|
PureBytes Links
Trading Reference Links
|
> -----Message d'origine-----
> De : Terry B. Rhodes [mailto:trhodes3@xxxxxxxxx]
> Envoye : mardi 19 novembre 2002 02:22
> A : omega-list@xxxxxxxxxx
> Objet : Re: Are the Neural Networks and Fuzzy logic of any use ?
>
>
>
> > Has anybody managed to come up with tradable intra-day system using neural
> > networks or fuzzy logic (for E-min S&P maybe).
Check our latest newsletter: It features an Oddball variant using neurofuzzy
logic
>
> I have read about 3 NN projects for index price prediction in the
> literature, 2 of the 3 claim to have achieved profitability.
>
You cannot predict the market IMHO, even with NN or FL.
Just do clever filtering, enough to build winning and stable systems
> There is a person on the realtraders list that is using NNs.
> Seems to me that he is always switching from one net to the other
> based on what works this week. Not very confidence inspiring.
Agreed
>
> Many claim NNs are nothing but overoptimisation engines. IMHO
> there is much truth in this, but it may not be such a bad thing.
> I would be quite comfortable with an NN that required weekly
> retraining (optimization) given sufficiently long term profitable
> backtesting.
It's not necessary to retrain. This is usually displacing the problem ( when to
retrzin and so on)
>
> I can think of at least 10 others that have bought such software,
> none of whom use it today. I know of no one using a successful
> intraday system based on a NN.
Mores than this...
>
> Some time ago I came up with approx 8 data series that I was
> pretty sure held important factors that influence prices
> intraday. It seemed to me that tweezing the significant
> relationships out of this data would take me months and was just
> the kind of job neural nets should be good at. I bought BrainCel,
> spent quite a bit of time reading up on neural nets, got
> everything working, found out that the initial nets gave REALLY
> bad signals and it would take several months of CPU time to
> search for the best net (assuming I got the data representation
> perfect first time, not bloody likely!). Some might claim genetic
> optimisation can avoid this time constraint, but this approach
> does not fully explore the solution space.
>
I also bought alla the available stuff in the 1992-93 years, with no real
success, then we started to develop our own tools
> IMHO...
>
> Garbage in - garbage out is a top level design issue for such
> systems. Randomly applying indicators and expecting the software
> to generate a profitable net is unlikely to work, no matter what
> the vendors say. There is no free lunch in this business.
In my next book, I have done and reported an interesting experience:
We use random data (reallly random), build charts, use indicators applied to it
and train the Safir-X software.
Then the system is applied to the main futures indexes.
Believe me or not , the ROA is more than 100 % yearly.
>
> Data representation is a key issue. Be ready to put significant
> time and effort into learning about this.
A good package does not need learning. It's again displacing the problem into
nifty settings.
>
> NN vendor's claims should be viewed the same as claims made by
> trading system vendors. Make sure you get a money back trial
> period for any software you buy. If you haven't developed NNs
> before a 1 month trial period will not be long enough for you to
> even get your feet on the ground.
This apply to most of the NN package where you cannot expect stable results
before days.
This do not apply to Safir-X that is able to find valid solution within minutes
and no learning curve.
Sincerely,
Pierre Orphelin
www.sirtrade.com
Tradestation 2000i, TradeStation 6 sales and support
Safir-X, neurofuzzy logic trading system builder
To subscribe to our FREE trading system newsletter:
http://www.sirtrade.com/newsletter.htm
Next issue: November 2002
|