|
PureBytes Links
Trading Reference Links
|
-----Message d'origine-----
De : Ron Hudson MCSD, MCDBA [mailto:ron@xxxxxxxxxxxxxxxxxx]
Envoye : jeudi 26 juin 2003 07:38
A : omega-list@xxxxxxxxxx
Objet : FW: Reoptimization and RevMark System
That's funny. Interestingly, you're the first one who's come out against the
reoptimization approach, so far. I've received 5 other responses from
people, some very experienced, who feel that it's a valid approach under
certain circumstances.
===
I must say that I'm not very much impressed by the equity curves publied on your
web site, unlesss I do not understand something, but to me a straight equity
curve for all the database tested is the proof of the validity, period.
Reoptimising is a valid answer if the reoptimization is included in the system
logic itself and if you can run it at any time, without having to choose the
optimization period. Any setting that you may need to turn this on /off on a
discretionary mode invalidates the method.
Like you, I am skeptical. I'm not sure what I'll decide, but I really
appreciate hearing all opinions on this.
One of the most interesting responses I've received is from someone who
feels that this kind of thing is the basis of Neural Nets, and that Neural
Nets have value. I would imagine that Pierre Orphelin has an interesting
opinion on this. I hope he sees this post, but I'll email him if he doesn't.
===
This has nothing to do with NNet, and probaly there is a confusion between
backpropagation training and optimization by classical curve fitting. All of
these process are trial and error during the development stage, but once done,
the system has to stay unmodified ( the sad thing is that you cannot optimize on
next week data that are unfortunately unavailable).
Reoptimization may work if done on a fully automated basis, where the
optimization routines are included in the system code itself ( a kind of
permanent walk formard testing). It's no mre than having an adaptive system
based on recent curve fiting.
There are other adaptive approach that are more robust and smarter . Fuzzy logic
is one of them. We have also provided a retraining option using fuzzy logic,
that improves a lot, but may also degrade under some cicumstances, so it's still
a research tool that we havbe lef to the user ( if he is able to find wen to
repotimize or not, this is the key to success).
Sincerely,
Pierre Orphelin
www.sirtrade.com
Tradestation 2000i, TradeStation 7 sales and support
Safir-X, neurofuzzy logic trading system builder
To subscribe to our FREE trading system newsletter:
http://www.sirtrade.com/newsletter.htm
|