Hi, probably asked before but can't find it so... is it possible when walking forward a portfolio to use the best optimization values for each symbol separately instead of the best portfolio value ? Thank you.
Guillaume
Your request has logical contradiction inside. If it is best value for portfolio backtest then it is not "separate".
On the other hand if you have separate best values for each symbol they only can come from individual optimization.
You can do walkforward on each symbol separately and then each symbol will have its own "best" values. This is would however result in heavy overfitting, so thing twice.
And how would that be done ?
Lets imagine you would have 10 symbols.
How would you individualy optimize each symbol, using walk forward, and then combine those results into one portfolio result ?
Would you do that using walk forward testing and saving those results for each symbol and then combine the 10 equity curves ? Because that would not be a real portfolio test.... you do not know when money is free and can be used for other symbols.
So is there another way to do this ?
I thought about writing the 10 optimization values to a file and reading those when doing the backtest on the complete portfolio for example....
Re-read my answer. Carefully. Without making your assumptions. Just read precisely what I wrote.
I wrote:
Your request has logical contradiction inside. If it is best value for portfolio backtest then it is not "separate".
On the other hand if you have separate best values for each symbol they only can come from individual optimization.
And that means that what original poster asks is illogical and you can't have a cake and eat it too. Either you are using individual thing or portfolio thing.
And no, I did not say anything about "combining" anything.
You would just have 10 different results / equity curves.
Thanks for the correction Tomasz. So it seems that I'm looking to do batch optimization/walk-forward testing, is there any efficient way to do that ?
Thinking further about this...
Let's say portfolio is symbol 1 & 2 and algo is MA period 10 or 20. We'll optimize on 10 years and walk-forward only 1 year. Let's say symbol 1 is more profitable with period 10 and symbol 2 with 20. You can then walk-forward (OOS) the portfolio with symbol 1 period 10 and symbol 2 period 20 with a rebalancing strategy and all the portfolio rules you like. I don't see why this would not be a portfolio simulation ? Currently it seems that all symbols have to use the same optimization values at every step of the OOS walk which prevents you from getting the "personnality" of each curve. I don't worry about curve fitting because it's the whole point of Walk-Forward to prevent this, you freeze the parameters...
When I read hastoy's question, I read:
is it possible when walking forward a portfolio to use the best optimization values for each symbol separately instead of the best portfolio value ?
So he, and I am also curious if that would work and if it is possible, would like to have
optimization values for each symbol reparately
Instead of
The bestportfolio value.
He does not ask for the best value for portfolio backtest. He asks for a separate value for each symbol.
So unless the OP edited his text, but his latest explanation does not imply so, I do not think that I read the question and answer wrong 
When I read your answer carefully, you clearly explain that you would have to do the walk forward optimization individualy for each symbol separately to get the best separate values.
And what OP then wants.... is not end up with 10 different results / equity curves.
He wants to use the trades from those separate optimizations be used in a portfolio test.
You did not say anything about combining, I did, because I think that is the end results OP is looking for.... Atleast I know I would be...
OP allready mentions that batch optimization / walk forward testing would probably be the way to go.... And also that he would like to combine those results in 1 big portfolio test that does rebalancing etc.
I have therefore the same question.... is there an efficient way to do all this.... Because like OP, I think this is very interesting....
Henri,
I've figured out the WF batch processing (but no portfolio), it's simple. You add a "New Batch" tab then double click in the window to create each new step. 1st add the action "Load Project" and select your file then "Set Current Symbol" and type in your 1st symbol name then add action "Walk-Forward Test". Then repeat step 2 & 3 for as many symbols as you want. Then run the batch (play button) and when finished open the Report Explorer and you get all the CAR values for each symbol and many more metrics. Unfortunately I still don't know how to walk forward a portfolio with separate optimization values... Let me know if you figure this one out.
Cheers
Unfortunatly I am far from even beginning so do not hold your breath 
I just find it a very interesting subject and think it is very nice to take all the value's you have to a portfolio level. I always had the idea that something like this had to be done with a CBT...
Maybe someone will jump in in the future who knows the answer because knowing AB... it should be possible some way....
This topic was automatically closed 100 days after the last reply. New replies are no longer allowed.