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

RE: Explorer plug-ins or Neural nets


  • To: "Philippe Poulat" <metastock@xxxxxxxxxxxxx>
  • Subject: RE: Explorer plug-ins or Neural nets
  • From: "Laurent GITTLER" <lgittler@xxxxxxxxxxx>
  • Date: Tue, 13 Apr 1999 09:20:28 -0400 (EDT)
  • In-reply-to: <3.0.32.19990409170010.006c2a58@xxxxxxxxxx-montp2.fr>

PureBytes Links

Trading Reference Links

Thank you Philippe for this valuable answer.

Laurent


-----Original Message-----
From: Philippe Poulat [mailto:poulat@xxxxxxxxxxxxxxxxxxxx]
Sent: Friday, April 09, 1999 8:08 PM
To: Laurent GITTLER
Subject: Re: Explorer plug-ins or Neural nets


Laurent,

At 11:41 08/04/99 +0100, you wrote:
>List,
>
>I am making customs indicators that work all together to detect buy and
>sells. Basically, I am using bollinger bands on indicators themselves. I am
>now using MACD, SAR, STOCHastics, RSI, StochRSI, DMI and ADX, CMO, Money
>Flow as defined by Bollinger.
>
>I have made binary waves indicators which give results but not exactly as I
>wish, because I would like to make more fuzzy logic relative to the
>indicator value. For example saying it is good to buy when RSI is crossing
>30 ascending is not really accurate. This is found when optimizing a
>Metastock system test using any conventional indicator. Sometimes the most
>accurate value is 29 or 27 or 33 and so on.
>
>I made signifiant progress in chart reading when applying Bollinger Bands
on
>the indicator itself.
>
>I am trying to make fuzzy logic approach, defining success factor ranges on
>each indicator I use. Saying for example when the RSI is rising and in the
>25-35 range, the stock will more probably rise then when RSI is in neutral
>zone (for example 35-65). This gives a fuzzy logic approch when applied
>globally on a lot of indicators.
>
>In Excel, it is far more easy to do rather then in Metastock which do not
>handle really well vectors (to compute the statistics with linear or non
>linea functions) and explorer and indicators formulas are limited in
formula
>text size.
>
>It is tedious and I find it very complicated to optimize.
>
>Possibility 1) I am considering either to plug-in an external function
>written in C or any programming language which can generate DLLs to the
>explorer and I was wondering if this is possible in Metastock, and how
>difficult it is to do this.
>
>Possibility 2) I am very doubtful about Neural nets because I think it is a
>great tool to recognize patterns (like Optical character Recognition or
>voice recognition), but I am not so sure it is a great tool to identify
good
>technical probabilities based on indicators. But it seems one quite popular
>branch in trading systems, so I was wondering how successfull is this
>approach. Has anybody tried Neural nets products like NeuroShell trader,
>Biocomp or NeuroForecaster+Genetica which seems to handle Fuzzy logic rules
>?

4 monthes ago, I bought SAFIR-X (http://www.sirtrade.com), a software using
neural nets and fuzzy-logic techniques applied to trading. Although, it is
more expensive than others software you cited it is very powerful and
friendly, and allowed real time retraining. However, strategies issued from
SAFIR can be run only in tradestation and can thus benefit of powerful of
easy-langage of TS.

>If you have tried using neural nets systems, are you still using them,

Yes, I recently find a valuable strategy (trading system) and I will soonly
enter the market with it.

> and how difficult it was to define new rules, was it a matter a days,
weeks or
>months ?

With this software, you have only to enter PERTINENT indicators (4 to 7-8)
on a representative period (including uptrend, downtrend, and trading
range) and if there is some valuable strategy to find, it will find it.
The software does all for you :
you defines only some parameters as default number of fuzzy sets, number of
iteration during  training, number of subfuzzyset, training / test period
and software defines the rules, the place of new subsets on which indicator
etc.... Each time a new strategy is better than a "parent one", it will be
kept on history list. At the end of this exploring phase you can check all
your strategies in the list on new historical data and you kept those which
are the best on all periods (or support ) tested (if there is any).
I invite you to go to the web site I mentionned in the beginning where you
can find further details and a full operational demo (limited only in
saving strategies obviously!). You can also see at this
site(http://www.sirtrade.com/fuzproof.htm) as example of what you can do
with this software. Technical support is no available for demo, obviously,
but I can help you if encountered some problems.
I spent three monthes full (very full) time to obtain a good trading
neuro-fuzzysystem. It depends many of your knowledge, skill, feeling with
indicators of technical analysis.

I send this message to you but feel free to put it on the list if you want,

Best regards,

Philippe Poulat