AndrewKent Posted January 23, 2011 Author Share Posted January 23, 2011 Re: Regression Graphs Don't worry guys, I'm on it. I would expect to have something to share in the next day or so. Quote Link to comment Share on other sites More sharing options...
Machine Posted January 23, 2011 Share Posted January 23, 2011 Re: Regression Graphs OK. Lets wait for AKent results. The above methods shown are taught to students who are studying computer science and statistical analysis at M.I.T Americas leading statistical research facility by Prof John Guttag and Prof Eric Grimson both withover 30 years of teaching computer science. (I have taken the course myself,its free you should try it) So instead of being so negative please enlighten us on your theory as Im sure Mulkis of the punters lounge is better placed to sort out such statistical problems. I wait your expert reply :ok Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted January 23, 2011 Author Share Posted January 23, 2011 Re: Regression Graphs Okay, thanks guys, I don't want to have a difference of opinions ruining this thread, I'm getting a lot out of it so far so can we keep things on topic? Mulkis, I appreciate your input however Monkeys Nest has backed up what he's saying with examples both in the form of his posts and his links (which I looked through today). If you believe this is wrong, could you go into the detail he has to back your point up? I would be interested to hear your views. Failing that, we're going with MN because he's given me direction, and for that I'm in your debt mate! I'll post up my findings asap... Quote Link to comment Share on other sites More sharing options...
Machine Posted January 23, 2011 Share Posted January 23, 2011 Re: Regression Graphs All the best with your trials :ok Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted January 23, 2011 Author Share Posted January 23, 2011 Quote Link to comment Share on other sites More sharing options...
Machine Posted January 24, 2011 Share Posted January 24, 2011 Re: Regression Graphs Hi andy Your primary aim is to replicate what actually happens with real life data then if similar/same results are achieved while testing fresh data you know you are on the right track and your predictive model is working how it should do. Q. are you using all your ratings or the product of 1 single rating for the above sample Quote Link to comment Share on other sites More sharing options...
ELO Evangelist Posted January 24, 2011 Share Posted January 24, 2011 Re: Regression Graphs Andrew, something else worth considering is that Paul Steele will have used a data set to formulate his 'Power Rating' model. If you are using this same data set to see if his model is profitable you are Texas sharp shooting. You need to find out when his book was written and only use data after this date. Quote Link to comment Share on other sites More sharing options...
Mulkis Posted January 24, 2011 Share Posted January 24, 2011 Re: Regression Graphs Okay' date=' so this is some VERY quick analysis on 35K games, the combination of Power Ratings, Game Form and Penetration Plus generated these bets...[/quote'] Nice. But your analysis missing one collumn - ROI% Sorry, for being so sceptical, but I've went through something simillar, i.e. analysing historical data from different angles and methods. It took a lot of time, a lot of hopes and false flags, but no positive results in longer run. I've also made exactly the same analysis as Monkey reccomends and backtesting with "winners" by odds range gave overal negative results. But I still looking forward for Andy's analysis, maybe I've done something wrong in my testing. Quote Link to comment Share on other sites More sharing options...
Machine Posted January 24, 2011 Share Posted January 24, 2011 Re: Regression Graphs Checking this way quickly, I'm seeing that I'm moving away from the red and starting to break even. If I take more time over this, I will start getting something worth while. Andy Getting within break even range is a sign that your model is heading in the right direction.(starting to hit the numbers in the right proportions) One way to test is to use only 1 bookmaker as your datum odds range ( I would use the one which constantly gives the worst average odds) then running a comparison against the better odds which were available (if any) and then test on clean data to see if it moves you into that elusive profit range. Quote Link to comment Share on other sites More sharing options...
Machine Posted January 24, 2011 Share Posted January 24, 2011 Re: Regression Graphs Nice. But your analysis missing one collumn - ROI% Sorry, for being so sceptical, but I've went through something simillar, i.e. analysing historical data from different angles and methods. It took a lot of time, a lot of hopes and false flags, but no positive results in longer run. I've also made exactly the same analysis as Monkey reccomends and backtesting with "winners" by odds range gave overal negative results. But I still looking forward for Andy's analysis, maybe I've done something wrong in my testing. Mulkis The first stage is to get your model replicating true events regardless of ROI if you can do this within 1-2% constantly then you have a workable model which is already reducing the books overound by approx 8-10%. If your model is poor your results will be poor. When I did a model the best I could get was + 2% which wasnt enough for me as my horseracing model was out performing it by quite a considerable amount so I ditched the football one and concentrated with my horseracing one. Andy seems to be getting his model in shape and he may see something that we haven't only time will tell . If you dont buy a ticket you cant win the raffle :ok Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted January 24, 2011 Author Share Posted January 24, 2011 Re: Regression Graphs Right guys, there are so many great ideas floating around in this thread I thought I better organise my thoughts a bit. Firstly. My system that comes up with the ratings is based on the three top rating systems from Paul Steele's book, "Profitable Football Betting". These are, Power Ratings (performance of a team over a season) Game Form (performance of a team in the last 5 matches) and Penetration Plus (a rating based on how well a team forces shots and corners). A match must meet a certain level in all three criteria before I consider the match for a bet. And it's also worth noting all my data comes from www.Football-Data.co.uk. Now, I think what I should provide for discussion next is two things. Firstly, a breakdown of odds. Using Monkey Nests approach, I will take five out of the ten seasons of analysis that I have, and establish the success rate based on odds. I'll probably use William Hill or Interwetten as most games after 2002 seem to have odds for these companies within my data. Secondly, I will test this out on three unrelated seasons (my clean data) using two approaches: Firstly, a flat staking system and secondly, an escalated staking system which I briefly looked at last night and could prove to give me that little extra if my system shows it has longer winning streaks. The other factor I want to look at in the same way as the odds is the leagues. Where I am going to look at what odds give a good success rate, I want to know what leagues the system performs best in. For example, this season, I can see laying the away team gives me 98% success rate in the conference, but a much lower success rate in the championship and bundesliga. Perhaps tailoring the successful odds to the league and thus tweaking the system to match the league might offer some more insights. Okay guys, almost home now. I'll post up some stats once I have done anlittle more work and let you see what I find. If this works out, I intend to start a thread for my picks, so those of you interested in that, hang in there, I'm getting round to it :-) All the best, Andy Quote Link to comment Share on other sites More sharing options...
Mulkis Posted January 24, 2011 Share Posted January 24, 2011 Re: Regression Graphs Sorry, maybe I missed something, but it is not clear for me on what specific criteria you made your bets? Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted January 24, 2011 Author Share Posted January 24, 2011 Re: Regression Graphs I said in the previous post that the criteria is based on the combination of three systems. To go into these three systems at this stage would detract away from the current discussion, which is what to do with the ratings after we've made them. I will happily go into detail about the systems if everyone reading this thread wants to but I would best start it in a new thread? Andy Quote Link to comment Share on other sites More sharing options...
Kanga Posted January 24, 2011 Share Posted January 24, 2011 Re: Regression Graphs Andy, Your table means very little in reality until you compare them against actual bookies odds, only that way can you see if your efforts have been profitable. I'd suggest checking the profitability of each of your ratings separately, and then in various combinations e.g where system 1 and 3 both pick the same outcome, this should give you 6 columns of interest. Quote Link to comment Share on other sites More sharing options...
TheLubet Posted January 24, 2011 Share Posted January 24, 2011 Re: Regression Graphs I'm just going to add something here as your using football-data.co.uk. The Odds on that site even though extensive are more often than not incorrect. So if you are using their odds as a benchmark for backtesting then beware! Also there have been the odd incorrect result posted. The odds posted are the ones used before the off, not the set odds when the book is calculated! The Lubet. Quote Link to comment Share on other sites More sharing options...
Datapunter Posted January 25, 2011 Share Posted January 25, 2011 Re: Regression Graphs I wouldn't call them incorrect but it is true you need to be carefull drawing conclusions. The available historic odds are a snapshot in time. They provide you with a basis to work on but no more than that. It's not because certain odds where available at some point in time pre-match that you as an individual would have been able to get those odds. Any backtesting must be followed by a real-life trial placing actual bets usually with minimal stakes. Note i'm not a believer in paper trading but maybe that's just me. Real life bets will tell you what odds you can get as an individual. The main difference between people is the timeframe. Someone placing bets early compared to someone who has the time and ability to wait until shortly before kickoff will get a different odds set and it may actually be the difference between profit or loss. Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted January 25, 2011 Author Share Posted January 25, 2011 Re: Regression Graphs The main difference between people is the timeframe. Is there an optimal time to place a bet? For example, just before kickoff when there is no speculation about team line-ups? What's your opinion? Andy Quote Link to comment Share on other sites More sharing options...
Datapunter Posted January 25, 2011 Share Posted January 25, 2011 Re: Regression Graphs The first question is when do you have time to place your bets ? Any optimising needs to occur inside that. Basically it works itself out after that. Very few people have the time to sit around monitoring markets and wait for the optimal prices to arrive. Pretty boring if nothing else. So simply look at what timeframe suits you and optimise based on that. If you can manage to monitor markets or you are around at the start of events then Betfair may be all you need. If you need to place bets some time before the events then get organised and get a dozen or so bookmaker accounts so you can shop around for decent prices. Especially if you want early prices Betfair simply does not have a lot of volume and the bookies are a better option. Quote Link to comment Share on other sites More sharing options...
Mulkis Posted January 26, 2011 Share Posted January 26, 2011 Re: Regression Graphs But I still see no decent system for Andy to place any bets at all... So, where the results? With ROI% please. Anything without the ROI% is empty discussioin about black matter - and Black Matter is invented by smart scientist when formulas do not sum up - to cover their failure:wall Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted January 26, 2011 Author Share Posted January 26, 2011 Re: Regression Graphs Sorry Mulkis I don't quite understand your post. At present I am in the "scoping" phase, and to that end, I am playing around with quite a few different ideas before settling on anything. This means I am still looking at different systems so I am no-where near giving "results" or ROI%. I can only imagine if I rush these things it'll be wrong and your fabled ROI% will be worth f*ck all. Could you explain what Black Matter is please? It sounds interesting. Andy Quote Link to comment Share on other sites More sharing options...
Mulkis Posted January 26, 2011 Share Posted January 26, 2011 Re: Regression Graphs Please consider my scepticism as encouragement for you to move forward with your endevour:cheers Quote Link to comment Share on other sites More sharing options...
Matthew Posted January 27, 2011 Share Posted January 27, 2011 Re: Regression Graphs Hi, I recall a thread by Mr OneMore that built up over time and drip fed the way to building a system. Laid down brick by brick. It is one of the most viewed threads ever on here and to this end I suspect Andrew's way of posting a progression of his system will be appreciated by many. Everyone has to learn somewhere. Some do it in a dark room until the early hours. Andrew is doing it in a more open forum. Imagine in 3 years time when we've all moved on and a 21yr old who has just figured that accumulators are worthless decides to get serious and finds Pual Steele's book. Yes he will complain that in the book Paul never measures the success of the 15 systems by p&l, but it will get him thinking about using Pen Plus, Power Ratings etc. Were that person to find this thread then they would probably save a lot of work - or at worst short cut to their way to a solution. Anyway - what I wanted to post and say is that it might be better to sample your data randomly and not season by season. The over-round in the odds available in the F-D data has dropped markedly from 2000-->2010. Good luck - following with interest. My tip is that you might find more success adapting the three systems you are using slightly. Every man and his dog has Paul Steele's ratings and the ability to calculate them. But only few have adapted them optimally. Quote Link to comment Share on other sites More sharing options...
AndrewKent Posted January 27, 2011 Author Share Posted January 27, 2011 Re: Regression Graphs Hi Matthew, Thanks for your post, I appreciate what you are saying and it is nice to see that people are "hitting the nail on the head" when it comes to spotting the purpose behind what I am doing. As it turns out, I'm enjoying the analysis side just as much as the betting (being a business analyst for a living certainly helps!) and considering I'm not looking at a get-rich-quick scheme, I'm happy to share what I learn on here. As long as it works for me, if it benefits everyone else along the way, all the better! I hear what you are all saying loud a clear: 1) Tweak/Optimize the systems 2) Use random test data to generate "ratings" and clean data to test on After re-reading sections of this book I'm actually finding that the three systems that Paul has picked were based on what picture they were painting about a game (long/short term form and ability to create chances). That doesn't make them the respective leaders in their individual fields. I had an idea that I wanted to run past you all. If I was to look at all 15 systems using the approach I mentioned in points 1 and 2, work out which were the top performers for each bet type (Back 1X2 or Lay1X2) and take it from there? If the moderators don't object, I might start a new thread devoted to this project (in the same way as Mr OneMore, which I read). Although I can build a the systems in Excel (as per the book) I might need help optimizing. I'm also prepared to share any work on Excel I do in return for useful input. What do you think? Andy Quote Link to comment Share on other sites More sharing options...
plkd Posted June 28, 2012 Share Posted June 28, 2012 Re: Regression Graphs I used LINEST funtion in Excel as follow =Linest(known_y, known_x, const, stats) with known_y: Home array {1,0,0...0,0,0} known_x: Power rating volume {4.25, 4.00...-3.00, -3.25} const: true stats: true ... and result are as below. Please tell me how to create your regression y = 0.1491x + 0.4344 for home win and so on Thank you [TABLE=width: 96] [TR] [TD=width: 64, bgcolor: transparent, align: right]-0,88226[/TD] [TD=width: 64, bgcolor: transparent, align: right]208,4411[/TD] [/TR] [TR] [TD=bgcolor: transparent, align: right]25,8779[/TD] [TD=bgcolor: transparent, align: right]59,29372[/TD] [/TR] [TR] [TD=bgcolor: transparent, align: right]4,01E-05[/TD] [TD=bgcolor: transparent, align: right]322,1773[/TD] [/TR] [TR] [TD=bgcolor: transparent, align: right]0,001162[/TD] [TD=bgcolor: transparent, align: right]29[/TD] [/TR] [TR] [TD=bgcolor: transparent, align: right]120,6488[/TD] [TD=bgcolor: transparent, align: right]3010147[/TD] [/TR] [/TABLE] Quote Link to comment Share on other sites More sharing options...
Machine Posted June 28, 2012 Share Posted June 28, 2012 Re: Regression Graphs I used LINEST funtion in Excel as follow =Linest(known_y, known_x, const, stats) with known_y: Home array {1,0,0...0,0,0} known_x: Power rating volume {4.25, 4.00...-3.00, -3.25} const: true stats: true ... and result are as below. Please tell me how to create your regression y = 0.1491x + 0.4344 for home win and so on Thank you [TABLE=width: 96] [TR] [TD=width: 64, bgcolor: transparent, align: right]-0,88226[/TD] [TD=width: 64, bgcolor: transparent, align: right]208,4411[/TD] [/TR] [TR] [TD=bgcolor: transparent, align: right]25,8779[/TD] [TD=bgcolor: transparent, align: right]59,29372[/TD] [/TR] [TR] [TD=bgcolor: transparent, align: right]4,01E-05[/TD] [TD=bgcolor: transparent, align: right]322,1773[/TD] [/TR] [TR] [TD=bgcolor: transparent, align: right]0,001162[/TD] [TD=bgcolor: transparent, align: right]29[/TD] [/TR] [TR] [TD=bgcolor: transparent, align: right]120,6488[/TD] [TD=bgcolor: transparent, align: right]3010147[/TD] [/TR] [/TABLE] Use the unshaded data in the Power Ratings Distribution sheet in post 24# Highlight cells F3 and G3 (or any 2 cells you want your slope & intercept in) For Home equation type in = Linest(B2:B16,A2:A16) Hit CTRL+Shift+Enter TA-DA.......... theres your slope & intercept ;) EG :ok Quote Link to comment Share on other sites More sharing options...
Machine Posted June 28, 2012 Share Posted June 28, 2012 Re: Regression Graphs For the draw the line of best fit is a curve (polynomial to an order of 2) Giving a Linest function of LINEST(C2:C16,A2:A16^{1,2}) ..........NB select 3 cells before typing the function (F7:H7 in example) Giving the formula......... -0.02831x^2 + -0.01127 + 0.289619 NB The formula in excel comes out in reverse ie A2:A16^{1,2} will come out value^2 + Value + intercept and not value + value^2 + intercept I dont know how he comes to the formula for aways as the data is a linear trend Y = -0.1223x + 0.3059 :ok 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.