|
PureBytes Links
Trading Reference Links
|
I would have to agree. Neural nets have been next to useless for predicting
market direction in my experience. The problem is that there is simply too
much random noise which buries any consistent patterns that would be there.
In my experience they all fall apart on unseen data.
It's possible if you knew what you were looking for that a net might be able
to exploit a small niche inefficiency, but that would presume that you knew
where to look for it. Nets are useless for identifying such inefficiencies.
If anything they prove the amount of random noise that any successful system
must be overcome.
----- Original Message -----
From: "Sven Napolean Montessori" <snm@xxxxxxxxxxxxxxx>
To: <pierre.orphelin@xxxxxxx>
Cc: <omega-list@xxxxxxxxxx>
Sent: Saturday, June 26, 2004 8:35 PM
Subject: Re: neural net expert
> From: "Pierre Orphelin" <pierre.orphelin@xxxxxxx>
> Date: Sat, 26 Jun 2004 21:40:51 +0200
>
> I still have to see a NN trained by GA that holds its promises over
huge unseen
> dataserie.
>
> Me too.
>
> GA is curve fitting at its best, but brings no intelligence at all to
the final
> system.
>
> You get whatever you put into the Fitness Function. Making a careful
> choice is a good idea.
>
> GA is also the easy path to real world like illusion.
>
> Hmm, you might want to check with the drug companies searching for the
> new billion dollar drug...
>
> Plug a GA on a NN and the NN will become intelligent.
> The key is in the belief...
>
> A GA and a NN are apples and oranges. One is a pattern classifier
> (without a null output) and the other is an optimization method for
> complex problems.
>
> I think you misunderstood my original comment and opinion of NN
> systems.
>
|