[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: More on optimization


  • To: metastock@xxxxxxxxxxxxx
  • Subject: Re: More on optimization
  • From: lists@xxxxxxxxxxx (rudolf stricker)
  • Date: Tue, 18 Jan 2000 15:27:36 -0800
  • In-reply-to: <LOBBLLPNIJFDCJKGEONHEEEDCCAA.ibarra10@xxxxxxxxxxx>

PureBytes Links

Trading Reference Links


Dear Angel,

On Mon, 17 Jan 2000 19:50:48 GMT, you wrote:

>you mentioned that you make some kind of genetic optimization with
>your sistems. This is one of the things I would like to explore now.
>For the moment I make the optimizacion scaning all the possible cases,
>but this ie too much and I would like to use other kind of
>optimization processes. I would appreciate very much if you can inform
>me where I can find any subrutine to meka "genetic" optimizacion using
>VB, or, at least where I can find the algorith to make the program.

Imo, a powerful optimization procedure is an important feature that
MetaStock should have, especially if it comes to highly nonlinear
system models with more than e.g. 5 parameters. In this context,
genetic algorithms should be the right choice, even if they may take a
lot of computing power.

Unfortunately, I did not find a way to make a genetic algorithm
working with MetaStock because of its "closed shop" behavior. So I
came back to Excel, but even there I did not find an appropriate GA
program to combine it with Excel.

Therefore, I have set up a home-made GA, consisting of some Excel
sheets driven by some VBA code. It surely is not the most effective
way to do GA-driven optimization, but it is simple (in terms of
programming effort) and it works:

My trading systems basically consist of some Excel sheets:
=> si.xls calculates a sequence of IN/OUT signals for a given time
frame, based on 5 to 7 "system parameters" dealing with "continuous"
ROCs.
=> tr.xls calculates from the IN/OUT signals a sequence of trades,
including some ratings, based on profit, number of wins/losses, shape
of loss distribution, etc.
=> op.xls provides sets of "system parameters" spread over an "area of
interest" and collects the best parameter sets.

GA-based optimization goes like this:

1. A test probe ("generation") of 100 parameter sets is prepared in
op.xls, based on given parameter boundaries and the best-rated
parameter sets found up to now.
2. For these 100 sets the IN/OUT signals are calculated by si.xls and
the rating is calculated by tr.xls.
3. The test probe of 100 parameter is sorted (together with a score
list of the 25 best-rated parameter sets from previous "generations").
4. The 25 best-rated parameter sets are transferred into the score
list.
5. Calculation stops, if there is no significant change (in terms of
rating) at the top of the score list anymore, otherwise the procedure
continues at 1)

For sure, there are several details and refinements in the procedure,
but the rough framework described above will give you an idea about my
GA-driven optimization routine.

Please, let me know if you need any further information.

mfg rudolf stricker
| Disclaimer: The views of this user are strictly his own.