Crouch Potato Posted July 29, 2009 Share Posted July 29, 2009 I'm currently testing a draw system that I have devised for English Football. So far I have only research the EPL, but over the four seasons I've tested there are on average just 25 bets per season. What I want to know/am concerned about is: 1) How much will the small bets/season ratio affect my ability to have trust in my numbers and results 2) I believe that my system parameters work a little differently in each of the divisions. How much trust can I put into a combined set of data? 3) I expect that from season to season, my numbers will differ a fair bit, particularly from league to league. What can I learn from the fact that from season to season, particularly in each league, my numbers are somewhat eratic? My results so far are (flat stakes -1pt per bet): EPL 2008/09: 32 Bets/10 Wins. Profit +4.18 (Yield = +13%) EPL 2007/08: 32 Bets/16 Wins. Profit +24.44 (Yield = +76.3%) EPL 2006/07: 19 Bets/9 Wins. Profit +15.65 (Yield = +82.36%) EPL 2005/06: 18 Bets/2 Wins. Profit -11.66 (Yield = -64.7%) Overall: 101 Bets/37 Wins. Profit +32.61 (Yield = +32.3%) I know 101 is comparatively, a very small sample size, but as I said before, there are very few games each year which fall in my defined parameters. Anyway, I've heard a Chi test is a good barometer in terms of testing the validity of data. I've searched the forums and tried to understand it, but I was wondering if maybe somebody could do a preliminary Chi result with the data I have so far? I have attached a spreadsheet - I believe the relevant information is there. In column 1 I have the season, column 2 the league (I'm obviously going to be adding to the file). Column 3 are the draw odds, column 4 is the book % (based on HW/D/AW odds for the match). All HW/D/AW odds were obtained from the same bookie, and wherever possible I have used Pinnacle's odds. Column 5 is the Win column, with a 1 denoting a win. I'm sure somebody out there understands this Chi-test better than I do :hope In the meantime, I'll start backtesting the Championship. Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Hi. I've set up a chi test calculator in excel so will pump your results and odds in total and see what it gives. Also using the same data I can produce a simple t test analysis. I'll have a look this morning. Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Results show the following Winning bets. Act: 37. Pred: 25 Loosibg bets: Act: 64. Pred: 76 Chi sq rating: 0.004438 (means there is a 99.56% probabiity that your results were not due to random luck) t test yield: 32.29%. With a 3.45% chance of achieving a negative yield Quote Link to comment Share on other sites More sharing options...
Crouch Potato Posted July 29, 2009 Author Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Hmm, thanks for that mate. I must say though, that probability seems too high to me for what seems to be such a small sample size? Or is that something that the chi-test doesn't factor in (Most of this stuff is beyond me). Anyway, I'll try and get the Championship tests completed sometime this week and maybe I can re-evaluate from there. Cheers again. Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) I've also ran your data through a sample size calculator. Your winning % from 101 games was 36.63% this gives a confidence level of 9.08 (based on 95% accuracy) and 11.95 (based on 99% accuracy) basically you can be 95% sure that had you placed these bets on 1,520 games rather than 101 your winning percentage would fall between 27.55% and 45.71% and 99% sure that your winning percentage would be between 24.68% and 48.58% hope this helps with your analysis Quote Link to comment Share on other sites More sharing options...
Crouch Potato Posted July 29, 2009 Author Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Cheers mate. Hopefully I can get the remaining 3 leagues done for the respective 4 seasons. Assuming an increase in the number of bets to say 40, I imagine there will be another 400-500 games to add to the analysis. Might get you to run through those numbers once I have them. Cheers again. Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Yes no problem. If you require any help or proofing for back testing I have 6-7 seasons worth of data for various leagues. I can run a simple report if you tell me your criteria. I understand if you don't want to "go public" just yet so feel free to email me at [email protected] kev Quote Link to comment Share on other sites More sharing options...
muppet77 Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) kev and crouch, great work. i usually find that any system i explore crashes and burns when i test other leagues, so good luck with it. the results should be interesting. kev, how do you go the extra step with chi squared? i can work out using excel the chi sq rating using wins and losses, expected vs actual, but how did you calculate the yield and t test stuff etc? Quote Link to comment Share on other sites More sharing options...
Crouch Potato Posted July 29, 2009 Author Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Cheers Kev. I've just sent you an email. Muppet - I seem to recall a preliminary check I did had the system profitable for 07/08 and 08/09 in the other leagues, though 06/07 was a shocker. I'll have to get the numbers in full and see what it says. Would not expect this to work in non-English leagues though. Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Hi muppet. The t test data draws from the same "raw" data as the chi sq it basically calculates the yield % for each game (a win at 3.84 = 284% a loss = -100%) you then average the total which gives overall yield for the probability of achieving negative yield it's a bit more complicated. You need to use square root of total games. Then divide the standard deviation of the total individual yield calculation by the square root figure. You then need to run a normal distribution function over that % and the average yield % Quote Link to comment Share on other sites More sharing options...
muppet77 Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) thanks, any idea how i can apply it to spread bet results? Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) If you have the odds and a win ( y/n or 1/0) next to each bet email me and I can see if it works. I don't really dabble in spread betting so wouldn't be able to tell you without seeing and testing your data. Quote Link to comment Share on other sites More sharing options...
muppet77 Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) well there are no odds. it links to my thread on corners. the firm will quote a spread on the estimated number of match corners. eg 10-10.5 (always half a corner spread) i can then choose to sell (bet under) @ 10 or buy (bet over) @ 10.5 if i sell at 1 unit a corner and there are 9 corners i win (10-9)= 1 unit if i buy and there are 9 corners i lose (10.5-9) = 1.5 units does this help? the hardest part in applying chi is calculating the 'expectation'. Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Perhaps if you could list all your bets with profit/loss on each I could do this. Feel free to email so we do not hijack crouch's thread Quote Link to comment Share on other sites More sharing options...
Crouch Potato Posted July 29, 2009 Author Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) You guys are fine to keep discussing it in here. It's really just a thread about how to analyse statistical information. I'm actually quite enjoying seeing all the various input (even if I don't under 99% of it!) Quote Link to comment Share on other sites More sharing options...
Crouch Potato Posted July 29, 2009 Author Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) First Championship Season is done. I'll edit the other 3 seasons in when I've done them and then post up another spreadsheet for Chi-testing. First season yield is sadly below EPL average, though that's hardly surprising. CCC 2008/09: 28 Bets/10 Wins. Profit +6.1 (Yield = +21.7%) Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Still a positive yield mate. Do you have to physically sort through the data? My excel and vb skills are pretty decent so if you have the raw data I maybe able to analyse it to your chosen criteria a bit quicker and more automated? Quote Link to comment Share on other sites More sharing options...
Crouch Potato Posted July 29, 2009 Author Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) It's all manual - one of the posters on here was quite helpful in pointing me to the right kind of program though, so it's actually pretty good. A lot of my time is actually spent looking up odds etc. - something I intend on possibly using as part of a filter criteria I have in mind. If I gave you a list of matches, would you be able to import the Pinnacle odds? I use betexplorer to get my odds. Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) What are the odds you need? If it is 1,2,X I have these already (using bookie averages) Quote Link to comment Share on other sites More sharing options...
muppet77 Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) 21% yield isn't bad at all crouchy!! looking forward to the rest. there must be an automated method? if your criteria involves statistics then these can be worked out using a programme. kev - attached are my spread bets. do you need to know my estimate of the corners for each game? (i have not included them but can do if needed) Quote Link to comment Share on other sites More sharing options...
Crouch Potato Posted July 29, 2009 Author Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) What are the odds you need? If it is 1' date='2,X I have these already (using bookie averages)[/quote'] Yeah, it's just 1,2,X, but I like to use pinnacle odds at all times. This is because I intend on making all bets with Pinnacle (I did some research into last year's games only and found Pinnacle almost always had the best odds, including Betfair, if you account for commission). As for my criteria, it's not really statistics. What I use I suppose could be converted or construed as statistics, but it requires a lot of manual recognition. It's a lot of work right now, but I only intend on going back to 05/06, so it's not like I have to do much once I'm done with the rest of the English leagues. Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Hi muppet. No I wouldn't need to know that. It's gonna be a tricky one (must admit I'm a spread novice). Normal t tests show any loss as -100%, but obviously your loss amount can vary. I'll have a proper look at this tonight, but out of interest how did you calculate your chi test for this?(I'm assuming you have) Quote Link to comment Share on other sites More sharing options...
muppet77 Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) well i took the average number of corners to be 11. the bookies spread is his back up or overround. in each case it is half a corner. so 0.5/11 = 4.55% overround so should the bookie win 54.55% and me the rest? but then some games are 'tied' - if the spread is 9.5-10 and i go 'over' 10 corners (buy) and there are exactly 10, then my profit/loss is zero. i think that IF the spread includes a whole number, the exact amount will occur around 1 in 8. ??? Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Muppet, I've calculated your avg yield at 72.10% and your chance of achieving a negative yield at 1.56%. However, i've never used spread betting results with T testing before so am not sure how accurate they are. I basically divided your return / stake. Then calculated this as a percentage yield and ran it through the t tester (using the square rooting / standard deviation and distribution sums spoken about before) Quote Link to comment Share on other sites More sharing options...
muppet77 Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) I have calculated yield as (profits - losses) divided by (profits + absolute losses). For example three bets. Two wins of ten units. One loss of ten units. Yield is 20 over 30 or 67%. My actual yield was calculated as 30%. What do you think of this method? Quote Link to comment Share on other sites More sharing options...
KThom Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Your actual yield should be total profit / total amount staked. I needed to calculate an individual yield to enable me to "replicate" odds for the purpose of t testing. Whether this has been done correctly or not i'm not sure. Quote Link to comment Share on other sites More sharing options...
slapdash Posted July 29, 2009 Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Chi sq rating: 0.004438 (means there is a 99.56% probabiity that your results were not due to random luck) Going a bit off-topic, but statistical tests like this don't tell you about the probability that your results are due to luck. They tell you about the probability that you would have done as well if they were due to luck. I know it sounds a bit pedantic, but they really are very different things: Suppose I had a "system" for predicting results which just consisted of tossing a coin (I'm not for a minute suggesting that Crouch Potato's system is anything like this, by the way!). After a decent number of games, it's quite possible that I'd have had reasonably good luck, and that a statistical test would tell me that there was only a 20% chance that I'd have done that well by luck alone. That doesn't mean that there's only a 20% chance that my results were due to luck: there's still pretty much a 100% chance that they were! Quote Link to comment Share on other sites More sharing options...
Crouch Potato Posted July 29, 2009 Author Share Posted July 29, 2009 Re: Data Analysis (Chi Square etc.) Suppose I had a "system" for predicting results which just consisted of tossing a coin (I'm not for a minute suggesting that Crouch Potato's system is anything like this, by the way!). I did try initially, but with the presence of the draw, I found that for it to be effective I had to use one of those three sided dice they use in dungeon and dragons :loon Jokes aside, your point is well made Slap-D. And in a very easy to understand example :ok That said, I suspect when I am done researching, from the beginning of 2005/06 season to now, I will only have about 400-500 games where my criteria was met. I don't really know anything about statistics or numbers, but it's my understanding that this is not a big sample size - a fact that concerns me in two ways: 1) It's hard to verify the strength of my system; and perhaps more importantly; 2) The limited number of bets that I am able to make each season increase the influence that chance has on my seasonal results. and 3) Yes, I know I said two ways, but I also said I am bad with numbers! 3) If I lack the volume of data to verify my system to the same level many other systems are ensured too, and I can't make many bets per season, how useful is such a system? Now, it should be noted at this point that this system is not a statistical one. I wouldn't even know where to begin with a statistical system, and couldn't even tell you what poisson distribution is or how it works. The point I'm trying to make is that my results aren't a result of a retro-fitted formula or best-fit equation. (There is no Post Hoc, Ergo Propter Hoc in play here!) Anyway, not having a clue about statistics and the like, I was wondering if what I just said somehow diluted or lessened the negative effects of a smaller sample size etc. PS - I hope that made sense. Quote Link to comment Share on other sites More sharing options...
muppet77 Posted July 30, 2009 Share Posted July 30, 2009 Re: Data Analysis (Chi Square etc.) Muppet, I've calculated your avg yield at 72.10% and your chance of achieving a negative yield at 1.56%. I basically divided your return / stake. calculating the stake in spread betting is the challenge. not sure that my yield is as high as 72.10%. how did you calculate my stake? the profit is easy to work out. Quote Link to comment Share on other sites More sharing options...
KThom Posted July 30, 2009 Share Posted July 30, 2009 Re: Data Analysis (Chi Square etc.) Slapdash - noted, and apologies for interperating the calculation incorrectly Muppet - I used your stake per corner and divided by the profit to get your yield. As mentioned not sure how applicable the calculation is to spreads but it was the most logical way I could think of 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.