Back tested system US Open 12 % yield, 12 years

[gonza3591]

So i found this site tennis data.co.uk download the result and odds for the last 12 U.S. Open, and tried to make a system based on bet in every match, chosing a minimum and maximum range of odds for favorite and underdog, and finally i think i got it. The thing is bet on underdog in certain range of odds depending if its the 1st round, 2nd, 3rd, semis...You can see the range of odds in the excel file. For chose the range of odds, write a macro who do it automatically, searching for the maximum profit, so its very easy. Then i back tested the system, copying the main file to another 10 and moving in every book, one year out of the range taked for the macro, so the good or bad profit in that year doesnt change the range of odds. And finally it goes from 169 bets and 25 % yield in the main file, to 133 bets and 12 % yield after back test the system Anyway i also tested the system in wimbledon, french and australian open and they were big losses, so i just let this here and you can give it a try, give your opinion..This was just made in 1 hour, while i take a rest from other project so im not 100 % sure the system works fine http://www.mediafire.com/download/yl1em9q1gdldnjn/Tenis(2).rar Sorry for bad english

[pinkisntwell]

Re: Back tested system US Open 12 % yield, 12 years Back testing is absolutely worthless, since you already know the results of the matches. Do yourself a favour and don't bet this system.

[methuselah]

Re: Back tested system US Open 12 % yield, 12 years IT is frequently said that back-testing is worthless, but I must disagree strongly. Isn,t that what we do when we create our systems from past results. Look at JTW,s successful H2H systems which are gleaned from past results, i.e. they are in essence back-tested.

[Grex]

Re: Back tested system US Open 12 % yield, 12 years

Back testing is absolutely worthless' date=' since you already know the results of the matches. Do yourself a favour and don't bet this system.[/quote'] Totally agree with Methuselah. I would imagine that the majority of systems posted on this forum were developed using back-testing (including my "Homes" and "Overs" systems). Only by back-testing can you establish "value" bets..... and if you can't establish theoretical "value", you will almost certainly lose in the long-term. Regarding your system, Gonza3591, you say you developed your cut-off criteria based on the last 12 years results (2000-2012??). I suggest you do your analysis on the results from 2000-2010, and then test it "live" on the results for 2011-2012. If the yield for the "live" testing is consistent with the previous theoretical results, then you may be onto a winner. If there is a sudden drop in performance for the "live" testing, then you have a problem.

[pinkisntwell]

Re: Back tested system US Open 12 % yield, 12 years

Totally agree with Methuselah. I would imagine that the majority of systems posted on this forum were developed using back-testing (including my "Homes" and "Overs" systems). Only by back-testing can you establish "value" bets..... and if you can't establish theoretical "value", you will almost certainly lose in the long-term. Regarding your system, Gonza3591, you say you developed your cut-off criteria based on the last 12 years results (2000-2012??). I suggest you do your analysis on the results from 2000-2010, and then test it "live" on the results for 2011-2012. If the yield for the "live" testing is consistent with the previous theoretical results, then you may be onto a winner. If there is a sudden drop in performance for the "live" testing, then you have a problem.

but that's no longer back testing.

[Grex]

Re: Back tested system US Open 12 % yield, 12 years

but that's no longer back testing.

I beg to differ, Pinkisntwell. All systems are developed based on historical data. Let´s assume you have a complete data set from 2000 to 2013. You can either: (a) develop the system on data for the period (say) 2004 to 2013, and then "back-test" it on data from 2000 to 2003....... OR (b) develop the system on data for the period (say) 2000 to 2009, and then "back-test" it (or, if you prefer "forward-test") on data from 2010 to 2013. If the positive yield remains constant throughout the period 2000 to 2013, then you have a good chance of success. I think option (b) is more appropriate as any change in trends (for example, if you are predicting O/U, changes in overall goals/game) evidenced in recent historical data are more likely to be replicated in the forthcoming season.

[plinge]

Re: Back tested system US Open 12 % yield, 12 years

Back testing is absolutely worthless

Done the right way it is anything but worthless.

since you already know the results of the matches. Do yourself a favour and don't bet this system.

I'd have to agree with this. Good luck, gonza, but bear in mind that if you scan enough data you are bound to find statistical anomalies. This is what I would term BackFITTING as opposed to BackTESTING/Historical Analysis. Unless you have a theory as to why certain odds brackets, in certain rounds, in a certain tournament should be successful (and it would want to be one remarkable theory) I would treat this very cautiously.

[Marazan]

Re: Back tested system US Open 12 % yield, 12 years

but that's no longer back testing.

You have a different concept of what back-testing means than the community at large then. Back testing is when you test your system on historical results, no more, no less. If your system is based on a model fitted to historical results then the traditional thing to do is to split the historical data into a training set and test set. This avoids the problem of "overfitting" which is what I think you are trying to get at.