BACKTESTING OPTIMIZATION PROBLEM...

New comment
 

Hello

I am trying to find some answers into some issue i am having while doing optimizations on an EA i have.

The main problem is i feel like i am doing optimization of the optimization. Meaning: For example with a simple crossover EA when

i choose the optimization data to test up to 500 pips stop loss it gives me the results and i find something that really suits me well, then

if i do another test with optimization stop loss at 800 pips it of course gives me different results but the good result from the last test where i

inputed 500 as my stop loss dosent appear in the new test where i used 800 pip stop loss as my input, is this clear? so i have to change

the stop loss step by step, 100, 200, 300 in the optimization settings to check each one, i would have thought if i put just 1000 stop loss

it would have given me for every step the best of the best...

Anybody know about this issue? could it be due to my specific EA perhaps?

Thanks 

 
ynachum: Anybody know about this issue? could it be due to my specific EA perhaps? 
Try optimizing a different EA like MacdSample which comes with the Terminal. If you don't experience similar issues then perhaps it-is your EA.
 
Ok I will try it out, thanks :)
 

I found that even the build in EA's create the same issues, it comes down to doing optimization of the optimization, which i can do, but its a long process, 

i have some good results, but i think there could be better and i dont under stand why the stratergy tester dosent automatically find it for you. Its like when you

put in a certain stop loss to check the optimization it finds the best combination with that stop loss, but the best combination with that stop loss dosent appear when you

put in another stop loss to check the optimization... :( 

 
i choose the optimization data to test up to 500 pips stop loss it gives me the results and i find something that really suits me well, then

I choose 100 pips take_profit. Start with 10, increase by 10, stop at 100. So 10,20,30,40 and so on....

 

Results.

 

Obviously take profit of 10 is the best result. The take-profit of 30 did-not show up because it was not good enough.

if i do another test with optimization stop loss at 800 pips it of course gives me different results but the good result from the last test where i

inputed 500 as my stop loss dosent appear

 So I increase my test from 100 to 200. Just like your from 500-800. Buy I still keep the starting point at 10.

 

Yet again the best value of 10 still appears in the list for me.

 

 
Long story short, don't use strategy optimizer to curve-fit because it leads to false-hope.
 

Thanks for the detailed reply it has really helped. I tried what you done and it didnt work exactly.

But in total what i have done is: I downloaded data from Duskcaopy and converted it so i can use it, then

i have been testing an EMA crossover EA starting from Daily time frame and going down, i have been testing with big stop loss and take profit differences

1 to 1000 with 1 steps etc and also from 2007 until now, 2009 until now etc etc... Do you think something could be wrong with the data or the EA or the

brokers platform that could be causing me not to see the previous best result in the new results?

Thank you very much for your help.

 
ynachum: ...  from 2007 until now, 2009 until now etc etc... 
Changing the Length of the test will influence your best result. If the setting made most of its profits between 2007-2008. Then within the next test you exclude this period. It's not gonna have anywhere near the same Luster. Sometimes 1-trade is all it takes for a trend-follower [as example] to make a winning or losing year.
New comment