Hi all,
Here I am, only had AB for 24 hours, and already
offering my 2c worth!
Thank you for making that file available Angelo.
I'm not sure if I am understanding this thread correctly.
If it's just a matter or running a list of tests against
a list of instruments - would that not be a reasonably
simple AFL/Batch run of some sort? ( Remember
I am a newbie! )
I think the beauty and power of running a
mulit system/multi instrument BT would be to optimise
the use of cash (and leverage ) in your account. If this
is the case, then I don't think that the pseudo code
you provided would do the job. I do appreciate that
it is a simplified version of the final product, but
looking at it I ask myself the following questions:
- Does the Unit Size script "know" how much each
other system/instrument strategy has allocated?
- Is the account "infinite" ? If it isn't an
infinite account, then the first systems in the loop
get the most of the allocation because all the fills
are checked before the next system is tested.
If the allocations are done via optimisation ie
running the above psuedo code for all possible
values returned from UnitSize, then
that becomes a huge task, but not one out of
the scope of the evolutionary optimisations
built into AB. This would give you a solution,
but I get the feel it would be a brute force
empirical one, and I think ( while it's all still
it the talking stages ), that we should perhaps
be discussing a more elegant parametrically
based solution. Go For Gold!
If you are trying to optimise account allocations
on a finite account then you would have to
prevent some good systems from trading simply
because they are either not as good or too
correlated to other systems. In some optimisations
you would have to include sub-optimal
systems because they are capable of smoothing
out the equity curve - like buying OTM puts. They
would lose most of the time, but when they won
they would do the equity curve a world of good.
This is definitely an area worth discussing at
length.
Regards
RZ
________________________________
From: ang_60 <[email protected]>
To: [email protected]
Sent: Friday, 8 May, 2009 11:44:17 PM
Subject: [amibroker] Re: testing multiple systems simultaneously
--- In amibro...@yahoogrou ps.com, "Paolo Cavatore" <pcavat...@. ..> wrote:
>
> the same logic used in portfolio backtesting should be used in >multi-systems
> portfolio backtesting.
>
>
>I should assign a proper positionsize of the equity line every time >I get a
>signal whatever system it comes from - for instance I can >always take a 2%
>position on every signal no matter if it comes from >system A or B if I'm not
>supposed to potentially get more than 50 >signals at the
>
>
>
Hi all,
here
http://www.filedrop per.com/loopform ultisystemmultim arket
you will find an almost- too-much simplified loop showing (*) how another
software has been programmed in order to get to the target (well…. not at 100%
but reasonably near…. I leave aside the details).
Just translate "instrument" with tickers of a portfolio, and "unit size" with
position sizing rules.
For the rest, I concur with most of what Paolo is saying (especially important
to me are his two sentences above).... I'm sure there's some misunderstanding
among the people in this thread, created because Internet is a wonderful mean
to talk to people everywhere in the world but sometimes it's really impossible
to reproduce the same efficacy on an eye-on-eye discussion.
PS Thanks to Paul and Benoitek for your precious inputs. They have not been
lost.
I've saved them on my PC and will be looking at their "ways to do" with great
attention.
Only…. this discussion has evolved on the possibility to get Tomasz think if
it's possible/economical ly efficient for him to get a "built in solution" …..
also because I know "low level CBT solutions" are not accessible to my present
knowledge …. and will not be in a not too distant future neither.
If it's not possible for now…. never mind ….. at the very least we have had a
really interesting discussion and some worthwhile suggestions.
Greetings,
Angelo.
(*) This flow chart is publicly available on the web, even by non-purchasers of
that product, so I'm confident I'm not violating any copyrights.
