What are your forward results after optimizing on recent data ?

New comment
 

I could figure this out on my own but it would take weeks or months of forward testing, so I'd just like to hear what your experience is with this.

Some indicator driven EAs can work quite well, but its well known that these need frequent optimizing to work.

So lets say I optimize on last 2-4 weeks of history data, and put it on forward test for say 1 week, then re-optimize like before.

Is this a viable approach and will the forward results typically be reasonably good ?

Have tried optimizing further back in history, and then backtested on later history, but the results were not all that good.

 
This approach might be viable if the markets are changing gradually but it will get caught out when there is a sudden change of characteristics. A robust EA should make a profit or not make a loss what ever the type of market and require very little adjustment over time. If an EA requires frequent adjustment is it really automatic?
 

but its well known that these need frequent optimizing to work

That is an assertion and unfortunately not a rule you can depend on

If the system really does sustain across all market patterns, it will need to be smarter, and as Ruptor says, not need frequent adjustment

Which, IMHO, would be no more than frequent curve-fitting to recent data

What you have to ask yourself, is.. is the EA a range trader, breakout or trend-follower

and.. how does it limit itself to its preferred trading situation?

Then focus on self-optimizing in real-time

As another spanner in the works, the more straight-through the data feed, the less meaningful is optimization & back-testing

FWIW

-BB-

 

DayTrader wrote >>

Have tried optimizing further back in history, and then backtested on later history, but the results were not all that good.

This would be as close as you can get without actual forward testing. One thing is for sure -- if you don't get good results doing this, you will not get good results in actual forward testing.

New comment