For a sustainable EA, how long period back test should be needed at least? (with 99% high quality back test data)

 
  • 18% (22)
  • 10% (12)
  • 12% (14)
  • 11% (13)
  • 12% (15)
  • 10% (12)
  • 2% (2)
  • 25% (30)
Total voters: 106
 

Dear experienced traders, 


How do you think for a sustainable EA, how long period back test will be needed at least?

I know the important thing might not the back test period, but it is still an important question to design strategies. 


If you are using many kinds of strategies, you can choose more than 2 choice. 


Thank you in advance and best regards


 
The high quality data seems only from 2011, so the longest choice is "Around 10 years ". 
 
What if I don't believe in back-testing?
 
I back test every year with the same starting capital every year, i see a lot of EA successful in recent year like 2020, 2019 but blow account in past year long time ago like 2013, 2012. A great EA should be successful (at least not blowing account) not only in recent year but in the past too at least 5 years from today.
 
Koh Kok Yeow:
What if I don't believe in back-testing?

You are right. 

It is a very necessary POV. 

Could you tell us why do you do not use back-test? 

Will you use an EA never passed any back-test? It looks not safe too...

 
Anak Agung:
I back test every year with the same starting capital every year, i see a lot of EA successful in recent year like 2020, 2019 but blow account in past year long time ago like 2013, 2012. A great EA should be successful (at least not blowing account) not only in recent year but in the past too at least 5 years from today.

Thank you so much, Anak. 

Why not more than 10 years but more than 5 years?

I mean longer might make more change be tested, does more years have any special meaning? 
 
Sky L:

Thank you so much, Anak. 

Why not more than 10 years but more than 5 years?

I mean longer might make more change be tested, does more years have any special meaning? 
Because there is not much data, in fact meta trader 4 only released in 2004 which is only 16 years ago. Forex market have changed compared to the past too, majority of forex trading now using algorithm which is bigger portion than the past so too much longer in the past would be irrelevant to current condition of the market. And if you see a failure in 10 years ago you wouldn't bought any EA at all because at some point an EA can become fail, its a matter of returning the investment before blowing up, if an EA can, it would still be considered good
 
Anak Agung:
Because there is not much data, in fact meta trader 4 only released in 2004 which is only 16 years ago. Forex market have changed compared to the past too, majority of forex trading now using algorithm which is bigger portion than the past so too much longer in the past would be irrelevant to current condition of the market. And if you see a failure in 10 years ago you wouldn't bought any EA at all because at some point an EA can become fail, its a matter of returning the investment before blowing up, if an EA can, it would still be considered good

Thank you, Anak.


I understand, but I still think around 10 years back-test is needed. 

MT5 start from 2011, almost ran 10 years. 

 
Sky L:
  • Less than 6 months will be fine
  • More than 6 months
  • More than 1 year
  • More than 2 years
  • More than 3 years
  • More than 5 years
  • More than 7 years
  • Around 10 years


Wrong approach in my opinion


I think the following things:

It depends also of the number of trades

to be succesful it should have always an earning in the time.

it shouldnt care of the pair

should have no input parameters

 
Sky L:

You are right. 

It is a very necessary POV. 

Could you tell us why do you do not use back-test? 

Will you use an EA never passed any back-test? It looks not safe too...

Market conditions have changed.  I want someone who is forward-looking, not backwards-looking.

Mutual funds always have this statement-past performance is not indicative of future performance.  What more can we derive by extrapolating current performance onto past history?