Tips? How to find the best parameters for an ea with strategy tester? How to use optimizer?

To add comments, please log in or register
Reilly Gailey
363
Reilly Gailey  

About the optimization option.

Example;

I run optimizer on 

01 january to 01 april (3 month test)

I use the best profit parameters of this 3 month test

and run it on 01 april to 01 may (1 month test)(the month directly after the 3 month test)

On the 1 month test I frequently lose money.

 

My question is how do I use the optimizer in a way that I can find parameters that will consistently work (for atleast a month after the months I ran the optimizer on), is there a trick? 

 

Any other tips and tricks with the strategy tester would be greatly appreciated.

 

Thanks!~ 

Stanislav Korotky
27253
Stanislav Korotky  
Try walk-forward. It's not supported by the tester directly. You need to do some work manually or use 3-d party products/scripts.
Fernando Carreiro
5048
Fernando Carreiro  
ReillyFox: About the optimization option.


Example; I run optimizer on 01 january to 01 april (3 month test) - I use the best profit parameters of this 3 month test and run it on 01 april to 01 may (1 month test)(the month directly after the 3 month test)

On the 1 month test I frequently lose money.

 My question is how do I use the optimizer in a way that I can find parameters that will consistently work (for atleast a month after the months I ran the optimizer on), is there a trick? 

 Any other tips and tricks with the strategy tester would be greatly appreciated.

Don't just look at profit. You have to analyse several metrics - some more important than others, like: Profit Factor, Drawdown, Recovery, etc.

Also actually "looking" at the equity curve to see that it is in fact "good" and especially not leveling off or curving down towards the end.

Reilly Gailey
363
Reilly Gailey  
Fernando Carreiro:

Don't just look at profit. You have to analyse several metrics - some more important than others, like: Profit Factor, Drawdown, Recovery, etc.

Also actually "looking" at the equity curve to see that it is in fact "good" and especially not leveling off or curving down towards the end.

yeah your right, Usually the ones with the highest profit are not the most consistently profitable parameter settings from my experience.
Reilly Gailey
363
Reilly Gailey  
Stanislav Korotky:
Try walk-forward. It's not supported by the tester directly. You need to do some work manually or use 3-d party products/scripts.
I read the wiki definition of walk forward and I think it gave me dyslexia.

Could you please explain what walk foward is?

Sounds like you optimize on a certain time window(eg 01/03/16 to 01/06/16), and use those optimization results on the next time window (01/06/16 to 01/09/16).

Which is what im already doing
Stanislav Korotky
27253
Stanislav Korotky  
ReillyFox:
I read the wiki definition of walk forward and I think it gave me dyslexia.

Could you please explain what walk foward is?

Sounds like you optimize on a certain time window(eg 01/03/16 to 01/06/16), and use those optimization results on the next time window (01/06/16 to 01/09/16).

Which is what im already doing

Yes you're right. Though there are nuances you've skipped. Optimization window is larger than the "next time" future step (you mentioned 3 months window and 1 month forward step in your original post, but the last message quoted above gives equal durations, which can be a problem). Also WFO implies that you check different window and step size combinations, and find the best (so called cluster optimization). Also you should do this on a long period (a single forward step is not enough). For example, if your forward step is 1 month, try to do the tests on a couple of years and build cumulative peformance on all the monthly steps (24). It's very hard to run such tests manually. Some automation is required (but not available out of the box in pure MT). You may perform a search through the forums, the market, or Internet to find existing options.

Anyway, WFO is not a magic. It only shows how your system would operate in a near to real life (market) ever changing situation. It's perfectly can be that the system is not robust and should be altered or dismissed completely.

To add comments, please log in or register