|
PureBytes Links
Trading Reference Links
|
Hi Mark,
I think that a good trading model must be simple.
I think the right way is to concentrate all your energy in understinding the
phase (NL) of the market as my friend Riccardo Ronco has done in his work
about the use of Chaos Theory in real market trading.
Yes , markets are non linear dynamic system and NN are a powerful tool to
study financial time series but also a simple trading system with two
simple entries: stochastic 80-20 and a channel breakout, can be also most
powerful than NN if you know the condition you need to validate trendy entry
or countertrend.
And you can program all under MS Excel without the need of K$ software.
More important is that if I look at a chart I can tell you excactly when and
why the model has taken the trade, a thing you can't do with NN. You can use
a decision tree algoritm like C.4.5. but you can't get the same rules, only
an approssimation.
Also Fuzzy logic seems a great tool but if you test hard it you will arrive
at the conclusion that they have the same problem of an overbought/oversold
oscillator during a strong trend.
Probably you can use a bit of NN and Fuzzy to develop only a part of your
strategy but personally I will never put my money on a semi black box.
For me 3 are the points:
- Know the market that has more potential profit.
- know what the market is doing now and apply the right rules.
- Concentrate the resources only on trades with high potential reward.
This is my opinion.
Nicola Prada
-----Messaggio originale-----
Da: Mark Jurik [mailto:mark@xxxxxxxxxxxx]
Inviato: domenica 13 agosto 2000 8.09
A: 'Omega List'
Oggetto: RE: Neural Networks for trading
>>
"Optimization never "learns"....it just follows the instructions provided
and a final combination of input parameters is
determined optimal."
<<
Why do you believe optimization never learns? In my opinion, not only is
learning a form of optimization, but that learning itself can be designed
to become more efficient over time, which is a form of meta-optimization.
>>
Bottom line: for complex data, neural nets are best and definitely more
efficient and elegant.
<<
Some I/O mapping techniques are more efficient than others, depending on
the relationship between the independent and dependent variables. NNs are
not categorically the best, most efficient nor most elegant. However, when
given proper input variables, NNs usually succeed in learning (optimizing)
the I/O map. It's the creation of those "proper" inputs that take the most
effort.
Regards,
Mark Jurik
Jurik Research
http://www.jurikres.com
|