|
PureBytes Links
Trading Reference Links
|
Neural nets can be constructed, trained and used in many ways, so it's
important to define your terms.
Neural nets have been proven in a variety of applications to be good at
pattern recognition (they are used to identify submarines based on sound
signatures, for example). So, theoretically one could construct a net to
locate classical chart patterns (triangles, rectangles, etc.) or other
trading 'set ups'.
Using a net for price prediction would be a whole different kettle of fish.
This would need to change along with whatever rules or principles are used
to predict prices.
So, the term 'neural net' by itself is nearly worthless. You'd have to
know a whole pile of information about construction and usage to determine
what is actually going on.
Trading with a tool that is not thoroughly understood is generally a recipe
for disaster, so do your homework before committing those dollars!
- Ed
At 04:55 PM 7/18/97 -0400, EKtwo@xxxxxxx wrote:
>Adrian:
>
>You and I have communicated privately, but I also wanted to add a few points
>about neural nets in general. Being eager to use anything in my trading that
>works, whether it is my company's vehicle or somebody elses, I tried neural
>nets awhile back. I agree that the problem is as you say, i.e. training a
>neural net seems no better than defining parameters in an oscillator based
>system. You're tied to using optimizing to a certain segment of time. If your
>optimized system or neural net is not adapting itself to properties of the
>market are constant, that you're dead meat. The system or neural net will
>always disintigrate.
>
>On the brighter side, though--I believe that there is a way to use neural
>nets that can be a practical adjunct to entry and exit methods. Some money
>managers for example use neural nets and retrain them every week. This is
>required because, as has been discussed on RealTraders before by Mr. Winski,
>myself, and another guy whose name I forget, the market is intelligently
>adaptive. It's like cockroaches which adapt themselves to the lastest
>insecticide formulation.
>
>Eddie Kwong
>
>customer service: RealTraders@xxxxxxxxxxxxxxxx
>****************************************************************
>AT OUR WEBSITE:
>Learn from the experts at RIT
>You can order books through RealTraders at a hefty discount
>Find yourself a Trading Partner
>****************************************************************
>RealTraders is sponsored by Kasanjian Research.
>Contact us at eddiekwong@xxxxxxxxxxxxxxx or 909 337-0816
>****************************************************************
>For digest format send message to:
>listproc@xxxxxxxxxxxxxx
>Turn on digest, type: set realtraders mail digest
>Turn off digest type: set realtraders mail ack
>*****************************************************************
>To get off this list:
>Send an email to:
>listproc@xxxxxxxxxxxxxx
>The first line in the body of the message should say:
>unsubscribe realtraders
>*****************************************************************
>RealTraders customer service: RealTraders@xxxxxxxxxxxxxxxx
>
>In a message dated 97-07-18 02:06:09 EDT, you write:
>
><< I am a novice from New Zealand ( not too far from the south pole )and I
> trade the US market based on end-of-day data.
>
> I recently spent thousands on purchasing neural network & intermarket
> analysis software from a company in Florida but I just can't seem to get
> any joy from the supposedly "predictive capabilities" of these junk
> software. They appear to be no more than glorified lagging indicators.
>
> Are there any out there who have had experience with such products? Any
> advice on how to benefit from them will be very much appreciated.
>
> Adrian
> >>
>
>
>
>
>
|