handicap second with -4.96 and Homes are profitable at 1.25. Anyway, mostly away bets on English football this midweek. Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 12, 2016 Author Share Posted January 12, 2016 Couldn't help it, odds looked to good, decided to add these as well. Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 12, 2016 Author Share Posted January 12, 2016 Meh, another disappointing day. Thanks to Payet I guess, defeated Bournemouth by himself. QPR close to upsetting Blackburn, conceded a late equalizer in 85th min...In other terms, what a shocker, Aston Villa actually won a game (first home win this season) - guess that had to happen at some point. 16.67% (4/24) hit rate on Away bets at average odds of 3.69 (expectation of 27% - expected 6.5 wins from the bookmakers, 31.8 % from my model - expected 7.6 wins). Either unlucky either large home field advantage is quickly coming strong into fashion again. Probably will limit bets in Championship and look to tweak the model in the following weeks, I think the number of recommendation is way to high in this league. Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 13, 2016 Author Share Posted January 13, 2016 Oh, welsh Xavi, you legend! Interesting to see that Arsenal odds went up to 2.96 at game time. Liverpool did line up a better team than expected, without having to play Lucas in defense, but still with so many players coming back from injuries I think the market overreacted to this one. Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 15, 2016 Author Share Posted January 15, 2016 Selections for this weekend: Quote Link to comment Share on other sites More sharing options...
Matthew Posted January 15, 2016 Share Posted January 15, 2016 Good luck. It'll turn around. Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 15, 2016 Author Share Posted January 15, 2016 Thanks Matthew, it's hard to do worse I guess. Some additional selections: Southampton - West Brom X2, 2.31(Pinnacle) , 2.05(Model) , X2 Atalanta - Inter, 4.16(Pinnacle), 3.50(Model), H Napoli - Sassuolo, , 1.79 (Pinnacle), NA(Model), 2AH + 1.75 Quote Link to comment Share on other sites More sharing options...
allen29 Posted January 16, 2016 Author Share Posted January 16, 2016 Added games from last post + 3 more "gems" found today. alani42 1 Quote Link to comment Share on other sites More sharing options...
Matthew Posted January 24, 2016 Share Posted January 24, 2016 Please tell me you haven't given up. I know it's hard when things don't work out as you'd hoped but I think it'll come right eventually, you just need time. I'd bet that you'd find a 66 game sample in your backtesting where things went just as badly. 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.